Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * xlog.c
4 : : * PostgreSQL write-ahead log manager
5 : : *
6 : : * The Write-Ahead Log (WAL) functionality is split into several source
7 : : * files, in addition to this one:
8 : : *
9 : : * xloginsert.c - Functions for constructing WAL records
10 : : * xlogrecovery.c - WAL recovery and standby code
11 : : * xlogreader.c - Facility for reading WAL files and parsing WAL records
12 : : * xlogutils.c - Helper functions for WAL redo routines
13 : : *
14 : : * This file contains functions for coordinating database startup and
15 : : * checkpointing, and managing the write-ahead log buffers when the
16 : : * system is running.
17 : : *
18 : : * StartupXLOG() is the main entry point of the startup process. It
19 : : * coordinates database startup, performing WAL recovery, and the
20 : : * transition from WAL recovery into normal operations.
21 : : *
22 : : * XLogInsertRecord() inserts a WAL record into the WAL buffers. Most
23 : : * callers should not call this directly, but use the functions in
24 : : * xloginsert.c to construct the WAL record. XLogFlush() can be used
25 : : * to force the WAL to disk.
26 : : *
27 : : * In addition to those, there are many other functions for interrogating
28 : : * the current system state, and for starting/stopping backups.
29 : : *
30 : : *
31 : : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
32 : : * Portions Copyright (c) 1994, Regents of the University of California
33 : : *
34 : : * src/backend/access/transam/xlog.c
35 : : *
36 : : *-------------------------------------------------------------------------
37 : : */
38 : :
39 : : #include "postgres.h"
40 : :
41 : : #include <ctype.h>
42 : : #include <math.h>
43 : : #include <time.h>
44 : : #include <fcntl.h>
45 : : #include <sys/stat.h>
46 : : #include <sys/time.h>
47 : : #include <unistd.h>
48 : :
49 : : #include "access/clog.h"
50 : : #include "access/commit_ts.h"
51 : : #include "access/heaptoast.h"
52 : : #include "access/multixact.h"
53 : : #include "access/rewriteheap.h"
54 : : #include "access/subtrans.h"
55 : : #include "access/timeline.h"
56 : : #include "access/transam.h"
57 : : #include "access/twophase.h"
58 : : #include "access/xact.h"
59 : : #include "access/xlog_internal.h"
60 : : #include "access/xlogarchive.h"
61 : : #include "access/xloginsert.h"
62 : : #include "access/xlogreader.h"
63 : : #include "access/xlogrecovery.h"
64 : : #include "access/xlogutils.h"
65 : : #include "access/xlogwait.h"
66 : : #include "backup/basebackup.h"
67 : : #include "catalog/catversion.h"
68 : : #include "catalog/pg_control.h"
69 : : #include "catalog/pg_database.h"
70 : : #include "common/controldata_utils.h"
71 : : #include "common/file_utils.h"
72 : : #include "executor/instrument.h"
73 : : #include "miscadmin.h"
74 : : #include "pg_trace.h"
75 : : #include "pgstat.h"
76 : : #include "port/atomics.h"
77 : : #include "postmaster/bgwriter.h"
78 : : #include "postmaster/startup.h"
79 : : #include "postmaster/walsummarizer.h"
80 : : #include "postmaster/walwriter.h"
81 : : #include "replication/origin.h"
82 : : #include "replication/slot.h"
83 : : #include "replication/snapbuild.h"
84 : : #include "replication/walreceiver.h"
85 : : #include "replication/walsender.h"
86 : : #include "storage/bufmgr.h"
87 : : #include "storage/fd.h"
88 : : #include "storage/ipc.h"
89 : : #include "storage/large_object.h"
90 : : #include "storage/latch.h"
91 : : #include "storage/predicate.h"
92 : : #include "storage/proc.h"
93 : : #include "storage/procarray.h"
94 : : #include "storage/reinit.h"
95 : : #include "storage/spin.h"
96 : : #include "storage/sync.h"
97 : : #include "utils/guc_hooks.h"
98 : : #include "utils/guc_tables.h"
99 : : #include "utils/injection_point.h"
100 : : #include "utils/pgstat_internal.h"
101 : : #include "utils/ps_status.h"
102 : : #include "utils/relmapper.h"
103 : : #include "utils/snapmgr.h"
104 : : #include "utils/timeout.h"
105 : : #include "utils/timestamp.h"
106 : : #include "utils/varlena.h"
107 : :
108 : : #ifdef WAL_DEBUG
109 : : #include "utils/memutils.h"
110 : : #endif
111 : :
112 : : /* timeline ID to be used when bootstrapping */
113 : : #define BootstrapTimeLineID 1
114 : :
115 : : /* User-settable parameters */
116 : : int max_wal_size_mb = 1024; /* 1 GB */
117 : : int min_wal_size_mb = 80; /* 80 MB */
118 : : int wal_keep_size_mb = 0;
119 : : int XLOGbuffers = -1;
120 : : int XLogArchiveTimeout = 0;
121 : : int XLogArchiveMode = ARCHIVE_MODE_OFF;
122 : : char *XLogArchiveCommand = NULL;
123 : : bool EnableHotStandby = false;
124 : : bool fullPageWrites = true;
125 : : bool wal_log_hints = false;
126 : : int wal_compression = WAL_COMPRESSION_NONE;
127 : : char *wal_consistency_checking_string = NULL;
128 : : bool *wal_consistency_checking = NULL;
129 : : bool wal_init_zero = true;
130 : : bool wal_recycle = true;
131 : : bool log_checkpoints = true;
132 : : int wal_sync_method = DEFAULT_WAL_SYNC_METHOD;
133 : : int wal_level = WAL_LEVEL_REPLICA;
134 : : int CommitDelay = 0; /* precommit delay in microseconds */
135 : : int CommitSiblings = 5; /* # concurrent xacts needed to sleep */
136 : : int wal_retrieve_retry_interval = 5000;
137 : : int max_slot_wal_keep_size_mb = -1;
138 : : int wal_decode_buffer_size = 512 * 1024;
139 : : bool track_wal_io_timing = false;
140 : :
141 : : #ifdef WAL_DEBUG
142 : : bool XLOG_DEBUG = false;
143 : : #endif
144 : :
145 : : int wal_segment_size = DEFAULT_XLOG_SEG_SIZE;
146 : :
147 : : /*
148 : : * Number of WAL insertion locks to use. A higher value allows more insertions
149 : : * to happen concurrently, but adds some CPU overhead to flushing the WAL,
150 : : * which needs to iterate all the locks.
151 : : */
152 : : #define NUM_XLOGINSERT_LOCKS 8
153 : :
154 : : /*
155 : : * Max distance from last checkpoint, before triggering a new xlog-based
156 : : * checkpoint.
157 : : */
158 : : int CheckPointSegments;
159 : :
160 : : /* Estimated distance between checkpoints, in bytes */
161 : : static double CheckPointDistanceEstimate = 0;
162 : : static double PrevCheckPointDistance = 0;
163 : :
164 : : /*
165 : : * Track whether there were any deferred checks for custom resource managers
166 : : * specified in wal_consistency_checking.
167 : : */
168 : : static bool check_wal_consistency_checking_deferred = false;
169 : :
170 : : /*
171 : : * GUC support
172 : : */
173 : : const struct config_enum_entry wal_sync_method_options[] = {
174 : : {"fsync", WAL_SYNC_METHOD_FSYNC, false},
175 : : #ifdef HAVE_FSYNC_WRITETHROUGH
176 : : {"fsync_writethrough", WAL_SYNC_METHOD_FSYNC_WRITETHROUGH, false},
177 : : #endif
178 : : {"fdatasync", WAL_SYNC_METHOD_FDATASYNC, false},
179 : : #ifdef O_SYNC
180 : : {"open_sync", WAL_SYNC_METHOD_OPEN, false},
181 : : #endif
182 : : #ifdef O_DSYNC
183 : : {"open_datasync", WAL_SYNC_METHOD_OPEN_DSYNC, false},
184 : : #endif
185 : : {NULL, 0, false}
186 : : };
187 : :
188 : :
189 : : /*
190 : : * Although only "on", "off", and "always" are documented,
191 : : * we accept all the likely variants of "on" and "off".
192 : : */
193 : : const struct config_enum_entry archive_mode_options[] = {
194 : : {"always", ARCHIVE_MODE_ALWAYS, false},
195 : : {"on", ARCHIVE_MODE_ON, false},
196 : : {"off", ARCHIVE_MODE_OFF, false},
197 : : {"true", ARCHIVE_MODE_ON, true},
198 : : {"false", ARCHIVE_MODE_OFF, true},
199 : : {"yes", ARCHIVE_MODE_ON, true},
200 : : {"no", ARCHIVE_MODE_OFF, true},
201 : : {"1", ARCHIVE_MODE_ON, true},
202 : : {"0", ARCHIVE_MODE_OFF, true},
203 : : {NULL, 0, false}
204 : : };
205 : :
206 : : /*
207 : : * Statistics for current checkpoint are collected in this global struct.
208 : : * Because only the checkpointer or a stand-alone backend can perform
209 : : * checkpoints, this will be unused in normal backends.
210 : : */
211 : : CheckpointStatsData CheckpointStats;
212 : :
213 : : /*
214 : : * During recovery, lastFullPageWrites keeps track of full_page_writes that
215 : : * the replayed WAL records indicate. It's initialized with full_page_writes
216 : : * that the recovery starting checkpoint record indicates, and then updated
217 : : * each time XLOG_FPW_CHANGE record is replayed.
218 : : */
219 : : static bool lastFullPageWrites;
220 : :
221 : : /*
222 : : * Local copy of the state tracked by SharedRecoveryState in shared memory,
223 : : * It is false if SharedRecoveryState is RECOVERY_STATE_DONE. True actually
224 : : * means "not known, need to check the shared state".
225 : : */
226 : : static bool LocalRecoveryInProgress = true;
227 : :
228 : : /*
229 : : * Local state for XLogInsertAllowed():
230 : : * 1: unconditionally allowed to insert XLOG
231 : : * 0: unconditionally not allowed to insert XLOG
232 : : * -1: must check RecoveryInProgress(); disallow until it is false
233 : : * Most processes start with -1 and transition to 1 after seeing that recovery
234 : : * is not in progress. But we can also force the value for special cases.
235 : : * The coding in XLogInsertAllowed() depends on the first two of these states
236 : : * being numerically the same as bool true and false.
237 : : */
238 : : static int LocalXLogInsertAllowed = -1;
239 : :
240 : : /*
241 : : * ProcLastRecPtr points to the start of the last XLOG record inserted by the
242 : : * current backend. It is updated for all inserts. XactLastRecEnd points to
243 : : * end+1 of the last record, and is reset when we end a top-level transaction,
244 : : * or start a new one; so it can be used to tell if the current transaction has
245 : : * created any XLOG records.
246 : : *
247 : : * While in parallel mode, this may not be fully up to date. When committing,
248 : : * a transaction can assume this covers all xlog records written either by the
249 : : * user backend or by any parallel worker which was present at any point during
250 : : * the transaction. But when aborting, or when still in parallel mode, other
251 : : * parallel backends may have written WAL records at later LSNs than the value
252 : : * stored here. The parallel leader advances its own copy, when necessary,
253 : : * in WaitForParallelWorkersToFinish.
254 : : */
255 : : XLogRecPtr ProcLastRecPtr = InvalidXLogRecPtr;
256 : : XLogRecPtr XactLastRecEnd = InvalidXLogRecPtr;
257 : : XLogRecPtr XactLastCommitEnd = InvalidXLogRecPtr;
258 : :
259 : : /*
260 : : * RedoRecPtr is this backend's local copy of the REDO record pointer
261 : : * (which is almost but not quite the same as a pointer to the most recent
262 : : * CHECKPOINT record). We update this from the shared-memory copy,
263 : : * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
264 : : * hold an insertion lock). See XLogInsertRecord for details. We are also
265 : : * allowed to update from XLogCtl->RedoRecPtr if we hold the info_lck;
266 : : * see GetRedoRecPtr.
267 : : *
268 : : * NB: Code that uses this variable must be prepared not only for the
269 : : * possibility that it may be arbitrarily out of date, but also for the
270 : : * possibility that it might be set to InvalidXLogRecPtr. We used to
271 : : * initialize it as a side effect of the first call to RecoveryInProgress(),
272 : : * which meant that most code that might use it could assume that it had a
273 : : * real if perhaps stale value. That's no longer the case.
274 : : */
275 : : static XLogRecPtr RedoRecPtr;
276 : :
277 : : /*
278 : : * doPageWrites is this backend's local copy of (fullPageWrites ||
279 : : * runningBackups > 0). It is used together with RedoRecPtr to decide whether
280 : : * a full-page image of a page need to be taken.
281 : : *
282 : : * NB: Initially this is false, and there's no guarantee that it will be
283 : : * initialized to any other value before it is first used. Any code that
284 : : * makes use of it must recheck the value after obtaining a WALInsertLock,
285 : : * and respond appropriately if it turns out that the previous value wasn't
286 : : * accurate.
287 : : */
288 : : static bool doPageWrites;
289 : :
290 : : /*----------
291 : : * Shared-memory data structures for XLOG control
292 : : *
293 : : * LogwrtRqst indicates a byte position that we need to write and/or fsync
294 : : * the log up to (all records before that point must be written or fsynced).
295 : : * The positions already written/fsynced are maintained in logWriteResult
296 : : * and logFlushResult using atomic access.
297 : : * In addition to the shared variable, each backend has a private copy of
298 : : * both in LogwrtResult, which is updated when convenient.
299 : : *
300 : : * The request bookkeeping is simpler: there is a shared XLogCtl->LogwrtRqst
301 : : * (protected by info_lck), but we don't need to cache any copies of it.
302 : : *
303 : : * info_lck is only held long enough to read/update the protected variables,
304 : : * so it's a plain spinlock. The other locks are held longer (potentially
305 : : * over I/O operations), so we use LWLocks for them. These locks are:
306 : : *
307 : : * WALBufMappingLock: must be held to replace a page in the WAL buffer cache.
308 : : * It is only held while initializing and changing the mapping. If the
309 : : * contents of the buffer being replaced haven't been written yet, the mapping
310 : : * lock is released while the write is done, and reacquired afterwards.
311 : : *
312 : : * WALWriteLock: must be held to write WAL buffers to disk (XLogWrite or
313 : : * XLogFlush).
314 : : *
315 : : * ControlFileLock: must be held to read/update control file or create
316 : : * new log file.
317 : : *
318 : : *----------
319 : : */
320 : :
321 : : typedef struct XLogwrtRqst
322 : : {
323 : : XLogRecPtr Write; /* last byte + 1 to write out */
324 : : XLogRecPtr Flush; /* last byte + 1 to flush */
325 : : } XLogwrtRqst;
326 : :
327 : : typedef struct XLogwrtResult
328 : : {
329 : : XLogRecPtr Write; /* last byte + 1 written out */
330 : : XLogRecPtr Flush; /* last byte + 1 flushed */
331 : : } XLogwrtResult;
332 : :
333 : : /*
334 : : * Inserting to WAL is protected by a small fixed number of WAL insertion
335 : : * locks. To insert to the WAL, you must hold one of the locks - it doesn't
336 : : * matter which one. To lock out other concurrent insertions, you must hold
337 : : * of them. Each WAL insertion lock consists of a lightweight lock, plus an
338 : : * indicator of how far the insertion has progressed (insertingAt).
339 : : *
340 : : * The insertingAt values are read when a process wants to flush WAL from
341 : : * the in-memory buffers to disk, to check that all the insertions to the
342 : : * region the process is about to write out have finished. You could simply
343 : : * wait for all currently in-progress insertions to finish, but the
344 : : * insertingAt indicator allows you to ignore insertions to later in the WAL,
345 : : * so that you only wait for the insertions that are modifying the buffers
346 : : * you're about to write out.
347 : : *
348 : : * This isn't just an optimization. If all the WAL buffers are dirty, an
349 : : * inserter that's holding a WAL insert lock might need to evict an old WAL
350 : : * buffer, which requires flushing the WAL. If it's possible for an inserter
351 : : * to block on another inserter unnecessarily, deadlock can arise when two
352 : : * inserters holding a WAL insert lock wait for each other to finish their
353 : : * insertion.
354 : : *
355 : : * Small WAL records that don't cross a page boundary never update the value,
356 : : * the WAL record is just copied to the page and the lock is released. But
357 : : * to avoid the deadlock-scenario explained above, the indicator is always
358 : : * updated before sleeping while holding an insertion lock.
359 : : *
360 : : * lastImportantAt contains the LSN of the last important WAL record inserted
361 : : * using a given lock. This value is used to detect if there has been
362 : : * important WAL activity since the last time some action, like a checkpoint,
363 : : * was performed - allowing to not repeat the action if not. The LSN is
364 : : * updated for all insertions, unless the XLOG_MARK_UNIMPORTANT flag was
365 : : * set. lastImportantAt is never cleared, only overwritten by the LSN of newer
366 : : * records. Tracking the WAL activity directly in WALInsertLock has the
367 : : * advantage of not needing any additional locks to update the value.
368 : : */
369 : : typedef struct
370 : : {
371 : : LWLock lock;
372 : : pg_atomic_uint64 insertingAt;
373 : : XLogRecPtr lastImportantAt;
374 : : } WALInsertLock;
375 : :
376 : : /*
377 : : * All the WAL insertion locks are allocated as an array in shared memory. We
378 : : * force the array stride to be a power of 2, which saves a few cycles in
379 : : * indexing, but more importantly also ensures that individual slots don't
380 : : * cross cache line boundaries. (Of course, we have to also ensure that the
381 : : * array start address is suitably aligned.)
382 : : */
383 : : typedef union WALInsertLockPadded
384 : : {
385 : : WALInsertLock l;
386 : : char pad[PG_CACHE_LINE_SIZE];
387 : : } WALInsertLockPadded;
388 : :
389 : : /*
390 : : * Session status of running backup, used for sanity checks in SQL-callable
391 : : * functions to start and stop backups.
392 : : */
393 : : static SessionBackupState sessionBackupState = SESSION_BACKUP_NONE;
394 : :
395 : : /*
396 : : * Shared state data for WAL insertion.
397 : : */
398 : : typedef struct XLogCtlInsert
399 : : {
400 : : slock_t insertpos_lck; /* protects CurrBytePos and PrevBytePos */
401 : :
402 : : /*
403 : : * CurrBytePos is the end of reserved WAL. The next record will be
404 : : * inserted at that position. PrevBytePos is the start position of the
405 : : * previously inserted (or rather, reserved) record - it is copied to the
406 : : * prev-link of the next record. These are stored as "usable byte
407 : : * positions" rather than XLogRecPtrs (see XLogBytePosToRecPtr()).
408 : : */
409 : : uint64 CurrBytePos;
410 : : uint64 PrevBytePos;
411 : :
412 : : /*
413 : : * Make sure the above heavily-contended spinlock and byte positions are
414 : : * on their own cache line. In particular, the RedoRecPtr and full page
415 : : * write variables below should be on a different cache line. They are
416 : : * read on every WAL insertion, but updated rarely, and we don't want
417 : : * those reads to steal the cache line containing Curr/PrevBytePos.
418 : : */
419 : : char pad[PG_CACHE_LINE_SIZE];
420 : :
421 : : /*
422 : : * fullPageWrites is the authoritative value used by all backends to
423 : : * determine whether to write full-page image to WAL. This shared value,
424 : : * instead of the process-local fullPageWrites, is required because, when
425 : : * full_page_writes is changed by SIGHUP, we must WAL-log it before it
426 : : * actually affects WAL-logging by backends. Checkpointer sets at startup
427 : : * or after SIGHUP.
428 : : *
429 : : * To read these fields, you must hold an insertion lock. To modify them,
430 : : * you must hold ALL the locks.
431 : : */
432 : : XLogRecPtr RedoRecPtr; /* current redo point for insertions */
433 : : bool fullPageWrites;
434 : :
435 : : /*
436 : : * runningBackups is a counter indicating the number of backups currently
437 : : * in progress. lastBackupStart is the latest checkpoint redo location
438 : : * used as a starting point for an online backup.
439 : : */
440 : : int runningBackups;
441 : : XLogRecPtr lastBackupStart;
442 : :
443 : : /*
444 : : * WAL insertion locks.
445 : : */
446 : : WALInsertLockPadded *WALInsertLocks;
447 : : } XLogCtlInsert;
448 : :
449 : : /*
450 : : * Total shared-memory state for XLOG.
451 : : */
452 : : typedef struct XLogCtlData
453 : : {
454 : : XLogCtlInsert Insert;
455 : :
456 : : /* Protected by info_lck: */
457 : : XLogwrtRqst LogwrtRqst;
458 : : XLogRecPtr RedoRecPtr; /* a recent copy of Insert->RedoRecPtr */
459 : : XLogRecPtr asyncXactLSN; /* LSN of newest async commit/abort */
460 : : XLogRecPtr replicationSlotMinLSN; /* oldest LSN needed by any slot */
461 : :
462 : : XLogSegNo lastRemovedSegNo; /* latest removed/recycled XLOG segment */
463 : :
464 : : /* Fake LSN counter, for unlogged relations. */
465 : : pg_atomic_uint64 unloggedLSN;
466 : :
467 : : /* Time and LSN of last xlog segment switch. Protected by WALWriteLock. */
468 : : pg_time_t lastSegSwitchTime;
469 : : XLogRecPtr lastSegSwitchLSN;
470 : :
471 : : /* These are accessed using atomics -- info_lck not needed */
472 : : pg_atomic_uint64 logInsertResult; /* last byte + 1 inserted to buffers */
473 : : pg_atomic_uint64 logWriteResult; /* last byte + 1 written out */
474 : : pg_atomic_uint64 logFlushResult; /* last byte + 1 flushed */
475 : :
476 : : /*
477 : : * Latest initialized page in the cache (last byte position + 1).
478 : : *
479 : : * To change the identity of a buffer (and InitializedUpTo), you need to
480 : : * hold WALBufMappingLock. To change the identity of a buffer that's
481 : : * still dirty, the old page needs to be written out first, and for that
482 : : * you need WALWriteLock, and you need to ensure that there are no
483 : : * in-progress insertions to the page by calling
484 : : * WaitXLogInsertionsToFinish().
485 : : */
486 : : XLogRecPtr InitializedUpTo;
487 : :
488 : : /*
489 : : * These values do not change after startup, although the pointed-to pages
490 : : * and xlblocks values certainly do. xlblocks values are protected by
491 : : * WALBufMappingLock.
492 : : */
493 : : char *pages; /* buffers for unwritten XLOG pages */
494 : : pg_atomic_uint64 *xlblocks; /* 1st byte ptr-s + XLOG_BLCKSZ */
495 : : int XLogCacheBlck; /* highest allocated xlog buffer index */
496 : :
497 : : /*
498 : : * InsertTimeLineID is the timeline into which new WAL is being inserted
499 : : * and flushed. It is zero during recovery, and does not change once set.
500 : : *
501 : : * If we create a new timeline when the system was started up,
502 : : * PrevTimeLineID is the old timeline's ID that we forked off from.
503 : : * Otherwise it's equal to InsertTimeLineID.
504 : : *
505 : : * We set these fields while holding info_lck. Most that reads these
506 : : * values knows that recovery is no longer in progress and so can safely
507 : : * read the value without a lock, but code that could be run either during
508 : : * or after recovery can take info_lck while reading these values.
509 : : */
510 : : TimeLineID InsertTimeLineID;
511 : : TimeLineID PrevTimeLineID;
512 : :
513 : : /*
514 : : * SharedRecoveryState indicates if we're still in crash or archive
515 : : * recovery. Protected by info_lck.
516 : : */
517 : : RecoveryState SharedRecoveryState;
518 : :
519 : : /*
520 : : * InstallXLogFileSegmentActive indicates whether the checkpointer should
521 : : * arrange for future segments by recycling and/or PreallocXlogFiles().
522 : : * Protected by ControlFileLock. Only the startup process changes it. If
523 : : * true, anyone can use InstallXLogFileSegment(). If false, the startup
524 : : * process owns the exclusive right to install segments, by reading from
525 : : * the archive and possibly replacing existing files.
526 : : */
527 : : bool InstallXLogFileSegmentActive;
528 : :
529 : : /*
530 : : * WalWriterSleeping indicates whether the WAL writer is currently in
531 : : * low-power mode (and hence should be nudged if an async commit occurs).
532 : : * Protected by info_lck.
533 : : */
534 : : bool WalWriterSleeping;
535 : :
536 : : /*
537 : : * During recovery, we keep a copy of the latest checkpoint record here.
538 : : * lastCheckPointRecPtr points to start of checkpoint record and
539 : : * lastCheckPointEndPtr points to end+1 of checkpoint record. Used by the
540 : : * checkpointer when it wants to create a restartpoint.
541 : : *
542 : : * Protected by info_lck.
543 : : */
544 : : XLogRecPtr lastCheckPointRecPtr;
545 : : XLogRecPtr lastCheckPointEndPtr;
546 : : CheckPoint lastCheckPoint;
547 : :
548 : : /*
549 : : * lastFpwDisableRecPtr points to the start of the last replayed
550 : : * XLOG_FPW_CHANGE record that instructs full_page_writes is disabled.
551 : : */
552 : : XLogRecPtr lastFpwDisableRecPtr;
553 : :
554 : : slock_t info_lck; /* locks shared variables shown above */
555 : : } XLogCtlData;
556 : :
557 : : /*
558 : : * Classification of XLogInsertRecord operations.
559 : : */
560 : : typedef enum
561 : : {
562 : : WALINSERT_NORMAL,
563 : : WALINSERT_SPECIAL_SWITCH,
564 : : WALINSERT_SPECIAL_CHECKPOINT
565 : : } WalInsertClass;
566 : :
567 : : static XLogCtlData *XLogCtl = NULL;
568 : :
569 : : /* a private copy of XLogCtl->Insert.WALInsertLocks, for convenience */
570 : : static WALInsertLockPadded *WALInsertLocks = NULL;
571 : :
572 : : /*
573 : : * We maintain an image of pg_control in shared memory.
574 : : */
575 : : static ControlFileData *ControlFile = NULL;
576 : :
577 : : /*
578 : : * Calculate the amount of space left on the page after 'endptr'. Beware
579 : : * multiple evaluation!
580 : : */
581 : : #define INSERT_FREESPACE(endptr) \
582 : : (((endptr) % XLOG_BLCKSZ == 0) ? 0 : (XLOG_BLCKSZ - (endptr) % XLOG_BLCKSZ))
583 : :
584 : : /* Macro to advance to next buffer index. */
585 : : #define NextBufIdx(idx) \
586 : : (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))
587 : :
588 : : /*
589 : : * XLogRecPtrToBufIdx returns the index of the WAL buffer that holds, or
590 : : * would hold if it was in cache, the page containing 'recptr'.
591 : : */
592 : : #define XLogRecPtrToBufIdx(recptr) \
593 : : (((recptr) / XLOG_BLCKSZ) % (XLogCtl->XLogCacheBlck + 1))
594 : :
595 : : /*
596 : : * These are the number of bytes in a WAL page usable for WAL data.
597 : : */
598 : : #define UsableBytesInPage (XLOG_BLCKSZ - SizeOfXLogShortPHD)
599 : :
600 : : /*
601 : : * Convert values of GUCs measured in megabytes to equiv. segment count.
602 : : * Rounds down.
603 : : */
604 : : #define ConvertToXSegs(x, segsize) XLogMBVarToSegs((x), (segsize))
605 : :
606 : : /* The number of bytes in a WAL segment usable for WAL data. */
607 : : static int UsableBytesInSegment;
608 : :
609 : : /*
610 : : * Private, possibly out-of-date copy of shared LogwrtResult.
611 : : * See discussion above.
612 : : */
613 : : static XLogwrtResult LogwrtResult = {0, 0};
614 : :
615 : : /*
616 : : * Update local copy of shared XLogCtl->log{Write,Flush}Result
617 : : *
618 : : * It's critical that Flush always trails Write, so the order of the reads is
619 : : * important, as is the barrier. See also XLogWrite.
620 : : */
621 : : #define RefreshXLogWriteResult(_target) \
622 : : do { \
623 : : _target.Flush = pg_atomic_read_u64(&XLogCtl->logFlushResult); \
624 : : pg_read_barrier(); \
625 : : _target.Write = pg_atomic_read_u64(&XLogCtl->logWriteResult); \
626 : : } while (0)
627 : :
628 : : /*
629 : : * openLogFile is -1 or a kernel FD for an open log file segment.
630 : : * openLogSegNo identifies the segment, and openLogTLI the corresponding TLI.
631 : : * These variables are only used to write the XLOG, and so will normally refer
632 : : * to the active segment.
633 : : *
634 : : * Note: call Reserve/ReleaseExternalFD to track consumption of this FD.
635 : : */
636 : : static int openLogFile = -1;
637 : : static XLogSegNo openLogSegNo = 0;
638 : : static TimeLineID openLogTLI = 0;
639 : :
640 : : /*
641 : : * Local copies of equivalent fields in the control file. When running
642 : : * crash recovery, LocalMinRecoveryPoint is set to InvalidXLogRecPtr as we
643 : : * expect to replay all the WAL available, and updateMinRecoveryPoint is
644 : : * switched to false to prevent any updates while replaying records.
645 : : * Those values are kept consistent as long as crash recovery runs.
646 : : */
647 : : static XLogRecPtr LocalMinRecoveryPoint;
648 : : static TimeLineID LocalMinRecoveryPointTLI;
649 : : static bool updateMinRecoveryPoint = true;
650 : :
651 : : /* For WALInsertLockAcquire/Release functions */
652 : : static int MyLockNo = 0;
653 : : static bool holdingAllLocks = false;
654 : :
655 : : #ifdef WAL_DEBUG
656 : : static MemoryContext walDebugCxt = NULL;
657 : : #endif
658 : :
659 : : static void CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI,
660 : : XLogRecPtr EndOfLog,
661 : : TimeLineID newTLI);
662 : : static void CheckRequiredParameterValues(void);
663 : : static void XLogReportParameters(void);
664 : : static int LocalSetXLogInsertAllowed(void);
665 : : static void CreateEndOfRecoveryRecord(void);
666 : : static XLogRecPtr CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn,
667 : : XLogRecPtr pagePtr,
668 : : TimeLineID newTLI);
669 : : static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
670 : : static void KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo);
671 : : static XLogRecPtr XLogGetReplicationSlotMinimumLSN(void);
672 : :
673 : : static void AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli,
674 : : bool opportunistic);
675 : : static void XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible);
676 : : static bool InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
677 : : bool find_free, XLogSegNo max_segno,
678 : : TimeLineID tli);
679 : : static void XLogFileClose(void);
680 : : static void PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli);
681 : : static void RemoveTempXlogFiles(void);
682 : : static void RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr,
683 : : XLogRecPtr endptr, TimeLineID insertTLI);
684 : : static void RemoveXlogFile(const struct dirent *segment_de,
685 : : XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo,
686 : : TimeLineID insertTLI);
687 : : static void UpdateLastRemovedPtr(char *filename);
688 : : static void ValidateXLOGDirectoryStructure(void);
689 : : static void CleanupBackupHistory(void);
690 : : static void UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force);
691 : : static bool PerformRecoveryXLogAction(void);
692 : : static void InitControlFile(uint64 sysidentifier, uint32 data_checksum_version);
693 : : static void WriteControlFile(void);
694 : : static void ReadControlFile(void);
695 : : static void UpdateControlFile(void);
696 : : static char *str_time(pg_time_t tnow, char *buf, size_t bufsize);
697 : :
698 : : static int get_sync_bit(int method);
699 : :
700 : : static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch,
701 : : XLogRecData *rdata,
702 : : XLogRecPtr StartPos, XLogRecPtr EndPos,
703 : : TimeLineID tli);
704 : : static void ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos,
705 : : XLogRecPtr *EndPos, XLogRecPtr *PrevPtr);
706 : : static bool ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos,
707 : : XLogRecPtr *PrevPtr);
708 : : static XLogRecPtr WaitXLogInsertionsToFinish(XLogRecPtr upto);
709 : : static char *GetXLogBuffer(XLogRecPtr ptr, TimeLineID tli);
710 : : static XLogRecPtr XLogBytePosToRecPtr(uint64 bytepos);
711 : : static XLogRecPtr XLogBytePosToEndRecPtr(uint64 bytepos);
712 : : static uint64 XLogRecPtrToBytePos(XLogRecPtr ptr);
713 : :
714 : : static void WALInsertLockAcquire(void);
715 : : static void WALInsertLockAcquireExclusive(void);
716 : : static void WALInsertLockRelease(void);
717 : : static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
718 : :
719 : : /*
720 : : * Insert an XLOG record represented by an already-constructed chain of data
721 : : * chunks. This is a low-level routine; to construct the WAL record header
722 : : * and data, use the higher-level routines in xloginsert.c.
723 : : *
724 : : * If 'fpw_lsn' is valid, it is the oldest LSN among the pages that this
725 : : * WAL record applies to, that were not included in the record as full page
726 : : * images. If fpw_lsn <= RedoRecPtr, the function does not perform the
727 : : * insertion and returns InvalidXLogRecPtr. The caller can then recalculate
728 : : * which pages need a full-page image, and retry. If fpw_lsn is invalid, the
729 : : * record is always inserted.
730 : : *
731 : : * 'flags' gives more in-depth control on the record being inserted. See
732 : : * XLogSetRecordFlags() for details.
733 : : *
734 : : * 'topxid_included' tells whether the top-transaction id is logged along with
735 : : * current subtransaction. See XLogRecordAssemble().
736 : : *
737 : : * The first XLogRecData in the chain must be for the record header, and its
738 : : * data must be MAXALIGNed. XLogInsertRecord fills in the xl_prev and
739 : : * xl_crc fields in the header, the rest of the header must already be filled
740 : : * by the caller.
741 : : *
742 : : * Returns XLOG pointer to end of record (beginning of next record).
743 : : * This can be used as LSN for data pages affected by the logged action.
744 : : * (LSN is the XLOG point up to which the XLOG must be flushed to disk
745 : : * before the data page can be written out. This implements the basic
746 : : * WAL rule "write the log before the data".)
747 : : */
748 : : XLogRecPtr
3283 andres@anarazel.de 749 :CBC 14169133 : XLogInsertRecord(XLogRecData *rdata,
750 : : XLogRecPtr fpw_lsn,
751 : : uint8 flags,
752 : : int num_fpi,
753 : : uint64 fpi_bytes,
754 : : bool topxid_included)
755 : : {
9037 bruce@momjian.us 756 : 14169133 : XLogCtlInsert *Insert = &XLogCtl->Insert;
757 : : pg_crc32c rdata_crc;
758 : : bool inserted;
4060 heikki.linnakangas@i 759 : 14169133 : XLogRecord *rechdr = (XLogRecord *) rdata->data;
3331 tgl@sss.pgh.pa.us 760 : 14169133 : uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
791 rhaas@postgresql.org 761 : 14169133 : WalInsertClass class = WALINSERT_NORMAL;
762 : : XLogRecPtr StartPos;
763 : : XLogRecPtr EndPos;
2653 akapila@postgresql.o 764 : 14169133 : bool prevDoPageWrites = doPageWrites;
765 : : TimeLineID insertTLI;
766 : :
767 : : /* Does this record type require special handling? */
791 rhaas@postgresql.org 768 [ + + ]: 14169133 : if (unlikely(rechdr->xl_rmid == RM_XLOG_ID))
769 : : {
770 [ + + ]: 220343 : if (info == XLOG_SWITCH)
771 : 710 : class = WALINSERT_SPECIAL_SWITCH;
772 [ + + ]: 219633 : else if (info == XLOG_CHECKPOINT_REDO)
773 : 915 : class = WALINSERT_SPECIAL_CHECKPOINT;
774 : : }
775 : :
776 : : /* we assume that all of the record header is in the first chunk */
4046 heikki.linnakangas@i 777 [ - + ]: 14169133 : Assert(rdata->len >= SizeOfXLogRecord);
778 : :
779 : : /* cross-check on whether we should be here or not */
6019 tgl@sss.pgh.pa.us 780 [ - + ]: 14169133 : if (!XLogInsertAllowed())
6019 tgl@sss.pgh.pa.us 781 [ # # ]:UBC 0 : elog(ERROR, "cannot make new WAL entries during recovery");
782 : :
783 : : /*
784 : : * Given that we're not in recovery, InsertTimeLineID is set and can't
785 : : * change, so we can read it without a lock.
786 : : */
1499 rhaas@postgresql.org 787 :CBC 14169133 : insertTLI = XLogCtl->InsertTimeLineID;
788 : :
789 : : /*----------
790 : : *
791 : : * We have now done all the preparatory work we can without holding a
792 : : * lock or modifying shared state. From here on, inserting the new WAL
793 : : * record to the shared WAL buffer cache is a two-step process:
794 : : *
795 : : * 1. Reserve the right amount of space from the WAL. The current head of
796 : : * reserved space is kept in Insert->CurrBytePos, and is protected by
797 : : * insertpos_lck.
798 : : *
799 : : * 2. Copy the record to the reserved WAL space. This involves finding the
800 : : * correct WAL buffer containing the reserved space, and copying the
801 : : * record in place. This can be done concurrently in multiple processes.
802 : : *
803 : : * To keep track of which insertions are still in-progress, each concurrent
804 : : * inserter acquires an insertion lock. In addition to just indicating that
805 : : * an insertion is in progress, the lock tells others how far the inserter
806 : : * has progressed. There is a small fixed number of insertion locks,
807 : : * determined by NUM_XLOGINSERT_LOCKS. When an inserter crosses a page
808 : : * boundary, it updates the value stored in the lock to the how far it has
809 : : * inserted, to allow the previous buffer to be flushed.
810 : : *
811 : : * Holding onto an insertion lock also protects RedoRecPtr and
812 : : * fullPageWrites from changing until the insertion is finished.
813 : : *
814 : : * Step 2 can usually be done completely in parallel. If the required WAL
815 : : * page is not initialized yet, you have to grab WALBufMappingLock to
816 : : * initialize it, but the WAL writer tries to do that ahead of insertions
817 : : * to avoid that from happening in the critical path.
818 : : *
819 : : *----------
820 : : */
5090 heikki.linnakangas@i 821 : 14169133 : START_CRIT_SECTION();
822 : :
791 rhaas@postgresql.org 823 [ + + ]: 14169133 : if (likely(class == WALINSERT_NORMAL))
824 : : {
800 825 : 14167508 : WALInsertLockAcquire();
826 : :
827 : : /*
828 : : * Check to see if my copy of RedoRecPtr is out of date. If so, may
829 : : * have to go back and have the caller recompute everything. This can
830 : : * only happen just after a checkpoint, so it's better to be slow in
831 : : * this case and fast otherwise.
832 : : *
833 : : * Also check to see if fullPageWrites was just turned on or there's a
834 : : * running backup (which forces full-page writes); if we weren't
835 : : * already doing full-page writes then go back and recompute.
836 : : *
837 : : * If we aren't doing full-page writes then RedoRecPtr doesn't
838 : : * actually affect the contents of the XLOG record, so we'll update
839 : : * our local copy but not force a recomputation. (If doPageWrites was
840 : : * just turned off, we could recompute the record without full pages,
841 : : * but we choose not to bother.)
842 : : */
843 [ + + ]: 14167508 : if (RedoRecPtr != Insert->RedoRecPtr)
844 : : {
845 [ - + ]: 6386 : Assert(RedoRecPtr < Insert->RedoRecPtr);
846 : 6386 : RedoRecPtr = Insert->RedoRecPtr;
847 : : }
848 [ + + + + ]: 14167508 : doPageWrites = (Insert->fullPageWrites || Insert->runningBackups > 0);
849 : :
850 [ + + ]: 14167508 : if (doPageWrites &&
851 [ + + + + ]: 13918652 : (!prevDoPageWrites ||
42 alvherre@kurilemu.de 852 [ + + ]:GNC 13167940 : (XLogRecPtrIsValid(fpw_lsn) && fpw_lsn <= RedoRecPtr)))
853 : : {
854 : : /*
855 : : * Oops, some buffer now needs to be backed up that the caller
856 : : * didn't back up. Start over.
857 : : */
800 rhaas@postgresql.org 858 :CBC 7106 : WALInsertLockRelease();
859 [ - + ]: 7106 : END_CRIT_SECTION();
860 : 7106 : return InvalidXLogRecPtr;
861 : : }
862 : :
863 : : /*
864 : : * Reserve space for the record in the WAL. This also sets the xl_prev
865 : : * pointer.
866 : : */
4060 heikki.linnakangas@i 867 : 14160402 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
868 : : &rechdr->xl_prev);
869 : :
870 : : /* Normal records are always inserted. */
4546 871 : 14160402 : inserted = true;
872 : : }
791 rhaas@postgresql.org 873 [ + + ]: 1625 : else if (class == WALINSERT_SPECIAL_SWITCH)
874 : : {
875 : : /*
876 : : * In order to insert an XLOG_SWITCH record, we need to hold all of
877 : : * the WAL insertion locks, not just one, so that no one else can
878 : : * begin inserting a record until we've figured out how much space
879 : : * remains in the current WAL segment and claimed all of it.
880 : : *
881 : : * Nonetheless, this case is simpler than the normal cases handled
882 : : * below, which must check for changes in doPageWrites and RedoRecPtr.
883 : : * Those checks are only needed for records that can contain buffer
884 : : * references, and an XLOG_SWITCH record never does.
885 : : */
42 alvherre@kurilemu.de 886 [ - + ]:GNC 710 : Assert(!XLogRecPtrIsValid(fpw_lsn));
800 rhaas@postgresql.org 887 :CBC 710 : WALInsertLockAcquireExclusive();
888 : 710 : inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
889 : : }
890 : : else
891 : : {
791 892 [ - + ]: 915 : Assert(class == WALINSERT_SPECIAL_CHECKPOINT);
893 : :
894 : : /*
895 : : * We need to update both the local and shared copies of RedoRecPtr,
896 : : * which means that we need to hold all the WAL insertion locks.
897 : : * However, there can't be any buffer references, so as above, we need
898 : : * not check RedoRecPtr before inserting the record; we just need to
899 : : * update it afterwards.
900 : : */
42 alvherre@kurilemu.de 901 [ - + ]:GNC 915 : Assert(!XLogRecPtrIsValid(fpw_lsn));
791 rhaas@postgresql.org 902 :CBC 915 : WALInsertLockAcquireExclusive();
903 : 915 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
904 : : &rechdr->xl_prev);
905 : 915 : RedoRecPtr = Insert->RedoRecPtr = StartPos;
906 : 915 : inserted = true;
907 : : }
908 : :
4546 heikki.linnakangas@i 909 [ + + ]: 14162027 : if (inserted)
910 : : {
911 : : /*
912 : : * Now that xl_prev has been filled in, calculate CRC of the record
913 : : * header.
914 : : */
4046 915 : 14161962 : rdata_crc = rechdr->xl_crc;
916 : 14161962 : COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
4062 917 : 14161962 : FIN_CRC32C(rdata_crc);
4546 918 : 14161962 : rechdr->xl_crc = rdata_crc;
919 : :
920 : : /*
921 : : * All the record data, including the header, is now ready to be
922 : : * inserted. Copy the record in the space reserved.
923 : : */
791 rhaas@postgresql.org 924 : 14161962 : CopyXLogRecordToWAL(rechdr->xl_tot_len,
925 : : class == WALINSERT_SPECIAL_SWITCH, rdata,
926 : : StartPos, EndPos, insertTLI);
927 : :
928 : : /*
929 : : * Unless record is flagged as not important, update LSN of last
930 : : * important record in the current slot. When holding all locks, just
931 : : * update the first one.
932 : : */
3283 andres@anarazel.de 933 [ + + ]: 14161962 : if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
934 : : {
3137 bruce@momjian.us 935 [ + + ]: 14071071 : int lockno = holdingAllLocks ? 0 : MyLockNo;
936 : :
3283 andres@anarazel.de 937 : 14071071 : WALInsertLocks[lockno].l.lastImportantAt = StartPos;
938 : : }
939 : : }
940 : : else
941 : : {
942 : : /*
943 : : * This was an xlog-switch record, but the current insert location was
944 : : * already exactly at the beginning of a segment, so there was no need
945 : : * to do anything.
946 : : */
947 : : }
948 : :
949 : : /*
950 : : * Done! Let others know that we're finished.
951 : : */
4290 heikki.linnakangas@i 952 : 14162027 : WALInsertLockRelease();
953 : :
4546 954 [ - + ]: 14162027 : END_CRIT_SECTION();
955 : :
1507 akapila@postgresql.o 956 : 14162027 : MarkCurrentTransactionIdLoggedIfAny();
957 : :
958 : : /*
959 : : * Mark top transaction id is logged (if needed) so that we should not try
960 : : * to log it again with the next WAL record in the current subtransaction.
961 : : */
962 [ + + ]: 14162027 : if (topxid_included)
963 : 219 : MarkSubxactTopXidLogged();
964 : :
965 : : /*
966 : : * Update shared LogwrtRqst.Write, if we crossed page boundary.
967 : : */
4546 heikki.linnakangas@i 968 [ + + ]: 14162027 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
969 : : {
4105 andres@anarazel.de 970 [ + + ]: 1685581 : SpinLockAcquire(&XLogCtl->info_lck);
971 : : /* advance global request to include new block(s) */
972 [ + + ]: 1685581 : if (XLogCtl->LogwrtRqst.Write < EndPos)
973 : 1615672 : XLogCtl->LogwrtRqst.Write = EndPos;
974 : 1685581 : SpinLockRelease(&XLogCtl->info_lck);
622 alvherre@alvh.no-ip. 975 : 1685581 : RefreshXLogWriteResult(LogwrtResult);
976 : : }
977 : :
978 : : /*
979 : : * If this was an XLOG_SWITCH record, flush the record and the empty
980 : : * padding space that fills the rest of the segment, and perform
981 : : * end-of-segment actions (eg, notifying archiver).
982 : : */
791 rhaas@postgresql.org 983 [ + + ]: 14162027 : if (class == WALINSERT_SPECIAL_SWITCH)
984 : : {
985 : : TRACE_POSTGRESQL_WAL_SWITCH();
4546 heikki.linnakangas@i 986 : 710 : XLogFlush(EndPos);
987 : :
988 : : /*
989 : : * Even though we reserved the rest of the segment for us, which is
990 : : * reflected in EndPos, we return a pointer to just the end of the
991 : : * xlog-switch record.
992 : : */
993 [ + + ]: 710 : if (inserted)
994 : : {
995 : 645 : EndPos = StartPos + SizeOfXLogRecord;
996 [ - + ]: 645 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
997 : : {
3012 andres@anarazel.de 998 :UBC 0 : uint64 offset = XLogSegmentOffset(EndPos, wal_segment_size);
999 : :
1000 [ # # ]: 0 : if (offset == EndPos % XLOG_BLCKSZ)
4546 heikki.linnakangas@i 1001 : 0 : EndPos += SizeOfXLogLongPHD;
1002 : : else
1003 : 0 : EndPos += SizeOfXLogShortPHD;
1004 : : }
1005 : : }
1006 : : }
1007 : :
1008 : : #ifdef WAL_DEBUG
1009 : : if (XLOG_DEBUG)
1010 : : {
1011 : : static XLogReaderState *debug_reader = NULL;
1012 : : XLogRecord *record;
1013 : : DecodedXLogRecord *decoded;
1014 : : StringInfoData buf;
1015 : : StringInfoData recordBuf;
1016 : : char *errormsg = NULL;
1017 : : MemoryContext oldCxt;
1018 : :
1019 : : oldCxt = MemoryContextSwitchTo(walDebugCxt);
1020 : :
1021 : : initStringInfo(&buf);
1022 : : appendStringInfo(&buf, "INSERT @ %X/%08X: ", LSN_FORMAT_ARGS(EndPos));
1023 : :
1024 : : /*
1025 : : * We have to piece together the WAL record data from the XLogRecData
1026 : : * entries, so that we can pass it to the rm_desc function as one
1027 : : * contiguous chunk.
1028 : : */
1029 : : initStringInfo(&recordBuf);
1030 : : for (; rdata != NULL; rdata = rdata->next)
1031 : : appendBinaryStringInfo(&recordBuf, rdata->data, rdata->len);
1032 : :
1033 : : /* We also need temporary space to decode the record. */
1034 : : record = (XLogRecord *) recordBuf.data;
1035 : : decoded = (DecodedXLogRecord *)
1036 : : palloc(DecodeXLogRecordRequiredSpace(record->xl_tot_len));
1037 : :
1038 : : if (!debug_reader)
1039 : : debug_reader = XLogReaderAllocate(wal_segment_size, NULL,
1040 : : XL_ROUTINE(.page_read = NULL,
1041 : : .segment_open = NULL,
1042 : : .segment_close = NULL),
1043 : : NULL);
1044 : : if (!debug_reader)
1045 : : {
1046 : : appendStringInfoString(&buf, "error decoding record: out of memory while allocating a WAL reading processor");
1047 : : }
1048 : : else if (!DecodeXLogRecord(debug_reader,
1049 : : decoded,
1050 : : record,
1051 : : EndPos,
1052 : : &errormsg))
1053 : : {
1054 : : appendStringInfo(&buf, "error decoding record: %s",
1055 : : errormsg ? errormsg : "no error message");
1056 : : }
1057 : : else
1058 : : {
1059 : : appendStringInfoString(&buf, " - ");
1060 : :
1061 : : debug_reader->record = decoded;
1062 : : xlog_outdesc(&buf, debug_reader);
1063 : : debug_reader->record = NULL;
1064 : : }
1065 : : elog(LOG, "%s", buf.data);
1066 : :
1067 : : pfree(decoded);
1068 : : pfree(buf.data);
1069 : : pfree(recordBuf.data);
1070 : : MemoryContextSwitchTo(oldCxt);
1071 : : }
1072 : : #endif
1073 : :
1074 : : /*
1075 : : * Update our global variables
1076 : : */
4546 heikki.linnakangas@i 1077 :CBC 14162027 : ProcLastRecPtr = StartPos;
1078 : 14162027 : XactLastRecEnd = EndPos;
1079 : :
1080 : : /* Report WAL traffic to the instrumentation. */
2084 akapila@postgresql.o 1081 [ + + ]: 14162027 : if (inserted)
1082 : : {
1083 : 14161962 : pgWalUsage.wal_bytes += rechdr->xl_tot_len;
1084 : 14161962 : pgWalUsage.wal_records++;
2053 1085 : 14161962 : pgWalUsage.wal_fpi += num_fpi;
50 michael@paquier.xyz 1086 :GNC 14161962 : pgWalUsage.wal_fpi_bytes += fpi_bytes;
1087 : :
1088 : : /* Required for the flush of pending stats WAL data */
143 michael@paquier.xyz 1089 :CBC 14161962 : pgstat_report_fixed = true;
1090 : : }
1091 : :
4546 heikki.linnakangas@i 1092 : 14162027 : return EndPos;
1093 : : }
1094 : :
1095 : : /*
1096 : : * Reserves the right amount of space for a record of given size from the WAL.
1097 : : * *StartPos is set to the beginning of the reserved section, *EndPos to
1098 : : * its end+1. *PrevPtr is set to the beginning of the previous record; it is
1099 : : * used to set the xl_prev of this record.
1100 : : *
1101 : : * This is the performance critical part of XLogInsert that must be serialized
1102 : : * across backends. The rest can happen mostly in parallel. Try to keep this
1103 : : * section as short as possible, insertpos_lck can be heavily contended on a
1104 : : * busy system.
1105 : : *
1106 : : * NB: The space calculation here must match the code in CopyXLogRecordToWAL,
1107 : : * where we actually copy the record to the reserved space.
1108 : : *
1109 : : * NB: Testing shows that XLogInsertRecord runs faster if this code is inlined;
1110 : : * however, because there are two call sites, the compiler is reluctant to
1111 : : * inline. We use pg_attribute_always_inline here to try to convince it.
1112 : : */
1113 : : static pg_attribute_always_inline void
1114 : 14161317 : ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos,
1115 : : XLogRecPtr *PrevPtr)
1116 : : {
4105 andres@anarazel.de 1117 : 14161317 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1118 : : uint64 startbytepos;
1119 : : uint64 endbytepos;
1120 : : uint64 prevbytepos;
1121 : :
4546 heikki.linnakangas@i 1122 : 14161317 : size = MAXALIGN(size);
1123 : :
1124 : : /* All (non xlog-switch) records should contain data. */
1125 [ - + ]: 14161317 : Assert(size > SizeOfXLogRecord);
1126 : :
1127 : : /*
1128 : : * The duration the spinlock needs to be held is minimized by minimizing
1129 : : * the calculations that have to be done while holding the lock. The
1130 : : * current tip of reserved WAL is kept in CurrBytePos, as a byte position
1131 : : * that only counts "usable" bytes in WAL, that is, it excludes all WAL
1132 : : * page headers. The mapping between "usable" byte positions and physical
1133 : : * positions (XLogRecPtrs) can be done outside the locked region, and
1134 : : * because the usable byte position doesn't include any headers, reserving
1135 : : * X bytes from WAL is almost as simple as "CurrBytePos += X".
1136 : : */
1137 [ + + ]: 14161317 : SpinLockAcquire(&Insert->insertpos_lck);
1138 : :
1139 : 14161317 : startbytepos = Insert->CurrBytePos;
1140 : 14161317 : endbytepos = startbytepos + size;
1141 : 14161317 : prevbytepos = Insert->PrevBytePos;
1142 : 14161317 : Insert->CurrBytePos = endbytepos;
1143 : 14161317 : Insert->PrevBytePos = startbytepos;
1144 : :
1145 : 14161317 : SpinLockRelease(&Insert->insertpos_lck);
1146 : :
1147 : 14161317 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1148 : 14161317 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1149 : 14161317 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1150 : :
1151 : : /*
1152 : : * Check that the conversions between "usable byte positions" and
1153 : : * XLogRecPtrs work consistently in both directions.
1154 : : */
1155 [ - + ]: 14161317 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1156 [ - + ]: 14161317 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1157 [ - + ]: 14161317 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1158 : 14161317 : }
1159 : :
1160 : : /*
1161 : : * Like ReserveXLogInsertLocation(), but for an xlog-switch record.
1162 : : *
1163 : : * A log-switch record is handled slightly differently. The rest of the
1164 : : * segment will be reserved for this insertion, as indicated by the returned
1165 : : * *EndPos value. However, if we are already at the beginning of the current
1166 : : * segment, *StartPos and *EndPos are set to the current location without
1167 : : * reserving any space, and the function returns false.
1168 : : */
1169 : : static bool
1170 : 710 : ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
1171 : : {
4105 andres@anarazel.de 1172 : 710 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1173 : : uint64 startbytepos;
1174 : : uint64 endbytepos;
1175 : : uint64 prevbytepos;
4046 heikki.linnakangas@i 1176 : 710 : uint32 size = MAXALIGN(SizeOfXLogRecord);
1177 : : XLogRecPtr ptr;
1178 : : uint32 segleft;
1179 : :
1180 : : /*
1181 : : * These calculations are a bit heavy-weight to be done while holding a
1182 : : * spinlock, but since we're holding all the WAL insertion locks, there
1183 : : * are no other inserters competing for it. GetXLogInsertRecPtr() does
1184 : : * compete for it, but that's not called very frequently.
1185 : : */
4546 1186 [ - + ]: 710 : SpinLockAcquire(&Insert->insertpos_lck);
1187 : :
1188 : 710 : startbytepos = Insert->CurrBytePos;
1189 : :
1190 : 710 : ptr = XLogBytePosToEndRecPtr(startbytepos);
3012 andres@anarazel.de 1191 [ + + ]: 710 : if (XLogSegmentOffset(ptr, wal_segment_size) == 0)
1192 : : {
4546 heikki.linnakangas@i 1193 : 65 : SpinLockRelease(&Insert->insertpos_lck);
1194 : 65 : *EndPos = *StartPos = ptr;
1195 : 65 : return false;
1196 : : }
1197 : :
1198 : 645 : endbytepos = startbytepos + size;
1199 : 645 : prevbytepos = Insert->PrevBytePos;
1200 : :
1201 : 645 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1202 : 645 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1203 : :
3012 andres@anarazel.de 1204 : 645 : segleft = wal_segment_size - XLogSegmentOffset(*EndPos, wal_segment_size);
1205 [ + - ]: 645 : if (segleft != wal_segment_size)
1206 : : {
1207 : : /* consume the rest of the segment */
4546 heikki.linnakangas@i 1208 : 645 : *EndPos += segleft;
1209 : 645 : endbytepos = XLogRecPtrToBytePos(*EndPos);
1210 : : }
1211 : 645 : Insert->CurrBytePos = endbytepos;
1212 : 645 : Insert->PrevBytePos = startbytepos;
1213 : :
1214 : 645 : SpinLockRelease(&Insert->insertpos_lck);
1215 : :
1216 : 645 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1217 : :
3012 andres@anarazel.de 1218 [ - + ]: 645 : Assert(XLogSegmentOffset(*EndPos, wal_segment_size) == 0);
4546 heikki.linnakangas@i 1219 [ - + ]: 645 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1220 [ - + ]: 645 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1221 [ - + ]: 645 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1222 : :
1223 : 645 : return true;
1224 : : }
1225 : :
1226 : : /*
1227 : : * Subroutine of XLogInsertRecord. Copies a WAL record to an already-reserved
1228 : : * area in the WAL.
1229 : : */
1230 : : static void
1231 : 14161962 : CopyXLogRecordToWAL(int write_len, bool isLogSwitch, XLogRecData *rdata,
1232 : : XLogRecPtr StartPos, XLogRecPtr EndPos, TimeLineID tli)
1233 : : {
1234 : : char *currpos;
1235 : : int freespace;
1236 : : int written;
1237 : : XLogRecPtr CurrPos;
1238 : : XLogPageHeader pagehdr;
1239 : :
1240 : : /*
1241 : : * Get a pointer to the right place in the right WAL buffer to start
1242 : : * inserting to.
1243 : : */
1244 : 14161962 : CurrPos = StartPos;
1504 rhaas@postgresql.org 1245 : 14161962 : currpos = GetXLogBuffer(CurrPos, tli);
4546 heikki.linnakangas@i 1246 [ + - ]: 14161962 : freespace = INSERT_FREESPACE(CurrPos);
1247 : :
1248 : : /*
1249 : : * there should be enough space for at least the first field (xl_tot_len)
1250 : : * on this page.
1251 : : */
1252 [ - + ]: 14161962 : Assert(freespace >= sizeof(uint32));
1253 : :
1254 : : /* Copy record data */
1255 : 14161962 : written = 0;
1256 [ + + ]: 67645175 : while (rdata != NULL)
1257 : : {
471 peter@eisentraut.org 1258 : 53483213 : const char *rdata_data = rdata->data;
4546 heikki.linnakangas@i 1259 : 53483213 : int rdata_len = rdata->len;
1260 : :
1261 [ + + ]: 55289202 : while (rdata_len > freespace)
1262 : : {
1263 : : /*
1264 : : * Write what fits on this page, and continue on the next page.
1265 : : */
1266 [ + + - + ]: 1805989 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || freespace == 0);
1267 : 1805989 : memcpy(currpos, rdata_data, freespace);
1268 : 1805989 : rdata_data += freespace;
1269 : 1805989 : rdata_len -= freespace;
1270 : 1805989 : written += freespace;
1271 : 1805989 : CurrPos += freespace;
1272 : :
1273 : : /*
1274 : : * Get pointer to beginning of next page, and set the xlp_rem_len
1275 : : * in the page header. Set XLP_FIRST_IS_CONTRECORD.
1276 : : *
1277 : : * It's safe to set the contrecord flag and xlp_rem_len without a
1278 : : * lock on the page. All the other flags were already set when the
1279 : : * page was initialized, in AdvanceXLInsertBuffer, and we're the
1280 : : * only backend that needs to set the contrecord flag.
1281 : : */
1504 rhaas@postgresql.org 1282 : 1805989 : currpos = GetXLogBuffer(CurrPos, tli);
4546 heikki.linnakangas@i 1283 : 1805989 : pagehdr = (XLogPageHeader) currpos;
1284 : 1805989 : pagehdr->xlp_rem_len = write_len - written;
1285 : 1805989 : pagehdr->xlp_info |= XLP_FIRST_IS_CONTRECORD;
1286 : :
1287 : : /* skip over the page header */
3012 andres@anarazel.de 1288 [ + + ]: 1805989 : if (XLogSegmentOffset(CurrPos, wal_segment_size) == 0)
1289 : : {
4546 heikki.linnakangas@i 1290 : 1200 : CurrPos += SizeOfXLogLongPHD;
1291 : 1200 : currpos += SizeOfXLogLongPHD;
1292 : : }
1293 : : else
1294 : : {
1295 : 1804789 : CurrPos += SizeOfXLogShortPHD;
1296 : 1804789 : currpos += SizeOfXLogShortPHD;
1297 : : }
1298 [ + - ]: 1805989 : freespace = INSERT_FREESPACE(CurrPos);
1299 : : }
1300 : :
1301 [ - + - - ]: 53483213 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || rdata_len == 0);
1302 : 53483213 : memcpy(currpos, rdata_data, rdata_len);
1303 : 53483213 : currpos += rdata_len;
1304 : 53483213 : CurrPos += rdata_len;
1305 : 53483213 : freespace -= rdata_len;
1306 : 53483213 : written += rdata_len;
1307 : :
1308 : 53483213 : rdata = rdata->next;
1309 : : }
1310 [ - + ]: 14161962 : Assert(written == write_len);
1311 : :
1312 : : /*
1313 : : * If this was an xlog-switch, it's not enough to write the switch record,
1314 : : * we also have to consume all the remaining space in the WAL segment. We
1315 : : * have already reserved that space, but we need to actually fill it.
1316 : : */
3012 andres@anarazel.de 1317 [ + + + - ]: 14161962 : if (isLogSwitch && XLogSegmentOffset(CurrPos, wal_segment_size) != 0)
1318 : : {
1319 : : /* An xlog-switch record doesn't contain any data besides the header */
4546 heikki.linnakangas@i 1320 [ - + ]: 645 : Assert(write_len == SizeOfXLogRecord);
1321 : :
1322 : : /* Assert that we did reserve the right amount of space */
3012 andres@anarazel.de 1323 [ - + ]: 645 : Assert(XLogSegmentOffset(EndPos, wal_segment_size) == 0);
1324 : :
1325 : : /* Use up all the remaining space on the current page */
4546 heikki.linnakangas@i 1326 : 645 : CurrPos += freespace;
1327 : :
1328 : : /*
1329 : : * Cause all remaining pages in the segment to be flushed, leaving the
1330 : : * XLog position where it should be, at the start of the next segment.
1331 : : * We do this one page at a time, to make sure we don't deadlock
1332 : : * against ourselves if wal_buffers < wal_segment_size.
1333 : : */
1334 [ + + ]: 538036 : while (CurrPos < EndPos)
1335 : : {
1336 : : /*
1337 : : * The minimal action to flush the page would be to call
1338 : : * WALInsertLockUpdateInsertingAt(CurrPos) followed by
1339 : : * AdvanceXLInsertBuffer(...). The page would be left initialized
1340 : : * mostly to zeros, except for the page header (always the short
1341 : : * variant, as this is never a segment's first page).
1342 : : *
1343 : : * The large vistas of zeros are good for compressibility, but the
1344 : : * headers interrupting them every XLOG_BLCKSZ (with values that
1345 : : * differ from page to page) are not. The effect varies with
1346 : : * compression tool, but bzip2 for instance compresses about an
1347 : : * order of magnitude worse if those headers are left in place.
1348 : : *
1349 : : * Rather than complicating AdvanceXLInsertBuffer itself (which is
1350 : : * called in heavily-loaded circumstances as well as this lightly-
1351 : : * loaded one) with variant behavior, we just use GetXLogBuffer
1352 : : * (which itself calls the two methods we need) to get the pointer
1353 : : * and zero most of the page. Then we just zero the page header.
1354 : : */
1504 rhaas@postgresql.org 1355 : 537391 : currpos = GetXLogBuffer(CurrPos, tli);
2820 tgl@sss.pgh.pa.us 1356 [ + - + - : 2149564 : MemSet(currpos, 0, SizeOfXLogShortPHD);
+ - + - +
+ ]
1357 : :
4546 heikki.linnakangas@i 1358 : 537391 : CurrPos += XLOG_BLCKSZ;
1359 : : }
1360 : : }
1361 : : else
1362 : : {
1363 : : /* Align the end position, so that the next record starts aligned */
4046 1364 : 14161317 : CurrPos = MAXALIGN64(CurrPos);
1365 : : }
1366 : :
4546 1367 [ - + ]: 14161962 : if (CurrPos != EndPos)
624 dgustafsson@postgres 1368 [ # # ]:UBC 0 : ereport(PANIC,
1369 : : errcode(ERRCODE_DATA_CORRUPTED),
1370 : : errmsg_internal("space reserved for WAL record does not match what was written"));
4546 heikki.linnakangas@i 1371 :CBC 14161962 : }
1372 : :
1373 : : /*
1374 : : * Acquire a WAL insertion lock, for inserting to WAL.
1375 : : */
1376 : : static void
4290 1377 : 14167518 : WALInsertLockAcquire(void)
1378 : : {
1379 : : bool immed;
1380 : :
1381 : : /*
1382 : : * It doesn't matter which of the WAL insertion locks we acquire, so try
1383 : : * the one we used last time. If the system isn't particularly busy, it's
1384 : : * a good bet that it's still available, and it's good to have some
1385 : : * affinity to a particular lock so that you don't unnecessarily bounce
1386 : : * cache lines between processes when there's no contention.
1387 : : *
1388 : : * If this is the first time through in this backend, pick a lock
1389 : : * (semi-)randomly. This allows the locks to be used evenly if you have a
1390 : : * lot of very short connections.
1391 : : */
1392 : : static int lockToTry = -1;
1393 : :
1394 [ + + ]: 14167518 : if (lockToTry == -1)
665 1395 : 7483 : lockToTry = MyProcNumber % NUM_XLOGINSERT_LOCKS;
4290 1396 : 14167518 : MyLockNo = lockToTry;
1397 : :
1398 : : /*
1399 : : * The insertingAt value is initially set to 0, as we don't know our
1400 : : * insert location yet.
1401 : : */
3793 andres@anarazel.de 1402 : 14167518 : immed = LWLockAcquire(&WALInsertLocks[MyLockNo].l.lock, LW_EXCLUSIVE);
4290 heikki.linnakangas@i 1403 [ + + ]: 14167518 : if (!immed)
1404 : : {
1405 : : /*
1406 : : * If we couldn't get the lock immediately, try another lock next
1407 : : * time. On a system with more insertion locks than concurrent
1408 : : * inserters, this causes all the inserters to eventually migrate to a
1409 : : * lock that no-one else is using. On a system with more inserters
1410 : : * than locks, it still helps to distribute the inserters evenly
1411 : : * across the locks.
1412 : : */
4096 1413 : 19253 : lockToTry = (lockToTry + 1) % NUM_XLOGINSERT_LOCKS;
1414 : : }
4546 1415 : 14167518 : }
1416 : :
1417 : : /*
1418 : : * Acquire all WAL insertion locks, to prevent other backends from inserting
1419 : : * to WAL.
1420 : : */
1421 : : static void
4290 1422 : 4356 : WALInsertLockAcquireExclusive(void)
1423 : : {
1424 : : int i;
1425 : :
1426 : : /*
1427 : : * When holding all the locks, all but the last lock's insertingAt
1428 : : * indicator is set to 0xFFFFFFFFFFFFFFFF, which is higher than any real
1429 : : * XLogRecPtr value, to make sure that no-one blocks waiting on those.
1430 : : */
4096 1431 [ + + ]: 34848 : for (i = 0; i < NUM_XLOGINSERT_LOCKS - 1; i++)
1432 : : {
3793 andres@anarazel.de 1433 : 30492 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1434 : 30492 : LWLockUpdateVar(&WALInsertLocks[i].l.lock,
1435 : 30492 : &WALInsertLocks[i].l.insertingAt,
1436 : : PG_UINT64_MAX);
1437 : : }
1438 : : /* Variable value reset to 0 at release */
1439 : 4356 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1440 : :
4290 heikki.linnakangas@i 1441 : 4356 : holdingAllLocks = true;
4546 1442 : 4356 : }
1443 : :
1444 : : /*
1445 : : * Release our insertion lock (or locks, if we're holding them all).
1446 : : *
1447 : : * NB: Reset all variables to 0, so they cause LWLockWaitForVar to block the
1448 : : * next time the lock is acquired.
1449 : : */
1450 : : static void
4290 1451 : 14171874 : WALInsertLockRelease(void)
1452 : : {
1453 [ + + ]: 14171874 : if (holdingAllLocks)
1454 : : {
1455 : : int i;
1456 : :
4096 1457 [ + + ]: 39204 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
3793 andres@anarazel.de 1458 : 34848 : LWLockReleaseClearVar(&WALInsertLocks[i].l.lock,
1459 : 34848 : &WALInsertLocks[i].l.insertingAt,
1460 : : 0);
1461 : :
4290 heikki.linnakangas@i 1462 : 4356 : holdingAllLocks = false;
1463 : : }
1464 : : else
1465 : : {
3793 andres@anarazel.de 1466 : 14167518 : LWLockReleaseClearVar(&WALInsertLocks[MyLockNo].l.lock,
1467 : 14167518 : &WALInsertLocks[MyLockNo].l.insertingAt,
1468 : : 0);
1469 : : }
4546 heikki.linnakangas@i 1470 : 14171874 : }
1471 : :
1472 : : /*
1473 : : * Update our insertingAt value, to let others know that we've finished
1474 : : * inserting up to that point.
1475 : : */
1476 : : static void
4290 1477 : 2284313 : WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt)
1478 : : {
1479 [ + + ]: 2284313 : if (holdingAllLocks)
1480 : : {
1481 : : /*
1482 : : * We use the last lock to mark our actual position, see comments in
1483 : : * WALInsertLockAcquireExclusive.
1484 : : */
4096 1485 : 536084 : LWLockUpdateVar(&WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.lock,
3102 tgl@sss.pgh.pa.us 1486 : 536084 : &WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.insertingAt,
1487 : : insertingAt);
1488 : : }
1489 : : else
4290 heikki.linnakangas@i 1490 : 1748229 : LWLockUpdateVar(&WALInsertLocks[MyLockNo].l.lock,
1491 : 1748229 : &WALInsertLocks[MyLockNo].l.insertingAt,
1492 : : insertingAt);
4546 1493 : 2284313 : }
1494 : :
1495 : : /*
1496 : : * Wait for any WAL insertions < upto to finish.
1497 : : *
1498 : : * Returns the location of the oldest insertion that is still in-progress.
1499 : : * Any WAL prior to that point has been fully copied into WAL buffers, and
1500 : : * can be flushed out to disk. Because this waits for any insertions older
1501 : : * than 'upto' to finish, the return value is always >= 'upto'.
1502 : : *
1503 : : * Note: When you are about to write out WAL, you must call this function
1504 : : * *before* acquiring WALWriteLock, to avoid deadlocks. This function might
1505 : : * need to wait for an insertion to finish (or at least advance to next
1506 : : * uninitialized page), and the inserter might need to evict an old WAL buffer
1507 : : * to make room for a new one, which in turn requires WALWriteLock.
1508 : : */
1509 : : static XLogRecPtr
1510 : 2168315 : WaitXLogInsertionsToFinish(XLogRecPtr upto)
1511 : : {
1512 : : uint64 bytepos;
1513 : : XLogRecPtr inserted;
1514 : : XLogRecPtr reservedUpto;
1515 : : XLogRecPtr finishedUpto;
4105 andres@anarazel.de 1516 : 2168315 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1517 : : int i;
1518 : :
4546 heikki.linnakangas@i 1519 [ - + ]: 2168315 : if (MyProc == NULL)
4546 heikki.linnakangas@i 1520 [ # # ]:UBC 0 : elog(PANIC, "cannot wait without a PGPROC structure");
1521 : :
1522 : : /*
1523 : : * Check if there's any work to do. Use a barrier to ensure we get the
1524 : : * freshest value.
1525 : : */
620 alvherre@alvh.no-ip. 1526 :CBC 2168315 : inserted = pg_atomic_read_membarrier_u64(&XLogCtl->logInsertResult);
1527 [ + + ]: 2168315 : if (upto <= inserted)
1528 : 1695312 : return inserted;
1529 : :
1530 : : /* Read the current insert position */
4546 heikki.linnakangas@i 1531 [ + + ]: 473003 : SpinLockAcquire(&Insert->insertpos_lck);
1532 : 473003 : bytepos = Insert->CurrBytePos;
1533 : 473003 : SpinLockRelease(&Insert->insertpos_lck);
1534 : 473003 : reservedUpto = XLogBytePosToEndRecPtr(bytepos);
1535 : :
1536 : : /*
1537 : : * No-one should request to flush a piece of WAL that hasn't even been
1538 : : * reserved yet. However, it can happen if there is a block with a bogus
1539 : : * LSN on disk, for example. XLogFlush checks for that situation and
1540 : : * complains, but only after the flush. Here we just assume that to mean
1541 : : * that all WAL that has been reserved needs to be finished. In this
1542 : : * corner-case, the return value can be smaller than 'upto' argument.
1543 : : */
1544 [ - + ]: 473003 : if (upto > reservedUpto)
1545 : : {
1840 peter@eisentraut.org 1546 [ # # ]:UBC 0 : ereport(LOG,
1547 : : errmsg("request to flush past end of generated WAL; request %X/%08X, current position %X/%08X",
1548 : : LSN_FORMAT_ARGS(upto), LSN_FORMAT_ARGS(reservedUpto)));
4546 heikki.linnakangas@i 1549 : 0 : upto = reservedUpto;
1550 : : }
1551 : :
1552 : : /*
1553 : : * Loop through all the locks, sleeping on any in-progress insert older
1554 : : * than 'upto'.
1555 : : *
1556 : : * finishedUpto is our return value, indicating the point upto which all
1557 : : * the WAL insertions have been finished. Initialize it to the head of
1558 : : * reserved WAL, and as we iterate through the insertion locks, back it
1559 : : * out for any insertion that's still in progress.
1560 : : */
4546 heikki.linnakangas@i 1561 :CBC 473003 : finishedUpto = reservedUpto;
4096 1562 [ + + ]: 4257027 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1563 : : {
4244 bruce@momjian.us 1564 : 3784024 : XLogRecPtr insertingat = InvalidXLogRecPtr;
1565 : :
1566 : : do
1567 : : {
1568 : : /*
1569 : : * See if this insertion is in progress. LWLockWaitForVar will
1570 : : * wait for the lock to be released, or for the 'value' to be set
1571 : : * by a LWLockUpdateVar call. When a lock is initially acquired,
1572 : : * its value is 0 (InvalidXLogRecPtr), which means that we don't
1573 : : * know where it's inserting yet. We will have to wait for it. If
1574 : : * it's a small insertion, the record will most likely fit on the
1575 : : * same page and the inserter will release the lock without ever
1576 : : * calling LWLockUpdateVar. But if it has to sleep, it will
1577 : : * advertise the insertion point with LWLockUpdateVar before
1578 : : * sleeping.
1579 : : *
1580 : : * In this loop we are only waiting for insertions that started
1581 : : * before WaitXLogInsertionsToFinish was called. The lack of
1582 : : * memory barriers in the loop means that we might see locks as
1583 : : * "unused" that have since become used. This is fine because
1584 : : * they only can be used for later insertions that we would not
1585 : : * want to wait on anyway. Not taking a lock to acquire the
1586 : : * current insertingAt value means that we might see older
1587 : : * insertingAt values. This is also fine, because if we read a
1588 : : * value too old, we will add ourselves to the wait queue, which
1589 : : * contains atomic operations.
1590 : : */
4290 heikki.linnakangas@i 1591 [ + + ]: 3922653 : if (LWLockWaitForVar(&WALInsertLocks[i].l.lock,
1592 : 3922653 : &WALInsertLocks[i].l.insertingAt,
1593 : : insertingat, &insertingat))
1594 : : {
1595 : : /* the lock was free, so no insertion in progress */
1596 : 2661276 : insertingat = InvalidXLogRecPtr;
1597 : 2661276 : break;
1598 : : }
1599 : :
1600 : : /*
1601 : : * This insertion is still in progress. Have to wait, unless the
1602 : : * inserter has proceeded past 'upto'.
1603 : : */
1604 [ + + ]: 1261377 : } while (insertingat < upto);
1605 : :
42 alvherre@kurilemu.de 1606 [ + + + + ]:GNC 3784024 : if (XLogRecPtrIsValid(insertingat) && insertingat < finishedUpto)
4290 heikki.linnakangas@i 1607 :CBC 422363 : finishedUpto = insertingat;
1608 : : }
1609 : :
1610 : : /*
1611 : : * Advance the limit we know to have been inserted and return the freshest
1612 : : * value we know of, which might be beyond what we requested if somebody
1613 : : * is concurrently doing this with an 'upto' pointer ahead of us.
1614 : : */
620 alvherre@alvh.no-ip. 1615 : 473003 : finishedUpto = pg_atomic_monotonic_advance_u64(&XLogCtl->logInsertResult,
1616 : : finishedUpto);
1617 : :
4546 heikki.linnakangas@i 1618 : 473003 : return finishedUpto;
1619 : : }
1620 : :
1621 : : /*
1622 : : * Get a pointer to the right location in the WAL buffer containing the
1623 : : * given XLogRecPtr.
1624 : : *
1625 : : * If the page is not initialized yet, it is initialized. That might require
1626 : : * evicting an old dirty buffer from the buffer cache, which means I/O.
1627 : : *
1628 : : * The caller must ensure that the page containing the requested location
1629 : : * isn't evicted yet, and won't be evicted. The way to ensure that is to
1630 : : * hold onto a WAL insertion lock with the insertingAt position set to
1631 : : * something <= ptr. GetXLogBuffer() will update insertingAt if it needs
1632 : : * to evict an old page from the buffer. (This means that once you call
1633 : : * GetXLogBuffer() with a given 'ptr', you must not access anything before
1634 : : * that point anymore, and must not call GetXLogBuffer() with an older 'ptr'
1635 : : * later, because older buffers might be recycled already)
1636 : : */
1637 : : static char *
1504 rhaas@postgresql.org 1638 : 16505352 : GetXLogBuffer(XLogRecPtr ptr, TimeLineID tli)
1639 : : {
1640 : : int idx;
1641 : : XLogRecPtr endptr;
1642 : : static uint64 cachedPage = 0;
1643 : : static char *cachedPos = NULL;
1644 : : XLogRecPtr expectedEndPtr;
1645 : :
1646 : : /*
1647 : : * Fast path for the common case that we need to access again the same
1648 : : * page as last time.
1649 : : */
4546 heikki.linnakangas@i 1650 [ + + ]: 16505352 : if (ptr / XLOG_BLCKSZ == cachedPage)
1651 : : {
1652 [ - + ]: 13801017 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1653 [ - + ]: 13801017 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1654 : 13801017 : return cachedPos + ptr % XLOG_BLCKSZ;
1655 : : }
1656 : :
1657 : : /*
1658 : : * The XLog buffer cache is organized so that a page is always loaded to a
1659 : : * particular buffer. That way we can easily calculate the buffer a given
1660 : : * page must be loaded into, from the XLogRecPtr alone.
1661 : : */
1662 : 2704335 : idx = XLogRecPtrToBufIdx(ptr);
1663 : :
1664 : : /*
1665 : : * See what page is loaded in the buffer at the moment. It could be the
1666 : : * page we're looking for, or something older. It can't be anything newer
1667 : : * - that would imply the page we're looking for has already been written
1668 : : * out to disk and evicted, and the caller is responsible for making sure
1669 : : * that doesn't happen.
1670 : : *
1671 : : * We don't hold a lock while we read the value. If someone is just about
1672 : : * to initialize or has just initialized the page, it's possible that we
1673 : : * get InvalidXLogRecPtr. That's ok, we'll grab the mapping lock (in
1674 : : * AdvanceXLInsertBuffer) and retry if we see anything other than the page
1675 : : * we're looking for.
1676 : : */
1677 : 2704335 : expectedEndPtr = ptr;
1678 : 2704335 : expectedEndPtr += XLOG_BLCKSZ - ptr % XLOG_BLCKSZ;
1679 : :
730 jdavis@postgresql.or 1680 : 2704335 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
4546 heikki.linnakangas@i 1681 [ + + ]: 2704335 : if (expectedEndPtr != endptr)
1682 : : {
1683 : : XLogRecPtr initializedUpto;
1684 : :
1685 : : /*
1686 : : * Before calling AdvanceXLInsertBuffer(), which can block, let others
1687 : : * know how far we're finished with inserting the record.
1688 : : *
1689 : : * NB: If 'ptr' points to just after the page header, advertise a
1690 : : * position at the beginning of the page rather than 'ptr' itself. If
1691 : : * there are no other insertions running, someone might try to flush
1692 : : * up to our advertised location. If we advertised a position after
1693 : : * the page header, someone might try to flush the page header, even
1694 : : * though page might actually not be initialized yet. As the first
1695 : : * inserter on the page, we are effectively responsible for making
1696 : : * sure that it's initialized, before we let insertingAt to move past
1697 : : * the page header.
1698 : : */
3791 1699 [ + + ]: 2284313 : if (ptr % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
3012 andres@anarazel.de 1700 [ + - ]: 5486 : XLogSegmentOffset(ptr, wal_segment_size) > XLOG_BLCKSZ)
3791 heikki.linnakangas@i 1701 : 5486 : initializedUpto = ptr - SizeOfXLogShortPHD;
1702 [ + + ]: 2278827 : else if (ptr % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
3012 andres@anarazel.de 1703 [ + + ]: 762 : XLogSegmentOffset(ptr, wal_segment_size) < XLOG_BLCKSZ)
3791 heikki.linnakangas@i 1704 : 535 : initializedUpto = ptr - SizeOfXLogLongPHD;
1705 : : else
1706 : 2278292 : initializedUpto = ptr;
1707 : :
1708 : 2284313 : WALInsertLockUpdateInsertingAt(initializedUpto);
1709 : :
1504 rhaas@postgresql.org 1710 : 2284313 : AdvanceXLInsertBuffer(ptr, tli, false);
730 jdavis@postgresql.or 1711 : 2284313 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1712 : :
4546 heikki.linnakangas@i 1713 [ - + ]: 2284313 : if (expectedEndPtr != endptr)
164 alvherre@kurilemu.de 1714 [ # # ]:UNC 0 : elog(PANIC, "could not find WAL buffer for %X/%08X",
1715 : : LSN_FORMAT_ARGS(ptr));
1716 : : }
1717 : : else
1718 : : {
1719 : : /*
1720 : : * Make sure the initialization of the page is visible to us, and
1721 : : * won't arrive later to overwrite the WAL data we write on the page.
1722 : : */
4546 heikki.linnakangas@i 1723 :CBC 420022 : pg_memory_barrier();
1724 : : }
1725 : :
1726 : : /*
1727 : : * Found the buffer holding this page. Return a pointer to the right
1728 : : * offset within the page.
1729 : : */
1730 : 2704335 : cachedPage = ptr / XLOG_BLCKSZ;
1731 : 2704335 : cachedPos = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1732 : :
1733 [ - + ]: 2704335 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1734 [ - + ]: 2704335 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1735 : :
1736 : 2704335 : return cachedPos + ptr % XLOG_BLCKSZ;
1737 : : }
1738 : :
1739 : : /*
1740 : : * Read WAL data directly from WAL buffers, if available. Returns the number
1741 : : * of bytes read successfully.
1742 : : *
1743 : : * Fewer than 'count' bytes may be read if some of the requested WAL data has
1744 : : * already been evicted.
1745 : : *
1746 : : * No locks are taken.
1747 : : *
1748 : : * Caller should ensure that it reads no further than LogwrtResult.Write
1749 : : * (which should have been updated by the caller when determining how far to
1750 : : * read). The 'tli' argument is only used as a convenient safety check so that
1751 : : * callers do not read from WAL buffers on a historical timeline.
1752 : : */
1753 : : Size
675 jdavis@postgresql.or 1754 : 102593 : WALReadFromBuffers(char *dstbuf, XLogRecPtr startptr, Size count,
1755 : : TimeLineID tli)
1756 : : {
1757 : 102593 : char *pdst = dstbuf;
1758 : 102593 : XLogRecPtr recptr = startptr;
1759 : : XLogRecPtr inserted;
671 1760 : 102593 : Size nbytes = count;
1761 : :
675 1762 [ + + + + ]: 102593 : if (RecoveryInProgress() || tli != GetWALInsertionTimeLine())
1763 : 719 : return 0;
1764 : :
42 alvherre@kurilemu.de 1765 [ - + ]:GNC 101874 : Assert(XLogRecPtrIsValid(startptr));
1766 : :
1767 : : /*
1768 : : * Caller should ensure that the requested data has been inserted into WAL
1769 : : * buffers before we try to read it.
1770 : : */
620 alvherre@alvh.no-ip. 1771 :CBC 101874 : inserted = pg_atomic_read_u64(&XLogCtl->logInsertResult);
1772 [ - + ]: 101874 : if (startptr + count > inserted)
620 alvherre@alvh.no-ip. 1773 [ # # ]:UBC 0 : ereport(ERROR,
1774 : : errmsg("cannot read past end of generated WAL: requested %X/%08X, current position %X/%08X",
1775 : : LSN_FORMAT_ARGS(startptr + count),
1776 : : LSN_FORMAT_ARGS(inserted)));
1777 : :
1778 : : /*
1779 : : * Loop through the buffers without a lock. For each buffer, atomically
1780 : : * read and verify the end pointer, then copy the data out, and finally
1781 : : * re-read and re-verify the end pointer.
1782 : : *
1783 : : * Once a page is evicted, it never returns to the WAL buffers, so if the
1784 : : * end pointer matches the expected end pointer before and after we copy
1785 : : * the data, then the right page must have been present during the data
1786 : : * copy. Read barriers are necessary to ensure that the data copy actually
1787 : : * happens between the two verification steps.
1788 : : *
1789 : : * If either verification fails, we simply terminate the loop and return
1790 : : * with the data that had been already copied out successfully.
1791 : : */
675 jdavis@postgresql.or 1792 [ + + ]:CBC 127680 : while (nbytes > 0)
1793 : : {
1794 : 120322 : uint32 offset = recptr % XLOG_BLCKSZ;
1795 : 120322 : int idx = XLogRecPtrToBufIdx(recptr);
1796 : : XLogRecPtr expectedEndPtr;
1797 : : XLogRecPtr endptr;
1798 : : const char *page;
1799 : : const char *psrc;
1800 : : Size npagebytes;
1801 : :
1802 : : /*
1803 : : * Calculate the end pointer we expect in the xlblocks array if the
1804 : : * correct page is present.
1805 : : */
1806 : 120322 : expectedEndPtr = recptr + (XLOG_BLCKSZ - offset);
1807 : :
1808 : : /*
1809 : : * First verification step: check that the correct page is present in
1810 : : * the WAL buffers.
1811 : : */
1812 : 120322 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1813 [ + + ]: 120322 : if (expectedEndPtr != endptr)
1814 : 94514 : break;
1815 : :
1816 : : /*
1817 : : * The correct page is present (or was at the time the endptr was
1818 : : * read; must re-verify later). Calculate pointer to source data and
1819 : : * determine how much data to read from this page.
1820 : : */
1821 : 25808 : page = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1822 : 25808 : psrc = page + offset;
1823 : 25808 : npagebytes = Min(nbytes, XLOG_BLCKSZ - offset);
1824 : :
1825 : : /*
1826 : : * Ensure that the data copy and the first verification step are not
1827 : : * reordered.
1828 : : */
1829 : 25808 : pg_read_barrier();
1830 : :
1831 : : /* data copy */
1832 : 25808 : memcpy(pdst, psrc, npagebytes);
1833 : :
1834 : : /*
1835 : : * Ensure that the data copy and the second verification step are not
1836 : : * reordered.
1837 : : */
1838 : 25808 : pg_read_barrier();
1839 : :
1840 : : /*
1841 : : * Second verification step: check that the page we read from wasn't
1842 : : * evicted while we were copying the data.
1843 : : */
1844 : 25808 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1845 [ + + ]: 25808 : if (expectedEndPtr != endptr)
675 jdavis@postgresql.or 1846 :GBC 2 : break;
1847 : :
675 jdavis@postgresql.or 1848 :CBC 25806 : pdst += npagebytes;
1849 : 25806 : recptr += npagebytes;
1850 : 25806 : nbytes -= npagebytes;
1851 : : }
1852 : :
1853 [ - + ]: 101874 : Assert(pdst - dstbuf <= count);
1854 : :
1855 : 101874 : return pdst - dstbuf;
1856 : : }
1857 : :
1858 : : /*
1859 : : * Converts a "usable byte position" to XLogRecPtr. A usable byte position
1860 : : * is the position starting from the beginning of WAL, excluding all WAL
1861 : : * page headers.
1862 : : */
1863 : : static XLogRecPtr
4546 heikki.linnakangas@i 1864 : 28326492 : XLogBytePosToRecPtr(uint64 bytepos)
1865 : : {
1866 : : uint64 fullsegs;
1867 : : uint64 fullpages;
1868 : : uint64 bytesleft;
1869 : : uint32 seg_offset;
1870 : : XLogRecPtr result;
1871 : :
1872 : 28326492 : fullsegs = bytepos / UsableBytesInSegment;
1873 : 28326492 : bytesleft = bytepos % UsableBytesInSegment;
1874 : :
1875 [ + + ]: 28326492 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1876 : : {
1877 : : /* fits on first page of segment */
1878 : 51301 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1879 : : }
1880 : : else
1881 : : {
1882 : : /* account for the first page on segment with long header */
1883 : 28275191 : seg_offset = XLOG_BLCKSZ;
1884 : 28275191 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1885 : :
1886 : 28275191 : fullpages = bytesleft / UsableBytesInPage;
1887 : 28275191 : bytesleft = bytesleft % UsableBytesInPage;
1888 : :
1889 : 28275191 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1890 : : }
1891 : :
2719 alvherre@alvh.no-ip. 1892 : 28326492 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1893 : :
4546 heikki.linnakangas@i 1894 : 28326492 : return result;
1895 : : }
1896 : :
1897 : : /*
1898 : : * Like XLogBytePosToRecPtr, but if the position is at a page boundary,
1899 : : * returns a pointer to the beginning of the page (ie. before page header),
1900 : : * not to where the first xlog record on that page would go to. This is used
1901 : : * when converting a pointer to the end of a record.
1902 : : */
1903 : : static XLogRecPtr
1904 : 14635675 : XLogBytePosToEndRecPtr(uint64 bytepos)
1905 : : {
1906 : : uint64 fullsegs;
1907 : : uint64 fullpages;
1908 : : uint64 bytesleft;
1909 : : uint32 seg_offset;
1910 : : XLogRecPtr result;
1911 : :
1912 : 14635675 : fullsegs = bytepos / UsableBytesInSegment;
1913 : 14635675 : bytesleft = bytepos % UsableBytesInSegment;
1914 : :
1915 [ + + ]: 14635675 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1916 : : {
1917 : : /* fits on first page of segment */
1918 [ + + ]: 82142 : if (bytesleft == 0)
1919 : 55233 : seg_offset = 0;
1920 : : else
1921 : 26909 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1922 : : }
1923 : : else
1924 : : {
1925 : : /* account for the first page on segment with long header */
1926 : 14553533 : seg_offset = XLOG_BLCKSZ;
1927 : 14553533 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1928 : :
1929 : 14553533 : fullpages = bytesleft / UsableBytesInPage;
1930 : 14553533 : bytesleft = bytesleft % UsableBytesInPage;
1931 : :
1932 [ + + ]: 14553533 : if (bytesleft == 0)
1933 : 14111 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft;
1934 : : else
1935 : 14539422 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1936 : : }
1937 : :
2719 alvherre@alvh.no-ip. 1938 : 14635675 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1939 : :
4546 heikki.linnakangas@i 1940 : 14635675 : return result;
1941 : : }
1942 : :
1943 : : /*
1944 : : * Convert an XLogRecPtr to a "usable byte position".
1945 : : */
1946 : : static uint64
1947 : 42488275 : XLogRecPtrToBytePos(XLogRecPtr ptr)
1948 : : {
1949 : : uint64 fullsegs;
1950 : : uint32 fullpages;
1951 : : uint32 offset;
1952 : : uint64 result;
1953 : :
3012 andres@anarazel.de 1954 : 42488275 : XLByteToSeg(ptr, fullsegs, wal_segment_size);
1955 : :
1956 : 42488275 : fullpages = (XLogSegmentOffset(ptr, wal_segment_size)) / XLOG_BLCKSZ;
4546 heikki.linnakangas@i 1957 : 42488275 : offset = ptr % XLOG_BLCKSZ;
1958 : :
1959 [ + + ]: 42488275 : if (fullpages == 0)
1960 : : {
1961 : 77671 : result = fullsegs * UsableBytesInSegment;
1962 [ + + ]: 77671 : if (offset > 0)
1963 : : {
1964 [ - + ]: 76333 : Assert(offset >= SizeOfXLogLongPHD);
1965 : 76333 : result += offset - SizeOfXLogLongPHD;
1966 : : }
1967 : : }
1968 : : else
1969 : : {
1970 : 42410604 : result = fullsegs * UsableBytesInSegment +
4244 bruce@momjian.us 1971 : 42410604 : (XLOG_BLCKSZ - SizeOfXLogLongPHD) + /* account for first page */
3102 tgl@sss.pgh.pa.us 1972 : 42410604 : (fullpages - 1) * UsableBytesInPage; /* full pages */
4546 heikki.linnakangas@i 1973 [ + + ]: 42410604 : if (offset > 0)
1974 : : {
1975 [ - + ]: 42396735 : Assert(offset >= SizeOfXLogShortPHD);
1976 : 42396735 : result += offset - SizeOfXLogShortPHD;
1977 : : }
1978 : : }
1979 : :
1980 : 42488275 : return result;
1981 : : }
1982 : :
1983 : : /*
1984 : : * Initialize XLOG buffers, writing out old buffers if they still contain
1985 : : * unwritten data, upto the page containing 'upto'. Or if 'opportunistic' is
1986 : : * true, initialize as many pages as we can without having to write out
1987 : : * unwritten data. Any new pages are initialized to zeros, with pages headers
1988 : : * initialized properly.
1989 : : */
1990 : : static void
1504 rhaas@postgresql.org 1991 : 2287601 : AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
1992 : : {
9046 tgl@sss.pgh.pa.us 1993 : 2287601 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1994 : : int nextidx;
1995 : : XLogRecPtr OldPageRqstPtr;
1996 : : XLogwrtRqst WriteRqst;
4546 heikki.linnakangas@i 1997 : 2287601 : XLogRecPtr NewPageEndPtr = InvalidXLogRecPtr;
1998 : : XLogRecPtr NewPageBeginPtr;
1999 : : XLogPageHeader NewPage;
1185 tgl@sss.pgh.pa.us 2000 : 2287601 : int npages pg_attribute_unused() = 0;
2001 : :
118 akorotkov@postgresql 2002 : 2287601 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2003 : :
2004 : : /*
2005 : : * Now that we have the lock, check if someone initialized the page
2006 : : * already.
2007 : : */
2008 [ + + + + ]: 6727569 : while (upto >= XLogCtl->InitializedUpTo || opportunistic)
2009 : : {
2010 : 4443256 : nextidx = XLogRecPtrToBufIdx(XLogCtl->InitializedUpTo);
2011 : :
2012 : : /*
2013 : : * Get ending-offset of the buffer page we need to replace (this may
2014 : : * be zero if the buffer hasn't been used yet). Fall through if it's
2015 : : * already written out.
2016 : : */
2017 : 4443256 : OldPageRqstPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[nextidx]);
2018 [ + + ]: 4443256 : if (LogwrtResult.Write < OldPageRqstPtr)
2019 : : {
2020 : : /*
2021 : : * Nope, got work to do. If we just want to pre-initialize as much
2022 : : * as we can without flushing, give up now.
2023 : : */
2024 [ + + ]: 2046956 : if (opportunistic)
2025 : 3288 : break;
2026 : :
2027 : : /* Advance shared memory write request position */
4105 andres@anarazel.de 2028 [ + + ]: 2043668 : SpinLockAcquire(&XLogCtl->info_lck);
2029 [ + + ]: 2043668 : if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
2030 : 509243 : XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
2031 : 2043668 : SpinLockRelease(&XLogCtl->info_lck);
2032 : :
2033 : : /*
2034 : : * Acquire an up-to-date LogwrtResult value and see if we still
2035 : : * need to write it or if someone else already did.
2036 : : */
622 alvherre@alvh.no-ip. 2037 : 2043668 : RefreshXLogWriteResult(LogwrtResult);
4546 heikki.linnakangas@i 2038 [ + + ]: 2043668 : if (LogwrtResult.Write < OldPageRqstPtr)
2039 : : {
2040 : : /*
2041 : : * Must acquire write lock. Release WALBufMappingLock first,
2042 : : * to make sure that all insertions that we need to wait for
2043 : : * can finish (up to this same position). Otherwise we risk
2044 : : * deadlock.
2045 : : */
118 akorotkov@postgresql 2046 : 2028836 : LWLockRelease(WALBufMappingLock);
2047 : :
4546 heikki.linnakangas@i 2048 : 2028836 : WaitXLogInsertionsToFinish(OldPageRqstPtr);
2049 : :
2050 : 2028836 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
2051 : :
624 alvherre@alvh.no-ip. 2052 : 2028836 : RefreshXLogWriteResult(LogwrtResult);
4546 heikki.linnakangas@i 2053 [ + + ]: 2028836 : if (LogwrtResult.Write >= OldPageRqstPtr)
2054 : : {
2055 : : /* OK, someone wrote it already */
2056 : 129388 : LWLockRelease(WALWriteLock);
2057 : : }
2058 : : else
2059 : : {
2060 : : /* Have to write it ourselves */
2061 : : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_START();
2062 : 1899448 : WriteRqst.Write = OldPageRqstPtr;
2063 : 1899448 : WriteRqst.Flush = 0;
1504 rhaas@postgresql.org 2064 : 1899448 : XLogWrite(WriteRqst, tli, false);
4546 heikki.linnakangas@i 2065 : 1899448 : LWLockRelease(WALWriteLock);
304 michael@paquier.xyz 2066 : 1899448 : pgWalUsage.wal_buffers_full++;
2067 : : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_DONE();
2068 : :
2069 : : /*
2070 : : * Required for the flush of pending stats WAL data, per
2071 : : * update of pgWalUsage.
2072 : : */
143 2073 : 1899448 : pgstat_report_fixed = true;
2074 : : }
2075 : : /* Re-acquire WALBufMappingLock and retry */
118 akorotkov@postgresql 2076 : 2028836 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2077 : 2028836 : continue;
2078 : : }
2079 : : }
2080 : :
2081 : : /*
2082 : : * Now the next buffer slot is free and we can set it up to be the
2083 : : * next output page.
2084 : : */
2085 : 2411132 : NewPageBeginPtr = XLogCtl->InitializedUpTo;
4546 heikki.linnakangas@i 2086 : 2411132 : NewPageEndPtr = NewPageBeginPtr + XLOG_BLCKSZ;
2087 : :
118 akorotkov@postgresql 2088 [ - + ]: 2411132 : Assert(XLogRecPtrToBufIdx(NewPageBeginPtr) == nextidx);
2089 : :
4546 heikki.linnakangas@i 2090 : 2411132 : NewPage = (XLogPageHeader) (XLogCtl->pages + nextidx * (Size) XLOG_BLCKSZ);
2091 : :
2092 : : /*
2093 : : * Mark the xlblock with InvalidXLogRecPtr and issue a write barrier
2094 : : * before initializing. Otherwise, the old page may be partially
2095 : : * zeroed but look valid.
2096 : : */
730 jdavis@postgresql.or 2097 : 2411132 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], InvalidXLogRecPtr);
2098 : 2411132 : pg_write_barrier();
2099 : :
2100 : : /*
2101 : : * Be sure to re-zero the buffer so that bytes beyond what we've
2102 : : * written will look like zeroes and not valid XLOG records...
2103 : : */
309 peter@eisentraut.org 2104 [ + - + - : 2411132 : MemSet(NewPage, 0, XLOG_BLCKSZ);
+ - - + -
- ]
2105 : :
2106 : : /*
2107 : : * Fill the new page's header
2108 : : */
3862 bruce@momjian.us 2109 : 2411132 : NewPage->xlp_magic = XLOG_PAGE_MAGIC;
2110 : :
2111 : : /* NewPage->xlp_info = 0; */ /* done by memset */
1504 rhaas@postgresql.org 2112 : 2411132 : NewPage->xlp_tli = tli;
3862 bruce@momjian.us 2113 : 2411132 : NewPage->xlp_pageaddr = NewPageBeginPtr;
2114 : :
2115 : : /* NewPage->xlp_rem_len = 0; */ /* done by memset */
2116 : :
2117 : : /*
2118 : : * If online backup is not in progress, mark the header to indicate
2119 : : * that WAL records beginning in this page have removable backup
2120 : : * blocks. This allows the WAL archiver to know whether it is safe to
2121 : : * compress archived WAL data by transforming full-block records into
2122 : : * the non-full-block format. It is sufficient to record this at the
2123 : : * page level because we force a page switch (in fact a segment
2124 : : * switch) when starting a backup, so the flag will be off before any
2125 : : * records can be written during the backup. At the end of a backup,
2126 : : * the last page will be marked as all unsafe when perhaps only part
2127 : : * is unsafe, but at worst the archiver would miss the opportunity to
2128 : : * compress a few records.
2129 : : */
1156 alvherre@alvh.no-ip. 2130 [ + + ]: 2411132 : if (Insert->runningBackups == 0)
3862 bruce@momjian.us 2131 : 2283405 : NewPage->xlp_info |= XLP_BKP_REMOVABLE;
2132 : :
2133 : : /*
2134 : : * If first page of an XLOG segment file, make it a long header.
2135 : : */
3012 andres@anarazel.de 2136 [ + + ]: 2411132 : if ((XLogSegmentOffset(NewPage->xlp_pageaddr, wal_segment_size)) == 0)
2137 : : {
4546 heikki.linnakangas@i 2138 : 1784 : XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
2139 : :
2140 : 1784 : NewLongPage->xlp_sysid = ControlFile->system_identifier;
3012 andres@anarazel.de 2141 : 1784 : NewLongPage->xlp_seg_size = wal_segment_size;
4546 heikki.linnakangas@i 2142 : 1784 : NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
3862 bruce@momjian.us 2143 : 1784 : NewPage->xlp_info |= XLP_LONG_HEADER;
2144 : : }
2145 : :
2146 : : /*
2147 : : * Make sure the initialization of the page becomes visible to others
2148 : : * before the xlblocks update. GetXLogBuffer() reads xlblocks without
2149 : : * holding a lock.
2150 : : */
4546 heikki.linnakangas@i 2151 : 2411132 : pg_write_barrier();
2152 : :
730 jdavis@postgresql.or 2153 : 2411132 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], NewPageEndPtr);
118 akorotkov@postgresql 2154 : 2411132 : XLogCtl->InitializedUpTo = NewPageEndPtr;
2155 : :
4546 heikki.linnakangas@i 2156 : 2411132 : npages++;
2157 : : }
118 akorotkov@postgresql 2158 : 2287601 : LWLockRelease(WALBufMappingLock);
2159 : :
2160 : : #ifdef WAL_DEBUG
2161 : : if (XLOG_DEBUG && npages > 0)
2162 : : {
2163 : : elog(DEBUG1, "initialized %d pages, up to %X/%08X",
2164 : : npages, LSN_FORMAT_ARGS(NewPageEndPtr));
2165 : : }
2166 : : #endif
9579 vadim4o@yahoo.com 2167 : 2287601 : }
2168 : :
2169 : : /*
2170 : : * Calculate CheckPointSegments based on max_wal_size_mb and
2171 : : * checkpoint_completion_target.
2172 : : */
2173 : : static void
3951 heikki.linnakangas@i 2174 : 7791 : CalculateCheckpointSegments(void)
2175 : : {
2176 : : double target;
2177 : :
2178 : : /*-------
2179 : : * Calculate the distance at which to trigger a checkpoint, to avoid
2180 : : * exceeding max_wal_size_mb. This is based on two assumptions:
2181 : : *
2182 : : * a) we keep WAL for only one checkpoint cycle (prior to PG11 we kept
2183 : : * WAL for two checkpoint cycles to allow us to recover from the
2184 : : * secondary checkpoint if the first checkpoint failed, though we
2185 : : * only did this on the primary anyway, not on standby. Keeping just
2186 : : * one checkpoint simplifies processing and reduces disk space in
2187 : : * many smaller databases.)
2188 : : * b) during checkpoint, we consume checkpoint_completion_target *
2189 : : * number of segments consumed between checkpoints.
2190 : : *-------
2191 : : */
3012 andres@anarazel.de 2192 : 7791 : target = (double) ConvertToXSegs(max_wal_size_mb, wal_segment_size) /
2963 simon@2ndQuadrant.co 2193 : 7791 : (1.0 + CheckPointCompletionTarget);
2194 : :
2195 : : /* round down */
3951 heikki.linnakangas@i 2196 : 7791 : CheckPointSegments = (int) target;
2197 : :
2198 [ + + ]: 7791 : if (CheckPointSegments < 1)
2199 : 10 : CheckPointSegments = 1;
2200 : 7791 : }
2201 : :
2202 : : void
2203 : 5705 : assign_max_wal_size(int newval, void *extra)
2204 : : {
3180 simon@2ndQuadrant.co 2205 : 5705 : max_wal_size_mb = newval;
3951 heikki.linnakangas@i 2206 : 5705 : CalculateCheckpointSegments();
2207 : 5705 : }
2208 : :
2209 : : void
2210 : 1109 : assign_checkpoint_completion_target(double newval, void *extra)
2211 : : {
2212 : 1109 : CheckPointCompletionTarget = newval;
2213 : 1109 : CalculateCheckpointSegments();
2214 : 1109 : }
2215 : :
2216 : : bool
843 peter@eisentraut.org 2217 : 2138 : check_wal_segment_size(int *newval, void **extra, GucSource source)
2218 : : {
2219 [ + - + - : 2138 : if (!IsValidWalSegSize(*newval))
+ - - + ]
2220 : : {
843 peter@eisentraut.org 2221 :UBC 0 : GUC_check_errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.");
2222 : 0 : return false;
2223 : : }
2224 : :
843 peter@eisentraut.org 2225 :CBC 2138 : return true;
2226 : : }
2227 : :
2228 : : /*
2229 : : * At a checkpoint, how many WAL segments to recycle as preallocated future
2230 : : * XLOG segments? Returns the highest segment that should be preallocated.
2231 : : */
2232 : : static XLogSegNo
2192 michael@paquier.xyz 2233 : 1732 : XLOGfileslop(XLogRecPtr lastredoptr)
2234 : : {
2235 : : XLogSegNo minSegNo;
2236 : : XLogSegNo maxSegNo;
2237 : : double distance;
2238 : : XLogSegNo recycleSegNo;
2239 : :
2240 : : /*
2241 : : * Calculate the segment numbers that min_wal_size_mb and max_wal_size_mb
2242 : : * correspond to. Always recycle enough segments to meet the minimum, and
2243 : : * remove enough segments to stay below the maximum.
2244 : : */
2245 : 1732 : minSegNo = lastredoptr / wal_segment_size +
3012 andres@anarazel.de 2246 : 1732 : ConvertToXSegs(min_wal_size_mb, wal_segment_size) - 1;
2192 michael@paquier.xyz 2247 : 1732 : maxSegNo = lastredoptr / wal_segment_size +
3012 andres@anarazel.de 2248 : 1732 : ConvertToXSegs(max_wal_size_mb, wal_segment_size) - 1;
2249 : :
2250 : : /*
2251 : : * Between those limits, recycle enough segments to get us through to the
2252 : : * estimated end of next checkpoint.
2253 : : *
2254 : : * To estimate where the next checkpoint will finish, assume that the
2255 : : * system runs steadily consuming CheckPointDistanceEstimate bytes between
2256 : : * every checkpoint.
2257 : : */
2963 simon@2ndQuadrant.co 2258 : 1732 : distance = (1.0 + CheckPointCompletionTarget) * CheckPointDistanceEstimate;
2259 : : /* add 10% for good measure. */
3951 heikki.linnakangas@i 2260 : 1732 : distance *= 1.10;
2261 : :
2192 michael@paquier.xyz 2262 : 1732 : recycleSegNo = (XLogSegNo) ceil(((double) lastredoptr + distance) /
2263 : : wal_segment_size);
2264 : :
3951 heikki.linnakangas@i 2265 [ + + ]: 1732 : if (recycleSegNo < minSegNo)
2266 : 1214 : recycleSegNo = minSegNo;
2267 [ + + ]: 1732 : if (recycleSegNo > maxSegNo)
2268 : 398 : recycleSegNo = maxSegNo;
2269 : :
2270 : 1732 : return recycleSegNo;
2271 : : }
2272 : :
2273 : : /*
2274 : : * Check whether we've consumed enough xlog space that a checkpoint is needed.
2275 : : *
2276 : : * new_segno indicates a log file that has just been filled up (or read
2277 : : * during recovery). We measure the distance from RedoRecPtr to new_segno
2278 : : * and see if that exceeds CheckPointSegments.
2279 : : *
2280 : : * Note: it is caller's responsibility that RedoRecPtr is up-to-date.
2281 : : */
2282 : : bool
4925 2283 : 4687 : XLogCheckpointNeeded(XLogSegNo new_segno)
2284 : : {
2285 : : XLogSegNo old_segno;
2286 : :
3012 andres@anarazel.de 2287 : 4687 : XLByteToSeg(RedoRecPtr, old_segno, wal_segment_size);
2288 : :
4925 heikki.linnakangas@i 2289 [ + + ]: 4687 : if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
6642 tgl@sss.pgh.pa.us 2290 : 2945 : return true;
2291 : 1742 : return false;
2292 : : }
2293 : :
2294 : : /*
2295 : : * Write and/or fsync the log at least as far as WriteRqst indicates.
2296 : : *
2297 : : * If flexible == true, we don't have to write as far as WriteRqst, but
2298 : : * may stop at any convenient boundary (such as a cache or logfile boundary).
2299 : : * This option allows us to avoid uselessly issuing multiple writes when a
2300 : : * single one would do.
2301 : : *
2302 : : * Must be called with WALWriteLock held. WaitXLogInsertionsToFinish(WriteRqst)
2303 : : * must be called before grabbing the lock, to make sure the data is ready to
2304 : : * write.
2305 : : */
2306 : : static void
1504 rhaas@postgresql.org 2307 : 2032723 : XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
2308 : : {
2309 : : bool ispartialpage;
2310 : : bool last_iteration;
2311 : : bool finishing_seg;
2312 : : int curridx;
2313 : : int npages;
2314 : : int startidx;
2315 : : uint32 startoffset;
2316 : :
2317 : : /* We should always be inside a critical section here */
7552 tgl@sss.pgh.pa.us 2318 [ - + ]: 2032723 : Assert(CritSectionCount > 0);
2319 : :
2320 : : /*
2321 : : * Update local LogwrtResult (caller probably did this already, but...)
2322 : : */
624 alvherre@alvh.no-ip. 2323 : 2032723 : RefreshXLogWriteResult(LogwrtResult);
2324 : :
2325 : : /*
2326 : : * Since successive pages in the xlog cache are consecutively allocated,
2327 : : * we can usually gather multiple pages together and issue just one
2328 : : * write() call. npages is the number of pages we have determined can be
2329 : : * written together; startidx is the cache block index of the first one,
2330 : : * and startoffset is the file offset at which it should go. The latter
2331 : : * two variables are only valid when npages > 0, but we must initialize
2332 : : * all of them to keep the compiler quiet.
2333 : : */
7423 tgl@sss.pgh.pa.us 2334 : 2032723 : npages = 0;
2335 : 2032723 : startidx = 0;
2336 : 2032723 : startoffset = 0;
2337 : :
2338 : : /*
2339 : : * Within the loop, curridx is the cache block index of the page to
2340 : : * consider writing. Begin at the buffer containing the next unwritten
2341 : : * page, or last partially written page.
2342 : : */
4537 heikki.linnakangas@i 2343 : 2032723 : curridx = XLogRecPtrToBufIdx(LogwrtResult.Write);
2344 : :
4738 alvherre@alvh.no-ip. 2345 [ + + ]: 4384654 : while (LogwrtResult.Write < WriteRqst.Write)
2346 : : {
2347 : : /*
2348 : : * Make sure we're not ahead of the insert process. This could happen
2349 : : * if we're passed a bogus WriteRqst.Write that is past the end of the
2350 : : * last page that's been initialized by AdvanceXLInsertBuffer.
2351 : : */
730 jdavis@postgresql.or 2352 : 2479551 : XLogRecPtr EndPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[curridx]);
2353 : :
4546 heikki.linnakangas@i 2354 [ - + ]: 2479551 : if (LogwrtResult.Write >= EndPtr)
164 alvherre@kurilemu.de 2355 [ # # ]:UNC 0 : elog(PANIC, "xlog write request %X/%08X is past end of log %X/%08X",
2356 : : LSN_FORMAT_ARGS(LogwrtResult.Write),
2357 : : LSN_FORMAT_ARGS(EndPtr));
2358 : :
2359 : : /* Advance LogwrtResult.Write to end of current buffer page */
4546 heikki.linnakangas@i 2360 :CBC 2479551 : LogwrtResult.Write = EndPtr;
4738 alvherre@alvh.no-ip. 2361 : 2479551 : ispartialpage = WriteRqst.Write < LogwrtResult.Write;
2362 : :
3012 andres@anarazel.de 2363 [ + + ]: 2479551 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2364 : : wal_segment_size))
2365 : : {
2366 : : /*
2367 : : * Switch to new logfile segment. We cannot have any pending
2368 : : * pages here (since we dump what we have at segment end).
2369 : : */
7423 tgl@sss.pgh.pa.us 2370 [ - + ]: 13252 : Assert(npages == 0);
9046 2371 [ + + ]: 13252 : if (openLogFile >= 0)
7126 bruce@momjian.us 2372 : 6359 : XLogFileClose();
3012 andres@anarazel.de 2373 : 13252 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2374 : : wal_segment_size);
1504 rhaas@postgresql.org 2375 : 13252 : openLogTLI = tli;
2376 : :
2377 : : /* create/use new log file */
2378 : 13252 : openLogFile = XLogFileInit(openLogSegNo, tli);
2124 tgl@sss.pgh.pa.us 2379 : 13252 : ReserveExternalFD();
2380 : : }
2381 : :
2382 : : /* Make sure we have the current logfile open */
9046 2383 [ - + ]: 2479551 : if (openLogFile < 0)
2384 : : {
3012 andres@anarazel.de 2385 :UBC 0 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2386 : : wal_segment_size);
1504 rhaas@postgresql.org 2387 : 0 : openLogTLI = tli;
2388 : 0 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2124 tgl@sss.pgh.pa.us 2389 : 0 : ReserveExternalFD();
2390 : : }
2391 : :
2392 : : /* Add current page to the set of pending pages-to-dump */
7423 tgl@sss.pgh.pa.us 2393 [ + + ]:CBC 2479551 : if (npages == 0)
2394 : : {
2395 : : /* first of group */
2396 : 2044859 : startidx = curridx;
3012 andres@anarazel.de 2397 : 2044859 : startoffset = XLogSegmentOffset(LogwrtResult.Write - XLOG_BLCKSZ,
2398 : : wal_segment_size);
2399 : : }
7423 tgl@sss.pgh.pa.us 2400 : 2479551 : npages++;
2401 : :
2402 : : /*
2403 : : * Dump the set if this will be the last loop iteration, or if we are
2404 : : * at the last page of the cache area (since the next page won't be
2405 : : * contiguous in memory), or if we are at the end of the logfile
2406 : : * segment.
2407 : : */
4738 alvherre@alvh.no-ip. 2408 : 2479551 : last_iteration = WriteRqst.Write <= LogwrtResult.Write;
2409 : :
7423 tgl@sss.pgh.pa.us 2410 [ + + ]: 4833914 : finishing_seg = !ispartialpage &&
3012 andres@anarazel.de 2411 [ + + ]: 2354363 : (startoffset + npages * XLOG_BLCKSZ) >= wal_segment_size;
2412 : :
7074 tgl@sss.pgh.pa.us 2413 [ + + ]: 2479551 : if (last_iteration ||
7423 2414 [ + + - + ]: 449454 : curridx == XLogCtl->XLogCacheBlck ||
2415 : : finishing_seg)
2416 : : {
2417 : : char *from;
2418 : : Size nbytes;
2419 : : Size nleft;
2420 : : ssize_t written;
2421 : : instr_time start;
2422 : :
2423 : : /* OK to write the page(s) */
7199 2424 : 2044859 : from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
2425 : 2044859 : nbytes = npages * (Size) XLOG_BLCKSZ;
4553 heikki.linnakangas@i 2426 : 2044859 : nleft = nbytes;
2427 : : do
2428 : : {
2429 : 2044859 : errno = 0;
2430 : :
2431 : : /*
2432 : : * Measure I/O timing to write WAL data, for pg_stat_io.
2433 : : */
295 michael@paquier.xyz 2434 : 2044859 : start = pgstat_prepare_io_time(track_wal_io_timing);
2435 : :
3197 rhaas@postgresql.org 2436 : 2044859 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
1176 tmunro@postgresql.or 2437 : 2044859 : written = pg_pwrite(openLogFile, from, nleft, startoffset);
3197 rhaas@postgresql.org 2438 : 2044859 : pgstat_report_wait_end();
2439 : :
317 michael@paquier.xyz 2440 : 2044859 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL,
2441 : : IOOP_WRITE, start, 1, written);
2442 : :
4553 heikki.linnakangas@i 2443 [ - + ]: 2044859 : if (written <= 0)
2444 : : {
2445 : : char xlogfname[MAXFNAMELEN];
2446 : : int save_errno;
2447 : :
4553 heikki.linnakangas@i 2448 [ # # ]:UBC 0 : if (errno == EINTR)
2449 : 0 : continue;
2450 : :
2207 michael@paquier.xyz 2451 : 0 : save_errno = errno;
1504 rhaas@postgresql.org 2452 : 0 : XLogFileName(xlogfname, tli, openLogSegNo,
2453 : : wal_segment_size);
2207 michael@paquier.xyz 2454 : 0 : errno = save_errno;
4553 heikki.linnakangas@i 2455 [ # # ]: 0 : ereport(PANIC,
2456 : : (errcode_for_file_access(),
2457 : : errmsg("could not write to log file \"%s\" at offset %u, length %zu: %m",
2458 : : xlogfname, startoffset, nleft)));
2459 : : }
4553 heikki.linnakangas@i 2460 :CBC 2044859 : nleft -= written;
2461 : 2044859 : from += written;
2598 tmunro@postgresql.or 2462 : 2044859 : startoffset += written;
4553 heikki.linnakangas@i 2463 [ - + ]: 2044859 : } while (nleft > 0);
2464 : :
7423 tgl@sss.pgh.pa.us 2465 : 2044859 : npages = 0;
2466 : :
2467 : : /*
2468 : : * If we just wrote the whole last page of a logfile segment,
2469 : : * fsync the segment immediately. This avoids having to go back
2470 : : * and re-open prior segments when an fsync request comes along
2471 : : * later. Doing it here ensures that one and only one backend will
2472 : : * perform this fsync.
2473 : : *
2474 : : * This is also the right place to notify the Archiver that the
2475 : : * segment is ready to copy to archival storage, and to update the
2476 : : * timer for archive_timeout, and to signal for a checkpoint if
2477 : : * too many logfile segments have been used since the last
2478 : : * checkpoint.
2479 : : */
4546 heikki.linnakangas@i 2480 [ + + ]: 2044859 : if (finishing_seg)
2481 : : {
1504 rhaas@postgresql.org 2482 : 1871 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2483 : :
2484 : : /* signal that we need to wakeup walsenders later */
4917 2485 : 1871 : WalSndWakeupRequest();
2486 : :
3102 tgl@sss.pgh.pa.us 2487 : 1871 : LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
2488 : :
7423 2489 [ + + - + : 1871 : if (XLogArchivingActive())
+ + ]
1504 rhaas@postgresql.org 2490 : 401 : XLogArchiveNotifySeg(openLogSegNo, tli);
2491 : :
4537 heikki.linnakangas@i 2492 : 1871 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
3283 andres@anarazel.de 2493 : 1871 : XLogCtl->lastSegSwitchLSN = LogwrtResult.Flush;
2494 : :
2495 : : /*
2496 : : * Request a checkpoint if we've consumed too much xlog since
2497 : : * the last one. For speed, we first check using the local
2498 : : * copy of RedoRecPtr, which might be out of date; if it looks
2499 : : * like a checkpoint is needed, forcibly update RedoRecPtr and
2500 : : * recheck.
2501 : : */
4925 heikki.linnakangas@i 2502 [ + + + + ]: 1871 : if (IsUnderPostmaster && XLogCheckpointNeeded(openLogSegNo))
2503 : : {
6642 tgl@sss.pgh.pa.us 2504 : 247 : (void) GetRedoRecPtr();
4925 heikki.linnakangas@i 2505 [ + + ]: 247 : if (XLogCheckpointNeeded(openLogSegNo))
6746 tgl@sss.pgh.pa.us 2506 : 197 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
2507 : : }
2508 : : }
2509 : : }
2510 : :
9046 2511 [ + + ]: 2479551 : if (ispartialpage)
2512 : : {
2513 : : /* Only asked to write a partial page */
2514 : 125188 : LogwrtResult.Write = WriteRqst.Write;
2515 : 125188 : break;
2516 : : }
7423 2517 [ + + ]: 2354363 : curridx = NextBufIdx(curridx);
2518 : :
2519 : : /* If flexible, break out of loop as soon as we wrote something */
2520 [ + + + + ]: 2354363 : if (flexible && npages == 0)
2521 : 2432 : break;
2522 : : }
2523 : :
2524 [ - + ]: 2032723 : Assert(npages == 0);
2525 : :
2526 : : /*
2527 : : * If asked to flush, do so
2528 : : */
4738 alvherre@alvh.no-ip. 2529 [ + + ]: 2032723 : if (LogwrtResult.Flush < WriteRqst.Flush &&
2530 [ + + ]: 132610 : LogwrtResult.Flush < LogwrtResult.Write)
2531 : : {
2532 : : /*
2533 : : * Could get here without iterating above loop, in which case we might
2534 : : * have no open file or the wrong one. However, we do not need to
2535 : : * fsync more than one file.
2536 : : */
797 nathan@postgresql.or 2537 [ + - ]: 132539 : if (wal_sync_method != WAL_SYNC_METHOD_OPEN &&
2538 [ + - ]: 132539 : wal_sync_method != WAL_SYNC_METHOD_OPEN_DSYNC)
2539 : : {
9043 tgl@sss.pgh.pa.us 2540 [ + + ]: 132539 : if (openLogFile >= 0 &&
3012 andres@anarazel.de 2541 [ + + ]: 132523 : !XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2542 : : wal_segment_size))
7126 bruce@momjian.us 2543 : 158 : XLogFileClose();
9043 tgl@sss.pgh.pa.us 2544 [ + + ]: 132539 : if (openLogFile < 0)
2545 : : {
3012 andres@anarazel.de 2546 : 174 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2547 : : wal_segment_size);
1504 rhaas@postgresql.org 2548 : 174 : openLogTLI = tli;
2549 : 174 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2124 tgl@sss.pgh.pa.us 2550 : 174 : ReserveExternalFD();
2551 : : }
2552 : :
1504 rhaas@postgresql.org 2553 : 132539 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2554 : : }
2555 : :
2556 : : /* signal that we need to wakeup walsenders later */
4917 2557 : 132539 : WalSndWakeupRequest();
2558 : :
9046 tgl@sss.pgh.pa.us 2559 : 132539 : LogwrtResult.Flush = LogwrtResult.Write;
2560 : : }
2561 : :
2562 : : /*
2563 : : * Update shared-memory status
2564 : : *
2565 : : * We make sure that the shared 'request' values do not fall behind the
2566 : : * 'result' values. This is not absolutely essential, but it saves some
2567 : : * code in a couple of places.
2568 : : */
622 alvherre@alvh.no-ip. 2569 [ + + ]: 2032723 : SpinLockAcquire(&XLogCtl->info_lck);
2570 [ + + ]: 2032723 : if (XLogCtl->LogwrtRqst.Write < LogwrtResult.Write)
2571 : 117345 : XLogCtl->LogwrtRqst.Write = LogwrtResult.Write;
2572 [ + + ]: 2032723 : if (XLogCtl->LogwrtRqst.Flush < LogwrtResult.Flush)
2573 : 134000 : XLogCtl->LogwrtRqst.Flush = LogwrtResult.Flush;
2574 : 2032723 : SpinLockRelease(&XLogCtl->info_lck);
2575 : :
2576 : : /*
2577 : : * We write Write first, bar, then Flush. When reading, the opposite must
2578 : : * be done (with a matching barrier in between), so that we always see a
2579 : : * Flush value that trails behind the Write value seen.
2580 : : */
2581 : 2032723 : pg_atomic_write_u64(&XLogCtl->logWriteResult, LogwrtResult.Write);
2582 : 2032723 : pg_write_barrier();
2583 : 2032723 : pg_atomic_write_u64(&XLogCtl->logFlushResult, LogwrtResult.Flush);
2584 : :
2585 : : #ifdef USE_ASSERT_CHECKING
2586 : : {
2587 : : XLogRecPtr Flush;
2588 : : XLogRecPtr Write;
2589 : : XLogRecPtr Insert;
2590 : :
2591 : 2032723 : Flush = pg_atomic_read_u64(&XLogCtl->logFlushResult);
2592 : 2032723 : pg_read_barrier();
2593 : 2032723 : Write = pg_atomic_read_u64(&XLogCtl->logWriteResult);
620 2594 : 2032723 : pg_read_barrier();
2595 : 2032723 : Insert = pg_atomic_read_u64(&XLogCtl->logInsertResult);
2596 : :
2597 : : /* WAL written to disk is always ahead of WAL flushed */
622 2598 [ - + ]: 2032723 : Assert(Write >= Flush);
2599 : :
2600 : : /* WAL inserted to buffers is always ahead of WAL written */
620 2601 [ - + ]: 2032723 : Assert(Insert >= Write);
2602 : : }
2603 : : #endif
9046 tgl@sss.pgh.pa.us 2604 : 2032723 : }
2605 : :
2606 : : /*
2607 : : * Record the LSN for an asynchronous transaction commit/abort
2608 : : * and nudge the WALWriter if there is work for it to do.
2609 : : * (This should not be called for synchronous commits.)
2610 : : */
2611 : : void
5621 simon@2ndQuadrant.co 2612 : 30585 : XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN)
2613 : : {
5149 2614 : 30585 : XLogRecPtr WriteRqstPtr = asyncXactLSN;
2615 : : bool sleeping;
752 heikki.linnakangas@i 2616 : 30585 : bool wakeup = false;
2617 : : XLogRecPtr prevAsyncXactLSN;
2618 : :
4105 andres@anarazel.de 2619 [ + + ]: 30585 : SpinLockAcquire(&XLogCtl->info_lck);
2620 : 30585 : sleeping = XLogCtl->WalWriterSleeping;
752 heikki.linnakangas@i 2621 : 30585 : prevAsyncXactLSN = XLogCtl->asyncXactLSN;
4105 andres@anarazel.de 2622 [ + + ]: 30585 : if (XLogCtl->asyncXactLSN < asyncXactLSN)
2623 : 30111 : XLogCtl->asyncXactLSN = asyncXactLSN;
2624 : 30585 : SpinLockRelease(&XLogCtl->info_lck);
2625 : :
2626 : : /*
2627 : : * If somebody else already called this function with a more aggressive
2628 : : * LSN, they will have done what we needed (and perhaps more).
2629 : : */
752 heikki.linnakangas@i 2630 [ + + ]: 30585 : if (asyncXactLSN <= prevAsyncXactLSN)
2631 : 474 : return;
2632 : :
2633 : : /*
2634 : : * If the WALWriter is sleeping, kick it to make it come out of low-power
2635 : : * mode, so that this async commit will reach disk within the expected
2636 : : * amount of time. Otherwise, determine whether it has enough WAL
2637 : : * available to flush, the same way that XLogBackgroundFlush() does.
2638 : : */
2639 [ + + ]: 30111 : if (sleeping)
2640 : 14 : wakeup = true;
2641 : : else
2642 : : {
2643 : : int flushblocks;
2644 : :
622 alvherre@alvh.no-ip. 2645 : 30097 : RefreshXLogWriteResult(LogwrtResult);
2646 : :
752 heikki.linnakangas@i 2647 : 30097 : flushblocks =
2648 : 30097 : WriteRqstPtr / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
2649 : :
2650 [ + - + + ]: 30097 : if (WalWriterFlushAfter == 0 || flushblocks >= WalWriterFlushAfter)
2651 : 3776 : wakeup = true;
2652 : : }
2653 : :
412 2654 [ + + ]: 30111 : if (wakeup)
2655 : : {
2656 : 3790 : volatile PROC_HDR *procglobal = ProcGlobal;
2657 : 3790 : ProcNumber walwriterProc = procglobal->walwriterProc;
2658 : :
2659 [ + + ]: 3790 : if (walwriterProc != INVALID_PROC_NUMBER)
2660 : 178 : SetLatch(&GetPGProcByNumber(walwriterProc)->procLatch);
2661 : : }
2662 : : }
2663 : :
2664 : : /*
2665 : : * Record the LSN up to which we can remove WAL because it's not required by
2666 : : * any replication slot.
2667 : : */
2668 : : void
4339 rhaas@postgresql.org 2669 : 39567 : XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn)
2670 : : {
4105 andres@anarazel.de 2671 [ + + ]: 39567 : SpinLockAcquire(&XLogCtl->info_lck);
2672 : 39567 : XLogCtl->replicationSlotMinLSN = lsn;
2673 : 39567 : SpinLockRelease(&XLogCtl->info_lck);
4339 rhaas@postgresql.org 2674 : 39567 : }
2675 : :
2676 : :
2677 : : /*
2678 : : * Return the oldest LSN we must retain to satisfy the needs of some
2679 : : * replication slot.
2680 : : */
2681 : : static XLogRecPtr
2682 : 2104 : XLogGetReplicationSlotMinimumLSN(void)
2683 : : {
2684 : : XLogRecPtr retval;
2685 : :
4105 andres@anarazel.de 2686 [ - + ]: 2104 : SpinLockAcquire(&XLogCtl->info_lck);
2687 : 2104 : retval = XLogCtl->replicationSlotMinLSN;
2688 : 2104 : SpinLockRelease(&XLogCtl->info_lck);
2689 : :
4339 rhaas@postgresql.org 2690 : 2104 : return retval;
2691 : : }
2692 : :
2693 : : /*
2694 : : * Advance minRecoveryPoint in control file.
2695 : : *
2696 : : * If we crash during recovery, we must reach this point again before the
2697 : : * database is consistent.
2698 : : *
2699 : : * If 'force' is true, 'lsn' argument is ignored. Otherwise, minRecoveryPoint
2700 : : * is only updated if it's not already greater than or equal to 'lsn'.
2701 : : */
2702 : : static void
6147 heikki.linnakangas@i 2703 : 107477 : UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
2704 : : {
2705 : : /* Quick check using our local copy of the variable */
1401 2706 [ + + + + : 107477 : if (!updateMinRecoveryPoint || (!force && lsn <= LocalMinRecoveryPoint))
+ + ]
6147 2707 : 100383 : return;
2708 : :
2709 : : /*
2710 : : * An invalid minRecoveryPoint means that we need to recover all the WAL,
2711 : : * i.e., we're doing crash recovery. We never modify the control file's
2712 : : * value in that case, so we can short-circuit future checks here too. The
2713 : : * local values of minRecoveryPoint and minRecoveryPointTLI should not be
2714 : : * updated until crash recovery finishes. We only do this for the startup
2715 : : * process as it should not update its own reference of minRecoveryPoint
2716 : : * until it has finished crash recovery to make sure that all WAL
2717 : : * available is replayed in this case. This also saves from extra locks
2718 : : * taken on the control file from the startup process.
2719 : : */
42 alvherre@kurilemu.de 2720 [ + + + + ]:GNC 7094 : if (!XLogRecPtrIsValid(LocalMinRecoveryPoint) && InRecovery)
2721 : : {
2723 michael@paquier.xyz 2722 :CBC 31 : updateMinRecoveryPoint = false;
2723 : 31 : return;
2724 : : }
2725 : :
6147 heikki.linnakangas@i 2726 : 7063 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2727 : :
2728 : : /* update local copy */
1401 2729 : 7063 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
2730 : 7063 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
2731 : :
42 alvherre@kurilemu.de 2732 [ + + ]:GNC 7063 : if (!XLogRecPtrIsValid(LocalMinRecoveryPoint))
2666 michael@paquier.xyz 2733 :CBC 3 : updateMinRecoveryPoint = false;
1401 heikki.linnakangas@i 2734 [ + + + + ]: 7060 : else if (force || LocalMinRecoveryPoint < lsn)
2735 : : {
2736 : : XLogRecPtr newMinRecoveryPoint;
2737 : : TimeLineID newMinRecoveryPointTLI;
2738 : :
2739 : : /*
2740 : : * To avoid having to update the control file too often, we update it
2741 : : * all the way to the last record being replayed, even though 'lsn'
2742 : : * would suffice for correctness. This also allows the 'force' case
2743 : : * to not need a valid 'lsn' value.
2744 : : *
2745 : : * Another important reason for doing it this way is that the passed
2746 : : * 'lsn' value could be bogus, i.e., past the end of available WAL, if
2747 : : * the caller got it from a corrupted heap page. Accepting such a
2748 : : * value as the min recovery point would prevent us from coming up at
2749 : : * all. Instead, we just log a warning and continue with recovery.
2750 : : * (See also the comments about corrupt LSNs in XLogFlush.)
2751 : : */
2752 : 5632 : newMinRecoveryPoint = GetCurrentReplayRecPtr(&newMinRecoveryPointTLI);
4738 alvherre@alvh.no-ip. 2753 [ + + - + ]: 5632 : if (!force && newMinRecoveryPoint < lsn)
6019 tgl@sss.pgh.pa.us 2754 [ # # ]:UBC 0 : elog(WARNING,
2755 : : "xlog min recovery request %X/%08X is past current point %X/%08X",
2756 : : LSN_FORMAT_ARGS(lsn), LSN_FORMAT_ARGS(newMinRecoveryPoint));
2757 : :
2758 : : /* update control file */
4738 alvherre@alvh.no-ip. 2759 [ + + ]:CBC 5632 : if (ControlFile->minRecoveryPoint < newMinRecoveryPoint)
2760 : : {
6147 heikki.linnakangas@i 2761 : 5282 : ControlFile->minRecoveryPoint = newMinRecoveryPoint;
4762 2762 : 5282 : ControlFile->minRecoveryPointTLI = newMinRecoveryPointTLI;
6147 2763 : 5282 : UpdateControlFile();
1401 2764 : 5282 : LocalMinRecoveryPoint = newMinRecoveryPoint;
2765 : 5282 : LocalMinRecoveryPointTLI = newMinRecoveryPointTLI;
2766 : :
6147 2767 [ + + ]: 5282 : ereport(DEBUG2,
2768 : : errmsg_internal("updated min recovery point to %X/%08X on timeline %u",
2769 : : LSN_FORMAT_ARGS(newMinRecoveryPoint),
2770 : : newMinRecoveryPointTLI));
2771 : : }
2772 : : }
2773 : 7063 : LWLockRelease(ControlFileLock);
2774 : : }
2775 : :
2776 : : /*
2777 : : * Ensure that all XLOG data through the given position is flushed to disk.
2778 : : *
2779 : : * NOTE: this differs from XLogWrite mainly in that the WALWriteLock is not
2780 : : * already held, and we try to avoid acquiring it if possible.
2781 : : */
2782 : : void
9046 tgl@sss.pgh.pa.us 2783 : 705372 : XLogFlush(XLogRecPtr record)
2784 : : {
2785 : : XLogRecPtr WriteRqstPtr;
2786 : : XLogwrtRqst WriteRqst;
1499 rhaas@postgresql.org 2787 : 705372 : TimeLineID insertTLI = XLogCtl->InsertTimeLineID;
2788 : :
2789 : : /*
2790 : : * During REDO, we are reading not writing WAL. Therefore, instead of
2791 : : * trying to flush the WAL, we should update minRecoveryPoint instead. We
2792 : : * test XLogInsertAllowed(), not InRecovery, because we need checkpointer
2793 : : * to act this way too, and because when it tries to write the
2794 : : * end-of-recovery checkpoint, it should indeed flush.
2795 : : */
6019 tgl@sss.pgh.pa.us 2796 [ + + ]: 705372 : if (!XLogInsertAllowed())
2797 : : {
6147 heikki.linnakangas@i 2798 : 107027 : UpdateMinRecoveryPoint(record, false);
9046 tgl@sss.pgh.pa.us 2799 : 560727 : return;
2800 : : }
2801 : :
2802 : : /* Quick exit if already known flushed */
4738 alvherre@alvh.no-ip. 2803 [ + + ]: 598345 : if (record <= LogwrtResult.Flush)
9046 tgl@sss.pgh.pa.us 2804 : 453700 : return;
2805 : :
2806 : : #ifdef WAL_DEBUG
2807 : : if (XLOG_DEBUG)
2808 : : elog(LOG, "xlog flush request %X/%08X; write %X/%08X; flush %X/%08X",
2809 : : LSN_FORMAT_ARGS(record),
2810 : : LSN_FORMAT_ARGS(LogwrtResult.Write),
2811 : : LSN_FORMAT_ARGS(LogwrtResult.Flush));
2812 : : #endif
2813 : :
2814 : 144645 : START_CRIT_SECTION();
2815 : :
2816 : : /*
2817 : : * Since fsync is usually a horribly expensive operation, we try to
2818 : : * piggyback as much data as we can on each fsync: if we see any more data
2819 : : * entered into the xlog buffer, we'll write and fsync that too, so that
2820 : : * the final value of LogwrtResult.Flush is as large as possible. This
2821 : : * gives us some chance of avoiding another fsync immediately after.
2822 : : */
2823 : :
2824 : : /* initialize to given target; may increase below */
2825 : 144645 : WriteRqstPtr = record;
2826 : :
2827 : : /*
2828 : : * Now wait until we get the write lock, or someone else does the flush
2829 : : * for us.
2830 : : */
2831 : : for (;;)
8756 2832 : 2413 : {
2833 : : XLogRecPtr insertpos;
2834 : :
2835 : : /* done already? */
622 alvherre@alvh.no-ip. 2836 : 147058 : RefreshXLogWriteResult(LogwrtResult);
4738 2837 [ + + ]: 147058 : if (record <= LogwrtResult.Flush)
5071 heikki.linnakangas@i 2838 : 10867 : break;
2839 : :
2840 : : /*
2841 : : * Before actually performing the write, wait for all in-flight
2842 : : * insertions to the pages we're about to write to finish.
2843 : : */
622 alvherre@alvh.no-ip. 2844 [ + + ]: 136191 : SpinLockAcquire(&XLogCtl->info_lck);
2845 [ + + ]: 136191 : if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2846 : 9601 : WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2847 : 136191 : SpinLockRelease(&XLogCtl->info_lck);
4546 heikki.linnakangas@i 2848 : 136191 : insertpos = WaitXLogInsertionsToFinish(WriteRqstPtr);
2849 : :
2850 : : /*
2851 : : * Try to get the write lock. If we can't get it immediately, wait
2852 : : * until it's released, and recheck if we still need to do the flush
2853 : : * or if the backend that held the lock did it for us already. This
2854 : : * helps to maintain a good rate of group committing when the system
2855 : : * is bottlenecked by the speed of fsyncing.
2856 : : */
5062 2857 [ + + ]: 136191 : if (!LWLockAcquireOrWait(WALWriteLock, LW_EXCLUSIVE))
2858 : : {
2859 : : /*
2860 : : * The lock is now free, but we didn't acquire it yet. Before we
2861 : : * do, loop back to check if someone else flushed the record for
2862 : : * us already.
2863 : : */
5071 2864 : 2413 : continue;
2865 : : }
2866 : :
2867 : : /* Got the lock; recheck whether request is satisfied */
624 alvherre@alvh.no-ip. 2868 : 133778 : RefreshXLogWriteResult(LogwrtResult);
4738 2869 [ + + ]: 133778 : if (record <= LogwrtResult.Flush)
2870 : : {
4917 rhaas@postgresql.org 2871 : 3723 : LWLockRelease(WALWriteLock);
2872 : 3723 : break;
2873 : : }
2874 : :
2875 : : /*
2876 : : * Sleep before flush! By adding a delay here, we may give further
2877 : : * backends the opportunity to join the backlog of group commit
2878 : : * followers; this can significantly improve transaction throughput,
2879 : : * at the risk of increasing transaction latency.
2880 : : *
2881 : : * We do not sleep if enableFsync is not turned on, nor if there are
2882 : : * fewer than CommitSiblings other backends with active transactions.
2883 : : */
2884 [ - + - - : 130055 : if (CommitDelay > 0 && enableFsync &&
- - ]
4917 rhaas@postgresql.org 2885 :UBC 0 : MinimumActiveBackends(CommitSiblings))
2886 : : {
9 heikki.linnakangas@i 2887 :UNC 0 : pgstat_report_wait_start(WAIT_EVENT_COMMIT_DELAY);
4917 rhaas@postgresql.org 2888 :UBC 0 : pg_usleep(CommitDelay);
9 heikki.linnakangas@i 2889 :UNC 0 : pgstat_report_wait_end();
2890 : :
2891 : : /*
2892 : : * Re-check how far we can now flush the WAL. It's generally not
2893 : : * safe to call WaitXLogInsertionsToFinish while holding
2894 : : * WALWriteLock, because an in-progress insertion might need to
2895 : : * also grab WALWriteLock to make progress. But we know that all
2896 : : * the insertions up to insertpos have already finished, because
2897 : : * that's what the earlier WaitXLogInsertionsToFinish() returned.
2898 : : * We're only calling it again to allow insertpos to be moved
2899 : : * further forward, not to actually wait for anyone.
2900 : : */
4546 heikki.linnakangas@i 2901 :UBC 0 : insertpos = WaitXLogInsertionsToFinish(insertpos);
2902 : : }
2903 : :
2904 : : /* try to write/flush later additions to XLOG as well */
4546 heikki.linnakangas@i 2905 :CBC 130055 : WriteRqst.Write = insertpos;
2906 : 130055 : WriteRqst.Flush = insertpos;
2907 : :
1504 rhaas@postgresql.org 2908 : 130055 : XLogWrite(WriteRqst, insertTLI, false);
2909 : :
8846 tgl@sss.pgh.pa.us 2910 : 130055 : LWLockRelease(WALWriteLock);
2911 : : /* done */
5071 heikki.linnakangas@i 2912 : 130055 : break;
2913 : : }
2914 : :
9046 tgl@sss.pgh.pa.us 2915 [ - + ]: 144645 : END_CRIT_SECTION();
2916 : :
2917 : : /* wake up walsenders now that we've released heavily contended locks */
985 andres@anarazel.de 2918 : 144645 : WalSndWakeupProcessRequests(true, !RecoveryInProgress());
2919 : :
2920 : : /*
2921 : : * If we still haven't flushed to the request point then we have a
2922 : : * problem; most likely, the requested flush point is past end of XLOG.
2923 : : * This has been seen to occur when a disk page has a corrupted LSN.
2924 : : *
2925 : : * Formerly we treated this as a PANIC condition, but that hurts the
2926 : : * system's robustness rather than helping it: we do not want to take down
2927 : : * the whole system due to corruption on one data page. In particular, if
2928 : : * the bad page is encountered again during recovery then we would be
2929 : : * unable to restart the database at all! (This scenario actually
2930 : : * happened in the field several times with 7.1 releases.) As of 8.4, bad
2931 : : * LSNs encountered during recovery are UpdateMinRecoveryPoint's problem;
2932 : : * the only time we can reach here during recovery is while flushing the
2933 : : * end-of-recovery checkpoint record, and we don't expect that to have a
2934 : : * bad LSN.
2935 : : *
2936 : : * Note that for calls from xact.c, the ERROR will be promoted to PANIC
2937 : : * since xact.c calls this routine inside a critical section. However,
2938 : : * calls from bufmgr.c are not within critical sections and so we will not
2939 : : * force a restart for a bad LSN on a data page.
2940 : : */
4738 alvherre@alvh.no-ip. 2941 [ - + ]: 144645 : if (LogwrtResult.Flush < record)
6019 tgl@sss.pgh.pa.us 2942 [ # # ]:UBC 0 : elog(ERROR,
2943 : : "xlog flush request %X/%08X is not satisfied --- flushed only to %X/%08X",
2944 : : LSN_FORMAT_ARGS(record),
2945 : : LSN_FORMAT_ARGS(LogwrtResult.Flush));
2946 : :
2947 : : /*
2948 : : * Cross-check XLogNeedsFlush(). Some of the checks of XLogFlush() and
2949 : : * XLogNeedsFlush() are duplicated, and this assertion ensures that these
2950 : : * remain consistent.
2951 : : */
90 michael@paquier.xyz 2952 [ - + ]:GNC 144645 : Assert(!XLogNeedsFlush(record));
2953 : : }
2954 : :
2955 : : /*
2956 : : * Write & flush xlog, but without specifying exactly where to.
2957 : : *
2958 : : * We normally write only completed blocks; but if there is nothing to do on
2959 : : * that basis, we check for unwritten async commits in the current incomplete
2960 : : * block, and write through the latest one of those. Thus, if async commits
2961 : : * are not being used, we will write complete blocks only.
2962 : : *
2963 : : * If, based on the above, there's anything to write we do so immediately. But
2964 : : * to avoid calling fsync, fdatasync et. al. at a rate that'd impact
2965 : : * concurrent IO, we only flush WAL every wal_writer_delay ms, or if there's
2966 : : * more than wal_writer_flush_after unflushed blocks.
2967 : : *
2968 : : * We can guarantee that async commits reach disk after at most three
2969 : : * wal_writer_delay cycles. (When flushing complete blocks, we allow XLogWrite
2970 : : * to write "flexibly", meaning it can stop at the end of the buffer ring;
2971 : : * this makes a difference only with very high load or long wal_writer_delay,
2972 : : * but imposes one extra cycle for the worst case for async commits.)
2973 : : *
2974 : : * This routine is invoked periodically by the background walwriter process.
2975 : : *
2976 : : * Returns true if there was any work to do, even if we skipped flushing due
2977 : : * to wal_writer_delay/wal_writer_flush_after.
2978 : : */
2979 : : bool
6722 tgl@sss.pgh.pa.us 2980 :CBC 10811 : XLogBackgroundFlush(void)
2981 : : {
2982 : : XLogwrtRqst WriteRqst;
2983 : 10811 : bool flexible = true;
2984 : : static TimestampTz lastflush;
2985 : : TimestampTz now;
2986 : : int flushblocks;
2987 : : TimeLineID insertTLI;
2988 : :
2989 : : /* XLOG doesn't need flushing during recovery */
6147 heikki.linnakangas@i 2990 [ - + ]: 10811 : if (RecoveryInProgress())
4972 tgl@sss.pgh.pa.us 2991 :UBC 0 : return false;
2992 : :
2993 : : /*
2994 : : * Since we're not in recovery, InsertTimeLineID is set and can't change,
2995 : : * so we can read it without a lock.
2996 : : */
1499 rhaas@postgresql.org 2997 :CBC 10811 : insertTLI = XLogCtl->InsertTimeLineID;
2998 : :
2999 : : /* read updated LogwrtRqst */
4105 andres@anarazel.de 3000 [ - + ]: 10811 : SpinLockAcquire(&XLogCtl->info_lck);
3594 3001 : 10811 : WriteRqst = XLogCtl->LogwrtRqst;
4105 3002 : 10811 : SpinLockRelease(&XLogCtl->info_lck);
3003 : :
3004 : : /* back off to last completed page boundary */
3594 3005 : 10811 : WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
3006 : :
3007 : : /* if we have already flushed that far, consider async commit records */
622 alvherre@alvh.no-ip. 3008 : 10811 : RefreshXLogWriteResult(LogwrtResult);
3594 andres@anarazel.de 3009 [ + + ]: 10811 : if (WriteRqst.Write <= LogwrtResult.Flush)
3010 : : {
4105 3011 [ - + ]: 8209 : SpinLockAcquire(&XLogCtl->info_lck);
3594 3012 : 8209 : WriteRqst.Write = XLogCtl->asyncXactLSN;
4105 3013 : 8209 : SpinLockRelease(&XLogCtl->info_lck);
6722 tgl@sss.pgh.pa.us 3014 : 8209 : flexible = false; /* ensure it all gets written */
3015 : : }
3016 : :
3017 : : /*
3018 : : * If already known flushed, we're done. Just need to check if we are
3019 : : * holding an open file handle to a logfile that's no longer in use,
3020 : : * preventing the file from being deleted.
3021 : : */
3594 andres@anarazel.de 3022 [ + + ]: 10811 : if (WriteRqst.Write <= LogwrtResult.Flush)
3023 : : {
5644 bruce@momjian.us 3024 [ + + ]: 7523 : if (openLogFile >= 0)
3025 : : {
3012 andres@anarazel.de 3026 [ + + ]: 4827 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
3027 : : wal_segment_size))
3028 : : {
5671 magnus@hagander.net 3029 : 185 : XLogFileClose();
3030 : : }
3031 : : }
4972 tgl@sss.pgh.pa.us 3032 : 7523 : return false;
3033 : : }
3034 : :
3035 : : /*
3036 : : * Determine how far to flush WAL, based on the wal_writer_delay and
3037 : : * wal_writer_flush_after GUCs.
3038 : : *
3039 : : * Note that XLogSetAsyncXactLSN() performs similar calculation based on
3040 : : * wal_writer_flush_after, to decide when to wake us up. Make sure the
3041 : : * logic is the same in both places if you change this.
3042 : : */
3594 andres@anarazel.de 3043 : 3288 : now = GetCurrentTimestamp();
752 heikki.linnakangas@i 3044 : 3288 : flushblocks =
3594 andres@anarazel.de 3045 : 3288 : WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
3046 : :
3047 [ + - + + ]: 3288 : if (WalWriterFlushAfter == 0 || lastflush == 0)
3048 : : {
3049 : : /* first call, or block based limits disabled */
3050 : 269 : WriteRqst.Flush = WriteRqst.Write;
3051 : 269 : lastflush = now;
3052 : : }
3053 [ + + ]: 3019 : else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
3054 : : {
3055 : : /*
3056 : : * Flush the writes at least every WalWriterDelay ms. This is
3057 : : * important to bound the amount of time it takes for an asynchronous
3058 : : * commit to hit disk.
3059 : : */
3060 : 2878 : WriteRqst.Flush = WriteRqst.Write;
3061 : 2878 : lastflush = now;
3062 : : }
752 heikki.linnakangas@i 3063 [ + + ]: 141 : else if (flushblocks >= WalWriterFlushAfter)
3064 : : {
3065 : : /* exceeded wal_writer_flush_after blocks, flush */
3594 andres@anarazel.de 3066 : 129 : WriteRqst.Flush = WriteRqst.Write;
3067 : 129 : lastflush = now;
3068 : : }
3069 : : else
3070 : : {
3071 : : /* no flushing, this time round */
3072 : 12 : WriteRqst.Flush = 0;
3073 : : }
3074 : :
3075 : : #ifdef WAL_DEBUG
3076 : : if (XLOG_DEBUG)
3077 : : elog(LOG, "xlog bg flush request write %X/%08X; flush: %X/%08X, current is write %X/%08X; flush %X/%08X",
3078 : : LSN_FORMAT_ARGS(WriteRqst.Write),
3079 : : LSN_FORMAT_ARGS(WriteRqst.Flush),
3080 : : LSN_FORMAT_ARGS(LogwrtResult.Write),
3081 : : LSN_FORMAT_ARGS(LogwrtResult.Flush));
3082 : : #endif
3083 : :
6722 tgl@sss.pgh.pa.us 3084 : 3288 : START_CRIT_SECTION();
3085 : :
3086 : : /* now wait for any in-progress insertions to finish and get write lock */
3594 andres@anarazel.de 3087 : 3288 : WaitXLogInsertionsToFinish(WriteRqst.Write);
6722 tgl@sss.pgh.pa.us 3088 : 3288 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
624 alvherre@alvh.no-ip. 3089 : 3288 : RefreshXLogWriteResult(LogwrtResult);
3594 andres@anarazel.de 3090 [ + + ]: 3288 : if (WriteRqst.Write > LogwrtResult.Write ||
3091 [ + + ]: 181 : WriteRqst.Flush > LogwrtResult.Flush)
3092 : : {
1504 rhaas@postgresql.org 3093 : 3220 : XLogWrite(WriteRqst, insertTLI, flexible);
3094 : : }
6722 tgl@sss.pgh.pa.us 3095 : 3288 : LWLockRelease(WALWriteLock);
3096 : :
3097 [ - + ]: 3288 : END_CRIT_SECTION();
3098 : :
3099 : : /* wake up walsenders now that we've released heavily contended locks */
985 andres@anarazel.de 3100 : 3288 : WalSndWakeupProcessRequests(true, !RecoveryInProgress());
3101 : :
3102 : : /*
3103 : : * Great, done. To take some work off the critical path, try to initialize
3104 : : * as many of the no-longer-needed WAL buffers for future use as we can.
3105 : : */
1504 rhaas@postgresql.org 3106 : 3288 : AdvanceXLInsertBuffer(InvalidXLogRecPtr, insertTLI, true);
3107 : :
3108 : : /*
3109 : : * If we determined that we need to write data, but somebody else
3110 : : * wrote/flushed already, it should be considered as being active, to
3111 : : * avoid hibernating too early.
3112 : : */
3594 andres@anarazel.de 3113 : 3288 : return true;
3114 : : }
3115 : :
3116 : : /*
3117 : : * Test whether XLOG data has been flushed up to (at least) the given
3118 : : * position, or whether the minimum recovery point has been updated past
3119 : : * the given position.
3120 : : *
3121 : : * Returns true if a flush is still needed, or if the minimum recovery point
3122 : : * must be updated.
3123 : : *
3124 : : * It is possible that someone else is already in the process of flushing
3125 : : * that far, or has updated the minimum recovery point up to the given
3126 : : * position.
3127 : : */
3128 : : bool
6777 tgl@sss.pgh.pa.us 3129 : 8869379 : XLogNeedsFlush(XLogRecPtr record)
3130 : : {
3131 : : /*
3132 : : * During recovery, we don't flush WAL but update minRecoveryPoint
3133 : : * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3134 : : * would need to be updated.
3135 : : *
3136 : : * Using XLogInsertAllowed() rather than RecoveryInProgress() matters for
3137 : : * the case of an end-of-recovery checkpoint, where WAL data is flushed.
3138 : : * This check should be consistent with the one in XLogFlush().
3139 : : */
90 michael@paquier.xyz 3140 [ + + ]:GNC 8869379 : if (!XLogInsertAllowed())
3141 : : {
3142 : : /* Quick exit if already known to be updated or cannot be updated */
79 3143 [ + - + + ]: 619319 : if (!updateMinRecoveryPoint || record <= LocalMinRecoveryPoint)
3144 : 610499 : return false;
3145 : :
3146 : : /*
3147 : : * An invalid minRecoveryPoint means that we need to recover all the
3148 : : * WAL, i.e., we're doing crash recovery. We never modify the control
3149 : : * file's value in that case, so we can short-circuit future checks
3150 : : * here too. This triggers a quick exit path for the startup process,
3151 : : * which cannot update its local copy of minRecoveryPoint as long as
3152 : : * it has not replayed all WAL available when doing crash recovery.
3153 : : */
42 alvherre@kurilemu.de 3154 [ + + - + ]: 8820 : if (!XLogRecPtrIsValid(LocalMinRecoveryPoint) && InRecovery)
3155 : : {
2723 michael@paquier.xyz 3156 :UBC 0 : updateMinRecoveryPoint = false;
5843 simon@2ndQuadrant.co 3157 :LBC (609784) : return false;
3158 : : }
3159 : :
3160 : : /*
3161 : : * Update local copy of minRecoveryPoint. But if the lock is busy,
3162 : : * just return a conservative guess.
3163 : : */
5843 simon@2ndQuadrant.co 3164 [ - + ]:CBC 8820 : if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
5843 simon@2ndQuadrant.co 3165 :UBC 0 : return true;
1401 heikki.linnakangas@i 3166 :CBC 8820 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
3167 : 8820 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
5843 simon@2ndQuadrant.co 3168 : 8820 : LWLockRelease(ControlFileLock);
3169 : :
3170 : : /*
3171 : : * Check minRecoveryPoint for any other process than the startup
3172 : : * process doing crash recovery, which should not update the control
3173 : : * file value if crash recovery is still running.
3174 : : */
42 alvherre@kurilemu.de 3175 [ - + ]:GNC 8820 : if (!XLogRecPtrIsValid(LocalMinRecoveryPoint))
2666 michael@paquier.xyz 3176 :UBC 0 : updateMinRecoveryPoint = false;
3177 : :
3178 : : /* check again */
1401 heikki.linnakangas@i 3179 [ + + - + ]:CBC 8820 : if (record <= LocalMinRecoveryPoint || !updateMinRecoveryPoint)
2666 michael@paquier.xyz 3180 : 73 : return false;
3181 : : else
3182 : 8747 : return true;
3183 : : }
3184 : :
3185 : : /* Quick exit if already known flushed */
4738 alvherre@alvh.no-ip. 3186 [ + + ]: 8250060 : if (record <= LogwrtResult.Flush)
6777 tgl@sss.pgh.pa.us 3187 : 8056587 : return false;
3188 : :
3189 : : /* read LogwrtResult and update local state */
624 alvherre@alvh.no-ip. 3190 : 193473 : RefreshXLogWriteResult(LogwrtResult);
3191 : :
3192 : : /* check again */
4738 3193 [ + + ]: 193473 : if (record <= LogwrtResult.Flush)
6777 tgl@sss.pgh.pa.us 3194 : 4069 : return false;
3195 : :
3196 : 189404 : return true;
3197 : : }
3198 : :
3199 : : /*
3200 : : * Try to make a given XLOG file segment exist.
3201 : : *
3202 : : * logsegno: identify segment.
3203 : : *
3204 : : * *added: on return, true if this call raised the number of extant segments.
3205 : : *
3206 : : * path: on return, this char[MAXPGPATH] has the path to the logsegno file.
3207 : : *
3208 : : * Returns -1 or FD of opened file. A -1 here is not an error; a caller
3209 : : * wanting an open segment should attempt to open "path", which usually will
3210 : : * succeed. (This is weird, but it's efficient for the callers.)
3211 : : */
3212 : : static int
1504 rhaas@postgresql.org 3213 : 14382 : XLogFileInitInternal(XLogSegNo logsegno, TimeLineID logtli,
3214 : : bool *added, char *path)
3215 : : {
3216 : : char tmppath[MAXPGPATH];
3217 : : XLogSegNo installed_segno;
3218 : : XLogSegNo max_segno;
3219 : : int fd;
3220 : : int save_errno;
985 tmunro@postgresql.or 3221 : 14382 : int open_flags = O_RDWR | O_CREAT | O_EXCL | PG_BINARY;
3222 : : instr_time io_start;
3223 : :
1504 rhaas@postgresql.org 3224 [ - + ]: 14382 : Assert(logtli != 0);
3225 : :
3226 : 14382 : XLogFilePath(path, logtli, logsegno, wal_segment_size);
3227 : :
3228 : : /*
3229 : : * Try to use existent file (checkpoint maker may have created it already)
3230 : : */
1634 noah@leadboat.com 3231 : 14382 : *added = false;
1021 tmunro@postgresql.or 3232 : 14382 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
797 nathan@postgresql.or 3233 : 14382 : get_sync_bit(wal_sync_method));
1634 noah@leadboat.com 3234 [ + + ]: 14382 : if (fd < 0)
3235 : : {
3236 [ - + ]: 1375 : if (errno != ENOENT)
1634 noah@leadboat.com 3237 [ # # ]:UBC 0 : ereport(ERROR,
3238 : : (errcode_for_file_access(),
3239 : : errmsg("could not open file \"%s\": %m", path)));
3240 : : }
3241 : : else
1634 noah@leadboat.com 3242 :CBC 13007 : return fd;
3243 : :
3244 : : /*
3245 : : * Initialize an empty (all zeroes) segment. NOTE: it is possible that
3246 : : * another process is doing the same thing. If so, we will end up
3247 : : * pre-creating an extra log segment. That seems OK, and better than
3248 : : * holding the lock throughout this lengthy process.
3249 : : */
6746 tgl@sss.pgh.pa.us 3250 [ + + ]: 1375 : elog(DEBUG2, "creating and filling new WAL file");
3251 : :
7472 3252 : 1375 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3253 : :
9042 3254 : 1375 : unlink(tmppath);
3255 : :
985 tmunro@postgresql.or 3256 [ - + ]: 1375 : if (io_direct_flags & IO_DIRECT_WAL_INIT)
985 tmunro@postgresql.or 3257 :UBC 0 : open_flags |= PG_O_DIRECT;
3258 : :
3259 : : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
985 tmunro@postgresql.or 3260 :CBC 1375 : fd = BasicOpenFile(tmppath, open_flags);
9579 vadim4o@yahoo.com 3261 [ - + ]: 1375 : if (fd < 0)
7552 tgl@sss.pgh.pa.us 3262 [ # # ]:UBC 0 : ereport(ERROR,
3263 : : (errcode_for_file_access(),
3264 : : errmsg("could not create file \"%s\": %m", tmppath)));
3265 : :
3266 : : /* Measure I/O timing when initializing segment */
295 michael@paquier.xyz 3267 :CBC 1375 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3268 : :
2452 tmunro@postgresql.or 3269 : 1375 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_WRITE);
3270 : 1375 : save_errno = 0;
3271 [ + - ]: 1375 : if (wal_init_zero)
3272 : : {
3273 : : ssize_t rc;
3274 : :
3275 : : /*
3276 : : * Zero-fill the file. With this setting, we do this the hard way to
3277 : : * ensure that all the file space has really been allocated. On
3278 : : * platforms that allow "holes" in files, just seeking to the end
3279 : : * doesn't allocate intermediate space. This way, we know that we
3280 : : * have all the space and (after the fsync below) that all the
3281 : : * indirect blocks are down on disk. Therefore, fdatasync(2) or
3282 : : * O_DSYNC will be sufficient to sync future writes to the log file.
3283 : : */
1018 michael@paquier.xyz 3284 : 1375 : rc = pg_pwrite_zeros(fd, wal_segment_size, 0);
3285 : :
1136 3286 [ - + ]: 1375 : if (rc < 0)
1136 michael@paquier.xyz 3287 :UBC 0 : save_errno = errno;
3288 : : }
3289 : : else
3290 : : {
3291 : : /*
3292 : : * Otherwise, seeking to the end and writing a solitary byte is
3293 : : * enough.
3294 : : */
4488 jdavis@postgresql.or 3295 : 0 : errno = 0;
1136 michael@paquier.xyz 3296 [ # # ]: 0 : if (pg_pwrite(fd, "\0", 1, wal_segment_size - 1) != 1)
3297 : : {
3298 : : /* if write didn't set errno, assume no disk space */
2452 tmunro@postgresql.or 3299 [ # # ]: 0 : save_errno = errno ? errno : ENOSPC;
3300 : : }
3301 : : }
2452 tmunro@postgresql.or 3302 :CBC 1375 : pgstat_report_wait_end();
3303 : :
3304 : : /*
3305 : : * A full segment worth of data is written when using wal_init_zero. One
3306 : : * byte is written when not using it.
3307 : : */
317 michael@paquier.xyz 3308 : 1375 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_INIT, IOOP_WRITE,
3309 : : io_start, 1,
3310 [ + - ]: 1375 : wal_init_zero ? wal_segment_size : 1);
3311 : :
2452 tmunro@postgresql.or 3312 [ - + ]: 1375 : if (save_errno)
3313 : : {
3314 : : /*
3315 : : * If we fail to make the file, delete it to release disk space
3316 : : */
2452 tmunro@postgresql.or 3317 :UBC 0 : unlink(tmppath);
3318 : :
3319 : 0 : close(fd);
3320 : :
3321 : 0 : errno = save_errno;
3322 : :
3323 [ # # ]: 0 : ereport(ERROR,
3324 : : (errcode_for_file_access(),
3325 : : errmsg("could not write to file \"%s\": %m", tmppath)));
3326 : : }
3327 : :
3328 : : /* Measure I/O timing when flushing segment */
295 michael@paquier.xyz 3329 :CBC 1375 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3330 : :
3197 rhaas@postgresql.org 3331 : 1375 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_SYNC);
9141 tgl@sss.pgh.pa.us 3332 [ - + ]: 1375 : if (pg_fsync(fd) != 0)
3333 : : {
1210 drowley@postgresql.o 3334 :UBC 0 : save_errno = errno;
4769 heikki.linnakangas@i 3335 : 0 : close(fd);
2733 michael@paquier.xyz 3336 : 0 : errno = save_errno;
7552 tgl@sss.pgh.pa.us 3337 [ # # ]: 0 : ereport(ERROR,
3338 : : (errcode_for_file_access(),
3339 : : errmsg("could not fsync file \"%s\": %m", tmppath)));
3340 : : }
3197 rhaas@postgresql.org 3341 :CBC 1375 : pgstat_report_wait_end();
3342 : :
317 michael@paquier.xyz 3343 : 1375 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_INIT,
3344 : : IOOP_FSYNC, io_start, 1, 0);
3345 : :
2357 peter@eisentraut.org 3346 [ - + ]: 1375 : if (close(fd) != 0)
7552 tgl@sss.pgh.pa.us 3347 [ # # ]:UBC 0 : ereport(ERROR,
3348 : : (errcode_for_file_access(),
3349 : : errmsg("could not close file \"%s\": %m", tmppath)));
3350 : :
3351 : : /*
3352 : : * Now move the segment into place with its final name. Cope with
3353 : : * possibility that someone else has created the file while we were
3354 : : * filling ours: if so, use ours to pre-create a future log segment.
3355 : : */
4925 heikki.linnakangas@i 3356 :CBC 1375 : installed_segno = logsegno;
3357 : :
3358 : : /*
3359 : : * XXX: What should we use as max_segno? We used to use XLOGfileslop when
3360 : : * that was a constant, but that was always a bit dubious: normally, at a
3361 : : * checkpoint, XLOGfileslop was the offset from the checkpoint record, but
3362 : : * here, it was the offset from the insert location. We can't do the
3363 : : * normal XLOGfileslop calculation here because we don't have access to
3364 : : * the prior checkpoint's redo location. So somewhat arbitrarily, just use
3365 : : * CheckPointSegments.
3366 : : */
3951 3367 : 1375 : max_segno = logsegno + CheckPointSegments;
1504 rhaas@postgresql.org 3368 [ + - ]: 1375 : if (InstallXLogFileSegment(&installed_segno, tmppath, true, max_segno,
3369 : : logtli))
3370 : : {
1634 noah@leadboat.com 3371 : 1375 : *added = true;
3372 [ + + ]: 1375 : elog(DEBUG2, "done creating and filling new WAL file");
3373 : : }
3374 : : else
3375 : : {
3376 : : /*
3377 : : * No need for any more future segments, or InstallXLogFileSegment()
3378 : : * failed to rename the file into place. If the rename failed, a
3379 : : * caller opening the file may fail.
3380 : : */
8918 tgl@sss.pgh.pa.us 3381 :UBC 0 : unlink(tmppath);
1634 noah@leadboat.com 3382 [ # # ]: 0 : elog(DEBUG2, "abandoned new WAL file");
3383 : : }
3384 : :
1634 noah@leadboat.com 3385 :CBC 1375 : return -1;
3386 : : }
3387 : :
3388 : : /*
3389 : : * Create a new XLOG file segment, or open a pre-existing one.
3390 : : *
3391 : : * logsegno: identify segment to be created/opened.
3392 : : *
3393 : : * Returns FD of opened file.
3394 : : *
3395 : : * Note: errors here are ERROR not PANIC because we might or might not be
3396 : : * inside a critical section (eg, during checkpoint there is no reason to
3397 : : * take down the system on failure). They will promote to PANIC if we are
3398 : : * in a critical section.
3399 : : */
3400 : : int
1504 rhaas@postgresql.org 3401 : 14150 : XLogFileInit(XLogSegNo logsegno, TimeLineID logtli)
3402 : : {
3403 : : bool ignore_added;
3404 : : char path[MAXPGPATH];
3405 : : int fd;
3406 : :
3407 [ - + ]: 14150 : Assert(logtli != 0);
3408 : :
3409 : 14150 : fd = XLogFileInitInternal(logsegno, logtli, &ignore_added, path);
1634 noah@leadboat.com 3410 [ + + ]: 14150 : if (fd >= 0)
3411 : 12845 : return fd;
3412 : :
3413 : : /* Now open original target segment (might not be file I just made) */
1021 tmunro@postgresql.or 3414 : 1305 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
797 nathan@postgresql.or 3415 : 1305 : get_sync_bit(wal_sync_method));
8918 tgl@sss.pgh.pa.us 3416 [ - + ]: 1305 : if (fd < 0)
7552 tgl@sss.pgh.pa.us 3417 [ # # ]:UBC 0 : ereport(ERROR,
3418 : : (errcode_for_file_access(),
3419 : : errmsg("could not open file \"%s\": %m", path)));
7281 neilc@samurai.com 3420 :CBC 1305 : return fd;
3421 : : }
3422 : :
3423 : : /*
3424 : : * Create a new XLOG file segment by copying a pre-existing one.
3425 : : *
3426 : : * destsegno: identify segment to be created.
3427 : : *
3428 : : * srcTLI, srcsegno: identify segment to be copied (could be from
3429 : : * a different timeline)
3430 : : *
3431 : : * upto: how much of the source file to copy (the rest is filled with
3432 : : * zeros)
3433 : : *
3434 : : * Currently this is only used during recovery, and so there are no locking
3435 : : * considerations. But we should be just as tense as XLogFileInit to avoid
3436 : : * emplacing a bogus file.
3437 : : */
3438 : : static void
1504 rhaas@postgresql.org 3439 : 39 : XLogFileCopy(TimeLineID destTLI, XLogSegNo destsegno,
3440 : : TimeLineID srcTLI, XLogSegNo srcsegno,
3441 : : int upto)
3442 : : {
3443 : : char path[MAXPGPATH];
3444 : : char tmppath[MAXPGPATH];
3445 : : PGAlignedXLogBlock buffer;
3446 : : int srcfd;
3447 : : int fd;
3448 : : int nbytes;
3449 : :
3450 : : /*
3451 : : * Open the source file
3452 : : */
3012 andres@anarazel.de 3453 : 39 : XLogFilePath(path, srcTLI, srcsegno, wal_segment_size);
3008 peter_e@gmx.net 3454 : 39 : srcfd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
7820 tgl@sss.pgh.pa.us 3455 [ - + ]: 39 : if (srcfd < 0)
7552 tgl@sss.pgh.pa.us 3456 [ # # ]:UBC 0 : ereport(ERROR,
3457 : : (errcode_for_file_access(),
3458 : : errmsg("could not open file \"%s\": %m", path)));
3459 : :
3460 : : /*
3461 : : * Copy into a temp file name.
3462 : : */
7472 tgl@sss.pgh.pa.us 3463 :CBC 39 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3464 : :
7820 3465 : 39 : unlink(tmppath);
3466 : :
3467 : : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3008 peter_e@gmx.net 3468 : 39 : fd = OpenTransientFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
7820 tgl@sss.pgh.pa.us 3469 [ - + ]: 39 : if (fd < 0)
7552 tgl@sss.pgh.pa.us 3470 [ # # ]:UBC 0 : ereport(ERROR,
3471 : : (errcode_for_file_access(),
3472 : : errmsg("could not create file \"%s\": %m", tmppath)));
3473 : :
3474 : : /*
3475 : : * Do the data copying.
3476 : : */
3012 andres@anarazel.de 3477 [ + + ]:CBC 79911 : for (nbytes = 0; nbytes < wal_segment_size; nbytes += sizeof(buffer))
3478 : : {
3479 : : int nread;
3480 : :
4018 heikki.linnakangas@i 3481 : 79872 : nread = upto - nbytes;
3482 : :
3483 : : /*
3484 : : * The part that is not read from the source file is filled with
3485 : : * zeros.
3486 : : */
3487 [ + + ]: 79872 : if (nread < sizeof(buffer))
2665 tgl@sss.pgh.pa.us 3488 : 39 : memset(buffer.data, 0, sizeof(buffer));
3489 : :
4018 heikki.linnakangas@i 3490 [ + + ]: 79872 : if (nread > 0)
3491 : : {
3492 : : int r;
3493 : :
3494 [ + + ]: 2198 : if (nread > sizeof(buffer))
3495 : 2159 : nread = sizeof(buffer);
3197 rhaas@postgresql.org 3496 : 2198 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_READ);
2665 tgl@sss.pgh.pa.us 3497 : 2198 : r = read(srcfd, buffer.data, nread);
2710 michael@paquier.xyz 3498 [ - + ]: 2198 : if (r != nread)
3499 : : {
2710 michael@paquier.xyz 3500 [ # # ]:UBC 0 : if (r < 0)
4018 heikki.linnakangas@i 3501 [ # # ]: 0 : ereport(ERROR,
3502 : : (errcode_for_file_access(),
3503 : : errmsg("could not read file \"%s\": %m",
3504 : : path)));
3505 : : else
3506 [ # # ]: 0 : ereport(ERROR,
3507 : : (errcode(ERRCODE_DATA_CORRUPTED),
3508 : : errmsg("could not read file \"%s\": read %d of %zu",
3509 : : path, r, (Size) nread)));
3510 : : }
3197 rhaas@postgresql.org 3511 :CBC 2198 : pgstat_report_wait_end();
3512 : : }
7820 tgl@sss.pgh.pa.us 3513 : 79872 : errno = 0;
3197 rhaas@postgresql.org 3514 : 79872 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_WRITE);
2665 tgl@sss.pgh.pa.us 3515 [ - + ]: 79872 : if ((int) write(fd, buffer.data, sizeof(buffer)) != (int) sizeof(buffer))
3516 : : {
7820 tgl@sss.pgh.pa.us 3517 :UBC 0 : int save_errno = errno;
3518 : :
3519 : : /*
3520 : : * If we fail to make the file, delete it to release disk space
3521 : : */
3522 : 0 : unlink(tmppath);
3523 : : /* if write didn't set errno, assume problem is no disk space */
3524 [ # # ]: 0 : errno = save_errno ? save_errno : ENOSPC;
3525 : :
7552 3526 [ # # ]: 0 : ereport(ERROR,
3527 : : (errcode_for_file_access(),
3528 : : errmsg("could not write to file \"%s\": %m", tmppath)));
3529 : : }
3197 rhaas@postgresql.org 3530 :CBC 79872 : pgstat_report_wait_end();
3531 : : }
3532 : :
3533 : 39 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_SYNC);
7820 tgl@sss.pgh.pa.us 3534 [ - + ]: 39 : if (pg_fsync(fd) != 0)
2586 tmunro@postgresql.or 3535 [ # # ]:UBC 0 : ereport(data_sync_elevel(ERROR),
3536 : : (errcode_for_file_access(),
3537 : : errmsg("could not fsync file \"%s\": %m", tmppath)));
3197 rhaas@postgresql.org 3538 :CBC 39 : pgstat_report_wait_end();
3539 : :
2357 peter@eisentraut.org 3540 [ - + ]: 39 : if (CloseTransientFile(fd) != 0)
7552 tgl@sss.pgh.pa.us 3541 [ # # ]:UBC 0 : ereport(ERROR,
3542 : : (errcode_for_file_access(),
3543 : : errmsg("could not close file \"%s\": %m", tmppath)));
3544 : :
2357 peter@eisentraut.org 3545 [ - + ]:CBC 39 : if (CloseTransientFile(srcfd) != 0)
2476 michael@paquier.xyz 3546 [ # # ]:UBC 0 : ereport(ERROR,
3547 : : (errcode_for_file_access(),
3548 : : errmsg("could not close file \"%s\": %m", path)));
3549 : :
3550 : : /*
3551 : : * Now move the segment into place with its final name.
3552 : : */
1504 rhaas@postgresql.org 3553 [ - + ]:CBC 39 : if (!InstallXLogFileSegment(&destsegno, tmppath, false, 0, destTLI))
3823 fujii@postgresql.org 3554 [ # # ]:UBC 0 : elog(ERROR, "InstallXLogFileSegment should not have failed");
7820 tgl@sss.pgh.pa.us 3555 :CBC 39 : }
3556 : :
3557 : : /*
3558 : : * Install a new XLOG segment file as a current or future log segment.
3559 : : *
3560 : : * This is used both to install a newly-created segment (which has a temp
3561 : : * filename while it's being created) and to recycle an old segment.
3562 : : *
3563 : : * *segno: identify segment to install as (or first possible target).
3564 : : * When find_free is true, this is modified on return to indicate the
3565 : : * actual installation location or last segment searched.
3566 : : *
3567 : : * tmppath: initial name of file to install. It will be renamed into place.
3568 : : *
3569 : : * find_free: if true, install the new segment at the first empty segno
3570 : : * number at or after the passed numbers. If false, install the new segment
3571 : : * exactly where specified, deleting any existing segment file there.
3572 : : *
3573 : : * max_segno: maximum segment number to install the new file as. Fail if no
3574 : : * free slot is found between *segno and max_segno. (Ignored when find_free
3575 : : * is false.)
3576 : : *
3577 : : * tli: The timeline on which the new segment should be installed.
3578 : : *
3579 : : * Returns true if the file was installed successfully. false indicates that
3580 : : * max_segno limit was exceeded, the startup process has disabled this
3581 : : * function for now, or an error occurred while renaming the file into place.
3582 : : */
3583 : : static bool
4925 heikki.linnakangas@i 3584 : 2979 : InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
3585 : : bool find_free, XLogSegNo max_segno, TimeLineID tli)
3586 : : {
3587 : : char path[MAXPGPATH];
3588 : : struct stat stat_buf;
3589 : :
1504 rhaas@postgresql.org 3590 [ - + ]: 2979 : Assert(tli != 0);
3591 : :
3592 : 2979 : XLogFilePath(path, tli, *segno, wal_segment_size);
3593 : :
1634 noah@leadboat.com 3594 : 2979 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
3595 [ - + ]: 2979 : if (!XLogCtl->InstallXLogFileSegmentActive)
3596 : : {
1634 noah@leadboat.com 3597 :UBC 0 : LWLockRelease(ControlFileLock);
3598 : 0 : return false;
3599 : : }
3600 : :
8918 tgl@sss.pgh.pa.us 3601 [ + + ]:CBC 2979 : if (!find_free)
3602 : : {
3603 : : /* Force installation: get rid of any pre-existing segment file */
3188 teodor@sigaev.ru 3604 : 39 : durable_unlink(path, DEBUG1);
3605 : : }
3606 : : else
3607 : : {
3608 : : /* Find a free slot to put it in */
8363 tgl@sss.pgh.pa.us 3609 [ + + ]: 4071 : while (stat(path, &stat_buf) == 0)
3610 : : {
3951 heikki.linnakangas@i 3611 [ + + ]: 1285 : if ((*segno) >= max_segno)
3612 : : {
3613 : : /* Failed to find a free slot within specified range */
1634 noah@leadboat.com 3614 : 154 : LWLockRelease(ControlFileLock);
8918 tgl@sss.pgh.pa.us 3615 : 154 : return false;
3616 : : }
4925 heikki.linnakangas@i 3617 : 1131 : (*segno)++;
1504 rhaas@postgresql.org 3618 : 1131 : XLogFilePath(path, tli, *segno, wal_segment_size);
3619 : : }
3620 : : }
3621 : :
1262 michael@paquier.xyz 3622 [ + - - + ]: 2825 : Assert(access(path, F_OK) != 0 && errno == ENOENT);
3623 [ - + ]: 2825 : if (durable_rename(tmppath, path, LOG) != 0)
3624 : : {
1634 noah@leadboat.com 3625 :UBC 0 : LWLockRelease(ControlFileLock);
3626 : : /* durable_rename already emitted log message */
5940 heikki.linnakangas@i 3627 : 0 : return false;
3628 : : }
3629 : :
1634 noah@leadboat.com 3630 :CBC 2825 : LWLockRelease(ControlFileLock);
3631 : :
8918 tgl@sss.pgh.pa.us 3632 : 2825 : return true;
3633 : : }
3634 : :
3635 : : /*
3636 : : * Open a pre-existing logfile segment for writing.
3637 : : */
3638 : : int
1504 rhaas@postgresql.org 3639 : 174 : XLogFileOpen(XLogSegNo segno, TimeLineID tli)
3640 : : {
3641 : : char path[MAXPGPATH];
3642 : : int fd;
3643 : :
3644 : 174 : XLogFilePath(path, tli, segno, wal_segment_size);
3645 : :
1021 tmunro@postgresql.or 3646 : 174 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
797 nathan@postgresql.or 3647 : 174 : get_sync_bit(wal_sync_method));
9579 vadim4o@yahoo.com 3648 [ - + ]: 174 : if (fd < 0)
8186 tgl@sss.pgh.pa.us 3649 [ # # ]:UBC 0 : ereport(PANIC,
3650 : : (errcode_for_file_access(),
3651 : : errmsg("could not open file \"%s\": %m", path)));
3652 : :
7820 tgl@sss.pgh.pa.us 3653 :CBC 174 : return fd;
3654 : : }
3655 : :
3656 : : /*
3657 : : * Close the current logfile segment for writing.
3658 : : */
3659 : : static void
7126 bruce@momjian.us 3660 : 6702 : XLogFileClose(void)
3661 : : {
3662 [ - + ]: 6702 : Assert(openLogFile >= 0);
3663 : :
3664 : : /*
3665 : : * WAL segment files will not be re-read in normal operation, so we advise
3666 : : * the OS to release any cached pages. But do not do so if WAL archiving
3667 : : * or streaming is active, because archiver and walsender process could
3668 : : * use the cache to read the WAL segment.
3669 : : */
3670 : : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
985 tmunro@postgresql.or 3671 [ + + + - ]: 6702 : if (!XLogIsNeeded() && (io_direct_flags & IO_DIRECT_WAL) == 0)
6185 tgl@sss.pgh.pa.us 3672 : 1671 : (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
3673 : : #endif
3674 : :
2357 peter@eisentraut.org 3675 [ - + ]: 6702 : if (close(openLogFile) != 0)
3676 : : {
3677 : : char xlogfname[MAXFNAMELEN];
2207 michael@paquier.xyz 3678 :UBC 0 : int save_errno = errno;
3679 : :
1504 rhaas@postgresql.org 3680 : 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo, wal_segment_size);
2207 michael@paquier.xyz 3681 : 0 : errno = save_errno;
7126 bruce@momjian.us 3682 [ # # ]: 0 : ereport(PANIC,
3683 : : (errcode_for_file_access(),
3684 : : errmsg("could not close file \"%s\": %m", xlogfname)));
3685 : : }
3686 : :
7126 bruce@momjian.us 3687 :CBC 6702 : openLogFile = -1;
2124 tgl@sss.pgh.pa.us 3688 : 6702 : ReleaseExternalFD();
7126 bruce@momjian.us 3689 : 6702 : }
3690 : :
3691 : : /*
3692 : : * Preallocate log files beyond the specified log endpoint.
3693 : : *
3694 : : * XXX this is currently extremely conservative, since it forces only one
3695 : : * future log segment to exist, and even that only if we are 75% done with
3696 : : * the current one. This is only appropriate for very low-WAL-volume systems.
3697 : : * High-volume systems will be OK once they've built up a sufficient set of
3698 : : * recycled log segments, but the startup transient is likely to include
3699 : : * a lot of segment creations by foreground processes, which is not so good.
3700 : : *
3701 : : * XLogFileInitInternal() can ereport(ERROR). All known causes indicate big
3702 : : * trouble; for example, a full filesystem is one cause. The checkpoint WAL
3703 : : * and/or ControlFile updates already completed. If a RequestCheckpoint()
3704 : : * initiated the present checkpoint and an ERROR ends this function, the
3705 : : * command that called RequestCheckpoint() fails. That's not ideal, but it's
3706 : : * not worth contorting more functions to use caller-specified elevel values.
3707 : : * (With or without RequestCheckpoint(), an ERROR forestalls some inessential
3708 : : * reporting and resource reclamation.)
3709 : : */
3710 : : static void
1504 rhaas@postgresql.org 3711 : 1987 : PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli)
3712 : : {
3713 : : XLogSegNo _logSegNo;
3714 : : int lf;
3715 : : bool added;
3716 : : char path[MAXPGPATH];
3717 : : uint64 offset;
3718 : :
1634 noah@leadboat.com 3719 [ + + ]: 1987 : if (!XLogCtl->InstallXLogFileSegmentActive)
3720 : 11 : return; /* unlocked check says no */
3721 : :
3012 andres@anarazel.de 3722 : 1976 : XLByteToPrevSeg(endptr, _logSegNo, wal_segment_size);
3723 : 1976 : offset = XLogSegmentOffset(endptr - 1, wal_segment_size);
3724 [ + + ]: 1976 : if (offset >= (uint32) (0.75 * wal_segment_size))
3725 : : {
4925 heikki.linnakangas@i 3726 : 232 : _logSegNo++;
1504 rhaas@postgresql.org 3727 : 232 : lf = XLogFileInitInternal(_logSegNo, tli, &added, path);
1634 noah@leadboat.com 3728 [ + + ]: 232 : if (lf >= 0)
3729 : 162 : close(lf);
3730 [ + + ]: 232 : if (added)
6746 tgl@sss.pgh.pa.us 3731 : 70 : CheckpointStats.ckpt_segs_added++;
3732 : : }
3733 : : }
3734 : :
3735 : : /*
3736 : : * Throws an error if the given log segment has already been removed or
3737 : : * recycled. The caller should only pass a segment that it knows to have
3738 : : * existed while the server has been running, as this function always
3739 : : * succeeds if no WAL segments have been removed since startup.
3740 : : * 'tli' is only used in the error message.
3741 : : *
3742 : : * Note: this function guarantees to keep errno unchanged on return.
3743 : : * This supports callers that use this to possibly deliver a better
3744 : : * error message about a missing file, while still being able to throw
3745 : : * a normal file-access error afterwards, if this does return.
3746 : : */
3747 : : void
4732 heikki.linnakangas@i 3748 : 122037 : CheckXLogRemoved(XLogSegNo segno, TimeLineID tli)
3749 : : {
2936 tgl@sss.pgh.pa.us 3750 : 122037 : int save_errno = errno;
3751 : : XLogSegNo lastRemovedSegNo;
3752 : :
4105 andres@anarazel.de 3753 [ + + ]: 122037 : SpinLockAcquire(&XLogCtl->info_lck);
3754 : 122037 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3755 : 122037 : SpinLockRelease(&XLogCtl->info_lck);
3756 : :
4732 heikki.linnakangas@i 3757 [ - + ]: 122037 : if (segno <= lastRemovedSegNo)
3758 : : {
3759 : : char filename[MAXFNAMELEN];
3760 : :
3012 andres@anarazel.de 3761 :UBC 0 : XLogFileName(filename, tli, segno, wal_segment_size);
2936 tgl@sss.pgh.pa.us 3762 : 0 : errno = save_errno;
4732 heikki.linnakangas@i 3763 [ # # ]: 0 : ereport(ERROR,
3764 : : (errcode_for_file_access(),
3765 : : errmsg("requested WAL segment %s has already been removed",
3766 : : filename)));
3767 : : }
2936 tgl@sss.pgh.pa.us 3768 :CBC 122037 : errno = save_errno;
5729 heikki.linnakangas@i 3769 : 122037 : }
3770 : :
3771 : : /*
3772 : : * Return the last WAL segment removed, or 0 if no segment has been removed
3773 : : * since startup.
3774 : : *
3775 : : * NB: the result can be out of date arbitrarily fast, the caller has to deal
3776 : : * with that.
3777 : : */
3778 : : XLogSegNo
4308 rhaas@postgresql.org 3779 : 963 : XLogGetLastRemovedSegno(void)
3780 : : {
3781 : : XLogSegNo lastRemovedSegNo;
3782 : :
4105 andres@anarazel.de 3783 [ - + ]: 963 : SpinLockAcquire(&XLogCtl->info_lck);
3784 : 963 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3785 : 963 : SpinLockRelease(&XLogCtl->info_lck);
3786 : :
4308 rhaas@postgresql.org 3787 : 963 : return lastRemovedSegNo;
3788 : : }
3789 : :
3790 : : /*
3791 : : * Return the oldest WAL segment on the given TLI that still exists in
3792 : : * XLOGDIR, or 0 if none.
3793 : : */
3794 : : XLogSegNo
729 3795 : 6 : XLogGetOldestSegno(TimeLineID tli)
3796 : : {
3797 : : DIR *xldir;
3798 : : struct dirent *xlde;
3799 : 6 : XLogSegNo oldest_segno = 0;
3800 : :
3801 : 6 : xldir = AllocateDir(XLOGDIR);
3802 [ + + ]: 40 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3803 : : {
3804 : : TimeLineID file_tli;
3805 : : XLogSegNo file_segno;
3806 : :
3807 : : /* Ignore files that are not XLOG segments. */
3808 [ + + ]: 34 : if (!IsXLogFileName(xlde->d_name))
3809 : 24 : continue;
3810 : :
3811 : : /* Parse filename to get TLI and segno. */
3812 : 10 : XLogFromFileName(xlde->d_name, &file_tli, &file_segno,
3813 : : wal_segment_size);
3814 : :
3815 : : /* Ignore anything that's not from the TLI of interest. */
3816 [ - + ]: 10 : if (tli != file_tli)
729 rhaas@postgresql.org 3817 :UBC 0 : continue;
3818 : :
3819 : : /* If it's the oldest so far, update oldest_segno. */
729 rhaas@postgresql.org 3820 [ + + + + ]:CBC 10 : if (oldest_segno == 0 || file_segno < oldest_segno)
3821 : 8 : oldest_segno = file_segno;
3822 : : }
3823 : :
3824 : 6 : FreeDir(xldir);
3825 : 6 : return oldest_segno;
3826 : : }
3827 : :
3828 : : /*
3829 : : * Update the last removed segno pointer in shared memory, to reflect that the
3830 : : * given XLOG file has been removed.
3831 : : */
3832 : : static void
5729 heikki.linnakangas@i 3833 : 2553 : UpdateLastRemovedPtr(char *filename)
3834 : : {
3835 : : uint32 tli;
3836 : : XLogSegNo segno;
3837 : :
3012 andres@anarazel.de 3838 : 2553 : XLogFromFileName(filename, &tli, &segno, wal_segment_size);
3839 : :
4105 3840 [ + + ]: 2553 : SpinLockAcquire(&XLogCtl->info_lck);
3841 [ + + ]: 2553 : if (segno > XLogCtl->lastRemovedSegNo)
3842 : 1136 : XLogCtl->lastRemovedSegNo = segno;
3843 : 2553 : SpinLockRelease(&XLogCtl->info_lck);
5729 heikki.linnakangas@i 3844 : 2553 : }
3845 : :
3846 : : /*
3847 : : * Remove all temporary log files in pg_wal
3848 : : *
3849 : : * This is called at the beginning of recovery after a previous crash,
3850 : : * at a point where no other processes write fresh WAL data.
3851 : : */
3852 : : static void
2715 michael@paquier.xyz 3853 : 172 : RemoveTempXlogFiles(void)
3854 : : {
3855 : : DIR *xldir;
3856 : : struct dirent *xlde;
3857 : :
3858 [ + + ]: 172 : elog(DEBUG2, "removing all temporary WAL segments");
3859 : :
3860 : 172 : xldir = AllocateDir(XLOGDIR);
3861 [ + + ]: 1111 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3862 : : {
3863 : : char path[MAXPGPATH];
3864 : :
3865 [ + - ]: 939 : if (strncmp(xlde->d_name, "xlogtemp.", 9) != 0)
3866 : 939 : continue;
3867 : :
2715 michael@paquier.xyz 3868 :UBC 0 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
3869 : 0 : unlink(path);
3870 [ # # ]: 0 : elog(DEBUG2, "removed temporary WAL segment \"%s\"", path);
3871 : : }
2715 michael@paquier.xyz 3872 :CBC 172 : FreeDir(xldir);
3873 : 172 : }
3874 : :
3875 : : /*
3876 : : * Recycle or remove all log files older or equal to passed segno.
3877 : : *
3878 : : * endptr is current (or recent) end of xlog, and lastredoptr is the
3879 : : * redo pointer of the last checkpoint. These are used to determine
3880 : : * whether we want to recycle rather than delete no-longer-wanted log files.
3881 : : *
3882 : : * insertTLI is the current timeline for XLOG insertion. Any recycled
3883 : : * segments should be reused for this timeline.
3884 : : */
3885 : : static void
1504 rhaas@postgresql.org 3886 : 1732 : RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr,
3887 : : TimeLineID insertTLI)
3888 : : {
3889 : : DIR *xldir;
3890 : : struct dirent *xlde;
3891 : : char lastoff[MAXFNAMELEN];
3892 : : XLogSegNo endlogSegNo;
3893 : : XLogSegNo recycleSegNo;
3894 : :
3895 : : /* Initialize info about where to try to recycle to */
1798 michael@paquier.xyz 3896 : 1732 : XLByteToSeg(endptr, endlogSegNo, wal_segment_size);
3897 : 1732 : recycleSegNo = XLOGfileslop(lastredoptr);
3898 : :
3899 : : /*
3900 : : * Construct a filename of the last segment to be kept. The timeline ID
3901 : : * doesn't matter, we ignore that in the comparison. (During recovery,
3902 : : * InsertTimeLineID isn't set, so we can't use that.)
3903 : : */
3012 andres@anarazel.de 3904 : 1732 : XLogFileName(lastoff, 0, segno, wal_segment_size);
3905 : :
5589 simon@2ndQuadrant.co 3906 [ + + ]: 1732 : elog(DEBUG2, "attempting to remove WAL segments older than log file %s",
3907 : : lastoff);
3908 : :
2936 tgl@sss.pgh.pa.us 3909 : 1732 : xldir = AllocateDir(XLOGDIR);
3910 : :
7472 3911 [ + + ]: 48667 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3912 : : {
3913 : : /* Ignore files that are not XLOG segments */
3877 heikki.linnakangas@i 3914 [ + + ]: 46935 : if (!IsXLogFileName(xlde->d_name) &&
3915 [ + + ]: 7369 : !IsPartialXLogFileName(xlde->d_name))
3902 3916 : 7365 : continue;
3917 : :
3918 : : /*
3919 : : * We ignore the timeline part of the XLOG segment identifiers in
3920 : : * deciding whether a segment is still needed. This ensures that we
3921 : : * won't prematurely remove a segment from a parent timeline. We could
3922 : : * probably be a little more proactive about removing segments of
3923 : : * non-parent timelines, but that would be a whole lot more
3924 : : * complicated.
3925 : : *
3926 : : * We use the alphanumeric sorting property of the filenames to decide
3927 : : * which ones are earlier than the lastoff segment.
3928 : : */
3929 [ + + ]: 39570 : if (strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
3930 : : {
4689 3931 [ + + ]: 33109 : if (XLogArchiveCheckDone(xlde->d_name))
3932 : : {
3933 : : /* Update the last removed location in shared memory first */
5729 3934 : 2553 : UpdateLastRemovedPtr(xlde->d_name);
3935 : :
1203 michael@paquier.xyz 3936 : 2553 : RemoveXlogFile(xlde, recycleSegNo, &endlogSegNo, insertTLI);
3937 : : }
3938 : : }
3939 : : }
3940 : :
3902 heikki.linnakangas@i 3941 : 1732 : FreeDir(xldir);
3942 : 1732 : }
3943 : :
3944 : : /*
3945 : : * Recycle or remove WAL files that are not part of the given timeline's
3946 : : * history.
3947 : : *
3948 : : * This is called during recovery, whenever we switch to follow a new
3949 : : * timeline, and at the end of recovery when we create a new timeline. We
3950 : : * wouldn't otherwise care about extra WAL files lying in pg_wal, but they
3951 : : * might be leftover pre-allocated or recycled WAL segments on the old timeline
3952 : : * that we haven't used yet, and contain garbage. If we just leave them in
3953 : : * pg_wal, they will eventually be archived, and we can't let that happen.
3954 : : * Files that belong to our timeline history are valid, because we have
3955 : : * successfully replayed them, but from others we can't be sure.
3956 : : *
3957 : : * 'switchpoint' is the current point in WAL where we switch to new timeline,
3958 : : * and 'newTLI' is the new timeline we switch to.
3959 : : */
3960 : : void
3961 : 59 : RemoveNonParentXlogFiles(XLogRecPtr switchpoint, TimeLineID newTLI)
3962 : : {
3963 : : DIR *xldir;
3964 : : struct dirent *xlde;
3965 : : char switchseg[MAXFNAMELEN];
3966 : : XLogSegNo endLogSegNo;
3967 : : XLogSegNo switchLogSegNo;
3968 : : XLogSegNo recycleSegNo;
3969 : :
3970 : : /*
3971 : : * Initialize info about where to begin the work. This will recycle,
3972 : : * somewhat arbitrarily, 10 future segments.
3973 : : */
1798 michael@paquier.xyz 3974 : 59 : XLByteToPrevSeg(switchpoint, switchLogSegNo, wal_segment_size);
3975 : 59 : XLByteToSeg(switchpoint, endLogSegNo, wal_segment_size);
3976 : 59 : recycleSegNo = endLogSegNo + 10;
3977 : :
3978 : : /*
3979 : : * Construct a filename of the last segment to be kept.
3980 : : */
3981 : 59 : XLogFileName(switchseg, newTLI, switchLogSegNo, wal_segment_size);
3982 : :
3902 heikki.linnakangas@i 3983 [ + + ]: 59 : elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
3984 : : switchseg);
3985 : :
2936 tgl@sss.pgh.pa.us 3986 : 59 : xldir = AllocateDir(XLOGDIR);
3987 : :
3902 heikki.linnakangas@i 3988 [ + + ]: 562 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3989 : : {
3990 : : /* Ignore files that are not XLOG segments */
3877 3991 [ + + ]: 503 : if (!IsXLogFileName(xlde->d_name))
3902 3992 : 313 : continue;
3993 : :
3994 : : /*
3995 : : * Remove files that are on a timeline older than the new one we're
3996 : : * switching to, but with a segment number >= the first segment on the
3997 : : * new timeline.
3998 : : */
3999 [ + + ]: 190 : if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
4000 [ + + ]: 123 : strcmp(xlde->d_name + 8, switchseg + 8) > 0)
4001 : : {
4002 : : /*
4003 : : * If the file has already been marked as .ready, however, don't
4004 : : * remove it yet. It should be OK to remove it - files that are
4005 : : * not part of our timeline history are not required for recovery
4006 : : * - but seems safer to let them be archived and removed later.
4007 : : */
4008 [ + - ]: 14 : if (!XLogArchiveIsReady(xlde->d_name))
1203 michael@paquier.xyz 4009 : 14 : RemoveXlogFile(xlde, recycleSegNo, &endLogSegNo, newTLI);
4010 : : }
4011 : : }
4012 : :
3902 heikki.linnakangas@i 4013 : 59 : FreeDir(xldir);
4014 : 59 : }
4015 : :
4016 : : /*
4017 : : * Recycle or remove a log file that's no longer needed.
4018 : : *
4019 : : * segment_de is the dirent structure of the segment to recycle or remove.
4020 : : * recycleSegNo is the segment number to recycle up to. endlogSegNo is
4021 : : * the segment number of the current (or recent) end of WAL.
4022 : : *
4023 : : * endlogSegNo gets incremented if the segment is recycled so as it is not
4024 : : * checked again with future callers of this function.
4025 : : *
4026 : : * insertTLI is the current timeline for XLOG insertion. Any recycled segments
4027 : : * should be used for this timeline.
4028 : : */
4029 : : static void
1203 michael@paquier.xyz 4030 : 2567 : RemoveXlogFile(const struct dirent *segment_de,
4031 : : XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo,
4032 : : TimeLineID insertTLI)
4033 : : {
4034 : : char path[MAXPGPATH];
4035 : : #ifdef WIN32
4036 : : char newpath[MAXPGPATH];
4037 : : #endif
4038 : 2567 : const char *segname = segment_de->d_name;
4039 : :
3902 heikki.linnakangas@i 4040 : 2567 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", segname);
4041 : :
4042 : : /*
4043 : : * Before deleting the file, see if it can be recycled as a future log
4044 : : * segment. Only recycle normal files, because we don't want to recycle
4045 : : * symbolic links pointing to a separate archive directory.
4046 : : */
2452 tmunro@postgresql.or 4047 [ + - ]: 2567 : if (wal_recycle &&
1798 michael@paquier.xyz 4048 [ + + ]: 2567 : *endlogSegNo <= recycleSegNo &&
1634 noah@leadboat.com 4049 [ + + + - ]: 3457 : XLogCtl->InstallXLogFileSegmentActive && /* callee rechecks this */
1203 michael@paquier.xyz 4050 [ + + ]: 3130 : get_dirent_type(path, segment_de, false, DEBUG2) == PGFILETYPE_REG &&
1798 4051 : 1565 : InstallXLogFileSegment(endlogSegNo, path,
4052 : : true, recycleSegNo, insertTLI))
4053 : : {
3902 heikki.linnakangas@i 4054 [ + + ]: 1411 : ereport(DEBUG2,
4055 : : (errmsg_internal("recycled write-ahead log file \"%s\"",
4056 : : segname)));
4057 : 1411 : CheckpointStats.ckpt_segs_recycled++;
4058 : : /* Needn't recheck that slot on future iterations */
1798 michael@paquier.xyz 4059 : 1411 : (*endlogSegNo)++;
4060 : : }
4061 : : else
4062 : : {
4063 : : /* No need for any more future segments, or recycling failed ... */
4064 : : int rc;
4065 : :
3902 heikki.linnakangas@i 4066 [ + + ]: 1156 : ereport(DEBUG2,
4067 : : (errmsg_internal("removing write-ahead log file \"%s\"",
4068 : : segname)));
4069 : :
4070 : : #ifdef WIN32
4071 : :
4072 : : /*
4073 : : * On Windows, if another process (e.g another backend) holds the file
4074 : : * open in FILE_SHARE_DELETE mode, unlink will succeed, but the file
4075 : : * will still show up in directory listing until the last handle is
4076 : : * closed. To avoid confusing the lingering deleted file for a live
4077 : : * WAL file that needs to be archived, rename it before deleting it.
4078 : : *
4079 : : * If another process holds the file open without FILE_SHARE_DELETE
4080 : : * flag, rename will fail. We'll try again at the next checkpoint.
4081 : : */
4082 : : snprintf(newpath, MAXPGPATH, "%s.deleted", path);
4083 : : if (rename(path, newpath) != 0)
4084 : : {
4085 : : ereport(LOG,
4086 : : (errcode_for_file_access(),
4087 : : errmsg("could not rename file \"%s\": %m",
4088 : : path)));
4089 : : return;
4090 : : }
4091 : : rc = durable_unlink(newpath, LOG);
4092 : : #else
3188 teodor@sigaev.ru 4093 : 1156 : rc = durable_unlink(path, LOG);
4094 : : #endif
3902 heikki.linnakangas@i 4095 [ - + ]: 1156 : if (rc != 0)
4096 : : {
4097 : : /* Message already logged by durable_unlink() */
3902 heikki.linnakangas@i 4098 :UBC 0 : return;
4099 : : }
3902 heikki.linnakangas@i 4100 :CBC 1156 : CheckpointStats.ckpt_segs_removed++;
4101 : : }
4102 : :
4103 : 2567 : XLogArchiveCleanup(segname);
4104 : : }
4105 : :
4106 : : /*
4107 : : * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
4108 : : * If the latter do not exist, recreate them.
4109 : : *
4110 : : * It is not the goal of this function to verify the contents of these
4111 : : * directories, but to help in cases where someone has performed a cluster
4112 : : * copy for PITR purposes but omitted pg_wal from the copy.
4113 : : *
4114 : : * We could also recreate pg_wal if it doesn't exist, but a deliberate
4115 : : * policy decision was made not to. It is fairly common for pg_wal to be
4116 : : * a symlink, and if that was the DBA's intent then automatically making a
4117 : : * plain directory would result in degraded performance with no notice.
4118 : : */
4119 : : static void
6248 tgl@sss.pgh.pa.us 4120 : 927 : ValidateXLOGDirectoryStructure(void)
4121 : : {
4122 : : char path[MAXPGPATH];
4123 : : struct stat stat_buf;
4124 : :
4125 : : /* Check for pg_wal; if it doesn't exist, error out */
4126 [ + - ]: 927 : if (stat(XLOGDIR, &stat_buf) != 0 ||
4127 [ - + ]: 927 : !S_ISDIR(stat_buf.st_mode))
6034 bruce@momjian.us 4128 [ # # ]:UBC 0 : ereport(FATAL,
4129 : : (errcode_for_file_access(),
4130 : : errmsg("required WAL directory \"%s\" does not exist",
4131 : : XLOGDIR)));
4132 : :
4133 : : /* Check for archive_status */
6248 tgl@sss.pgh.pa.us 4134 :CBC 927 : snprintf(path, MAXPGPATH, XLOGDIR "/archive_status");
4135 [ + + ]: 927 : if (stat(path, &stat_buf) == 0)
4136 : : {
4137 : : /* Check for weird cases where it exists but isn't a directory */
4138 [ - + ]: 926 : if (!S_ISDIR(stat_buf.st_mode))
6034 bruce@momjian.us 4139 [ # # ]:UBC 0 : ereport(FATAL,
4140 : : (errcode_for_file_access(),
4141 : : errmsg("required WAL directory \"%s\" does not exist",
4142 : : path)));
4143 : : }
4144 : : else
4145 : : {
6248 tgl@sss.pgh.pa.us 4146 [ + - ]:CBC 1 : ereport(LOG,
4147 : : (errmsg("creating missing WAL directory \"%s\"", path)));
2812 sfrost@snowman.net 4148 [ - + ]: 1 : if (MakePGDirectory(path) < 0)
6034 bruce@momjian.us 4149 [ # # ]:UBC 0 : ereport(FATAL,
4150 : : (errcode_for_file_access(),
4151 : : errmsg("could not create missing directory \"%s\": %m",
4152 : : path)));
4153 : : }
4154 : :
4155 : : /* Check for summaries */
729 rhaas@postgresql.org 4156 :CBC 927 : snprintf(path, MAXPGPATH, XLOGDIR "/summaries");
4157 [ + + ]: 927 : if (stat(path, &stat_buf) == 0)
4158 : : {
4159 : : /* Check for weird cases where it exists but isn't a directory */
4160 [ - + ]: 926 : if (!S_ISDIR(stat_buf.st_mode))
729 rhaas@postgresql.org 4161 [ # # ]:UBC 0 : ereport(FATAL,
4162 : : (errmsg("required WAL directory \"%s\" does not exist",
4163 : : path)));
4164 : : }
4165 : : else
4166 : : {
729 rhaas@postgresql.org 4167 [ + - ]:CBC 1 : ereport(LOG,
4168 : : (errmsg("creating missing WAL directory \"%s\"", path)));
4169 [ - + ]: 1 : if (MakePGDirectory(path) < 0)
729 rhaas@postgresql.org 4170 [ # # ]:UBC 0 : ereport(FATAL,
4171 : : (errmsg("could not create missing directory \"%s\": %m",
4172 : : path)));
4173 : : }
6248 tgl@sss.pgh.pa.us 4174 :CBC 927 : }
4175 : :
4176 : : /*
4177 : : * Remove previous backup history files. This also retries creation of
4178 : : * .ready files for any backup history files for which XLogArchiveNotify
4179 : : * failed earlier.
4180 : : */
4181 : : static void
7119 4182 : 155 : CleanupBackupHistory(void)
4183 : : {
4184 : : DIR *xldir;
4185 : : struct dirent *xlde;
4186 : : char path[MAXPGPATH + sizeof(XLOGDIR)];
4187 : :
7472 4188 : 155 : xldir = AllocateDir(XLOGDIR);
4189 : :
4190 [ + + ]: 1554 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4191 : : {
3877 heikki.linnakangas@i 4192 [ + + ]: 1244 : if (IsBackupHistoryFileName(xlde->d_name))
4193 : : {
6310 tgl@sss.pgh.pa.us 4194 [ + + ]: 163 : if (XLogArchiveCheckDone(xlde->d_name))
4195 : : {
3142 peter_e@gmx.net 4196 [ + + ]: 131 : elog(DEBUG2, "removing WAL backup history file \"%s\"",
4197 : : xlde->d_name);
3173 4198 : 131 : snprintf(path, sizeof(path), XLOGDIR "/%s", xlde->d_name);
7491 bruce@momjian.us 4199 : 131 : unlink(path);
4200 : 131 : XLogArchiveCleanup(xlde->d_name);
4201 : : }
4202 : : }
4203 : : }
4204 : :
4205 : 155 : FreeDir(xldir);
4206 : 155 : }
4207 : :
4208 : : /*
4209 : : * I/O routines for pg_control
4210 : : *
4211 : : * *ControlFile is a buffer in shared memory that holds an image of the
4212 : : * contents of pg_control. WriteControlFile() initializes pg_control
4213 : : * given a preloaded buffer, ReadControlFile() loads the buffer from
4214 : : * the pg_control file (during postmaster or standalone-backend startup),
4215 : : * and UpdateControlFile() rewrites pg_control after we modify xlog state.
4216 : : * InitControlFile() fills the buffer with initial values.
4217 : : *
4218 : : * For simplicity, WriteControlFile() initializes the fields of pg_control
4219 : : * that are related to checking backend/database compatibility, and
4220 : : * ReadControlFile() verifies they are correct. We could split out the
4221 : : * I/O and compatibility-check functions, but there seems no need currently.
4222 : : */
4223 : :
4224 : : static void
513 peter@eisentraut.org 4225 : 51 : InitControlFile(uint64 sysidentifier, uint32 data_checksum_version)
4226 : : {
4227 : : char mock_auth_nonce[MOCK_AUTH_NONCE_LEN];
4228 : :
4229 : : /*
4230 : : * Generate a random nonce. This is used for authentication requests that
4231 : : * will fail because the user does not exist. The nonce is used to create
4232 : : * a genuine-looking password challenge for the non-existent user, in lieu
4233 : : * of an actual stored password.
4234 : : */
1401 heikki.linnakangas@i 4235 [ - + ]: 51 : if (!pg_strong_random(mock_auth_nonce, MOCK_AUTH_NONCE_LEN))
1401 heikki.linnakangas@i 4236 [ # # ]:UBC 0 : ereport(PANIC,
4237 : : (errcode(ERRCODE_INTERNAL_ERROR),
4238 : : errmsg("could not generate secret authorization token")));
4239 : :
1401 heikki.linnakangas@i 4240 :CBC 51 : memset(ControlFile, 0, sizeof(ControlFileData));
4241 : : /* Initialize pg_control status fields */
4242 : 51 : ControlFile->system_identifier = sysidentifier;
4243 : 51 : memcpy(ControlFile->mock_authentication_nonce, mock_auth_nonce, MOCK_AUTH_NONCE_LEN);
4244 : 51 : ControlFile->state = DB_SHUTDOWNED;
4245 : 51 : ControlFile->unloggedLSN = FirstNormalUnloggedLSN;
4246 : :
4247 : : /* Set important parameter values for use when replaying WAL */
2131 peter@eisentraut.org 4248 : 51 : ControlFile->MaxConnections = MaxConnections;
4249 : 51 : ControlFile->max_worker_processes = max_worker_processes;
4250 : 51 : ControlFile->max_wal_senders = max_wal_senders;
4251 : 51 : ControlFile->max_prepared_xacts = max_prepared_xacts;
4252 : 51 : ControlFile->max_locks_per_xact = max_locks_per_xact;
4253 : 51 : ControlFile->wal_level = wal_level;
4254 : 51 : ControlFile->wal_log_hints = wal_log_hints;
4255 : 51 : ControlFile->track_commit_timestamp = track_commit_timestamp;
513 4256 : 51 : ControlFile->data_checksum_version = data_checksum_version;
2131 4257 : 51 : }
4258 : :
4259 : : static void
9154 tgl@sss.pgh.pa.us 4260 : 51 : WriteControlFile(void)
4261 : : {
4262 : : int fd;
4263 : : char buffer[PG_CONTROL_FILE_SIZE]; /* need not be aligned */
4264 : :
4265 : : /*
4266 : : * Initialize version and compatibility-check fields
4267 : : */
9046 4268 : 51 : ControlFile->pg_control_version = PG_CONTROL_VERSION;
4269 : 51 : ControlFile->catalog_version_no = CATALOG_VERSION_NO;
4270 : :
7381 4271 : 51 : ControlFile->maxAlign = MAXIMUM_ALIGNOF;
4272 : 51 : ControlFile->floatFormat = FLOATFORMAT_VALUE;
4273 : :
9154 4274 : 51 : ControlFile->blcksz = BLCKSZ;
4275 : 51 : ControlFile->relseg_size = RELSEG_SIZE;
38 heikki.linnakangas@i 4276 :GNC 51 : ControlFile->slru_pages_per_segment = SLRU_PAGES_PER_SEGMENT;
7199 tgl@sss.pgh.pa.us 4277 :CBC 51 : ControlFile->xlog_blcksz = XLOG_BLCKSZ;
3012 andres@anarazel.de 4278 : 51 : ControlFile->xlog_seg_size = wal_segment_size;
4279 : :
8642 lockhart@fourpalms.o 4280 : 51 : ControlFile->nameDataLen = NAMEDATALEN;
7569 tgl@sss.pgh.pa.us 4281 : 51 : ControlFile->indexMaxKeys = INDEX_MAX_KEYS;
4282 : :
6834 4283 : 51 : ControlFile->toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
4214 4284 : 51 : ControlFile->loblksize = LOBLKSIZE;
4285 : :
127 tgl@sss.pgh.pa.us 4286 :GNC 51 : ControlFile->float8ByVal = true; /* vestigial */
4287 : :
4288 : : /*
4289 : : * Initialize the default 'char' signedness.
4290 : : *
4291 : : * The signedness of the char type is implementation-defined. For instance
4292 : : * on x86 architecture CPUs, the char data type is typically treated as
4293 : : * signed by default, whereas on aarch architecture CPUs, it is typically
4294 : : * treated as unsigned by default. In v17 or earlier, we accidentally let
4295 : : * C implementation signedness affect persistent data. This led to
4296 : : * inconsistent results when comparing char data across different
4297 : : * platforms.
4298 : : *
4299 : : * This flag can be used as a hint to ensure consistent behavior for
4300 : : * pre-v18 data files that store data sorted by the 'char' type on disk,
4301 : : * especially in cross-platform replication scenarios.
4302 : : *
4303 : : * Newly created database clusters unconditionally set the default char
4304 : : * signedness to true. pg_upgrade changes this flag for clusters that were
4305 : : * initialized on signedness=false platforms. As a result,
4306 : : * signedness=false setting will become rare over time. If we had known
4307 : : * about this problem during the last development cycle that forced initdb
4308 : : * (v8.3), we would have made all clusters signed or all clusters
4309 : : * unsigned. Making pg_upgrade the only source of signedness=false will
4310 : : * cause the population of database clusters to converge toward that
4311 : : * retrospective ideal.
4312 : : */
300 msawada@postgresql.o 4313 :CBC 51 : ControlFile->default_char_signedness = true;
4314 : :
4315 : : /* Contents are protected with a CRC */
4062 heikki.linnakangas@i 4316 : 51 : INIT_CRC32C(ControlFile->crc);
4317 : 51 : COMP_CRC32C(ControlFile->crc,
4318 : : ControlFile,
4319 : : offsetof(ControlFileData, crc));
4320 : 51 : FIN_CRC32C(ControlFile->crc);
4321 : :
4322 : : /*
4323 : : * We write out PG_CONTROL_FILE_SIZE bytes into pg_control, zero-padding
4324 : : * the excess over sizeof(ControlFileData). This reduces the odds of
4325 : : * premature-EOF errors when reading pg_control. We'll still fail when we
4326 : : * check the contents of the file, but hopefully with a more specific
4327 : : * error than "couldn't read pg_control".
4328 : : */
3074 tgl@sss.pgh.pa.us 4329 : 51 : memset(buffer, 0, PG_CONTROL_FILE_SIZE);
9154 4330 : 51 : memcpy(buffer, ControlFile, sizeof(ControlFileData));
4331 : :
7472 4332 : 51 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4333 : : O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
9154 4334 [ - + ]: 51 : if (fd < 0)
8186 tgl@sss.pgh.pa.us 4335 [ # # ]:UBC 0 : ereport(PANIC,
4336 : : (errcode_for_file_access(),
4337 : : errmsg("could not create file \"%s\": %m",
4338 : : XLOG_CONTROL_FILE)));
4339 : :
8961 tgl@sss.pgh.pa.us 4340 :CBC 51 : errno = 0;
3197 rhaas@postgresql.org 4341 : 51 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE);
3074 tgl@sss.pgh.pa.us 4342 [ - + ]: 51 : if (write(fd, buffer, PG_CONTROL_FILE_SIZE) != PG_CONTROL_FILE_SIZE)
4343 : : {
4344 : : /* if write didn't set errno, assume problem is no disk space */
8961 tgl@sss.pgh.pa.us 4345 [ # # ]:UBC 0 : if (errno == 0)
4346 : 0 : errno = ENOSPC;
8186 4347 [ # # ]: 0 : ereport(PANIC,
4348 : : (errcode_for_file_access(),
4349 : : errmsg("could not write to file \"%s\": %m",
4350 : : XLOG_CONTROL_FILE)));
4351 : : }
3197 rhaas@postgresql.org 4352 :CBC 51 : pgstat_report_wait_end();
4353 : :
4354 : 51 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC);
9141 tgl@sss.pgh.pa.us 4355 [ - + ]: 51 : if (pg_fsync(fd) != 0)
8186 tgl@sss.pgh.pa.us 4356 [ # # ]:UBC 0 : ereport(PANIC,
4357 : : (errcode_for_file_access(),
4358 : : errmsg("could not fsync file \"%s\": %m",
4359 : : XLOG_CONTROL_FILE)));
3197 rhaas@postgresql.org 4360 :CBC 51 : pgstat_report_wait_end();
4361 : :
2357 peter@eisentraut.org 4362 [ - + ]: 51 : if (close(fd) != 0)
7997 tgl@sss.pgh.pa.us 4363 [ # # ]:UBC 0 : ereport(PANIC,
4364 : : (errcode_for_file_access(),
4365 : : errmsg("could not close file \"%s\": %m",
4366 : : XLOG_CONTROL_FILE)));
9154 tgl@sss.pgh.pa.us 4367 :CBC 51 : }
4368 : :
4369 : : static void
4370 : 977 : ReadControlFile(void)
4371 : : {
4372 : : pg_crc32c crc;
4373 : : int fd;
4374 : : char wal_segsz_str[20];
4375 : : int r;
4376 : :
4377 : : /*
4378 : : * Read data...
4379 : : */
7472 4380 : 977 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4381 : : O_RDWR | PG_BINARY);
9154 4382 [ - + ]: 977 : if (fd < 0)
8186 tgl@sss.pgh.pa.us 4383 [ # # ]:UBC 0 : ereport(PANIC,
4384 : : (errcode_for_file_access(),
4385 : : errmsg("could not open file \"%s\": %m",
4386 : : XLOG_CONTROL_FILE)));
4387 : :
3197 rhaas@postgresql.org 4388 :CBC 977 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_READ);
2771 magnus@hagander.net 4389 : 977 : r = read(fd, ControlFile, sizeof(ControlFileData));
4390 [ - + ]: 977 : if (r != sizeof(ControlFileData))
4391 : : {
2771 magnus@hagander.net 4392 [ # # ]:UBC 0 : if (r < 0)
4393 [ # # ]: 0 : ereport(PANIC,
4394 : : (errcode_for_file_access(),
4395 : : errmsg("could not read file \"%s\": %m",
4396 : : XLOG_CONTROL_FILE)));
4397 : : else
4398 [ # # ]: 0 : ereport(PANIC,
4399 : : (errcode(ERRCODE_DATA_CORRUPTED),
4400 : : errmsg("could not read file \"%s\": read %d of %zu",
4401 : : XLOG_CONTROL_FILE, r, sizeof(ControlFileData))));
4402 : : }
3197 rhaas@postgresql.org 4403 :CBC 977 : pgstat_report_wait_end();
4404 : :
9154 tgl@sss.pgh.pa.us 4405 : 977 : close(fd);
4406 : :
4407 : : /*
4408 : : * Check for expected pg_control format version. If this is wrong, the
4409 : : * CRC check will likely fail because we'll be checking the wrong number
4410 : : * of bytes. Complaining about wrong version will probably be more
4411 : : * enlightening than complaining about wrong CRC.
4412 : : */
4413 : :
6541 peter_e@gmx.net 4414 [ - + - - : 977 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION && ControlFile->pg_control_version % 65536 == 0 && ControlFile->pg_control_version / 65536 != 0)
- - ]
6541 peter_e@gmx.net 4415 [ # # ]:UBC 0 : ereport(FATAL,
4416 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4417 : : errmsg("database files are incompatible with server"),
4418 : : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
4419 : : " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
4420 : : ControlFile->pg_control_version, ControlFile->pg_control_version,
4421 : : PG_CONTROL_VERSION, PG_CONTROL_VERSION),
4422 : : errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
4423 : :
9046 tgl@sss.pgh.pa.us 4424 [ - + ]:CBC 977 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION)
8186 tgl@sss.pgh.pa.us 4425 [ # # ]:UBC 0 : ereport(FATAL,
4426 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4427 : : errmsg("database files are incompatible with server"),
4428 : : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
4429 : : " but the server was compiled with PG_CONTROL_VERSION %d.",
4430 : : ControlFile->pg_control_version, PG_CONTROL_VERSION),
4431 : : errhint("It looks like you need to initdb.")));
4432 : :
4433 : : /* Now check the CRC. */
4062 heikki.linnakangas@i 4434 :CBC 977 : INIT_CRC32C(crc);
4435 : 977 : COMP_CRC32C(crc,
4436 : : ControlFile,
4437 : : offsetof(ControlFileData, crc));
4438 : 977 : FIN_CRC32C(crc);
4439 : :
4440 [ - + ]: 977 : if (!EQ_CRC32C(crc, ControlFile->crc))
8186 tgl@sss.pgh.pa.us 4441 [ # # ]:UBC 0 : ereport(FATAL,
4442 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4443 : : errmsg("incorrect checksum in control file")));
4444 : :
4445 : : /*
4446 : : * Do compatibility checking immediately. If the database isn't
4447 : : * compatible with the backend executable, we want to abort before we can
4448 : : * possibly do any damage.
4449 : : */
9046 tgl@sss.pgh.pa.us 4450 [ - + ]:CBC 977 : if (ControlFile->catalog_version_no != CATALOG_VERSION_NO)
8186 tgl@sss.pgh.pa.us 4451 [ # # ]:UBC 0 : ereport(FATAL,
4452 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4453 : : errmsg("database files are incompatible with server"),
4454 : : /* translator: %s is a variable name and %d is its value */
4455 : : errdetail("The database cluster was initialized with %s %d,"
4456 : : " but the server was compiled with %s %d.",
4457 : : "CATALOG_VERSION_NO", ControlFile->catalog_version_no,
4458 : : "CATALOG_VERSION_NO", CATALOG_VERSION_NO),
4459 : : errhint("It looks like you need to initdb.")));
7381 tgl@sss.pgh.pa.us 4460 [ - + ]:CBC 977 : if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
7381 tgl@sss.pgh.pa.us 4461 [ # # ]:UBC 0 : ereport(FATAL,
4462 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4463 : : errmsg("database files are incompatible with server"),
4464 : : /* translator: %s is a variable name and %d is its value */
4465 : : errdetail("The database cluster was initialized with %s %d,"
4466 : : " but the server was compiled with %s %d.",
4467 : : "MAXALIGN", ControlFile->maxAlign,
4468 : : "MAXALIGN", MAXIMUM_ALIGNOF),
4469 : : errhint("It looks like you need to initdb.")));
7381 tgl@sss.pgh.pa.us 4470 [ - + ]:CBC 977 : if (ControlFile->floatFormat != FLOATFORMAT_VALUE)
7381 tgl@sss.pgh.pa.us 4471 [ # # ]:UBC 0 : ereport(FATAL,
4472 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4473 : : errmsg("database files are incompatible with server"),
4474 : : errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
4475 : : errhint("It looks like you need to initdb.")));
9154 tgl@sss.pgh.pa.us 4476 [ - + ]:CBC 977 : if (ControlFile->blcksz != BLCKSZ)
8186 tgl@sss.pgh.pa.us 4477 [ # # ]:UBC 0 : ereport(FATAL,
4478 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4479 : : errmsg("database files are incompatible with server"),
4480 : : /* translator: %s is a variable name and %d is its value */
4481 : : errdetail("The database cluster was initialized with %s %d,"
4482 : : " but the server was compiled with %s %d.",
4483 : : "BLCKSZ", ControlFile->blcksz,
4484 : : "BLCKSZ", BLCKSZ),
4485 : : errhint("It looks like you need to recompile or initdb.")));
9154 tgl@sss.pgh.pa.us 4486 [ - + ]:CBC 977 : if (ControlFile->relseg_size != RELSEG_SIZE)
8186 tgl@sss.pgh.pa.us 4487 [ # # ]:UBC 0 : ereport(FATAL,
4488 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4489 : : errmsg("database files are incompatible with server"),
4490 : : /* translator: %s is a variable name and %d is its value */
4491 : : errdetail("The database cluster was initialized with %s %d,"
4492 : : " but the server was compiled with %s %d.",
4493 : : "RELSEG_SIZE", ControlFile->relseg_size,
4494 : : "RELSEG_SIZE", RELSEG_SIZE),
4495 : : errhint("It looks like you need to recompile or initdb.")));
38 heikki.linnakangas@i 4496 [ - + ]:GNC 977 : if (ControlFile->slru_pages_per_segment != SLRU_PAGES_PER_SEGMENT)
38 heikki.linnakangas@i 4497 [ # # ]:UNC 0 : ereport(FATAL,
4498 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4499 : : errmsg("database files are incompatible with server"),
4500 : : /* translator: %s is a variable name and %d is its value */
4501 : : errdetail("The database cluster was initialized with %s %d,"
4502 : : " but the server was compiled with %s %d.",
4503 : : "SLRU_PAGES_PER_SEGMENT", ControlFile->slru_pages_per_segment,
4504 : : "SLRU_PAGES_PER_SEGMENT", SLRU_PAGES_PER_SEGMENT),
4505 : : errhint("It looks like you need to recompile or initdb.")));
7199 tgl@sss.pgh.pa.us 4506 [ - + ]:CBC 977 : if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
7199 tgl@sss.pgh.pa.us 4507 [ # # ]:UBC 0 : ereport(FATAL,
4508 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4509 : : errmsg("database files are incompatible with server"),
4510 : : /* translator: %s is a variable name and %d is its value */
4511 : : errdetail("The database cluster was initialized with %s %d,"
4512 : : " but the server was compiled with %s %d.",
4513 : : "XLOG_BLCKSZ", ControlFile->xlog_blcksz,
4514 : : "XLOG_BLCKSZ", XLOG_BLCKSZ),
4515 : : errhint("It looks like you need to recompile or initdb.")));
8642 lockhart@fourpalms.o 4516 [ - + ]:CBC 977 : if (ControlFile->nameDataLen != NAMEDATALEN)
8186 tgl@sss.pgh.pa.us 4517 [ # # ]:UBC 0 : ereport(FATAL,
4518 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4519 : : errmsg("database files are incompatible with server"),
4520 : : /* translator: %s is a variable name and %d is its value */
4521 : : errdetail("The database cluster was initialized with %s %d,"
4522 : : " but the server was compiled with %s %d.",
4523 : : "NAMEDATALEN", ControlFile->nameDataLen,
4524 : : "NAMEDATALEN", NAMEDATALEN),
4525 : : errhint("It looks like you need to recompile or initdb.")));
7569 tgl@sss.pgh.pa.us 4526 [ - + ]:CBC 977 : if (ControlFile->indexMaxKeys != INDEX_MAX_KEYS)
8186 tgl@sss.pgh.pa.us 4527 [ # # ]:UBC 0 : ereport(FATAL,
4528 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4529 : : errmsg("database files are incompatible with server"),
4530 : : /* translator: %s is a variable name and %d is its value */
4531 : : errdetail("The database cluster was initialized with %s %d,"
4532 : : " but the server was compiled with %s %d.",
4533 : : "INDEX_MAX_KEYS", ControlFile->indexMaxKeys,
4534 : : "INDEX_MAX_KEYS", INDEX_MAX_KEYS),
4535 : : errhint("It looks like you need to recompile or initdb.")));
6834 tgl@sss.pgh.pa.us 4536 [ - + ]:CBC 977 : if (ControlFile->toast_max_chunk_size != TOAST_MAX_CHUNK_SIZE)
6834 tgl@sss.pgh.pa.us 4537 [ # # ]:UBC 0 : ereport(FATAL,
4538 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4539 : : errmsg("database files are incompatible with server"),
4540 : : /* translator: %s is a variable name and %d is its value */
4541 : : errdetail("The database cluster was initialized with %s %d,"
4542 : : " but the server was compiled with %s %d.",
4543 : : "TOAST_MAX_CHUNK_SIZE", ControlFile->toast_max_chunk_size,
4544 : : "TOAST_MAX_CHUNK_SIZE", (int) TOAST_MAX_CHUNK_SIZE),
4545 : : errhint("It looks like you need to recompile or initdb.")));
4214 tgl@sss.pgh.pa.us 4546 [ - + ]:CBC 977 : if (ControlFile->loblksize != LOBLKSIZE)
4214 tgl@sss.pgh.pa.us 4547 [ # # ]:UBC 0 : ereport(FATAL,
4548 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4549 : : errmsg("database files are incompatible with server"),
4550 : : /* translator: %s is a variable name and %d is its value */
4551 : : errdetail("The database cluster was initialized with %s %d,"
4552 : : " but the server was compiled with %s %d.",
4553 : : "LOBLKSIZE", ControlFile->loblksize,
4554 : : "LOBLKSIZE", (int) LOBLKSIZE),
4555 : : errhint("It looks like you need to recompile or initdb.")));
4556 : :
127 tgl@sss.pgh.pa.us 4557 [ - + ]:GNC 977 : Assert(ControlFile->float8ByVal); /* vestigial, not worth an error msg */
4558 : :
3012 andres@anarazel.de 4559 :CBC 977 : wal_segment_size = ControlFile->xlog_seg_size;
4560 : :
4561 [ + - + - : 977 : if (!IsValidWalSegSize(wal_segment_size))
+ - - + ]
3012 andres@anarazel.de 4562 [ # # ]:UBC 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4563 : : errmsg_plural("invalid WAL segment size in control file (%d byte)",
4564 : : "invalid WAL segment size in control file (%d bytes)",
4565 : : wal_segment_size,
4566 : : wal_segment_size),
4567 : : errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.")));
4568 : :
3012 andres@anarazel.de 4569 :CBC 977 : snprintf(wal_segsz_str, sizeof(wal_segsz_str), "%d", wal_segment_size);
4570 : 977 : SetConfigOption("wal_segment_size", wal_segsz_str, PGC_INTERNAL,
4571 : : PGC_S_DYNAMIC_DEFAULT);
4572 : :
4573 : : /* check and update variables dependent on wal_segment_size */
4574 [ - + ]: 977 : if (ConvertToXSegs(min_wal_size_mb, wal_segment_size) < 2)
3012 andres@anarazel.de 4575 [ # # ]:UBC 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4576 : : /* translator: both %s are GUC names */
4577 : : errmsg("\"%s\" must be at least twice \"%s\"",
4578 : : "min_wal_size", "wal_segment_size")));
4579 : :
3012 andres@anarazel.de 4580 [ - + ]:CBC 977 : if (ConvertToXSegs(max_wal_size_mb, wal_segment_size) < 2)
3012 andres@anarazel.de 4581 [ # # ]:UBC 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4582 : : /* translator: both %s are GUC names */
4583 : : errmsg("\"%s\" must be at least twice \"%s\"",
4584 : : "max_wal_size", "wal_segment_size")));
4585 : :
3012 andres@anarazel.de 4586 :CBC 977 : UsableBytesInSegment =
4587 : 977 : (wal_segment_size / XLOG_BLCKSZ * UsableBytesInPage) -
4588 : : (SizeOfXLogLongPHD - SizeOfXLogShortPHD);
4589 : :
4590 : 977 : CalculateCheckpointSegments();
4591 : :
4592 : : /* Make the initdb settings visible as GUC variables, too */
2810 magnus@hagander.net 4593 [ + + ]: 977 : SetConfigOption("data_checksums", DataChecksumsEnabled() ? "yes" : "no",
4594 : : PGC_INTERNAL, PGC_S_DYNAMIC_DEFAULT);
9154 tgl@sss.pgh.pa.us 4595 : 977 : }
4596 : :
4597 : : /*
4598 : : * Utility wrapper to update the control file. Note that the control
4599 : : * file gets flushed.
4600 : : */
4601 : : static void
4602 : 9322 : UpdateControlFile(void)
4603 : : {
2453 peter@eisentraut.org 4604 : 9322 : update_controlfile(DataDir, ControlFile, true);
9570 vadim4o@yahoo.com 4605 : 9322 : }
4606 : :
4607 : : /*
4608 : : * Returns the unique system identifier from control file.
4609 : : */
4610 : : uint64
5816 heikki.linnakangas@i 4611 : 1372 : GetSystemIdentifier(void)
4612 : : {
4613 [ - + ]: 1372 : Assert(ControlFile != NULL);
4614 : 1372 : return ControlFile->system_identifier;
4615 : : }
4616 : :
4617 : : /*
4618 : : * Returns the random nonce from control file.
4619 : : */
4620 : : char *
3208 4621 : 1 : GetMockAuthenticationNonce(void)
4622 : : {
4623 [ - + ]: 1 : Assert(ControlFile != NULL);
4624 : 1 : return ControlFile->mock_authentication_nonce;
4625 : : }
4626 : :
4627 : : /*
4628 : : * Are checksums enabled for data pages?
4629 : : */
4630 : : bool
2810 magnus@hagander.net 4631 : 9288954 : DataChecksumsEnabled(void)
4632 : : {
4654 simon@2ndQuadrant.co 4633 [ - + ]: 9288954 : Assert(ControlFile != NULL);
4615 4634 : 9288954 : return (ControlFile->data_checksum_version > 0);
4635 : : }
4636 : :
4637 : : /*
4638 : : * Return true if the cluster was initialized on a platform where the
4639 : : * default signedness of char is "signed". This function exists for code
4640 : : * that deals with pre-v18 data files that store data sorted by the 'char'
4641 : : * type on disk (e.g., GIN and GiST indexes). See the comments in
4642 : : * WriteControlFile() for details.
4643 : : */
4644 : : bool
300 msawada@postgresql.o 4645 : 3 : GetDefaultCharSignedness(void)
4646 : : {
4647 : 3 : return ControlFile->default_char_signedness;
4648 : : }
4649 : :
4650 : : /*
4651 : : * Returns a fake LSN for unlogged relations.
4652 : : *
4653 : : * Each call generates an LSN that is greater than any previous value
4654 : : * returned. The current counter value is saved and restored across clean
4655 : : * shutdowns, but like unlogged relations, does not survive a crash. This can
4656 : : * be used in lieu of real LSN values returned by XLogInsert, if you need an
4657 : : * LSN-like increasing sequence of numbers without writing any WAL.
4658 : : */
4659 : : XLogRecPtr
4693 heikki.linnakangas@i 4660 : 33 : GetFakeLSNForUnloggedRel(void)
4661 : : {
658 nathan@postgresql.or 4662 : 33 : return pg_atomic_fetch_add_u64(&XLogCtl->unloggedLSN, 1);
4663 : : }
4664 : :
4665 : : /*
4666 : : * Auto-tune the number of XLOG buffers.
4667 : : *
4668 : : * The preferred setting for wal_buffers is about 3% of shared_buffers, with
4669 : : * a maximum of one XLOG segment (there is little reason to think that more
4670 : : * is helpful, at least so long as we force an fsync when switching log files)
4671 : : * and a minimum of 8 blocks (which was the default value prior to PostgreSQL
4672 : : * 9.1, when auto-tuning was added).
4673 : : *
4674 : : * This should not be called until NBuffers has received its final value.
4675 : : */
4676 : : static int
5369 tgl@sss.pgh.pa.us 4677 : 1070 : XLOGChooseNumBuffers(void)
4678 : : {
4679 : : int xbuffers;
4680 : :
4681 : 1070 : xbuffers = NBuffers / 32;
3012 andres@anarazel.de 4682 [ + + ]: 1070 : if (xbuffers > (wal_segment_size / XLOG_BLCKSZ))
4683 : 24 : xbuffers = (wal_segment_size / XLOG_BLCKSZ);
5369 tgl@sss.pgh.pa.us 4684 [ + + ]: 1070 : if (xbuffers < 8)
4685 : 414 : xbuffers = 8;
4686 : 1070 : return xbuffers;
4687 : : }
4688 : :
4689 : : /*
4690 : : * GUC check_hook for wal_buffers
4691 : : */
4692 : : bool
4693 : 2179 : check_wal_buffers(int *newval, void **extra, GucSource source)
4694 : : {
4695 : : /*
4696 : : * -1 indicates a request for auto-tune.
4697 : : */
4698 [ + + ]: 2179 : if (*newval == -1)
4699 : : {
4700 : : /*
4701 : : * If we haven't yet changed the boot_val default of -1, just let it
4702 : : * be. We'll fix it when XLOGShmemSize is called.
4703 : : */
4704 [ + - ]: 1109 : if (XLOGbuffers == -1)
4705 : 1109 : return true;
4706 : :
4707 : : /* Otherwise, substitute the auto-tune value */
5369 tgl@sss.pgh.pa.us 4708 :UBC 0 : *newval = XLOGChooseNumBuffers();
4709 : : }
4710 : :
4711 : : /*
4712 : : * We clamp manually-set values to at least 4 blocks. Prior to PostgreSQL
4713 : : * 9.1, a minimum of 4 was enforced by guc.c, but since that is no longer
4714 : : * the case, we just silently treat such values as a request for the
4715 : : * minimum. (We could throw an error instead, but that doesn't seem very
4716 : : * helpful.)
4717 : : */
5369 tgl@sss.pgh.pa.us 4718 [ - + ]:CBC 1070 : if (*newval < 4)
5369 tgl@sss.pgh.pa.us 4719 :UBC 0 : *newval = 4;
4720 : :
5369 tgl@sss.pgh.pa.us 4721 :CBC 1070 : return true;
4722 : : }
4723 : :
4724 : : /*
4725 : : * GUC check_hook for wal_consistency_checking
4726 : : */
4727 : : bool
1192 4728 : 2061 : check_wal_consistency_checking(char **newval, void **extra, GucSource source)
4729 : : {
4730 : : char *rawstring;
4731 : : List *elemlist;
4732 : : ListCell *l;
4733 : : bool newwalconsistency[RM_MAX_ID + 1];
4734 : :
4735 : : /* Initialize the array */
4736 [ + - + - : 68013 : MemSet(newwalconsistency, 0, (RM_MAX_ID + 1) * sizeof(bool));
+ - + - +
+ ]
4737 : :
4738 : : /* Need a modifiable copy of string */
4739 : 2061 : rawstring = pstrdup(*newval);
4740 : :
4741 : : /* Parse string into list of identifiers */
4742 [ - + ]: 2061 : if (!SplitIdentifierString(rawstring, ',', &elemlist))
4743 : : {
4744 : : /* syntax error in list */
1192 tgl@sss.pgh.pa.us 4745 :UBC 0 : GUC_check_errdetail("List syntax is invalid.");
4746 : 0 : pfree(rawstring);
4747 : 0 : list_free(elemlist);
4748 : 0 : return false;
4749 : : }
4750 : :
1192 tgl@sss.pgh.pa.us 4751 [ + + + + :CBC 2540 : foreach(l, elemlist)
+ + ]
4752 : : {
4753 : 479 : char *tok = (char *) lfirst(l);
4754 : : int rmid;
4755 : :
4756 : : /* Check for 'all'. */
4757 [ + + ]: 479 : if (pg_strcasecmp(tok, "all") == 0)
4758 : : {
4759 [ + + ]: 122589 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4760 [ + + + + ]: 122112 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL)
4761 : 4770 : newwalconsistency[rmid] = true;
4762 : : }
4763 : : else
4764 : : {
4765 : : /* Check if the token matches any known resource manager. */
4766 : 2 : bool found = false;
4767 : :
4768 [ + - ]: 36 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4769 : : {
4770 [ + - + + : 54 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL &&
+ + ]
4771 : 18 : pg_strcasecmp(tok, GetRmgr(rmid).rm_name) == 0)
4772 : : {
4773 : 2 : newwalconsistency[rmid] = true;
4774 : 2 : found = true;
4775 : 2 : break;
4776 : : }
4777 : : }
4778 [ - + ]: 2 : if (!found)
4779 : : {
4780 : : /*
4781 : : * During startup, it might be a not-yet-loaded custom
4782 : : * resource manager. Defer checking until
4783 : : * InitializeWalConsistencyChecking().
4784 : : */
1192 tgl@sss.pgh.pa.us 4785 [ # # ]:UBC 0 : if (!process_shared_preload_libraries_done)
4786 : : {
4787 : 0 : check_wal_consistency_checking_deferred = true;
4788 : : }
4789 : : else
4790 : : {
4791 : 0 : GUC_check_errdetail("Unrecognized key word: \"%s\".", tok);
4792 : 0 : pfree(rawstring);
4793 : 0 : list_free(elemlist);
4794 : 0 : return false;
4795 : : }
4796 : : }
4797 : : }
4798 : : }
4799 : :
1192 tgl@sss.pgh.pa.us 4800 :CBC 2061 : pfree(rawstring);
4801 : 2061 : list_free(elemlist);
4802 : :
4803 : : /* assign new value */
266 dgustafsson@postgres 4804 : 2061 : *extra = guc_malloc(LOG, (RM_MAX_ID + 1) * sizeof(bool));
4805 [ - + ]: 2061 : if (!*extra)
266 dgustafsson@postgres 4806 :UBC 0 : return false;
1192 tgl@sss.pgh.pa.us 4807 :CBC 2061 : memcpy(*extra, newwalconsistency, (RM_MAX_ID + 1) * sizeof(bool));
4808 : 2061 : return true;
4809 : : }
4810 : :
4811 : : /*
4812 : : * GUC assign_hook for wal_consistency_checking
4813 : : */
4814 : : void
4815 : 2060 : assign_wal_consistency_checking(const char *newval, void *extra)
4816 : : {
4817 : : /*
4818 : : * If some checks were deferred, it's possible that the checks will fail
4819 : : * later during InitializeWalConsistencyChecking(). But in that case, the
4820 : : * postmaster will exit anyway, so it's safe to proceed with the
4821 : : * assignment.
4822 : : *
4823 : : * Any built-in resource managers specified are assigned immediately,
4824 : : * which affects WAL created before shared_preload_libraries are
4825 : : * processed. Any custom resource managers specified won't be assigned
4826 : : * until after shared_preload_libraries are processed, but that's OK
4827 : : * because WAL for a custom resource manager can't be written before the
4828 : : * module is loaded anyway.
4829 : : */
4830 : 2060 : wal_consistency_checking = extra;
4831 : 2060 : }
4832 : :
4833 : : /*
4834 : : * InitializeWalConsistencyChecking: run after loading custom resource managers
4835 : : *
4836 : : * If any unknown resource managers were specified in the
4837 : : * wal_consistency_checking GUC, processing was deferred. Now that
4838 : : * shared_preload_libraries have been loaded, process wal_consistency_checking
4839 : : * again.
4840 : : */
4841 : : void
4842 : 917 : InitializeWalConsistencyChecking(void)
4843 : : {
4844 [ - + ]: 917 : Assert(process_shared_preload_libraries_done);
4845 : :
4846 [ - + ]: 917 : if (check_wal_consistency_checking_deferred)
4847 : : {
4848 : : struct config_generic *guc;
4849 : :
1192 tgl@sss.pgh.pa.us 4850 :UBC 0 : guc = find_option("wal_consistency_checking", false, false, ERROR);
4851 : :
4852 : 0 : check_wal_consistency_checking_deferred = false;
4853 : :
4854 : 0 : set_config_option_ext("wal_consistency_checking",
4855 : : wal_consistency_checking_string,
4856 : : guc->scontext, guc->source, guc->srole,
4857 : : GUC_ACTION_SET, true, ERROR, false);
4858 : :
4859 : : /* checking should not be deferred again */
4860 [ # # ]: 0 : Assert(!check_wal_consistency_checking_deferred);
4861 : : }
1192 tgl@sss.pgh.pa.us 4862 :CBC 917 : }
4863 : :
4864 : : /*
4865 : : * GUC show_hook for archive_command
4866 : : */
4867 : : const char *
4868 : 1693 : show_archive_command(void)
4869 : : {
4870 [ + + - + : 1693 : if (XLogArchivingActive())
+ + ]
4871 : 2 : return XLogArchiveCommand;
4872 : : else
4873 : 1691 : return "(disabled)";
4874 : : }
4875 : :
4876 : : /*
4877 : : * GUC show_hook for in_hot_standby
4878 : : */
4879 : : const char *
4880 : 14294 : show_in_hot_standby(void)
4881 : : {
4882 : : /*
4883 : : * We display the actual state based on shared memory, so that this GUC
4884 : : * reports up-to-date state if examined intra-query. The underlying
4885 : : * variable (in_hot_standby_guc) changes only when we transmit a new value
4886 : : * to the client.
4887 : : */
4888 [ + + ]: 14294 : return RecoveryInProgress() ? "on" : "off";
4889 : : }
4890 : :
4891 : : /*
4892 : : * Read the control file, set respective GUCs.
4893 : : *
4894 : : * This is to be called during startup, including a crash recovery cycle,
4895 : : * unless in bootstrap mode, where no control file yet exists. As there's no
4896 : : * usable shared memory yet (its sizing can depend on the contents of the
4897 : : * control file!), first store the contents in local memory. XLOGShmemInit()
4898 : : * will then copy it to shared memory later.
4899 : : *
4900 : : * reset just controls whether previous contents are to be expected (in the
4901 : : * reset case, there's a dangling pointer into old shared memory), or not.
4902 : : */
4903 : : void
3014 andres@anarazel.de 4904 : 926 : LocalProcessControlFile(bool reset)
4905 : : {
4906 [ + + - + ]: 926 : Assert(reset || ControlFile == NULL);
8 michael@paquier.xyz 4907 :GNC 926 : ControlFile = palloc_object(ControlFileData);
3018 andres@anarazel.de 4908 :CBC 926 : ReadControlFile();
4909 : 926 : }
4910 : :
4911 : : /*
4912 : : * Get the wal_level from the control file. For a standby, this value should be
4913 : : * considered as its active wal_level, because it may be different from what
4914 : : * was originally configured on standby.
4915 : : */
4916 : : WalLevel
985 4917 : 1 : GetActiveWalLevelOnStandby(void)
4918 : : {
4919 : 1 : return ControlFile->wal_level;
4920 : : }
4921 : :
4922 : : /*
4923 : : * Initialization of shared memory for XLOG
4924 : : */
4925 : : Size
9158 peter_e@gmx.net 4926 : 3061 : XLOGShmemSize(void)
4927 : : {
4928 : : Size size;
4929 : :
4930 : : /*
4931 : : * If the value of wal_buffers is -1, use the preferred auto-tune value.
4932 : : * This isn't an amazingly clean place to do this, but we must wait till
4933 : : * NBuffers has received its final value, and must do it before using the
4934 : : * value of XLOGbuffers to do anything important.
4935 : : *
4936 : : * We prefer to report this value's source as PGC_S_DYNAMIC_DEFAULT.
4937 : : * However, if the DBA explicitly set wal_buffers = -1 in the config file,
4938 : : * then PGC_S_DYNAMIC_DEFAULT will fail to override that and we must force
4939 : : * the matter with PGC_S_OVERRIDE.
4940 : : */
5369 tgl@sss.pgh.pa.us 4941 [ + + ]: 3061 : if (XLOGbuffers == -1)
4942 : : {
4943 : : char buf[32];
4944 : :
4945 : 1070 : snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
1289 4946 : 1070 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4947 : : PGC_S_DYNAMIC_DEFAULT);
4948 [ - + ]: 1070 : if (XLOGbuffers == -1) /* failed to apply it? */
1289 tgl@sss.pgh.pa.us 4949 :UBC 0 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4950 : : PGC_S_OVERRIDE);
4951 : : }
5444 tgl@sss.pgh.pa.us 4952 [ - + ]:CBC 3061 : Assert(XLOGbuffers > 0);
4953 : :
4954 : : /* XLogCtl */
7425 4955 : 3061 : size = sizeof(XLogCtlData);
4956 : :
4957 : : /* WAL insertion locks, plus alignment */
4096 heikki.linnakangas@i 4958 : 3061 : size = add_size(size, mul_size(sizeof(WALInsertLockPadded), NUM_XLOGINSERT_LOCKS + 1));
4959 : : /* xlblocks array */
730 jdavis@postgresql.or 4960 : 3061 : size = add_size(size, mul_size(sizeof(pg_atomic_uint64), XLOGbuffers));
4961 : : /* extra alignment padding for XLOG I/O buffers */
985 tmunro@postgresql.or 4962 : 3061 : size = add_size(size, Max(XLOG_BLCKSZ, PG_IO_ALIGN_SIZE));
4963 : : /* and the buffers themselves */
7199 tgl@sss.pgh.pa.us 4964 : 3061 : size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4965 : :
4966 : : /*
4967 : : * Note: we don't count ControlFileData, it comes out of the "slop factor"
4968 : : * added by CreateSharedMemoryAndSemaphores. This lets us use this
4969 : : * routine again below to compute the actual allocation size.
4970 : : */
4971 : :
7425 4972 : 3061 : return size;
4973 : : }
4974 : :
4975 : : void
9570 vadim4o@yahoo.com 4976 : 1071 : XLOGShmemInit(void)
4977 : : {
4978 : : bool foundCFile,
4979 : : foundXLog;
4980 : : char *allocptr;
4981 : : int i;
4982 : : ControlFileData *localControlFile;
4983 : :
4984 : : #ifdef WAL_DEBUG
4985 : :
4986 : : /*
4987 : : * Create a memory context for WAL debugging that's exempt from the normal
4988 : : * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
4989 : : * an allocation fails, but wal_debug is not for production use anyway.
4990 : : */
4991 : : if (walDebugCxt == NULL)
4992 : : {
4993 : : walDebugCxt = AllocSetContextCreate(TopMemoryContext,
4994 : : "WAL Debug",
4995 : : ALLOCSET_DEFAULT_SIZES);
4996 : : MemoryContextAllowInCriticalSection(walDebugCxt, true);
4997 : : }
4998 : : #endif
4999 : :
5000 : :
3014 andres@anarazel.de 5001 : 1071 : XLogCtl = (XLogCtlData *)
5002 : 1071 : ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
5003 : :
3018 5004 : 1071 : localControlFile = ControlFile;
9154 tgl@sss.pgh.pa.us 5005 : 1071 : ControlFile = (ControlFileData *)
8034 bruce@momjian.us 5006 : 1071 : ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
5007 : :
7423 tgl@sss.pgh.pa.us 5008 [ + - - + ]: 1071 : if (foundCFile || foundXLog)
5009 : : {
5010 : : /* both should be present or neither */
7423 tgl@sss.pgh.pa.us 5011 [ # # # # ]:UBC 0 : Assert(foundCFile && foundXLog);
5012 : :
5013 : : /* Initialize local copy of WALInsertLocks */
4165 rhaas@postgresql.org 5014 : 0 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks;
5015 : :
3014 andres@anarazel.de 5016 [ # # ]: 0 : if (localControlFile)
5017 : 0 : pfree(localControlFile);
8034 bruce@momjian.us 5018 : 0 : return;
5019 : : }
9046 tgl@sss.pgh.pa.us 5020 :CBC 1071 : memset(XLogCtl, 0, sizeof(XLogCtlData));
5021 : :
5022 : : /*
5023 : : * Already have read control file locally, unless in bootstrap mode. Move
5024 : : * contents into shared memory.
5025 : : */
3014 andres@anarazel.de 5026 [ + + ]: 1071 : if (localControlFile)
5027 : : {
5028 : 918 : memcpy(ControlFile, localControlFile, sizeof(ControlFileData));
5029 : 918 : pfree(localControlFile);
5030 : : }
5031 : :
5032 : : /*
5033 : : * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
5034 : : * multiple of the alignment for same, so no extra alignment padding is
5035 : : * needed here.
5036 : : */
4546 heikki.linnakangas@i 5037 : 1071 : allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
730 jdavis@postgresql.or 5038 : 1071 : XLogCtl->xlblocks = (pg_atomic_uint64 *) allocptr;
5039 : 1071 : allocptr += sizeof(pg_atomic_uint64) * XLOGbuffers;
5040 : :
5041 [ + + ]: 306012 : for (i = 0; i < XLOGbuffers; i++)
5042 : : {
5043 : 304941 : pg_atomic_init_u64(&XLogCtl->xlblocks[i], InvalidXLogRecPtr);
5044 : : }
5045 : :
5046 : : /* WAL insertion locks. Ensure they're aligned to the full padded size */
4290 heikki.linnakangas@i 5047 : 1071 : allocptr += sizeof(WALInsertLockPadded) -
3102 tgl@sss.pgh.pa.us 5048 : 1071 : ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
4290 heikki.linnakangas@i 5049 : 1071 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks =
5050 : : (WALInsertLockPadded *) allocptr;
4096 5051 : 1071 : allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
5052 : :
5053 [ + + ]: 9639 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
5054 : : {
3656 rhaas@postgresql.org 5055 : 8568 : LWLockInitialize(&WALInsertLocks[i].l.lock, LWTRANCHE_WAL_INSERT);
877 michael@paquier.xyz 5056 : 8568 : pg_atomic_init_u64(&WALInsertLocks[i].l.insertingAt, InvalidXLogRecPtr);
3283 andres@anarazel.de 5057 : 8568 : WALInsertLocks[i].l.lastImportantAt = InvalidXLogRecPtr;
5058 : : }
5059 : :
5060 : : /*
5061 : : * Align the start of the page buffers to a full xlog block size boundary.
5062 : : * This simplifies some calculations in XLOG insertion. It is also
5063 : : * required for O_DIRECT.
5064 : : */
4546 heikki.linnakangas@i 5065 : 1071 : allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
7425 tgl@sss.pgh.pa.us 5066 : 1071 : XLogCtl->pages = allocptr;
7199 5067 : 1071 : memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
5068 : :
5069 : : /*
5070 : : * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
5071 : : * in additional info.)
5072 : : */
9046 5073 : 1071 : XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
2064 michael@paquier.xyz 5074 : 1071 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
1634 noah@leadboat.com 5075 : 1071 : XLogCtl->InstallXLogFileSegmentActive = false;
4972 tgl@sss.pgh.pa.us 5076 : 1071 : XLogCtl->WalWriterSleeping = false;
5077 : :
4546 heikki.linnakangas@i 5078 : 1071 : SpinLockInit(&XLogCtl->Insert.insertpos_lck);
8846 tgl@sss.pgh.pa.us 5079 : 1071 : SpinLockInit(&XLogCtl->info_lck);
620 alvherre@alvh.no-ip. 5080 : 1071 : pg_atomic_init_u64(&XLogCtl->logInsertResult, InvalidXLogRecPtr);
622 5081 : 1071 : pg_atomic_init_u64(&XLogCtl->logWriteResult, InvalidXLogRecPtr);
5082 : 1071 : pg_atomic_init_u64(&XLogCtl->logFlushResult, InvalidXLogRecPtr);
658 nathan@postgresql.or 5083 : 1071 : pg_atomic_init_u64(&XLogCtl->unloggedLSN, InvalidXLogRecPtr);
5084 : : }
5085 : :
5086 : : /*
5087 : : * This func must be called ONCE on system install. It creates pg_control
5088 : : * and the initial XLOG segment.
5089 : : */
5090 : : void
513 peter@eisentraut.org 5091 : 51 : BootStrapXLOG(uint32 data_checksum_version)
5092 : : {
5093 : : CheckPoint checkPoint;
5094 : : PGAlignedXLogBlock buffer;
5095 : : XLogPageHeader page;
5096 : : XLogLongPageHeader longpage;
5097 : : XLogRecord *record;
5098 : : char *recptr;
5099 : : uint64 sysidentifier;
5100 : : struct timeval tv;
5101 : : pg_crc32c crc;
5102 : :
5103 : : /* allow ordinary WAL segment creation, like StartupXLOG() would */
1219 michael@paquier.xyz 5104 : 51 : SetInstallXLogFileSegmentActive();
5105 : :
5106 : : /*
5107 : : * Select a hopefully-unique system identifier code for this installation.
5108 : : * We use the result of gettimeofday(), including the fractional seconds
5109 : : * field, as being about as unique as we can easily get. (Think not to
5110 : : * use random(), since it hasn't been seeded and there's no portable way
5111 : : * to seed it other than the system clock value...) The upper half of the
5112 : : * uint64 value is just the tv_sec part, while the lower half contains the
5113 : : * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
5114 : : * PID for a little extra uniqueness. A person knowing this encoding can
5115 : : * determine the initialization time of the installation, which could
5116 : : * perhaps be useful sometimes.
5117 : : */
7981 tgl@sss.pgh.pa.us 5118 : 51 : gettimeofday(&tv, NULL);
5119 : 51 : sysidentifier = ((uint64) tv.tv_sec) << 32;
4254 5120 : 51 : sysidentifier |= ((uint64) tv.tv_usec) << 12;
5121 : 51 : sysidentifier |= getpid() & 0xFFF;
5122 : :
10 peter@eisentraut.org 5123 :GNC 51 : memset(&buffer, 0, sizeof buffer);
5124 : 51 : page = (XLogPageHeader) &buffer;
5125 : :
5126 : : /*
5127 : : * Set up information for the initial checkpoint record
5128 : : *
5129 : : * The initial checkpoint record is written to the beginning of the WAL
5130 : : * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
5131 : : * used, so that we can use 0/0 to mean "before any valid WAL segment".
5132 : : */
3012 andres@anarazel.de 5133 :CBC 51 : checkPoint.redo = wal_segment_size + SizeOfXLogLongPHD;
1504 rhaas@postgresql.org 5134 : 51 : checkPoint.ThisTimeLineID = BootstrapTimeLineID;
5135 : 51 : checkPoint.PrevTimeLineID = BootstrapTimeLineID;
5076 simon@2ndQuadrant.co 5136 : 51 : checkPoint.fullPageWrites = fullPageWrites;
514 rhaas@postgresql.org 5137 : 51 : checkPoint.wal_level = wal_level;
5138 : : checkPoint.nextXid =
2457 tmunro@postgresql.or 5139 : 51 : FullTransactionIdFromEpochAndXid(0, FirstNormalTransactionId);
1617 tgl@sss.pgh.pa.us 5140 : 51 : checkPoint.nextOid = FirstGenbkiObjectId;
7539 5141 : 51 : checkPoint.nextMulti = FirstMultiXactId;
9 heikki.linnakangas@i 5142 :GNC 51 : checkPoint.nextMultiOffset = 1;
5953 tgl@sss.pgh.pa.us 5143 :CBC 51 : checkPoint.oldestXid = FirstNormalTransactionId;
1337 5144 : 51 : checkPoint.oldestXidDB = Template1DbOid;
4712 alvherre@alvh.no-ip. 5145 : 51 : checkPoint.oldestMulti = FirstMultiXactId;
1337 tgl@sss.pgh.pa.us 5146 : 51 : checkPoint.oldestMultiDB = Template1DbOid;
3643 mail@joeconway.com 5147 : 51 : checkPoint.oldestCommitTsXid = InvalidTransactionId;
5148 : 51 : checkPoint.newestCommitTsXid = InvalidTransactionId;
6514 tgl@sss.pgh.pa.us 5149 : 51 : checkPoint.time = (pg_time_t) time(NULL);
5843 simon@2ndQuadrant.co 5150 : 51 : checkPoint.oldestActiveXid = InvalidTransactionId;
5151 : :
741 heikki.linnakangas@i 5152 : 51 : TransamVariables->nextXid = checkPoint.nextXid;
5153 : 51 : TransamVariables->nextOid = checkPoint.nextOid;
5154 : 51 : TransamVariables->oidCount = 0;
7498 tgl@sss.pgh.pa.us 5155 : 51 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
3192 rhaas@postgresql.org 5156 : 51 : AdvanceOldestClogXid(checkPoint.oldestXid);
5783 tgl@sss.pgh.pa.us 5157 : 51 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
9 heikki.linnakangas@i 5158 :GNC 51 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB);
4033 alvherre@alvh.no-ip. 5159 :CBC 51 : SetCommitTsLimit(InvalidTransactionId, InvalidTransactionId);
5160 : :
5161 : : /* Set up the XLOG page header */
9570 vadim4o@yahoo.com 5162 : 51 : page->xlp_magic = XLOG_PAGE_MAGIC;
7820 tgl@sss.pgh.pa.us 5163 : 51 : page->xlp_info = XLP_LONG_HEADER;
1504 rhaas@postgresql.org 5164 : 51 : page->xlp_tli = BootstrapTimeLineID;
3012 andres@anarazel.de 5165 : 51 : page->xlp_pageaddr = wal_segment_size;
7820 tgl@sss.pgh.pa.us 5166 : 51 : longpage = (XLogLongPageHeader) page;
5167 : 51 : longpage->xlp_sysid = sysidentifier;
3012 andres@anarazel.de 5168 : 51 : longpage->xlp_seg_size = wal_segment_size;
7197 tgl@sss.pgh.pa.us 5169 : 51 : longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5170 : :
5171 : : /* Insert the initial checkpoint record */
4046 heikki.linnakangas@i 5172 : 51 : recptr = ((char *) page + SizeOfXLogLongPHD);
5173 : 51 : record = (XLogRecord *) recptr;
4925 5174 : 51 : record->xl_prev = 0;
9570 vadim4o@yahoo.com 5175 : 51 : record->xl_xid = InvalidTransactionId;
4046 heikki.linnakangas@i 5176 : 51 : record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
9046 tgl@sss.pgh.pa.us 5177 : 51 : record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
9570 vadim4o@yahoo.com 5178 : 51 : record->xl_rmid = RM_XLOG_ID;
4046 heikki.linnakangas@i 5179 : 51 : recptr += SizeOfXLogRecord;
5180 : : /* fill the XLogRecordDataHeaderShort struct */
3187 tgl@sss.pgh.pa.us 5181 : 51 : *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
4046 heikki.linnakangas@i 5182 : 51 : *(recptr++) = sizeof(checkPoint);
5183 : 51 : memcpy(recptr, &checkPoint, sizeof(checkPoint));
5184 : 51 : recptr += sizeof(checkPoint);
5185 [ - + ]: 51 : Assert(recptr - (char *) record == record->xl_tot_len);
5186 : :
4062 5187 : 51 : INIT_CRC32C(crc);
4046 5188 : 51 : COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
4062 5189 : 51 : COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5190 : 51 : FIN_CRC32C(crc);
9121 vadim4o@yahoo.com 5191 : 51 : record->xl_crc = crc;
5192 : :
5193 : : /* Create first XLOG segment file */
1504 rhaas@postgresql.org 5194 : 51 : openLogTLI = BootstrapTimeLineID;
5195 : 51 : openLogFile = XLogFileInit(1, BootstrapTimeLineID);
5196 : :
5197 : : /*
5198 : : * We needn't bother with Reserve/ReleaseExternalFD here, since we'll
5199 : : * close the file again in a moment.
5200 : : */
5201 : :
5202 : : /* Write the first page with the initial record */
8961 tgl@sss.pgh.pa.us 5203 : 51 : errno = 0;
3197 rhaas@postgresql.org 5204 : 51 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
10 peter@eisentraut.org 5205 [ - + ]:GNC 51 : if (write(openLogFile, &buffer, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5206 : : {
5207 : : /* if write didn't set errno, assume problem is no disk space */
8961 tgl@sss.pgh.pa.us 5208 [ # # ]:UBC 0 : if (errno == 0)
5209 : 0 : errno = ENOSPC;
8186 5210 [ # # ]: 0 : ereport(PANIC,
5211 : : (errcode_for_file_access(),
5212 : : errmsg("could not write bootstrap write-ahead log file: %m")));
5213 : : }
3197 rhaas@postgresql.org 5214 :CBC 51 : pgstat_report_wait_end();
5215 : :
5216 : 51 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
9046 tgl@sss.pgh.pa.us 5217 [ - + ]: 51 : if (pg_fsync(openLogFile) != 0)
8186 tgl@sss.pgh.pa.us 5218 [ # # ]:UBC 0 : ereport(PANIC,
5219 : : (errcode_for_file_access(),
5220 : : errmsg("could not fsync bootstrap write-ahead log file: %m")));
3197 rhaas@postgresql.org 5221 :CBC 51 : pgstat_report_wait_end();
5222 : :
2357 peter@eisentraut.org 5223 [ - + ]: 51 : if (close(openLogFile) != 0)
7997 tgl@sss.pgh.pa.us 5224 [ # # ]:UBC 0 : ereport(PANIC,
5225 : : (errcode_for_file_access(),
5226 : : errmsg("could not close bootstrap write-ahead log file: %m")));
5227 : :
9046 tgl@sss.pgh.pa.us 5228 :CBC 51 : openLogFile = -1;
5229 : :
5230 : : /* Now create pg_control */
513 peter@eisentraut.org 5231 : 51 : InitControlFile(sysidentifier, data_checksum_version);
9046 tgl@sss.pgh.pa.us 5232 : 51 : ControlFile->time = checkPoint.time;
9570 vadim4o@yahoo.com 5233 : 51 : ControlFile->checkPoint = checkPoint.redo;
9046 tgl@sss.pgh.pa.us 5234 : 51 : ControlFile->checkPointCopy = checkPoint;
5235 : :
5236 : : /* some additional ControlFile fields are set in WriteControlFile() */
9154 5237 : 51 : WriteControlFile();
5238 : :
5239 : : /* Bootstrap the commit log, too */
8881 5240 : 51 : BootStrapCLOG();
4033 alvherre@alvh.no-ip. 5241 : 51 : BootStrapCommitTs();
7840 tgl@sss.pgh.pa.us 5242 : 51 : BootStrapSUBTRANS();
7539 5243 : 51 : BootStrapMultiXact();
5244 : :
5245 : : /*
5246 : : * Force control file to be read - in contrast to normal processing we'd
5247 : : * otherwise never run the checks and GUC related initializations therein.
5248 : : */
3018 andres@anarazel.de 5249 : 51 : ReadControlFile();
9570 vadim4o@yahoo.com 5250 : 51 : }
5251 : :
5252 : : static char *
138 tgl@sss.pgh.pa.us 5253 :GNC 815 : str_time(pg_time_t tnow, char *buf, size_t bufsize)
5254 : : {
5255 : 815 : pg_strftime(buf, bufsize,
5256 : : "%Y-%m-%d %H:%M:%S %Z",
6711 tgl@sss.pgh.pa.us 5257 :CBC 815 : pg_localtime(&tnow, log_timezone));
5258 : :
9158 peter_e@gmx.net 5259 : 815 : return buf;
5260 : : }
5261 : :
5262 : : /*
5263 : : * Initialize the first WAL segment on new timeline.
5264 : : */
5265 : : static void
1401 heikki.linnakangas@i 5266 : 49 : XLogInitNewTimeline(TimeLineID endTLI, XLogRecPtr endOfLog, TimeLineID newTLI)
5267 : : {
5268 : : char xlogfname[MAXFNAMELEN];
5269 : : XLogSegNo endLogSegNo;
5270 : : XLogSegNo startLogSegNo;
5271 : :
5272 : : /* we always switch to a new timeline after archive recovery */
1504 rhaas@postgresql.org 5273 [ - + ]: 49 : Assert(endTLI != newTLI);
5274 : :
5275 : : /*
5276 : : * Update min recovery point one last time.
5277 : : */
6020 heikki.linnakangas@i 5278 : 49 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5279 : :
5280 : : /*
5281 : : * Calculate the last segment on the old timeline, and the first segment
5282 : : * on the new timeline. If the switch happens in the middle of a segment,
5283 : : * they are the same, but if the switch happens exactly at a segment
5284 : : * boundary, startLogSegNo will be endLogSegNo + 1.
5285 : : */
3012 andres@anarazel.de 5286 : 49 : XLByteToPrevSeg(endOfLog, endLogSegNo, wal_segment_size);
5287 : 49 : XLByteToSeg(endOfLog, startLogSegNo, wal_segment_size);
5288 : :
5289 : : /*
5290 : : * Initialize the starting WAL segment for the new timeline. If the switch
5291 : : * happens in the middle of a segment, copy data from the last WAL segment
5292 : : * of the old timeline up to the switch point, to the starting WAL segment
5293 : : * on the new timeline.
5294 : : */
4018 heikki.linnakangas@i 5295 [ + + ]: 49 : if (endLogSegNo == startLogSegNo)
5296 : : {
5297 : : /*
5298 : : * Make a copy of the file on the new timeline.
5299 : : *
5300 : : * Writing WAL isn't allowed yet, so there are no locking
5301 : : * considerations. But we should be just as tense as XLogFileInit to
5302 : : * avoid emplacing a bogus file.
5303 : : */
1504 rhaas@postgresql.org 5304 : 39 : XLogFileCopy(newTLI, endLogSegNo, endTLI, endLogSegNo,
3012 andres@anarazel.de 5305 : 39 : XLogSegmentOffset(endOfLog, wal_segment_size));
5306 : : }
5307 : : else
5308 : : {
5309 : : /*
5310 : : * The switch happened at a segment boundary, so just create the next
5311 : : * segment on the new timeline.
5312 : : */
5313 : : int fd;
5314 : :
1504 rhaas@postgresql.org 5315 : 10 : fd = XLogFileInit(startLogSegNo, newTLI);
5316 : :
2357 peter@eisentraut.org 5317 [ - + ]: 10 : if (close(fd) != 0)
5318 : : {
2207 michael@paquier.xyz 5319 :UBC 0 : int save_errno = errno;
5320 : :
1504 rhaas@postgresql.org 5321 : 0 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
2207 michael@paquier.xyz 5322 : 0 : errno = save_errno;
4015 heikki.linnakangas@i 5323 [ # # ]: 0 : ereport(ERROR,
5324 : : (errcode_for_file_access(),
5325 : : errmsg("could not close file \"%s\": %m", xlogfname)));
5326 : : }
5327 : : }
5328 : :
5329 : : /*
5330 : : * Let's just make real sure there are not .ready or .done flags posted
5331 : : * for the new segment.
5332 : : */
1504 rhaas@postgresql.org 5333 :CBC 49 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
4074 fujii@postgresql.org 5334 : 49 : XLogArchiveCleanup(xlogfname);
7822 tgl@sss.pgh.pa.us 5335 : 49 : }
5336 : :
5337 : : /*
5338 : : * Perform cleanup actions at the conclusion of archive recovery.
5339 : : */
5340 : : static void
1504 rhaas@postgresql.org 5341 : 49 : CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog,
5342 : : TimeLineID newTLI)
5343 : : {
5344 : : /*
5345 : : * Execute the recovery_end_command, if any.
5346 : : */
1527 5347 [ + - + + ]: 49 : if (recoveryEndCommand && strcmp(recoveryEndCommand, "") != 0)
1046 michael@paquier.xyz 5348 : 2 : ExecuteRecoveryCommand(recoveryEndCommand,
5349 : : "recovery_end_command",
5350 : : true,
5351 : : WAIT_EVENT_RECOVERY_END_COMMAND);
5352 : :
5353 : : /*
5354 : : * We switched to a new timeline. Clean up segments on the old timeline.
5355 : : *
5356 : : * If there are any higher-numbered segments on the old timeline, remove
5357 : : * them. They might contain valid WAL, but they might also be
5358 : : * pre-allocated files containing garbage. In any case, they are not part
5359 : : * of the new timeline's history so we don't need them.
5360 : : */
1504 rhaas@postgresql.org 5361 : 49 : RemoveNonParentXlogFiles(EndOfLog, newTLI);
5362 : :
5363 : : /*
5364 : : * If the switch happened in the middle of a segment, what to do with the
5365 : : * last, partial segment on the old timeline? If we don't archive it, and
5366 : : * the server that created the WAL never archives it either (e.g. because
5367 : : * it was hit by a meteor), it will never make it to the archive. That's
5368 : : * OK from our point of view, because the new segment that we created with
5369 : : * the new TLI contains all the WAL from the old timeline up to the switch
5370 : : * point. But if you later try to do PITR to the "missing" WAL on the old
5371 : : * timeline, recovery won't find it in the archive. It's physically
5372 : : * present in the new file with new TLI, but recovery won't look there
5373 : : * when it's recovering to the older timeline. On the other hand, if we
5374 : : * archive the partial segment, and the original server on that timeline
5375 : : * is still running and archives the completed version of the same segment
5376 : : * later, it will fail. (We used to do that in 9.4 and below, and it
5377 : : * caused such problems).
5378 : : *
5379 : : * As a compromise, we rename the last segment with the .partial suffix,
5380 : : * and archive it. Archive recovery will never try to read .partial
5381 : : * segments, so they will normally go unused. But in the odd PITR case,
5382 : : * the administrator can copy them manually to the pg_wal directory
5383 : : * (removing the suffix). They can be useful in debugging, too.
5384 : : *
5385 : : * If a .done or .ready file already exists for the old timeline, however,
5386 : : * we had already determined that the segment is complete, so we can let
5387 : : * it be archived normally. (In particular, if it was restored from the
5388 : : * archive to begin with, it's expected to have a .done file).
5389 : : */
1527 5390 [ + + + + ]: 88 : if (XLogSegmentOffset(EndOfLog, wal_segment_size) != 0 &&
5391 [ + + - + ]: 39 : XLogArchivingActive())
5392 : : {
5393 : : char origfname[MAXFNAMELEN];
5394 : : XLogSegNo endLogSegNo;
5395 : :
5396 : 9 : XLByteToPrevSeg(EndOfLog, endLogSegNo, wal_segment_size);
5397 : 9 : XLogFileName(origfname, EndOfLogTLI, endLogSegNo, wal_segment_size);
5398 : :
5399 [ + + ]: 9 : if (!XLogArchiveIsReadyOrDone(origfname))
5400 : : {
5401 : : char origpath[MAXPGPATH];
5402 : : char partialfname[MAXFNAMELEN];
5403 : : char partialpath[MAXPGPATH];
5404 : :
5405 : : /*
5406 : : * If we're summarizing WAL, we can't rename the partial file
5407 : : * until the summarizer finishes with it, else it will fail.
5408 : : */
510 5409 [ + + ]: 5 : if (summarize_wal)
5410 : 1 : WaitForWalSummarization(EndOfLog);
5411 : :
1527 5412 : 5 : XLogFilePath(origpath, EndOfLogTLI, endLogSegNo, wal_segment_size);
5413 : 5 : snprintf(partialfname, MAXFNAMELEN, "%s.partial", origfname);
5414 : 5 : snprintf(partialpath, MAXPGPATH, "%s.partial", origpath);
5415 : :
5416 : : /*
5417 : : * Make sure there's no .done or .ready file for the .partial
5418 : : * file.
5419 : : */
5420 : 5 : XLogArchiveCleanup(partialfname);
5421 : :
5422 : 5 : durable_rename(origpath, partialpath, ERROR);
5423 : 5 : XLogArchiveNotify(partialfname);
5424 : : }
5425 : : }
5426 : 49 : }
5427 : :
5428 : : /*
5429 : : * Check to see if required parameters are set high enough on this server
5430 : : * for various aspects of recovery operation.
5431 : : *
5432 : : * Note that all the parameters which this function tests need to be
5433 : : * listed in Administrator's Overview section in high-availability.sgml.
5434 : : * If you change them, don't forget to update the list.
5435 : : */
5436 : : static void
1401 heikki.linnakangas@i 5437 : 238 : CheckRequiredParameterValues(void)
5438 : : {
5439 : : /*
5440 : : * For archive recovery, the WAL must be generated with at least 'replica'
5441 : : * wal_level.
5442 : : */
5443 [ + + + + ]: 238 : if (ArchiveRecoveryRequested && ControlFile->wal_level == WAL_LEVEL_MINIMAL)
5444 : : {
5445 [ + - ]: 2 : ereport(FATAL,
5446 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5447 : : errmsg("WAL was generated with \"wal_level=minimal\", cannot continue recovering"),
5448 : : errdetail("This happens if you temporarily set \"wal_level=minimal\" on the server."),
5449 : : errhint("Use a backup taken after setting \"wal_level\" to higher than \"minimal\".")));
5450 : : }
5451 : :
5452 : : /*
5453 : : * For Hot Standby, the WAL must be generated with 'replica' mode, and we
5454 : : * must have at least as many backend slots as the primary.
5455 : : */
4306 5456 [ + + + + ]: 236 : if (ArchiveRecoveryRequested && EnableHotStandby)
5457 : : {
5458 : : /* We ignore autovacuum_worker_slots when we make this test. */
5713 5459 : 118 : RecoveryRequiresIntParameter("max_connections",
5460 : : MaxConnections,
5712 tgl@sss.pgh.pa.us 5461 : 118 : ControlFile->MaxConnections);
4550 rhaas@postgresql.org 5462 : 118 : RecoveryRequiresIntParameter("max_worker_processes",
5463 : : max_worker_processes,
5464 : 118 : ControlFile->max_worker_processes);
2501 michael@paquier.xyz 5465 : 118 : RecoveryRequiresIntParameter("max_wal_senders",
5466 : : max_wal_senders,
5467 : 118 : ControlFile->max_wal_senders);
4852 tgl@sss.pgh.pa.us 5468 : 118 : RecoveryRequiresIntParameter("max_prepared_transactions",
5469 : : max_prepared_xacts,
5712 5470 : 118 : ControlFile->max_prepared_xacts);
4852 5471 : 118 : RecoveryRequiresIntParameter("max_locks_per_transaction",
5472 : : max_locks_per_xact,
5712 5473 : 118 : ControlFile->max_locks_per_xact);
5474 : : }
5843 simon@2ndQuadrant.co 5475 : 236 : }
5476 : :
5477 : : /*
5478 : : * This must be called ONCE during postmaster or standalone-backend startup
5479 : : */
5480 : : void
9046 tgl@sss.pgh.pa.us 5481 : 927 : StartupXLOG(void)
5482 : : {
5483 : : XLogCtlInsert *Insert;
5484 : : CheckPoint checkPoint;
5485 : : bool wasShutdown;
5486 : : bool didCrash;
5487 : : bool haveTblspcMap;
5488 : : bool haveBackupLabel;
5489 : : XLogRecPtr EndOfLog;
5490 : : TimeLineID EndOfLogTLI;
5491 : : TimeLineID newTLI;
5492 : : bool performedWalRecovery;
5493 : : EndOfWalRecoveryInfo *endOfRecoveryInfo;
5494 : : XLogRecPtr abortedRecPtr;
5495 : : XLogRecPtr missingContrecPtr;
5496 : : TransactionId oldestActiveXID;
1968 fujii@postgresql.org 5497 : 927 : bool promoted = false;
5498 : : char timebuf[128];
5499 : :
5500 : : /*
5501 : : * We should have an aux process resource owner to use, and we should not
5502 : : * be in a transaction that's installed some other resowner.
5503 : : */
2710 tgl@sss.pgh.pa.us 5504 [ - + ]: 927 : Assert(AuxProcessResourceOwner != NULL);
5505 [ + - - + ]: 927 : Assert(CurrentResourceOwner == NULL ||
5506 : : CurrentResourceOwner == AuxProcessResourceOwner);
5507 : 927 : CurrentResourceOwner = AuxProcessResourceOwner;
5508 : :
5509 : : /*
5510 : : * Check that contents look valid.
5511 : : */
2232 peter@eisentraut.org 5512 [ - + ]: 927 : if (!XRecOffIsValid(ControlFile->checkPoint))
8186 tgl@sss.pgh.pa.us 5513 [ # # ]:UBC 0 : ereport(FATAL,
5514 : : (errcode(ERRCODE_DATA_CORRUPTED),
5515 : : errmsg("control file contains invalid checkpoint location")));
5516 : :
2232 peter@eisentraut.org 5517 [ + + - - :CBC 927 : switch (ControlFile->state)
+ + - ]
5518 : : {
5519 : 727 : case DB_SHUTDOWNED:
5520 : :
5521 : : /*
5522 : : * This is the expected case, so don't be chatty in standalone
5523 : : * mode
5524 : : */
5525 [ + + + + ]: 727 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
5526 : : (errmsg("database system was shut down at %s",
5527 : : str_time(ControlFile->time,
5528 : : timebuf, sizeof(timebuf)))));
5529 : 727 : break;
5530 : :
5531 : 28 : case DB_SHUTDOWNED_IN_RECOVERY:
5532 [ + - ]: 28 : ereport(LOG,
5533 : : (errmsg("database system was shut down in recovery at %s",
5534 : : str_time(ControlFile->time,
5535 : : timebuf, sizeof(timebuf)))));
5536 : 28 : break;
5537 : :
2232 peter@eisentraut.org 5538 :UBC 0 : case DB_SHUTDOWNING:
5539 [ # # ]: 0 : ereport(LOG,
5540 : : (errmsg("database system shutdown was interrupted; last known up at %s",
5541 : : str_time(ControlFile->time,
5542 : : timebuf, sizeof(timebuf)))));
5543 : 0 : break;
5544 : :
5545 : 0 : case DB_IN_CRASH_RECOVERY:
5546 [ # # ]: 0 : ereport(LOG,
5547 : : (errmsg("database system was interrupted while in recovery at %s",
5548 : : str_time(ControlFile->time,
5549 : : timebuf, sizeof(timebuf))),
5550 : : errhint("This probably means that some data is corrupted and"
5551 : : " you will have to use the last backup for recovery.")));
5552 : 0 : break;
5553 : :
2232 peter@eisentraut.org 5554 :CBC 6 : case DB_IN_ARCHIVE_RECOVERY:
5555 [ + - ]: 6 : ereport(LOG,
5556 : : (errmsg("database system was interrupted while in recovery at log time %s",
5557 : : str_time(ControlFile->checkPointCopy.time,
5558 : : timebuf, sizeof(timebuf))),
5559 : : errhint("If this has occurred more than once some data might be corrupted"
5560 : : " and you might need to choose an earlier recovery target.")));
5561 : 6 : break;
5562 : :
5563 : 166 : case DB_IN_PRODUCTION:
5564 [ + - ]: 166 : ereport(LOG,
5565 : : (errmsg("database system was interrupted; last known up at %s",
5566 : : str_time(ControlFile->time,
5567 : : timebuf, sizeof(timebuf)))));
5568 : 166 : break;
5569 : :
2232 peter@eisentraut.org 5570 :UBC 0 : default:
5571 [ # # ]: 0 : ereport(FATAL,
5572 : : (errcode(ERRCODE_DATA_CORRUPTED),
5573 : : errmsg("control file contains invalid database cluster state")));
5574 : : }
5575 : :
5576 : : /* This is just to allow attaching to startup process with a debugger */
5577 : : #ifdef XLOG_REPLAY_DELAY
5578 : : if (ControlFile->state != DB_SHUTDOWNED)
5579 : : pg_usleep(60000000L);
5580 : : #endif
5581 : :
5582 : : /*
5583 : : * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
5584 : : * In cases where someone has performed a copy for PITR, these directories
5585 : : * may have been excluded and need to be re-created.
5586 : : */
6248 tgl@sss.pgh.pa.us 5587 :CBC 927 : ValidateXLOGDirectoryStructure();
5588 : :
5589 : : /* Set up timeout handler needed to report startup progress. */
1515 rhaas@postgresql.org 5590 [ + + ]: 927 : if (!IsBootstrapProcessingMode())
5591 : 876 : RegisterTimeout(STARTUP_PROGRESS_TIMEOUT,
5592 : : startup_progress_timeout_handler);
5593 : :
5594 : : /*----------
5595 : : * If we previously crashed, perform a couple of actions:
5596 : : *
5597 : : * - The pg_wal directory may still include some temporary WAL segments
5598 : : * used when creating a new segment, so perform some clean up to not
5599 : : * bloat this path. This is done first as there is no point to sync
5600 : : * this temporary data.
5601 : : *
5602 : : * - There might be data which we had written, intending to fsync it, but
5603 : : * which we had not actually fsync'd yet. Therefore, a power failure in
5604 : : * the near future might cause earlier unflushed writes to be lost, even
5605 : : * though more recent data written to disk from here on would be
5606 : : * persisted. To avoid that, fsync the entire data directory.
5607 : : */
1401 heikki.linnakangas@i 5608 [ + + ]: 927 : if (ControlFile->state != DB_SHUTDOWNED &&
5609 [ + + ]: 200 : ControlFile->state != DB_SHUTDOWNED_IN_RECOVERY)
5610 : : {
5611 : 172 : RemoveTempXlogFiles();
5612 : 172 : SyncDataDirectory();
1352 andres@anarazel.de 5613 : 172 : didCrash = true;
5614 : : }
5615 : : else
5616 : 755 : didCrash = false;
5617 : :
5618 : : /*
5619 : : * Prepare for WAL recovery if needed.
5620 : : *
5621 : : * InitWalRecovery analyzes the control file and the backup label file, if
5622 : : * any. It updates the in-memory ControlFile buffer according to the
5623 : : * starting checkpoint, and sets InRecovery and ArchiveRecoveryRequested.
5624 : : * It also applies the tablespace map file, if any.
5625 : : */
1401 heikki.linnakangas@i 5626 : 927 : InitWalRecovery(ControlFile, &wasShutdown,
5627 : : &haveBackupLabel, &haveTblspcMap);
5628 : 927 : checkPoint = ControlFile->checkPointCopy;
5629 : :
5630 : : /* initialize shared memory variables from the checkpoint record */
741 5631 : 927 : TransamVariables->nextXid = checkPoint.nextXid;
5632 : 927 : TransamVariables->nextOid = checkPoint.nextOid;
5633 : 927 : TransamVariables->oidCount = 0;
7498 tgl@sss.pgh.pa.us 5634 : 927 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
3192 rhaas@postgresql.org 5635 : 927 : AdvanceOldestClogXid(checkPoint.oldestXid);
5783 tgl@sss.pgh.pa.us 5636 : 927 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
9 heikki.linnakangas@i 5637 :GNC 927 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB);
3643 mail@joeconway.com 5638 :CBC 927 : SetCommitTsLimit(checkPoint.oldestCommitTsXid,
5639 : : checkPoint.newestCommitTsXid);
5640 : :
5641 : : /*
5642 : : * Clear out any old relcache cache files. This is *necessary* if we do
5643 : : * any WAL replay, since that would probably result in the cache files
5644 : : * being out of sync with database reality. In theory we could leave them
5645 : : * in place if the database had been cleanly shut down, but it seems
5646 : : * safest to just remove them always and let them be rebuilt during the
5647 : : * first backend startup. These files needs to be removed from all
5648 : : * directories including pg_tblspc, however the symlinks are created only
5649 : : * after reading tablespace_map file in case of archive recovery from
5650 : : * backup, so needs to clear old relcache files here after creating
5651 : : * symlinks.
5652 : : */
1401 heikki.linnakangas@i 5653 : 927 : RelationCacheInitFileRemove();
5654 : :
5655 : : /*
5656 : : * Initialize replication slots, before there's a chance to remove
5657 : : * required resources.
5658 : : */
4207 andres@anarazel.de 5659 : 927 : StartupReplicationSlots();
5660 : :
5661 : : /*
5662 : : * Startup logical state, needs to be setup now so we have proper data
5663 : : * during crash recovery.
5664 : : */
4308 rhaas@postgresql.org 5665 : 927 : StartupReorderBuffer();
5666 : :
5667 : : /*
5668 : : * Startup CLOG. This must be done after TransamVariables->nextXid has
5669 : : * been initialized and before we accept connections or begin WAL replay.
5670 : : */
1786 5671 : 927 : StartupCLOG();
5672 : :
5673 : : /*
5674 : : * Startup MultiXact. We need to do this early to be able to replay
5675 : : * truncations.
5676 : : */
4402 alvherre@alvh.no-ip. 5677 : 927 : StartupMultiXact();
5678 : :
5679 : : /*
5680 : : * Ditto for commit timestamps. Activate the facility if the setting is
5681 : : * enabled in the control file, as there should be no tracking of commit
5682 : : * timestamps done when the setting was disabled. This facility can be
5683 : : * started or stopped when replaying a XLOG_PARAMETER_CHANGE record.
5684 : : */
2640 michael@paquier.xyz 5685 [ + + ]: 927 : if (ControlFile->track_commit_timestamp)
3660 alvherre@alvh.no-ip. 5686 : 13 : StartupCommitTs();
5687 : :
5688 : : /*
5689 : : * Recover knowledge about replay progress of known replication partners.
5690 : : */
3886 andres@anarazel.de 5691 : 927 : StartupReplicationOrigin();
5692 : :
5693 : : /*
5694 : : * Initialize unlogged LSN. On a clean shutdown, it's restored from the
5695 : : * control file. On recovery, all unlogged relations are blown away, so
5696 : : * the unlogged LSN counter can be reset too.
5697 : : */
4693 heikki.linnakangas@i 5698 [ + + ]: 927 : if (ControlFile->state == DB_SHUTDOWNED)
658 nathan@postgresql.or 5699 : 720 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5700 : 720 : ControlFile->unloggedLSN);
5701 : : else
5702 : 207 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5703 : : FirstNormalUnloggedLSN);
5704 : :
5705 : : /*
5706 : : * Copy any missing timeline history files between 'now' and the recovery
5707 : : * target timeline from archive to pg_wal. While we don't need those files
5708 : : * ourselves - the history file of the recovery target timeline covers all
5709 : : * the previous timelines in the history too - a cascading standby server
5710 : : * might be interested in them. Or, if you archive the WAL from this
5711 : : * server to a different archive than the primary, it'd be good for all
5712 : : * the history files to get archived there after failover, so that you can
5713 : : * use one of the old timelines as a PITR target. Timeline history files
5714 : : * are small, so it's better to copy them unnecessarily than not copy them
5715 : : * and regret later.
5716 : : */
1401 heikki.linnakangas@i 5717 : 927 : restoreTimeLineHistoryFiles(checkPoint.ThisTimeLineID, recoveryTargetTLI);
5718 : :
5719 : : /*
5720 : : * Before running in recovery, scan pg_twophase and fill in its status to
5721 : : * be able to work on entries generated by redo. Doing a scan before
5722 : : * taking any recovery action has the merit to discard any 2PC files that
5723 : : * are newer than the first record to replay, saving from any conflicts at
5724 : : * replay. This avoids as well any subsequent scans when doing recovery
5725 : : * of the on-disk two-phase data.
5726 : : */
3180 simon@2ndQuadrant.co 5727 : 927 : restoreTwoPhaseData();
5728 : :
5729 : : /*
5730 : : * When starting with crash recovery, reset pgstat data - it might not be
5731 : : * valid. Otherwise restore pgstat data. It's safe to do this here,
5732 : : * because postmaster will not yet have started any other processes.
5733 : : *
5734 : : * NB: Restoring replication slot stats relies on slot state to have
5735 : : * already been restored from disk.
5736 : : *
5737 : : * TODO: With a bit of extra work we could just start with a pgstat file
5738 : : * associated with the checkpoint redo location we're starting from.
5739 : : */
1352 andres@anarazel.de 5740 [ + + ]: 927 : if (didCrash)
5741 : 172 : pgstat_discard_stats();
5742 : : else
276 michael@paquier.xyz 5743 : 755 : pgstat_restore_stats();
5744 : :
5076 simon@2ndQuadrant.co 5745 : 927 : lastFullPageWrites = checkPoint.fullPageWrites;
5746 : :
4546 heikki.linnakangas@i 5747 : 927 : RedoRecPtr = XLogCtl->RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
4060 5748 : 927 : doPageWrites = lastFullPageWrites;
5749 : :
5750 : : /* REDO */
7806 tgl@sss.pgh.pa.us 5751 [ + + ]: 927 : if (InRecovery)
5752 : : {
5753 : : /* Initialize state for RecoveryInProgress() */
1401 heikki.linnakangas@i 5754 [ - + ]: 207 : SpinLockAcquire(&XLogCtl->info_lck);
5755 [ + + ]: 207 : if (InArchiveRecovery)
5756 : 108 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
5757 : : else
5758 : 99 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
5759 : 207 : SpinLockRelease(&XLogCtl->info_lck);
5760 : :
5761 : : /*
5762 : : * Update pg_control to show that we are recovering and to show the
5763 : : * selected checkpoint as the place we are starting from. We also mark
5764 : : * pg_control with any minimum recovery stop point obtained from a
5765 : : * backup history file.
5766 : : *
5767 : : * No need to hold ControlFileLock yet, we aren't up far enough.
5768 : : */
5769 : 207 : UpdateControlFile();
5770 : :
5771 : : /*
5772 : : * If there was a backup label file, it's done its job and the info
5773 : : * has now been propagated into pg_control. We must get rid of the
5774 : : * label file so that if we crash during recovery, we'll pick up at
5775 : : * the latest recovery restartpoint instead of going all the way back
5776 : : * to the backup start point. It seems prudent though to just rename
5777 : : * the file out of the way rather than delete it completely.
5778 : : */
5779 [ + + ]: 207 : if (haveBackupLabel)
5780 : : {
5781 : 71 : unlink(BACKUP_LABEL_OLD);
5782 : 71 : durable_rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, FATAL);
5783 : : }
5784 : :
5785 : : /*
5786 : : * If there was a tablespace_map file, it's done its job and the
5787 : : * symlinks have been created. We must get rid of the map file so
5788 : : * that if we crash during recovery, we don't create symlinks again.
5789 : : * It seems prudent though to just rename the file out of the way
5790 : : * rather than delete it completely.
5791 : : */
5792 [ + + ]: 207 : if (haveTblspcMap)
5793 : : {
5794 : 2 : unlink(TABLESPACE_MAP_OLD);
5795 : 2 : durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, FATAL);
5796 : : }
5797 : :
5798 : : /*
5799 : : * Initialize our local copy of minRecoveryPoint. When doing crash
5800 : : * recovery we want to replay up to the end of WAL. Particularly, in
5801 : : * the case of a promoted standby minRecoveryPoint value in the
5802 : : * control file is only updated after the first checkpoint. However,
5803 : : * if the instance crashes before the first post-recovery checkpoint
5804 : : * is completed then recovery will use a stale location causing the
5805 : : * startup process to think that there are still invalid page
5806 : : * references when checking for data consistency.
5807 : : */
2723 michael@paquier.xyz 5808 [ + + ]: 207 : if (InArchiveRecovery)
5809 : : {
1401 heikki.linnakangas@i 5810 : 108 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
5811 : 108 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
5812 : : }
5813 : : else
5814 : : {
5815 : 99 : LocalMinRecoveryPoint = InvalidXLogRecPtr;
5816 : 99 : LocalMinRecoveryPointTLI = 0;
5817 : : }
5818 : :
5819 : : /* Check that the GUCs used to generate the WAL allow recovery */
5713 5820 : 207 : CheckRequiredParameterValues();
5821 : :
5822 : : /*
5823 : : * We're in recovery, so unlogged relations may be trashed and must be
5824 : : * reset. This should be done BEFORE allowing Hot Standby
5825 : : * connections, so that read-only backends don't try to read whatever
5826 : : * garbage is left over from before.
5827 : : */
5468 rhaas@postgresql.org 5828 : 207 : ResetUnloggedRelations(UNLOGGED_RELATION_CLEANUP);
5829 : :
5830 : : /*
5831 : : * Likewise, delete any saved transaction snapshot files that got left
5832 : : * behind by crashed backends.
5833 : : */
5171 tgl@sss.pgh.pa.us 5834 : 207 : DeleteAllExportedSnapshotFiles();
5835 : :
5836 : : /*
5837 : : * Initialize for Hot Standby, if enabled. We won't let backends in
5838 : : * yet, not until we've reached the min recovery point specified in
5839 : : * control file and we've established a recovery snapshot from a
5840 : : * running-xacts WAL record.
5841 : : */
4682 heikki.linnakangas@i 5842 [ + + + + ]: 207 : if (ArchiveRecoveryRequested && EnableHotStandby)
5843 : : {
5844 : : TransactionId *xids;
5845 : : int nxids;
5846 : :
5788 5847 [ + + ]: 102 : ereport(DEBUG1,
5848 : : (errmsg_internal("initializing for hot standby")));
5849 : :
5843 simon@2ndQuadrant.co 5850 : 102 : InitRecoveryTransactionEnvironment();
5851 : :
5852 [ + + ]: 102 : if (wasShutdown)
5853 : 26 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
5854 : : else
5855 : 76 : oldestActiveXID = checkPoint.oldestActiveXid;
5856 [ - + ]: 102 : Assert(TransactionIdIsValid(oldestActiveXID));
5857 : :
5858 : : /* Tell procarray about the range of xids it has to deal with */
741 heikki.linnakangas@i 5859 : 102 : ProcArrayInitRecovery(XidFromFullTransactionId(TransamVariables->nextXid));
5860 : :
5861 : : /*
5862 : : * Startup subtrans only. CLOG, MultiXact and commit timestamp
5863 : : * have already been started up and other SLRUs are not maintained
5864 : : * during recovery and need not be started yet.
5865 : : */
5843 simon@2ndQuadrant.co 5866 : 102 : StartupSUBTRANS(oldestActiveXID);
5867 : :
5868 : : /*
5869 : : * If we're beginning at a shutdown checkpoint, we know that
5870 : : * nothing was running on the primary at this point. So fake-up an
5871 : : * empty running-xacts record and use that here and now. Recover
5872 : : * additional standby state for prepared transactions.
5873 : : */
5728 heikki.linnakangas@i 5874 [ + + ]: 102 : if (wasShutdown)
5875 : : {
5876 : : RunningTransactionsData running;
5877 : : TransactionId latestCompletedXid;
5878 : :
5879 : : /* Update pg_subtrans entries for any prepared transactions */
539 5880 : 26 : StandbyRecoverPreparedTransactions();
5881 : :
5882 : : /*
5883 : : * Construct a RunningTransactions snapshot representing a
5884 : : * shut down server, with only prepared transactions still
5885 : : * alive. We're never overflowed at this point because all
5886 : : * subxids are listed with their parent prepared transactions.
5887 : : */
5728 5888 : 26 : running.xcnt = nxids;
4764 simon@2ndQuadrant.co 5889 : 26 : running.subxcnt = 0;
539 heikki.linnakangas@i 5890 : 26 : running.subxid_status = SUBXIDS_IN_SUBTRANS;
1955 andres@anarazel.de 5891 : 26 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5728 heikki.linnakangas@i 5892 : 26 : running.oldestRunningXid = oldestActiveXID;
1955 andres@anarazel.de 5893 : 26 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5698 simon@2ndQuadrant.co 5894 [ - + ]: 26 : TransactionIdRetreat(latestCompletedXid);
5697 5895 [ - + ]: 26 : Assert(TransactionIdIsNormal(latestCompletedXid));
5698 5896 : 26 : running.latestCompletedXid = latestCompletedXid;
5728 heikki.linnakangas@i 5897 : 26 : running.xids = xids;
5898 : :
5899 : 26 : ProcArrayApplyRecoveryInfo(&running);
5900 : : }
5901 : : }
5902 : :
5903 : : /*
5904 : : * We're all set for replaying the WAL now. Do it.
5905 : : */
1401 5906 : 207 : PerformWalRecovery();
5907 : 152 : performedWalRecovery = true;
5908 : : }
5909 : : else
1397 5910 : 720 : performedWalRecovery = false;
5911 : :
5912 : : /*
5913 : : * Finish WAL recovery.
5914 : : */
1401 5915 : 872 : endOfRecoveryInfo = FinishWalRecovery();
5916 : 872 : EndOfLog = endOfRecoveryInfo->endOfLog;
5917 : 872 : EndOfLogTLI = endOfRecoveryInfo->endOfLogTLI;
5918 : 872 : abortedRecPtr = endOfRecoveryInfo->abortedRecPtr;
5919 : 872 : missingContrecPtr = endOfRecoveryInfo->missingContrecPtr;
5920 : :
5921 : : /*
5922 : : * Reset ps status display, so as no information related to recovery shows
5923 : : * up.
5924 : : */
1183 michael@paquier.xyz 5925 : 872 : set_ps_display("");
5926 : :
5927 : : /*
5928 : : * When recovering from a backup (we are in recovery, and archive recovery
5929 : : * was requested), complain if we did not roll forward far enough to reach
5930 : : * the point where the database is consistent. For regular online
5931 : : * backup-from-primary, that means reaching the end-of-backup WAL record
5932 : : * (at which point we reset backupStartPoint to be Invalid), for
5933 : : * backup-from-replica (which can't inject records into the WAL stream),
5934 : : * that point is when we reach the minRecoveryPoint in pg_control (which
5935 : : * we purposefully copy last when backing up from a replica). For
5936 : : * pg_rewind (which creates a backup_label with a method of "pg_rewind")
5937 : : * or snapshot-style backups (which don't), backupEndRequired will be set
5938 : : * to false.
5939 : : *
5940 : : * Note: it is indeed okay to look at the local variable
5941 : : * LocalMinRecoveryPoint here, even though ControlFile->minRecoveryPoint
5942 : : * might be further ahead --- ControlFile->minRecoveryPoint cannot have
5943 : : * been advanced beyond the WAL we processed.
5944 : : */
5377 heikki.linnakangas@i 5945 [ + + ]: 872 : if (InRecovery &&
1401 5946 [ + - ]: 152 : (EndOfLog < LocalMinRecoveryPoint ||
42 alvherre@kurilemu.de 5947 [ - + ]:GNC 152 : XLogRecPtrIsValid(ControlFile->backupStartPoint)))
5948 : : {
5949 : : /*
5950 : : * Ran off end of WAL before reaching end-of-backup WAL record, or
5951 : : * minRecoveryPoint. That's a bad sign, indicating that you tried to
5952 : : * recover from an online backup but never called pg_backup_stop(), or
5953 : : * you didn't archive all the WAL needed.
5954 : : */
4682 heikki.linnakangas@i 5955 [ # # # # ]:UBC 0 : if (ArchiveRecoveryRequested || ControlFile->backupEndRequired)
5956 : : {
42 alvherre@kurilemu.de 5957 [ # # # # ]:UNC 0 : if (XLogRecPtrIsValid(ControlFile->backupStartPoint) || ControlFile->backupEndRequired)
5244 heikki.linnakangas@i 5958 [ # # ]:UBC 0 : ereport(FATAL,
5959 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5960 : : errmsg("WAL ends before end of online backup"),
5961 : : errhint("All WAL generated while online backup was taken must be available at recovery.")));
5962 : : else
5363 5963 [ # # ]: 0 : ereport(FATAL,
5964 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5965 : : errmsg("WAL ends before consistent recovery point")));
5966 : : }
5967 : : }
5968 : :
5969 : : /*
5970 : : * Reset unlogged relations to the contents of their INIT fork. This is
5971 : : * done AFTER recovery is complete so as to include any unlogged relations
5972 : : * created during recovery, but BEFORE recovery is marked as having
5973 : : * completed successfully. Otherwise we'd not retry if any of the post
5974 : : * end-of-recovery steps fail.
5975 : : */
1401 heikki.linnakangas@i 5976 [ + + ]:CBC 872 : if (InRecovery)
5977 : 152 : ResetUnloggedRelations(UNLOGGED_RELATION_INIT);
5978 : :
5979 : : /*
5980 : : * Pre-scan prepared transactions to find out the range of XIDs present.
5981 : : * This information is not quite needed yet, but it is positioned here so
5982 : : * as potential problems are detected before any on-disk change is done.
5983 : : */
2719 michael@paquier.xyz 5984 : 872 : oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
5985 : :
5986 : : /*
5987 : : * Allow ordinary WAL segment creation before possibly switching to a new
5988 : : * timeline, which creates a new segment, and after the last ReadRecord().
5989 : : */
1219 5990 : 872 : SetInstallXLogFileSegmentActive();
5991 : :
5992 : : /*
5993 : : * Consider whether we need to assign a new timeline ID.
5994 : : *
5995 : : * If we did archive recovery, we always assign a new ID. This handles a
5996 : : * couple of issues. If we stopped short of the end of WAL during
5997 : : * recovery, then we are clearly generating a new timeline and must assign
5998 : : * it a unique new ID. Even if we ran to the end, modifying the current
5999 : : * last segment is problematic because it may result in trying to
6000 : : * overwrite an already-archived copy of that segment, and we encourage
6001 : : * DBAs to make their archive_commands reject that. We can dodge the
6002 : : * problem by making the new active segment have a new timeline ID.
6003 : : *
6004 : : * In a normal crash recovery, we can just extend the timeline we were in.
6005 : : */
1401 heikki.linnakangas@i 6006 : 872 : newTLI = endOfRecoveryInfo->lastRecTLI;
4682 6007 [ + + ]: 872 : if (ArchiveRecoveryRequested)
6008 : : {
1499 rhaas@postgresql.org 6009 : 49 : newTLI = findNewestTimeLine(recoveryTargetTLI) + 1;
7820 tgl@sss.pgh.pa.us 6010 [ + - ]: 49 : ereport(LOG,
6011 : : (errmsg("selected new timeline ID: %u", newTLI)));
6012 : :
6013 : : /*
6014 : : * Make a writable copy of the last WAL segment. (Note that we also
6015 : : * have a copy of the last block of the old WAL in
6016 : : * endOfRecovery->lastPage; we will use that below.)
6017 : : */
1401 heikki.linnakangas@i 6018 : 49 : XLogInitNewTimeline(EndOfLogTLI, EndOfLog, newTLI);
6019 : :
6020 : : /*
6021 : : * Remove the signal files out of the way, so that we don't
6022 : : * accidentally re-enter archive recovery mode in a subsequent crash.
6023 : : */
6024 [ + + ]: 49 : if (endOfRecoveryInfo->standby_signal_file_found)
6025 : 46 : durable_unlink(STANDBY_SIGNAL_FILE, FATAL);
6026 : :
6027 [ + + ]: 49 : if (endOfRecoveryInfo->recovery_signal_file_found)
6028 : 3 : durable_unlink(RECOVERY_SIGNAL_FILE, FATAL);
6029 : :
6030 : : /*
6031 : : * Write the timeline history file, and have it archived. After this
6032 : : * point (or rather, as soon as the file is archived), the timeline
6033 : : * will appear as "taken" in the WAL archive and to any standby
6034 : : * servers. If we crash before actually switching to the new
6035 : : * timeline, standby servers will nevertheless think that we switched
6036 : : * to the new timeline, and will try to connect to the new timeline.
6037 : : * To minimize the window for that, try to do as little as possible
6038 : : * between here and writing the end-of-recovery record.
6039 : : */
1499 rhaas@postgresql.org 6040 : 49 : writeTimeLineHistory(newTLI, recoveryTargetTLI,
6041 : : EndOfLog, endOfRecoveryInfo->recoveryStopReason);
6042 : :
1401 heikki.linnakangas@i 6043 [ + - ]: 49 : ereport(LOG,
6044 : : (errmsg("archive recovery complete")));
6045 : : }
6046 : :
6047 : : /* Save the selected TimeLineID in shared memory, too */
510 rhaas@postgresql.org 6048 [ - + ]: 872 : SpinLockAcquire(&XLogCtl->info_lck);
1499 6049 : 872 : XLogCtl->InsertTimeLineID = newTLI;
1401 heikki.linnakangas@i 6050 : 872 : XLogCtl->PrevTimeLineID = endOfRecoveryInfo->lastRecTLI;
510 rhaas@postgresql.org 6051 : 872 : SpinLockRelease(&XLogCtl->info_lck);
6052 : :
6053 : : /*
6054 : : * Actually, if WAL ended in an incomplete record, skip the parts that
6055 : : * made it through and start writing after the portion that persisted.
6056 : : * (It's critical to first write an OVERWRITE_CONTRECORD message, which
6057 : : * we'll do as soon as we're open for writing new WAL.)
6058 : : */
42 alvherre@kurilemu.de 6059 [ + + ]:GNC 872 : if (XLogRecPtrIsValid(missingContrecPtr))
6060 : : {
6061 : : /*
6062 : : * We should only have a missingContrecPtr if we're not switching to a
6063 : : * new timeline. When a timeline switch occurs, WAL is copied from the
6064 : : * old timeline to the new only up to the end of the last complete
6065 : : * record, so there can't be an incomplete WAL record that we need to
6066 : : * disregard.
6067 : : */
1207 rhaas@postgresql.org 6068 [ - + ]:CBC 10 : Assert(newTLI == endOfRecoveryInfo->lastRecTLI);
42 alvherre@kurilemu.de 6069 [ - + ]:GNC 10 : Assert(XLogRecPtrIsValid(abortedRecPtr));
1541 alvherre@alvh.no-ip. 6070 :CBC 10 : EndOfLog = missingContrecPtr;
6071 : : }
6072 : :
6073 : : /*
6074 : : * Prepare to write WAL starting at EndOfLog location, and init xlog
6075 : : * buffer cache using the block containing the last record from the
6076 : : * previous incarnation.
6077 : : */
9182 vadim4o@yahoo.com 6078 : 872 : Insert = &XLogCtl->Insert;
1401 heikki.linnakangas@i 6079 : 872 : Insert->PrevBytePos = XLogRecPtrToBytePos(endOfRecoveryInfo->lastRec);
4537 6080 : 872 : Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
6081 : :
6082 : : /*
6083 : : * Tricky point here: lastPage contains the *last* block that the LastRec
6084 : : * record spans, not the one it starts in. The last block is indeed the
6085 : : * one we want to use.
6086 : : */
6087 [ + + ]: 872 : if (EndOfLog % XLOG_BLCKSZ != 0)
6088 : : {
6089 : : char *page;
6090 : : int len;
6091 : : int firstIdx;
6092 : :
6093 : 845 : firstIdx = XLogRecPtrToBufIdx(EndOfLog);
1401 6094 : 845 : len = EndOfLog - endOfRecoveryInfo->lastPageBeginPtr;
6095 [ - + ]: 845 : Assert(len < XLOG_BLCKSZ);
6096 : :
6097 : : /* Copy the valid part of the last block, and zero the rest */
4537 6098 : 845 : page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
1401 6099 : 845 : memcpy(page, endOfRecoveryInfo->lastPage, len);
4537 6100 : 845 : memset(page + len, 0, XLOG_BLCKSZ - len);
6101 : :
730 jdavis@postgresql.or 6102 : 845 : pg_atomic_write_u64(&XLogCtl->xlblocks[firstIdx], endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ);
118 akorotkov@postgresql 6103 : 845 : XLogCtl->InitializedUpTo = endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ;
6104 : : }
6105 : : else
6106 : : {
6107 : : /*
6108 : : * There is no partial block to copy. Just set InitializedUpTo, and
6109 : : * let the first attempt to insert a log record to initialize the next
6110 : : * buffer.
6111 : : */
6112 : 27 : XLogCtl->InitializedUpTo = EndOfLog;
6113 : : }
6114 : :
6115 : : /*
6116 : : * Update local and shared status. This is OK to do without any locks
6117 : : * because no other process can be reading or writing WAL yet.
6118 : : */
4537 heikki.linnakangas@i 6119 : 872 : LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
620 alvherre@alvh.no-ip. 6120 : 872 : pg_atomic_write_u64(&XLogCtl->logInsertResult, EndOfLog);
622 6121 : 872 : pg_atomic_write_u64(&XLogCtl->logWriteResult, EndOfLog);
6122 : 872 : pg_atomic_write_u64(&XLogCtl->logFlushResult, EndOfLog);
4537 heikki.linnakangas@i 6123 : 872 : XLogCtl->LogwrtRqst.Write = EndOfLog;
6124 : 872 : XLogCtl->LogwrtRqst.Flush = EndOfLog;
6125 : :
6126 : : /*
6127 : : * Preallocate additional log files, if wanted.
6128 : : */
1499 rhaas@postgresql.org 6129 : 872 : PreallocXlogFiles(EndOfLog, newTLI);
6130 : :
6131 : : /*
6132 : : * Okay, we're officially UP.
6133 : : */
9182 vadim4o@yahoo.com 6134 : 872 : InRecovery = false;
6135 : :
6136 : : /* start the archive_timeout timer and LSN running */
4537 heikki.linnakangas@i 6137 : 872 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
3283 andres@anarazel.de 6138 : 872 : XLogCtl->lastSegSwitchLSN = EndOfLog;
6139 : :
6140 : : /* also initialize latestCompletedXid, to nextXid - 1 */
5064 tgl@sss.pgh.pa.us 6141 : 872 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
741 heikki.linnakangas@i 6142 : 872 : TransamVariables->latestCompletedXid = TransamVariables->nextXid;
6143 : 872 : FullTransactionIdRetreat(&TransamVariables->latestCompletedXid);
5064 tgl@sss.pgh.pa.us 6144 : 872 : LWLockRelease(ProcArrayLock);
6145 : :
6146 : : /*
6147 : : * Start up subtrans, if not already done for hot standby. (commit
6148 : : * timestamps are started below, if necessary.)
6149 : : */
5843 simon@2ndQuadrant.co 6150 [ + + ]: 872 : if (standbyState == STANDBY_DISABLED)
6151 : 823 : StartupSUBTRANS(oldestActiveXID);
6152 : :
6153 : : /*
6154 : : * Perform end of recovery actions for any SLRUs that need it.
6155 : : */
5160 6156 : 872 : TrimCLOG();
4402 alvherre@alvh.no-ip. 6157 : 872 : TrimMultiXact();
6158 : :
6159 : : /*
6160 : : * Reload shared-memory state for prepared transactions. This needs to
6161 : : * happen before renaming the last partial segment of the old timeline as
6162 : : * it may be possible that we have to recover some transactions from it.
6163 : : */
7489 tgl@sss.pgh.pa.us 6164 : 872 : RecoverPreparedTransactions();
6165 : :
6166 : : /* Shut down xlogreader */
1401 heikki.linnakangas@i 6167 : 872 : ShutdownWalRecovery();
6168 : :
6169 : : /* Enable WAL writes for this backend only. */
1526 rhaas@postgresql.org 6170 : 872 : LocalSetXLogInsertAllowed();
6171 : :
6172 : : /* If necessary, write overwrite-contrecord before doing anything else */
42 alvherre@kurilemu.de 6173 [ + + ]:GNC 872 : if (XLogRecPtrIsValid(abortedRecPtr))
6174 : : {
6175 [ - + ]: 10 : Assert(XLogRecPtrIsValid(missingContrecPtr));
1401 heikki.linnakangas@i 6176 :CBC 10 : CreateOverwriteContrecordRecord(abortedRecPtr, missingContrecPtr, newTLI);
6177 : : }
6178 : :
6179 : : /*
6180 : : * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
6181 : : * record before resource manager writes cleanup WAL records or checkpoint
6182 : : * record is written.
6183 : : */
1526 rhaas@postgresql.org 6184 : 872 : Insert->fullPageWrites = lastFullPageWrites;
6185 : 872 : UpdateFullPageWrites();
6186 : :
6187 : : /*
6188 : : * Emit checkpoint or end-of-recovery record in XLOG, if required.
6189 : : */
1401 heikki.linnakangas@i 6190 [ + + ]: 872 : if (performedWalRecovery)
1526 rhaas@postgresql.org 6191 : 152 : promoted = PerformRecoveryXLogAction();
6192 : :
6193 : : /*
6194 : : * If any of the critical GUCs have changed, log them before we allow
6195 : : * backends to write WAL.
6196 : : */
5713 heikki.linnakangas@i 6197 : 872 : XLogReportParameters();
6198 : :
6199 : : /* If this is archive recovery, perform post-recovery cleanup actions. */
1515 rhaas@postgresql.org 6200 [ + + ]: 872 : if (ArchiveRecoveryRequested)
1499 6201 : 49 : CleanupAfterArchiveRecovery(EndOfLogTLI, EndOfLog, newTLI);
6202 : :
6203 : : /*
6204 : : * Local WAL inserts enabled, so it's time to finish initialization of
6205 : : * commit timestamp.
6206 : : */
4033 alvherre@alvh.no-ip. 6207 : 872 : CompleteCommitTsInitialization();
6208 : :
6209 : : /* Clean up EndOfWalRecoveryInfo data to appease Valgrind leak checking */
138 tgl@sss.pgh.pa.us 6210 [ + + ]:GNC 872 : if (endOfRecoveryInfo->lastPage)
6211 : 855 : pfree(endOfRecoveryInfo->lastPage);
6212 : 872 : pfree(endOfRecoveryInfo->recoveryStopReason);
6213 : 872 : pfree(endOfRecoveryInfo);
6214 : :
6215 : : /*
6216 : : * All done with end-of-recovery actions.
6217 : : *
6218 : : * Now allow backends to write WAL and update the control file status in
6219 : : * consequence. SharedRecoveryState, that controls if backends can write
6220 : : * WAL, is updated while holding ControlFileLock to prevent other backends
6221 : : * to look at an inconsistent state of the control file in shared memory.
6222 : : * There is still a small window during which backends can write WAL and
6223 : : * the control file is still referring to a system not in DB_IN_PRODUCTION
6224 : : * state while looking at the on-disk control file.
6225 : : *
6226 : : * Also, we use info_lck to update SharedRecoveryState to ensure that
6227 : : * there are no race conditions concerning visibility of other recent
6228 : : * updates to shared memory.
6229 : : */
3432 peter_e@gmx.net 6230 :CBC 872 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6231 : 872 : ControlFile->state = DB_IN_PRODUCTION;
6232 : :
4105 andres@anarazel.de 6233 [ - + ]: 872 : SpinLockAcquire(&XLogCtl->info_lck);
2064 michael@paquier.xyz 6234 : 872 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_DONE;
4105 andres@anarazel.de 6235 : 872 : SpinLockRelease(&XLogCtl->info_lck);
6236 : :
3432 peter_e@gmx.net 6237 : 872 : UpdateControlFile();
6238 : 872 : LWLockRelease(ControlFileLock);
6239 : :
6240 : : /*
6241 : : * Wake up all waiters for replay LSN. They need to report an error that
6242 : : * recovery was ended before reaching the target LSN.
6243 : : */
43 akorotkov@postgresql 6244 :GNC 872 : WaitLSNWakeup(WAIT_LSN_TYPE_REPLAY, InvalidXLogRecPtr);
6245 : :
6246 : : /*
6247 : : * Shutdown the recovery environment. This must occur after
6248 : : * RecoverPreparedTransactions() (see notes in lock_twophase_recover())
6249 : : * and after switching SharedRecoveryState to RECOVERY_STATE_DONE so as
6250 : : * any session building a snapshot will not rely on KnownAssignedXids as
6251 : : * RecoveryInProgress() would return false at this stage. This is
6252 : : * particularly critical for prepared 2PC transactions, that would still
6253 : : * need to be included in snapshots once recovery has ended.
6254 : : */
1536 michael@paquier.xyz 6255 [ + + ]:CBC 872 : if (standbyState != STANDBY_DISABLED)
6256 : 49 : ShutdownRecoveryTransactionEnvironment();
6257 : :
6258 : : /*
6259 : : * If there were cascading standby servers connected to us, nudge any wal
6260 : : * sender processes to notice that we've been promoted.
6261 : : */
985 andres@anarazel.de 6262 : 872 : WalSndWakeup(true, true);
6263 : :
6264 : : /*
6265 : : * If this was a promotion, request an (online) checkpoint now. This isn't
6266 : : * required for consistency, but the last restartpoint might be far back,
6267 : : * and in case of a crash, recovering from it might take a longer than is
6268 : : * appropriate now that we're not in standby mode anymore.
6269 : : */
1968 fujii@postgresql.org 6270 [ + + ]: 872 : if (promoted)
4594 simon@2ndQuadrant.co 6271 : 42 : RequestCheckpoint(CHECKPOINT_FORCE);
6147 heikki.linnakangas@i 6272 : 872 : }
6273 : :
6274 : : /*
6275 : : * Callback from PerformWalRecovery(), called when we switch from crash
6276 : : * recovery to archive recovery mode. Updates the control file accordingly.
6277 : : */
6278 : : void
1401 6279 : 2 : SwitchIntoArchiveRecovery(XLogRecPtr EndRecPtr, TimeLineID replayTLI)
6280 : : {
6281 : : /* initialize minRecoveryPoint to this record */
6282 : 2 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6283 : 2 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
6284 [ + - ]: 2 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6285 : : {
6286 : 2 : ControlFile->minRecoveryPoint = EndRecPtr;
6287 : 2 : ControlFile->minRecoveryPointTLI = replayTLI;
6288 : : }
6289 : : /* update local copy */
6290 : 2 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
6291 : 2 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
6292 : :
6293 : : /*
6294 : : * The startup process can update its local copy of minRecoveryPoint from
6295 : : * this point.
6296 : : */
6297 : 2 : updateMinRecoveryPoint = true;
6298 : :
6299 : 2 : UpdateControlFile();
6300 : :
6301 : : /*
6302 : : * We update SharedRecoveryState while holding the lock on ControlFileLock
6303 : : * so both states are consistent in shared memory.
6304 : : */
6305 [ - + ]: 2 : SpinLockAcquire(&XLogCtl->info_lck);
6306 : 2 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
6307 : 2 : SpinLockRelease(&XLogCtl->info_lck);
6308 : :
6309 : 2 : LWLockRelease(ControlFileLock);
6310 : 2 : }
6311 : :
6312 : : /*
6313 : : * Callback from PerformWalRecovery(), called when we reach the end of backup.
6314 : : * Updates the control file accordingly.
6315 : : */
6316 : : void
6317 : 71 : ReachedEndOfBackup(XLogRecPtr EndRecPtr, TimeLineID tli)
6318 : : {
6319 : : /*
6320 : : * We have reached the end of base backup, as indicated by pg_control. The
6321 : : * data on disk is now consistent (unless minRecoveryPoint is further
6322 : : * ahead, which can happen if we crashed during previous recovery). Reset
6323 : : * backupStartPoint and backupEndPoint, and update minRecoveryPoint to
6324 : : * make sure we don't allow starting up at an earlier point even if
6325 : : * recovery is stopped and restarted soon after this.
6326 : : */
6327 : 71 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6328 : :
6329 [ + + ]: 71 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6330 : : {
6331 : 67 : ControlFile->minRecoveryPoint = EndRecPtr;
6332 : 67 : ControlFile->minRecoveryPointTLI = tli;
6333 : : }
6334 : :
6335 : 71 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
6336 : 71 : ControlFile->backupEndPoint = InvalidXLogRecPtr;
6337 : 71 : ControlFile->backupEndRequired = false;
6338 : 71 : UpdateControlFile();
6339 : :
6340 : 71 : LWLockRelease(ControlFileLock);
5728 6341 : 71 : }
6342 : :
6343 : : /*
6344 : : * Perform whatever XLOG actions are necessary at end of REDO.
6345 : : *
6346 : : * The goal here is to make sure that we'll be able to recover properly if
6347 : : * we crash again. If we choose to write a checkpoint, we'll write a shutdown
6348 : : * checkpoint rather than an on-line one. This is not particularly critical,
6349 : : * but since we may be assigning a new TLI, using a shutdown checkpoint allows
6350 : : * us to have the rule that TLI only changes in shutdown checkpoints, which
6351 : : * allows some extra error checking in xlog_redo.
6352 : : */
6353 : : static bool
1527 rhaas@postgresql.org 6354 : 152 : PerformRecoveryXLogAction(void)
6355 : : {
6356 : 152 : bool promoted = false;
6357 : :
6358 : : /*
6359 : : * Perform a checkpoint to update all our recovery activity to disk.
6360 : : *
6361 : : * Note that we write a shutdown checkpoint rather than an on-line one.
6362 : : * This is not particularly critical, but since we may be assigning a new
6363 : : * TLI, using a shutdown checkpoint allows us to have the rule that TLI
6364 : : * only changes in shutdown checkpoints, which allows some extra error
6365 : : * checking in xlog_redo.
6366 : : *
6367 : : * In promotion, only create a lightweight end-of-recovery record instead
6368 : : * of a full checkpoint. A checkpoint is requested later, after we're
6369 : : * fully out of recovery mode and already accepting queries.
6370 : : */
6371 [ + + + - : 201 : if (ArchiveRecoveryRequested && IsUnderPostmaster &&
+ + ]
1401 heikki.linnakangas@i 6372 : 49 : PromoteIsTriggered())
6373 : : {
1527 rhaas@postgresql.org 6374 : 42 : promoted = true;
6375 : :
6376 : : /*
6377 : : * Insert a special WAL record to mark the end of recovery, since we
6378 : : * aren't doing a checkpoint. That means that the checkpointer process
6379 : : * may likely be in the middle of a time-smoothed restartpoint and
6380 : : * could continue to be for minutes after this. That sounds strange,
6381 : : * but the effect is roughly the same and it would be stranger to try
6382 : : * to come out of the restartpoint and then checkpoint. We request a
6383 : : * checkpoint later anyway, just for safety.
6384 : : */
6385 : 42 : CreateEndOfRecoveryRecord();
6386 : : }
6387 : : else
6388 : : {
6389 : 110 : RequestCheckpoint(CHECKPOINT_END_OF_RECOVERY |
6390 : : CHECKPOINT_FAST |
6391 : : CHECKPOINT_WAIT);
6392 : : }
6393 : :
6394 : 152 : return promoted;
6395 : : }
6396 : :
6397 : : /*
6398 : : * Is the system still in recovery?
6399 : : *
6400 : : * Unlike testing InRecovery, this works in any process that's connected to
6401 : : * shared memory.
6402 : : */
6403 : : bool
6147 heikki.linnakangas@i 6404 : 53346111 : RecoveryInProgress(void)
6405 : : {
6406 : : /*
6407 : : * We check shared state each time only until we leave recovery mode. We
6408 : : * can't re-enter recovery, so there's no need to keep checking after the
6409 : : * shared variable has once been seen false.
6410 : : */
6411 [ + + ]: 53346111 : if (!LocalRecoveryInProgress)
6412 : 51180432 : return false;
6413 : : else
6414 : : {
6415 : : /*
6416 : : * use volatile pointer to make sure we make a fresh read of the
6417 : : * shared variable.
6418 : : */
6419 : 2165679 : volatile XLogCtlData *xlogctl = XLogCtl;
6420 : :
2064 michael@paquier.xyz 6421 : 2165679 : LocalRecoveryInProgress = (xlogctl->SharedRecoveryState != RECOVERY_STATE_DONE);
6422 : :
6423 : : /*
6424 : : * Note: We don't need a memory barrier when we're still in recovery.
6425 : : * We might exit recovery immediately after return, so the caller
6426 : : * can't rely on 'true' meaning that we're still in recovery anyway.
6427 : : */
6428 : :
6147 heikki.linnakangas@i 6429 : 2165679 : return LocalRecoveryInProgress;
6430 : : }
6431 : : }
6432 : :
6433 : : /*
6434 : : * Returns current recovery state from shared memory.
6435 : : *
6436 : : * This returned state is kept consistent with the contents of the control
6437 : : * file. See details about the possible values of RecoveryState in xlog.h.
6438 : : */
6439 : : RecoveryState
2064 michael@paquier.xyz 6440 : 31119 : GetRecoveryState(void)
6441 : : {
6442 : : RecoveryState retval;
6443 : :
6444 [ - + ]: 31119 : SpinLockAcquire(&XLogCtl->info_lck);
6445 : 31119 : retval = XLogCtl->SharedRecoveryState;
6446 : 31119 : SpinLockRelease(&XLogCtl->info_lck);
6447 : :
6448 : 31119 : return retval;
6449 : : }
6450 : :
6451 : : /*
6452 : : * Is this process allowed to insert new WAL records?
6453 : : *
6454 : : * Ordinarily this is essentially equivalent to !RecoveryInProgress().
6455 : : * But we also have provisions for forcing the result "true" or "false"
6456 : : * within specific processes regardless of the global state.
6457 : : */
6458 : : bool
6019 tgl@sss.pgh.pa.us 6459 : 38545451 : XLogInsertAllowed(void)
6460 : : {
6461 : : /*
6462 : : * If value is "unconditionally true" or "unconditionally false", just
6463 : : * return it. This provides the normal fast path once recovery is known
6464 : : * done.
6465 : : */
6466 [ + + ]: 38545451 : if (LocalXLogInsertAllowed >= 0)
6467 : 37810347 : return (bool) LocalXLogInsertAllowed;
6468 : :
6469 : : /*
6470 : : * Else, must check to see if we're still in recovery.
6471 : : */
6472 [ + + ]: 735104 : if (RecoveryInProgress())
6473 : 726346 : return false;
6474 : :
6475 : : /*
6476 : : * On exit from recovery, reset to "unconditionally true", since there is
6477 : : * no need to keep checking.
6478 : : */
6479 : 8758 : LocalXLogInsertAllowed = 1;
6480 : 8758 : return true;
6481 : : }
6482 : :
6483 : : /*
6484 : : * Make XLogInsertAllowed() return true in the current process only.
6485 : : *
6486 : : * Note: it is allowed to switch LocalXLogInsertAllowed back to -1 later,
6487 : : * and even call LocalSetXLogInsertAllowed() again after that.
6488 : : *
6489 : : * Returns the previous value of LocalXLogInsertAllowed.
6490 : : */
6491 : : static int
6492 : 900 : LocalSetXLogInsertAllowed(void)
6493 : : {
1401 heikki.linnakangas@i 6494 : 900 : int oldXLogAllowed = LocalXLogInsertAllowed;
6495 : :
6019 tgl@sss.pgh.pa.us 6496 : 900 : LocalXLogInsertAllowed = 1;
6497 : :
1515 rhaas@postgresql.org 6498 : 900 : return oldXLogAllowed;
6499 : : }
6500 : :
6501 : : /*
6502 : : * Return the current Redo pointer from shared memory.
6503 : : *
6504 : : * As a side-effect, the local RedoRecPtr copy is updated.
6505 : : */
6506 : : XLogRecPtr
9121 vadim4o@yahoo.com 6507 : 213108 : GetRedoRecPtr(void)
6508 : : {
6509 : : XLogRecPtr ptr;
6510 : :
6511 : : /*
6512 : : * The possibly not up-to-date copy in XlogCtl is enough. Even if we
6513 : : * grabbed a WAL insertion lock to read the authoritative value in
6514 : : * Insert->RedoRecPtr, someone might update it just after we've released
6515 : : * the lock.
6516 : : */
4105 andres@anarazel.de 6517 [ + + ]: 213108 : SpinLockAcquire(&XLogCtl->info_lck);
6518 : 213108 : ptr = XLogCtl->RedoRecPtr;
6519 : 213108 : SpinLockRelease(&XLogCtl->info_lck);
6520 : :
4546 heikki.linnakangas@i 6521 [ + + ]: 213108 : if (RedoRecPtr < ptr)
6522 : 1612 : RedoRecPtr = ptr;
6523 : :
8679 tgl@sss.pgh.pa.us 6524 : 213108 : return RedoRecPtr;
6525 : : }
6526 : :
6527 : : /*
6528 : : * Return information needed to decide whether a modified block needs a
6529 : : * full-page image to be included in the WAL record.
6530 : : *
6531 : : * The returned values are cached copies from backend-private memory, and
6532 : : * possibly out-of-date or, indeed, uninitialized, in which case they will
6533 : : * be InvalidXLogRecPtr and false, respectively. XLogInsertRecord will
6534 : : * re-check them against up-to-date values, while holding the WAL insert lock.
6535 : : */
6536 : : void
4060 heikki.linnakangas@i 6537 : 14318113 : GetFullPageWriteInfo(XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
6538 : : {
6539 : 14318113 : *RedoRecPtr_p = RedoRecPtr;
6540 : 14318113 : *doPageWrites_p = doPageWrites;
6541 : 14318113 : }
6542 : :
6543 : : /*
6544 : : * GetInsertRecPtr -- Returns the current insert position.
6545 : : *
6546 : : * NOTE: The value *actually* returned is the position of the last full
6547 : : * xlog page. It lags behind the real insert position by at most 1 page.
6548 : : * For that, we don't need to scan through WAL insertion locks, and an
6549 : : * approximation is enough for the current usage of this function.
6550 : : */
6551 : : XLogRecPtr
6748 tgl@sss.pgh.pa.us 6552 : 7248 : GetInsertRecPtr(void)
6553 : : {
6554 : : XLogRecPtr recptr;
6555 : :
4105 andres@anarazel.de 6556 [ + + ]: 7248 : SpinLockAcquire(&XLogCtl->info_lck);
6557 : 7248 : recptr = XLogCtl->LogwrtRqst.Write;
6558 : 7248 : SpinLockRelease(&XLogCtl->info_lck);
6559 : :
6748 tgl@sss.pgh.pa.us 6560 : 7248 : return recptr;
6561 : : }
6562 : :
6563 : : /*
6564 : : * GetFlushRecPtr -- Returns the current flush position, ie, the last WAL
6565 : : * position known to be fsync'd to disk. This should only be used on a
6566 : : * system that is known not to be in recovery.
6567 : : */
6568 : : XLogRecPtr
1504 rhaas@postgresql.org 6569 : 182804 : GetFlushRecPtr(TimeLineID *insertTLI)
6570 : : {
1499 6571 [ - + ]: 182804 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6572 : :
624 alvherre@alvh.no-ip. 6573 : 182804 : RefreshXLogWriteResult(LogwrtResult);
6574 : :
6575 : : /*
6576 : : * If we're writing and flushing WAL, the time line can't be changing, so
6577 : : * no lock is required.
6578 : : */
1504 rhaas@postgresql.org 6579 [ + + ]: 182804 : if (insertTLI)
1499 6580 : 23109 : *insertTLI = XLogCtl->InsertTimeLineID;
6581 : :
3628 simon@2ndQuadrant.co 6582 : 182804 : return LogwrtResult.Flush;
6583 : : }
6584 : :
6585 : : /*
6586 : : * GetWALInsertionTimeLine -- Returns the current timeline of a system that
6587 : : * is not in recovery.
6588 : : */
6589 : : TimeLineID
1504 rhaas@postgresql.org 6590 : 111328 : GetWALInsertionTimeLine(void)
6591 : : {
6592 [ - + ]: 111328 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6593 : :
6594 : : /* Since the value can't be changing, no lock is required. */
1499 6595 : 111328 : return XLogCtl->InsertTimeLineID;
6596 : : }
6597 : :
6598 : : /*
6599 : : * GetWALInsertionTimeLineIfSet -- If the system is not in recovery, returns
6600 : : * the WAL insertion timeline; else, returns 0. Wherever possible, use
6601 : : * GetWALInsertionTimeLine() instead, since it's cheaper. Note that this
6602 : : * function decides recovery has ended as soon as the insert TLI is set, which
6603 : : * happens before we set XLogCtl->SharedRecoveryState to RECOVERY_STATE_DONE.
6604 : : */
6605 : : TimeLineID
510 rhaas@postgresql.org 6606 :UBC 0 : GetWALInsertionTimeLineIfSet(void)
6607 : : {
6608 : : TimeLineID insertTLI;
6609 : :
6610 [ # # ]: 0 : SpinLockAcquire(&XLogCtl->info_lck);
6611 : 0 : insertTLI = XLogCtl->InsertTimeLineID;
6612 : 0 : SpinLockRelease(&XLogCtl->info_lck);
6613 : :
6614 : 0 : return insertTLI;
6615 : : }
6616 : :
6617 : : /*
6618 : : * GetLastImportantRecPtr -- Returns the LSN of the last important record
6619 : : * inserted. All records not explicitly marked as unimportant are considered
6620 : : * important.
6621 : : *
6622 : : * The LSN is determined by computing the maximum of
6623 : : * WALInsertLocks[i].lastImportantAt.
6624 : : */
6625 : : XLogRecPtr
3283 andres@anarazel.de 6626 :CBC 1578 : GetLastImportantRecPtr(void)
6627 : : {
6628 : 1578 : XLogRecPtr res = InvalidXLogRecPtr;
6629 : : int i;
6630 : :
6631 [ + + ]: 14202 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
6632 : : {
6633 : : XLogRecPtr last_important;
6634 : :
6635 : : /*
6636 : : * Need to take a lock to prevent torn reads of the LSN, which are
6637 : : * possible on some of the supported platforms. WAL insert locks only
6638 : : * support exclusive mode, so we have to use that.
6639 : : */
6640 : 12624 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
6641 : 12624 : last_important = WALInsertLocks[i].l.lastImportantAt;
6642 : 12624 : LWLockRelease(&WALInsertLocks[i].l.lock);
6643 : :
6644 [ + + ]: 12624 : if (res < last_important)
6645 : 2740 : res = last_important;
6646 : : }
6647 : :
6648 : 1578 : return res;
6649 : : }
6650 : :
6651 : : /*
6652 : : * Get the time and LSN of the last xlog segment switch
6653 : : */
6654 : : pg_time_t
3283 andres@anarazel.de 6655 :UBC 0 : GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
6656 : : {
6657 : : pg_time_t result;
6658 : :
6659 : : /* Need WALWriteLock, but shared lock is sufficient */
7063 tgl@sss.pgh.pa.us 6660 : 0 : LWLockAcquire(WALWriteLock, LW_SHARED);
4537 heikki.linnakangas@i 6661 : 0 : result = XLogCtl->lastSegSwitchTime;
3283 andres@anarazel.de 6662 : 0 : *lastSwitchLSN = XLogCtl->lastSegSwitchLSN;
7063 tgl@sss.pgh.pa.us 6663 : 0 : LWLockRelease(WALWriteLock);
6664 : :
6665 : 0 : return result;
6666 : : }
6667 : :
6668 : : /*
6669 : : * This must be called ONCE during postmaster or standalone-backend shutdown
6670 : : */
6671 : : void
8042 peter_e@gmx.net 6672 :CBC 641 : ShutdownXLOG(int code, Datum arg)
6673 : : {
6674 : : /*
6675 : : * We should have an aux process resource owner to use, and we should not
6676 : : * be in a transaction that's installed some other resowner.
6677 : : */
2710 tgl@sss.pgh.pa.us 6678 [ - + ]: 641 : Assert(AuxProcessResourceOwner != NULL);
6679 [ + + - + ]: 641 : Assert(CurrentResourceOwner == NULL ||
6680 : : CurrentResourceOwner == AuxProcessResourceOwner);
6681 : 641 : CurrentResourceOwner = AuxProcessResourceOwner;
6682 : :
6683 : : /* Don't be chatty in standalone mode */
4571 6684 [ + + + + ]: 641 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
6685 : : (errmsg("shutting down")));
6686 : :
6687 : : /*
6688 : : * Signal walsenders to move to stopping state.
6689 : : */
3118 andres@anarazel.de 6690 : 641 : WalSndInitStopping();
6691 : :
6692 : : /*
6693 : : * Wait for WAL senders to be in stopping state. This prevents commands
6694 : : * from writing new WAL.
6695 : : */
6696 : 641 : WalSndWaitStopping();
6697 : :
6147 heikki.linnakangas@i 6698 [ + + ]: 641 : if (RecoveryInProgress())
160 nathan@postgresql.or 6699 :GNC 52 : CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_FAST);
6700 : : else
6701 : : {
6702 : : /*
6703 : : * If archiving is enabled, rotate the last XLOG file so that all the
6704 : : * remaining records are archived (postmaster wakes up the archiver
6705 : : * process one more time at the end of shutdown). The checkpoint
6706 : : * record will go to the next XLOG file and won't be archived (yet).
6707 : : */
1414 rhaas@postgresql.org 6708 [ + + - + :CBC 589 : if (XLogArchivingActive())
+ + ]
3283 andres@anarazel.de 6709 : 12 : RequestXLogSwitch(false);
6710 : :
160 nathan@postgresql.or 6711 :GNC 589 : CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_FAST);
6712 : : }
9570 vadim4o@yahoo.com 6713 :CBC 641 : }
6714 : :
6715 : : /*
6716 : : * Log start of a checkpoint.
6717 : : */
6718 : : static void
6147 heikki.linnakangas@i 6719 : 1434 : LogCheckpointStart(int flags, bool restartpoint)
6720 : : {
1840 peter@eisentraut.org 6721 [ + + ]: 1434 : if (restartpoint)
6722 [ + - - + : 200 : ereport(LOG,
- + + + +
+ + + + +
- + + + ]
6723 : : /* translator: the placeholders show checkpoint options */
6724 : : (errmsg("restartpoint starting:%s%s%s%s%s%s%s%s",
6725 : : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6726 : : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6727 : : (flags & CHECKPOINT_FAST) ? " fast" : "",
6728 : : (flags & CHECKPOINT_FORCE) ? " force" : "",
6729 : : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6730 : : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6731 : : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6732 : : (flags & CHECKPOINT_FLUSH_UNLOGGED) ? " flush-unlogged" : "")));
6733 : : else
6734 [ + - + + : 1234 : ereport(LOG,
- + + + +
+ + + + +
+ + + + ]
6735 : : /* translator: the placeholders show checkpoint options */
6736 : : (errmsg("checkpoint starting:%s%s%s%s%s%s%s%s",
6737 : : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6738 : : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6739 : : (flags & CHECKPOINT_FAST) ? " fast" : "",
6740 : : (flags & CHECKPOINT_FORCE) ? " force" : "",
6741 : : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6742 : : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6743 : : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6744 : : (flags & CHECKPOINT_FLUSH_UNLOGGED) ? " flush-unlogged" : "")));
6746 tgl@sss.pgh.pa.us 6745 : 1434 : }
6746 : :
6747 : : /*
6748 : : * Log end of a checkpoint.
6749 : : */
6750 : : static void
6147 heikki.linnakangas@i 6751 : 1732 : LogCheckpointEnd(bool restartpoint)
6752 : : {
6753 : : long write_msecs,
6754 : : sync_msecs,
6755 : : total_msecs,
6756 : : longest_msecs,
6757 : : average_msecs;
6758 : : uint64 average_sync_time;
6759 : :
6746 tgl@sss.pgh.pa.us 6760 : 1732 : CheckpointStats.ckpt_end_t = GetCurrentTimestamp();
6761 : :
1864 6762 : 1732 : write_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_write_t,
6763 : : CheckpointStats.ckpt_sync_t);
6764 : :
6765 : 1732 : sync_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_sync_t,
6766 : : CheckpointStats.ckpt_sync_end_t);
6767 : :
6768 : : /* Accumulate checkpoint timing summary data, in milliseconds. */
780 michael@paquier.xyz 6769 : 1732 : PendingCheckpointerStats.write_time += write_msecs;
6770 : 1732 : PendingCheckpointerStats.sync_time += sync_msecs;
6771 : :
6772 : : /*
6773 : : * All of the published timing statistics are accounted for. Only
6774 : : * continue if a log message is to be written.
6775 : : */
5005 rhaas@postgresql.org 6776 [ + + ]: 1732 : if (!log_checkpoints)
6777 : 298 : return;
6778 : :
1864 tgl@sss.pgh.pa.us 6779 : 1434 : total_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_start_t,
6780 : : CheckpointStats.ckpt_end_t);
6781 : :
6782 : : /*
6783 : : * Timing values returned from CheckpointStats are in microseconds.
6784 : : * Convert to milliseconds for consistent printing.
6785 : : */
6786 : 1434 : longest_msecs = (long) ((CheckpointStats.ckpt_longest_sync + 999) / 1000);
6787 : :
5483 rhaas@postgresql.org 6788 : 1434 : average_sync_time = 0;
5366 bruce@momjian.us 6789 [ - + ]: 1434 : if (CheckpointStats.ckpt_sync_rels > 0)
5483 rhaas@postgresql.org 6790 :UBC 0 : average_sync_time = CheckpointStats.ckpt_agg_sync_time /
6791 : 0 : CheckpointStats.ckpt_sync_rels;
1864 tgl@sss.pgh.pa.us 6792 :CBC 1434 : average_msecs = (long) ((average_sync_time + 999) / 1000);
6793 : :
6794 : : /*
6795 : : * ControlFileLock is not required to see ControlFile->checkPoint and
6796 : : * ->checkPointCopy here as we are the only updator of those variables at
6797 : : * this moment.
6798 : : */
1840 peter@eisentraut.org 6799 [ + + ]: 1434 : if (restartpoint)
6800 [ + - ]: 200 : ereport(LOG,
6801 : : (errmsg("restartpoint complete: wrote %d buffers (%.1f%%), "
6802 : : "wrote %d SLRU buffers; %d WAL file(s) added, "
6803 : : "%d removed, %d recycled; write=%ld.%03d s, "
6804 : : "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6805 : : "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6806 : : "estimate=%d kB; lsn=%X/%08X, redo lsn=%X/%08X",
6807 : : CheckpointStats.ckpt_bufs_written,
6808 : : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6809 : : CheckpointStats.ckpt_slru_written,
6810 : : CheckpointStats.ckpt_segs_added,
6811 : : CheckpointStats.ckpt_segs_removed,
6812 : : CheckpointStats.ckpt_segs_recycled,
6813 : : write_msecs / 1000, (int) (write_msecs % 1000),
6814 : : sync_msecs / 1000, (int) (sync_msecs % 1000),
6815 : : total_msecs / 1000, (int) (total_msecs % 1000),
6816 : : CheckpointStats.ckpt_sync_rels,
6817 : : longest_msecs / 1000, (int) (longest_msecs % 1000),
6818 : : average_msecs / 1000, (int) (average_msecs % 1000),
6819 : : (int) (PrevCheckPointDistance / 1024.0),
6820 : : (int) (CheckPointDistanceEstimate / 1024.0),
6821 : : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6822 : : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6823 : : else
6824 [ + - ]: 1234 : ereport(LOG,
6825 : : (errmsg("checkpoint complete: wrote %d buffers (%.1f%%), "
6826 : : "wrote %d SLRU buffers; %d WAL file(s) added, "
6827 : : "%d removed, %d recycled; write=%ld.%03d s, "
6828 : : "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6829 : : "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6830 : : "estimate=%d kB; lsn=%X/%08X, redo lsn=%X/%08X",
6831 : : CheckpointStats.ckpt_bufs_written,
6832 : : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6833 : : CheckpointStats.ckpt_slru_written,
6834 : : CheckpointStats.ckpt_segs_added,
6835 : : CheckpointStats.ckpt_segs_removed,
6836 : : CheckpointStats.ckpt_segs_recycled,
6837 : : write_msecs / 1000, (int) (write_msecs % 1000),
6838 : : sync_msecs / 1000, (int) (sync_msecs % 1000),
6839 : : total_msecs / 1000, (int) (total_msecs % 1000),
6840 : : CheckpointStats.ckpt_sync_rels,
6841 : : longest_msecs / 1000, (int) (longest_msecs % 1000),
6842 : : average_msecs / 1000, (int) (average_msecs % 1000),
6843 : : (int) (PrevCheckPointDistance / 1024.0),
6844 : : (int) (CheckPointDistanceEstimate / 1024.0),
6845 : : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6846 : : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6847 : : }
6848 : :
6849 : : /*
6850 : : * Update the estimate of distance between checkpoints.
6851 : : *
6852 : : * The estimate is used to calculate the number of WAL segments to keep
6853 : : * preallocated, see XLOGfileslop().
6854 : : */
6855 : : static void
3951 heikki.linnakangas@i 6856 : 1732 : UpdateCheckPointDistanceEstimate(uint64 nbytes)
6857 : : {
6858 : : /*
6859 : : * To estimate the number of segments consumed between checkpoints, keep a
6860 : : * moving average of the amount of WAL generated in previous checkpoint
6861 : : * cycles. However, if the load is bursty, with quiet periods and busy
6862 : : * periods, we want to cater for the peak load. So instead of a plain
6863 : : * moving average, let the average decline slowly if the previous cycle
6864 : : * used less WAL than estimated, but bump it up immediately if it used
6865 : : * more.
6866 : : *
6867 : : * When checkpoints are triggered by max_wal_size, this should converge to
6868 : : * CheckpointSegments * wal_segment_size,
6869 : : *
6870 : : * Note: This doesn't pay any attention to what caused the checkpoint.
6871 : : * Checkpoints triggered manually with CHECKPOINT command, or by e.g.
6872 : : * starting a base backup, are counted the same as those created
6873 : : * automatically. The slow-decline will largely mask them out, if they are
6874 : : * not frequent. If they are frequent, it seems reasonable to count them
6875 : : * in as any others; if you issue a manual checkpoint every 5 minutes and
6876 : : * never let a timed checkpoint happen, it makes sense to base the
6877 : : * preallocation on that 5 minute interval rather than whatever
6878 : : * checkpoint_timeout is set to.
6879 : : */
6880 : 1732 : PrevCheckPointDistance = nbytes;
6881 [ + + ]: 1732 : if (CheckPointDistanceEstimate < nbytes)
6882 : 727 : CheckPointDistanceEstimate = nbytes;
6883 : : else
6884 : 1005 : CheckPointDistanceEstimate =
6885 : 1005 : (0.90 * CheckPointDistanceEstimate + 0.10 * (double) nbytes);
6746 tgl@sss.pgh.pa.us 6886 : 1732 : }
6887 : :
6888 : : /*
6889 : : * Update the ps display for a process running a checkpoint. Note that
6890 : : * this routine should not do any allocations so as it can be called
6891 : : * from a critical section.
6892 : : */
6893 : : static void
1830 michael@paquier.xyz 6894 : 3464 : update_checkpoint_display(int flags, bool restartpoint, bool reset)
6895 : : {
6896 : : /*
6897 : : * The status is reported only for end-of-recovery and shutdown
6898 : : * checkpoints or shutdown restartpoints. Updating the ps display is
6899 : : * useful in those situations as it may not be possible to rely on
6900 : : * pg_stat_activity to see the status of the checkpointer or the startup
6901 : : * process.
6902 : : */
6903 [ + + ]: 3464 : if ((flags & (CHECKPOINT_END_OF_RECOVERY | CHECKPOINT_IS_SHUTDOWN)) == 0)
6904 : 2188 : return;
6905 : :
6906 [ + + ]: 1276 : if (reset)
6907 : 638 : set_ps_display("");
6908 : : else
6909 : : {
6910 : : char activitymsg[128];
6911 : :
6912 [ + + ]: 1914 : snprintf(activitymsg, sizeof(activitymsg), "performing %s%s%s",
6913 [ + + ]: 638 : (flags & CHECKPOINT_END_OF_RECOVERY) ? "end-of-recovery " : "",
6914 [ + + ]: 638 : (flags & CHECKPOINT_IS_SHUTDOWN) ? "shutdown " : "",
6915 : : restartpoint ? "restartpoint" : "checkpoint");
6916 : 638 : set_ps_display(activitymsg);
6917 : : }
6918 : : }
6919 : :
6920 : :
6921 : : /*
6922 : : * Perform a checkpoint --- either during shutdown, or on-the-fly
6923 : : *
6924 : : * flags is a bitwise OR of the following:
6925 : : * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
6926 : : * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
6927 : : * CHECKPOINT_FAST: finish the checkpoint ASAP, ignoring
6928 : : * checkpoint_completion_target parameter.
6929 : : * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
6930 : : * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
6931 : : * CHECKPOINT_END_OF_RECOVERY).
6932 : : * CHECKPOINT_FLUSH_UNLOGGED: also flush buffers of unlogged tables.
6933 : : *
6934 : : * Note: flags contains other bits, of interest here only for logging purposes.
6935 : : * In particular note that this routine is synchronous and does not pay
6936 : : * attention to CHECKPOINT_WAIT.
6937 : : *
6938 : : * If !shutdown then we are writing an online checkpoint. An XLOG_CHECKPOINT_REDO
6939 : : * record is inserted into WAL at the logical location of the checkpoint, before
6940 : : * flushing anything to disk, and when the checkpoint is eventually completed,
6941 : : * and it is from this point that WAL replay will begin in the case of a recovery
6942 : : * from this checkpoint. Once everything is written to disk, an
6943 : : * XLOG_CHECKPOINT_ONLINE record is written to complete the checkpoint, and
6944 : : * points back to the earlier XLOG_CHECKPOINT_REDO record. This mechanism allows
6945 : : * other write-ahead log records to be written while the checkpoint is in
6946 : : * progress, but we must be very careful about order of operations. This function
6947 : : * may take many minutes to execute on a busy system.
6948 : : *
6949 : : * On the other hand, when shutdown is true, concurrent insertion into the
6950 : : * write-ahead log is impossible, so there is no need for two separate records.
6951 : : * In this case, we only insert an XLOG_CHECKPOINT_SHUTDOWN record, and it's
6952 : : * both the record marking the completion of the checkpoint and the location
6953 : : * from which WAL replay would begin if needed.
6954 : : *
6955 : : * Returns true if a new checkpoint was performed, or false if it was skipped
6956 : : * because the system was idle.
6957 : : */
6958 : : bool
6748 tgl@sss.pgh.pa.us 6959 : 1532 : CreateCheckPoint(int flags)
6960 : : {
6961 : : bool shutdown;
6962 : : CheckPoint checkPoint;
6963 : : XLogRecPtr recptr;
6964 : : XLogSegNo _logSegNo;
9381 bruce@momjian.us 6965 : 1532 : XLogCtlInsert *Insert = &XLogCtl->Insert;
6966 : : uint32 freespace;
6967 : : XLogRecPtr PriorRedoPtr;
6968 : : XLogRecPtr last_important_lsn;
6969 : : VirtualTransactionId *vxids;
6970 : : int nvxids;
1515 rhaas@postgresql.org 6971 : 1532 : int oldXLogAllowed = 0;
6972 : :
6973 : : /*
6974 : : * An end-of-recovery checkpoint is really a shutdown checkpoint, just
6975 : : * issued at a different time.
6976 : : */
6019 tgl@sss.pgh.pa.us 6977 [ + + ]: 1532 : if (flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY))
6020 heikki.linnakangas@i 6978 : 617 : shutdown = true;
6979 : : else
6980 : 915 : shutdown = false;
6981 : :
6982 : : /* sanity check */
6019 tgl@sss.pgh.pa.us 6983 [ + + - + ]: 1532 : if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
6019 tgl@sss.pgh.pa.us 6984 [ # # ]:UBC 0 : elog(ERROR, "can't create a checkpoint during recovery");
6985 : :
6986 : : /*
6987 : : * Prepare to accumulate statistics.
6988 : : *
6989 : : * Note: because it is possible for log_checkpoints to change while a
6990 : : * checkpoint proceeds, we always accumulate stats, even if
6991 : : * log_checkpoints is currently off.
6992 : : */
6746 tgl@sss.pgh.pa.us 6993 [ + - + - :CBC 16852 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ - + - +
+ ]
6994 : 1532 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
6995 : :
6996 : : /*
6997 : : * Let smgr prepare for checkpoint; this has to happen outside the
6998 : : * critical section and before we determine the REDO pointer. Note that
6999 : : * smgr must not do anything that'd have to be undone if we decide no
7000 : : * checkpoint is needed.
7001 : : */
1373 tmunro@postgresql.or 7002 : 1532 : SyncPreCheckpoint();
7003 : :
7004 : : /* Run these points outside the critical section. */
7005 : : INJECTION_POINT("create-checkpoint-initial", NULL);
7006 : : INJECTION_POINT_LOAD("create-checkpoint-run");
7007 : :
7008 : : /*
7009 : : * Use a critical section to force system panic if we have trouble.
7010 : : */
8846 tgl@sss.pgh.pa.us 7011 : 1532 : START_CRIT_SECTION();
7012 : :
9579 vadim4o@yahoo.com 7013 [ + + ]: 1532 : if (shutdown)
7014 : : {
6147 heikki.linnakangas@i 7015 : 617 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9579 vadim4o@yahoo.com 7016 : 617 : ControlFile->state = DB_SHUTDOWNING;
7017 : 617 : UpdateControlFile();
6147 heikki.linnakangas@i 7018 : 617 : LWLockRelease(ControlFileLock);
7019 : : }
7020 : :
7021 : : /* Begin filling in the checkpoint WAL record */
8246 tgl@sss.pgh.pa.us 7022 [ + - + - : 18384 : MemSet(&checkPoint, 0, sizeof(checkPoint));
+ - + - +
+ ]
6514 7023 : 1532 : checkPoint.time = (pg_time_t) time(NULL);
7024 : :
7025 : : /*
7026 : : * For Hot Standby, derive the oldestActiveXid before we fix the redo
7027 : : * pointer. This allows us to begin accumulating changes to assemble our
7028 : : * starting snapshot of locks and transactions.
7029 : : */
5160 simon@2ndQuadrant.co 7030 [ + + + + ]: 1532 : if (!shutdown && XLogStandbyInfoActive())
148 akapila@postgresql.o 7031 :GNC 864 : checkPoint.oldestActiveXid = GetOldestActiveTransactionId(false, true);
7032 : : else
5160 simon@2ndQuadrant.co 7033 :CBC 668 : checkPoint.oldestActiveXid = InvalidTransactionId;
7034 : :
7035 : : /*
7036 : : * Get location of last important record before acquiring insert locks (as
7037 : : * GetLastImportantRecPtr() also locks WAL locks).
7038 : : */
3283 andres@anarazel.de 7039 : 1532 : last_important_lsn = GetLastImportantRecPtr();
7040 : :
7041 : : /*
7042 : : * If this isn't a shutdown or forced checkpoint, and if there has been no
7043 : : * WAL activity requiring a checkpoint, skip it. The idea here is to
7044 : : * avoid inserting duplicate checkpoints when the system is idle.
7045 : : */
6020 heikki.linnakangas@i 7046 [ + + ]: 1532 : if ((flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY |
7047 : : CHECKPOINT_FORCE)) == 0)
7048 : : {
3283 andres@anarazel.de 7049 [ - + ]: 190 : if (last_important_lsn == ControlFile->checkPoint)
7050 : : {
9046 tgl@sss.pgh.pa.us 7051 [ # # ]:LBC (1) : END_CRIT_SECTION();
3283 andres@anarazel.de 7052 [ # # ]: (1) : ereport(DEBUG1,
7053 : : (errmsg_internal("checkpoint skipped because system is idle")));
444 fujii@postgresql.org 7054 : (1) : return false;
7055 : : }
7056 : : }
7057 : :
7058 : : /*
7059 : : * An end-of-recovery checkpoint is created before anyone is allowed to
7060 : : * write WAL. To allow us to write the checkpoint record, temporarily
7061 : : * enable XLogInsertAllowed.
7062 : : */
5957 heikki.linnakangas@i 7063 [ + + ]:CBC 1532 : if (flags & CHECKPOINT_END_OF_RECOVERY)
1515 rhaas@postgresql.org 7064 : 28 : oldXLogAllowed = LocalSetXLogInsertAllowed();
7065 : :
1499 7066 : 1532 : checkPoint.ThisTimeLineID = XLogCtl->InsertTimeLineID;
4693 heikki.linnakangas@i 7067 [ + + ]: 1532 : if (flags & CHECKPOINT_END_OF_RECOVERY)
7068 : 28 : checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
7069 : : else
1504 rhaas@postgresql.org 7070 : 1504 : checkPoint.PrevTimeLineID = checkPoint.ThisTimeLineID;
7071 : :
7072 : : /*
7073 : : * We must block concurrent insertions while examining insert state.
7074 : : */
791 7075 : 1532 : WALInsertLockAcquireExclusive();
7076 : :
7077 : 1532 : checkPoint.fullPageWrites = Insert->fullPageWrites;
518 7078 : 1532 : checkPoint.wal_level = wal_level;
7079 : :
791 7080 [ + + ]: 1532 : if (shutdown)
7081 : : {
7082 : 617 : XLogRecPtr curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
7083 : :
7084 : : /*
7085 : : * Compute new REDO record ptr = location of next XLOG record.
7086 : : *
7087 : : * Since this is a shutdown checkpoint, there can't be any concurrent
7088 : : * WAL insertion.
7089 : : */
7090 [ + - ]: 617 : freespace = INSERT_FREESPACE(curInsert);
7091 [ - + ]: 617 : if (freespace == 0)
7092 : : {
791 rhaas@postgresql.org 7093 [ # # ]:UBC 0 : if (XLogSegmentOffset(curInsert, wal_segment_size) == 0)
7094 : 0 : curInsert += SizeOfXLogLongPHD;
7095 : : else
7096 : 0 : curInsert += SizeOfXLogShortPHD;
7097 : : }
791 rhaas@postgresql.org 7098 :CBC 617 : checkPoint.redo = curInsert;
7099 : :
7100 : : /*
7101 : : * Here we update the shared RedoRecPtr for future XLogInsert calls;
7102 : : * this must be done while holding all the insertion locks.
7103 : : *
7104 : : * Note: if we fail to complete the checkpoint, RedoRecPtr will be
7105 : : * left pointing past where it really needs to point. This is okay;
7106 : : * the only consequence is that XLogInsert might back up whole buffers
7107 : : * that it didn't really need to. We can't postpone advancing
7108 : : * RedoRecPtr because XLogInserts that happen while we are dumping
7109 : : * buffers must assume that their buffer changes are not included in
7110 : : * the checkpoint.
7111 : : */
7112 : 617 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
7113 : : }
7114 : :
7115 : : /*
7116 : : * Now we can release the WAL insertion locks, allowing other xacts to
7117 : : * proceed while we are flushing disk buffers.
7118 : : */
4290 heikki.linnakangas@i 7119 : 1532 : WALInsertLockRelease();
7120 : :
7121 : : /*
7122 : : * If this is an online checkpoint, we have not yet determined the redo
7123 : : * point. We do so now by inserting the special XLOG_CHECKPOINT_REDO
7124 : : * record; the LSN at which it starts becomes the new redo pointer. We
7125 : : * don't do this for a shutdown checkpoint, because in that case no WAL
7126 : : * can be written between the redo point and the insertion of the
7127 : : * checkpoint record itself, so the checkpoint record itself serves to
7128 : : * mark the redo point.
7129 : : */
791 rhaas@postgresql.org 7130 [ + + ]: 1532 : if (!shutdown)
7131 : : {
7132 : : /* Include WAL level in record for WAL summarizer's benefit. */
7133 : 915 : XLogBeginInsert();
310 peter@eisentraut.org 7134 : 915 : XLogRegisterData(&wal_level, sizeof(wal_level));
791 rhaas@postgresql.org 7135 : 915 : (void) XLogInsert(RM_XLOG_ID, XLOG_CHECKPOINT_REDO);
7136 : :
7137 : : /*
7138 : : * XLogInsertRecord will have updated XLogCtl->Insert.RedoRecPtr in
7139 : : * shared memory and RedoRecPtr in backend-local memory, but we need
7140 : : * to copy that into the record that will be inserted when the
7141 : : * checkpoint is complete.
7142 : : */
7143 : 915 : checkPoint.redo = RedoRecPtr;
7144 : : }
7145 : :
7146 : : /* Update the info_lck-protected copy of RedoRecPtr as well */
4105 andres@anarazel.de 7147 [ - + ]: 1532 : SpinLockAcquire(&XLogCtl->info_lck);
7148 : 1532 : XLogCtl->RedoRecPtr = checkPoint.redo;
7149 : 1532 : SpinLockRelease(&XLogCtl->info_lck);
7150 : :
7151 : : /*
7152 : : * If enabled, log checkpoint start. We postpone this until now so as not
7153 : : * to log anything if we decided to skip the checkpoint.
7154 : : */
6746 tgl@sss.pgh.pa.us 7155 [ + + ]: 1532 : if (log_checkpoints)
6147 heikki.linnakangas@i 7156 : 1234 : LogCheckpointStart(flags, false);
7157 : :
7158 : : INJECTION_POINT_CACHED("create-checkpoint-run", NULL);
7159 : :
7160 : : /* Update the process title */
1830 michael@paquier.xyz 7161 : 1532 : update_checkpoint_display(flags, false, false);
7162 : :
7163 : : TRACE_POSTGRESQL_CHECKPOINT_START(flags);
7164 : :
7165 : : /*
7166 : : * Get the other info we need for the checkpoint record.
7167 : : *
7168 : : * We don't need to save oldestClogXid in the checkpoint, it only matters
7169 : : * for the short period in which clog is being truncated, and if we crash
7170 : : * during that we'll redo the clog truncation and fix up oldestClogXid
7171 : : * there.
7172 : : */
4276 heikki.linnakangas@i 7173 : 1532 : LWLockAcquire(XidGenLock, LW_SHARED);
741 7174 : 1532 : checkPoint.nextXid = TransamVariables->nextXid;
7175 : 1532 : checkPoint.oldestXid = TransamVariables->oldestXid;
7176 : 1532 : checkPoint.oldestXidDB = TransamVariables->oldestXidDB;
4276 7177 : 1532 : LWLockRelease(XidGenLock);
7178 : :
4033 alvherre@alvh.no-ip. 7179 : 1532 : LWLockAcquire(CommitTsLock, LW_SHARED);
741 heikki.linnakangas@i 7180 : 1532 : checkPoint.oldestCommitTsXid = TransamVariables->oldestCommitTsXid;
7181 : 1532 : checkPoint.newestCommitTsXid = TransamVariables->newestCommitTsXid;
4033 alvherre@alvh.no-ip. 7182 : 1532 : LWLockRelease(CommitTsLock);
7183 : :
4276 heikki.linnakangas@i 7184 : 1532 : LWLockAcquire(OidGenLock, LW_SHARED);
741 7185 : 1532 : checkPoint.nextOid = TransamVariables->nextOid;
4276 7186 [ + + ]: 1532 : if (!shutdown)
741 7187 : 915 : checkPoint.nextOid += TransamVariables->oidCount;
4276 7188 : 1532 : LWLockRelease(OidGenLock);
7189 : :
7190 : 1532 : MultiXactGetCheckptMulti(shutdown,
7191 : : &checkPoint.nextMulti,
7192 : : &checkPoint.nextMultiOffset,
7193 : : &checkPoint.oldestMulti,
7194 : : &checkPoint.oldestMultiDB);
7195 : :
7196 : : /*
7197 : : * Having constructed the checkpoint record, ensure all shmem disk buffers
7198 : : * and commit-log buffers are flushed to disk.
7199 : : *
7200 : : * This I/O could fail for various reasons. If so, we will fail to
7201 : : * complete the checkpoint, but there is no reason to force a system
7202 : : * panic. Accordingly, exit critical section while doing it.
7203 : : */
7204 [ - + ]: 1532 : END_CRIT_SECTION();
7205 : :
7206 : : /*
7207 : : * In some cases there are groups of actions that must all occur on one
7208 : : * side or the other of a checkpoint record. Before flushing the
7209 : : * checkpoint record we must explicitly wait for any backend currently
7210 : : * performing those groups of actions.
7211 : : *
7212 : : * One example is end of transaction, so we must wait for any transactions
7213 : : * that are currently in commit critical sections. If an xact inserted
7214 : : * its commit record into XLOG just before the REDO point, then a crash
7215 : : * restart from the REDO point would not replay that record, which means
7216 : : * that our flushing had better include the xact's update of pg_xact. So
7217 : : * we wait till he's out of his commit critical section before proceeding.
7218 : : * See notes in RecordTransactionCommit().
7219 : : *
7220 : : * Because we've already released the insertion locks, this test is a bit
7221 : : * fuzzy: it is possible that we will wait for xacts we didn't really need
7222 : : * to wait for. But the delay should be short and it seems better to make
7223 : : * checkpoint take a bit longer than to hold off insertions longer than
7224 : : * necessary. (In fact, the whole reason we have this issue is that xact.c
7225 : : * does commit record XLOG insertion and clog update as two separate steps
7226 : : * protected by different locks, but again that seems best on grounds of
7227 : : * minimizing lock contention.)
7228 : : *
7229 : : * A transaction that has not yet set delayChkptFlags when we look cannot
7230 : : * be at risk, since it has not inserted its commit record yet; and one
7231 : : * that's already cleared it is not at risk either, since it's done fixing
7232 : : * clog and we will correctly flush the update below. So we cannot miss
7233 : : * any xacts we need to wait for.
7234 : : */
1365 rhaas@postgresql.org 7235 : 1532 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_START);
4763 simon@2ndQuadrant.co 7236 [ + + ]: 1532 : if (nvxids > 0)
7237 : : {
7238 : : do
7239 : : {
7240 : : /*
7241 : : * Keep absorbing fsync requests while we wait. There could even
7242 : : * be a deadlock if we don't, if the process that prevents the
7243 : : * checkpoint is trying to add a request to the queue.
7244 : : */
545 heikki.linnakangas@i 7245 : 45 : AbsorbSyncRequests();
7246 : :
797 tmunro@postgresql.or 7247 : 45 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_START);
6608 bruce@momjian.us 7248 : 45 : pg_usleep(10000L); /* wait for 10 msec */
797 tmunro@postgresql.or 7249 : 45 : pgstat_report_wait_end();
1365 rhaas@postgresql.org 7250 [ + + ]: 45 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7251 : : DELAY_CHKPT_START));
7252 : : }
4763 simon@2ndQuadrant.co 7253 : 1532 : pfree(vxids);
7254 : :
6748 tgl@sss.pgh.pa.us 7255 : 1532 : CheckPointGuts(checkPoint.redo, flags);
7256 : :
1365 rhaas@postgresql.org 7257 : 1532 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_COMPLETE);
7258 [ - + ]: 1532 : if (nvxids > 0)
7259 : : {
7260 : : do
7261 : : {
545 heikki.linnakangas@i 7262 :UBC 0 : AbsorbSyncRequests();
7263 : :
797 tmunro@postgresql.or 7264 : 0 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_COMPLETE);
1365 rhaas@postgresql.org 7265 : 0 : pg_usleep(10000L); /* wait for 10 msec */
797 tmunro@postgresql.or 7266 : 0 : pgstat_report_wait_end();
1365 rhaas@postgresql.org 7267 [ # # ]: 0 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7268 : : DELAY_CHKPT_COMPLETE));
7269 : : }
1365 rhaas@postgresql.org 7270 :CBC 1532 : pfree(vxids);
7271 : :
7272 : : /*
7273 : : * Take a snapshot of running transactions and write this to WAL. This
7274 : : * allows us to reconstruct the state of running transactions during
7275 : : * archive recovery, if required. Skip, if this info disabled.
7276 : : *
7277 : : * If we are shutting down, or Startup process is completing crash
7278 : : * recovery we don't need to write running xact data.
7279 : : */
5843 simon@2ndQuadrant.co 7280 [ + + + + ]: 1532 : if (!shutdown && XLogStandbyInfoActive())
4764 tgl@sss.pgh.pa.us 7281 : 864 : LogStandbySnapshot();
7282 : :
8258 7283 : 1532 : START_CRIT_SECTION();
7284 : :
7285 : : /*
7286 : : * Now insert the checkpoint record into XLOG.
7287 : : */
4046 heikki.linnakangas@i 7288 : 1532 : XLogBeginInsert();
310 peter@eisentraut.org 7289 : 1532 : XLogRegisterData(&checkPoint, sizeof(checkPoint));
9046 tgl@sss.pgh.pa.us 7290 [ + + ]: 1532 : recptr = XLogInsert(RM_XLOG_ID,
7291 : : shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
7292 : : XLOG_CHECKPOINT_ONLINE);
7293 : :
7294 : 1532 : XLogFlush(recptr);
7295 : :
7296 : : /*
7297 : : * We mustn't write any new WAL after a shutdown checkpoint, or it will be
7298 : : * overwritten at next startup. No-one should even try, this just allows
7299 : : * sanity-checking. In the case of an end-of-recovery checkpoint, we want
7300 : : * to just temporarily disable writing until the system has exited
7301 : : * recovery.
7302 : : */
6019 7303 [ + + ]: 1532 : if (shutdown)
7304 : : {
7305 [ + + ]: 617 : if (flags & CHECKPOINT_END_OF_RECOVERY)
1515 rhaas@postgresql.org 7306 : 28 : LocalXLogInsertAllowed = oldXLogAllowed;
7307 : : else
5774 bruce@momjian.us 7308 : 589 : LocalXLogInsertAllowed = 0; /* never again write WAL */
7309 : : }
7310 : :
7311 : : /*
7312 : : * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
7313 : : * = end of actual checkpoint record.
7314 : : */
4738 alvherre@alvh.no-ip. 7315 [ + + - + ]: 1532 : if (shutdown && checkPoint.redo != ProcLastRecPtr)
8186 tgl@sss.pgh.pa.us 7316 [ # # ]:UBC 0 : ereport(PANIC,
7317 : : (errmsg("concurrent write-ahead log activity while database system is shutting down")));
7318 : :
7319 : : /*
7320 : : * Remember the prior checkpoint's redo ptr for
7321 : : * UpdateCheckPointDistanceEstimate()
7322 : : */
3951 heikki.linnakangas@i 7323 :CBC 1532 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7324 : :
7325 : : /*
7326 : : * Update the control file.
7327 : : */
8846 tgl@sss.pgh.pa.us 7328 : 1532 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9579 vadim4o@yahoo.com 7329 [ + + ]: 1532 : if (shutdown)
7330 : 617 : ControlFile->state = DB_SHUTDOWNED;
9046 tgl@sss.pgh.pa.us 7331 : 1532 : ControlFile->checkPoint = ProcLastRecPtr;
7332 : 1532 : ControlFile->checkPointCopy = checkPoint;
7333 : : /* crash recovery should always recover to the end of WAL */
4739 alvherre@alvh.no-ip. 7334 : 1532 : ControlFile->minRecoveryPoint = InvalidXLogRecPtr;
4762 heikki.linnakangas@i 7335 : 1532 : ControlFile->minRecoveryPointTLI = 0;
7336 : :
7337 : : /*
7338 : : * Persist unloggedLSN value. It's reset on crash recovery, so this goes
7339 : : * unused on non-shutdown checkpoints, but seems useful to store it always
7340 : : * for debugging purposes.
7341 : : */
658 nathan@postgresql.or 7342 : 1532 : ControlFile->unloggedLSN = pg_atomic_read_membarrier_u64(&XLogCtl->unloggedLSN);
7343 : :
9579 vadim4o@yahoo.com 7344 : 1532 : UpdateControlFile();
8846 tgl@sss.pgh.pa.us 7345 : 1532 : LWLockRelease(ControlFileLock);
7346 : :
7347 : : /*
7348 : : * We are now done with critical updates; no need for system panic if we
7349 : : * have trouble while fooling with old log segments.
7350 : : */
8258 7351 [ - + ]: 1532 : END_CRIT_SECTION();
7352 : :
7353 : : /*
7354 : : * WAL summaries end when the next XLOG_CHECKPOINT_REDO or
7355 : : * XLOG_CHECKPOINT_SHUTDOWN record is reached. This is the first point
7356 : : * where (a) we're not inside of a critical section and (b) we can be
7357 : : * certain that the relevant record has been flushed to disk, which must
7358 : : * happen before it can be summarized.
7359 : : *
7360 : : * If this is a shutdown checkpoint, then this happens reasonably
7361 : : * promptly: we've only just inserted and flushed the
7362 : : * XLOG_CHECKPOINT_SHUTDOWN record. If this is not a shutdown checkpoint,
7363 : : * then this might not be very prompt at all: the XLOG_CHECKPOINT_REDO
7364 : : * record was written before we began flushing data to disk, and that
7365 : : * could be many minutes ago at this point. However, we don't XLogFlush()
7366 : : * after inserting that record, so we're not guaranteed that it's on disk
7367 : : * until after the above call that flushes the XLOG_CHECKPOINT_ONLINE
7368 : : * record.
7369 : : */
412 heikki.linnakangas@i 7370 : 1532 : WakeupWalSummarizer();
7371 : :
7372 : : /*
7373 : : * Let smgr do post-checkpoint cleanup (eg, deleting old files).
7374 : : */
2450 tmunro@postgresql.or 7375 : 1532 : SyncPostCheckpoint();
7376 : :
7377 : : /*
7378 : : * Update the average distance between checkpoints if the prior checkpoint
7379 : : * exists.
7380 : : */
42 alvherre@kurilemu.de 7381 [ + - ]:GNC 1532 : if (XLogRecPtrIsValid(PriorRedoPtr))
3951 heikki.linnakangas@i 7382 :CBC 1532 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7383 : :
7384 : : INJECTION_POINT("checkpoint-before-old-wal-removal", NULL);
7385 : :
7386 : : /*
7387 : : * Delete old log files, those no longer needed for last checkpoint to
7388 : : * prevent the disk holding the xlog from growing full.
7389 : : */
2704 michael@paquier.xyz 7390 : 1532 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7391 : 1532 : KeepLogSeg(recptr, &_logSegNo);
302 akapila@postgresql.o 7392 [ + + ]: 1532 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED | RS_INVAL_IDLE_TIMEOUT,
7393 : : _logSegNo, InvalidOid,
7394 : : InvalidTransactionId))
7395 : : {
7396 : : /*
7397 : : * Some slots have been invalidated; recalculate the old-segment
7398 : : * horizon, starting again from RedoRecPtr.
7399 : : */
1616 alvherre@alvh.no-ip. 7400 : 3 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7401 : 3 : KeepLogSeg(recptr, &_logSegNo);
7402 : : }
2704 michael@paquier.xyz 7403 : 1532 : _logSegNo--;
1504 rhaas@postgresql.org 7404 : 1532 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, recptr,
7405 : : checkPoint.ThisTimeLineID);
7406 : :
7407 : : /*
7408 : : * Make more log segments if needed. (Do this after recycling old log
7409 : : * segments, since that may supply some of the needed files.)
7410 : : */
9046 tgl@sss.pgh.pa.us 7411 [ + + ]: 1532 : if (!shutdown)
1504 rhaas@postgresql.org 7412 : 915 : PreallocXlogFiles(recptr, checkPoint.ThisTimeLineID);
7413 : :
7414 : : /*
7415 : : * Truncate pg_subtrans if possible. We can throw away all data before
7416 : : * the oldest XMIN of any running transaction. No future transaction will
7417 : : * attempt to reference any pg_subtrans entry older than that (see Asserts
7418 : : * in subtrans.c). During recovery, though, we mustn't do this because
7419 : : * StartupSUBTRANS hasn't been called yet.
7420 : : */
6019 tgl@sss.pgh.pa.us 7421 [ + + ]: 1532 : if (!RecoveryInProgress())
1954 andres@anarazel.de 7422 : 1504 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7423 : :
7424 : : /* Real work is done; log and update stats. */
5005 rhaas@postgresql.org 7425 : 1532 : LogCheckpointEnd(false);
7426 : :
7427 : : /* Reset the process title */
1830 michael@paquier.xyz 7428 : 1532 : update_checkpoint_display(flags, false, true);
7429 : :
7430 : : TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
7431 : : NBuffers,
7432 : : CheckpointStats.ckpt_segs_added,
7433 : : CheckpointStats.ckpt_segs_removed,
7434 : : CheckpointStats.ckpt_segs_recycled);
7435 : :
444 fujii@postgresql.org 7436 : 1532 : return true;
7437 : : }
7438 : :
7439 : : /*
7440 : : * Mark the end of recovery in WAL though without running a full checkpoint.
7441 : : * We can expect that a restartpoint is likely to be in progress as we
7442 : : * do this, though we are unwilling to wait for it to complete.
7443 : : *
7444 : : * CreateRestartPoint() allows for the case where recovery may end before
7445 : : * the restartpoint completes so there is no concern of concurrent behaviour.
7446 : : */
7447 : : static void
4706 simon@2ndQuadrant.co 7448 : 42 : CreateEndOfRecoveryRecord(void)
7449 : : {
7450 : : xl_end_of_recovery xlrec;
7451 : : XLogRecPtr recptr;
7452 : :
7453 : : /* sanity check */
7454 [ - + ]: 42 : if (!RecoveryInProgress())
4706 simon@2ndQuadrant.co 7455 [ # # ]:UBC 0 : elog(ERROR, "can only be used to end recovery");
7456 : :
4017 heikki.linnakangas@i 7457 :CBC 42 : xlrec.end_time = GetCurrentTimestamp();
518 rhaas@postgresql.org 7458 : 42 : xlrec.wal_level = wal_level;
7459 : :
4290 heikki.linnakangas@i 7460 : 42 : WALInsertLockAcquireExclusive();
1499 rhaas@postgresql.org 7461 : 42 : xlrec.ThisTimeLineID = XLogCtl->InsertTimeLineID;
4693 heikki.linnakangas@i 7462 : 42 : xlrec.PrevTimeLineID = XLogCtl->PrevTimeLineID;
4290 7463 : 42 : WALInsertLockRelease();
7464 : :
4706 simon@2ndQuadrant.co 7465 : 42 : START_CRIT_SECTION();
7466 : :
4046 heikki.linnakangas@i 7467 : 42 : XLogBeginInsert();
310 peter@eisentraut.org 7468 : 42 : XLogRegisterData(&xlrec, sizeof(xl_end_of_recovery));
4046 heikki.linnakangas@i 7469 : 42 : recptr = XLogInsert(RM_XLOG_ID, XLOG_END_OF_RECOVERY);
7470 : :
4704 simon@2ndQuadrant.co 7471 : 42 : XLogFlush(recptr);
7472 : :
7473 : : /*
7474 : : * Update the control file so that crash recovery can follow the timeline
7475 : : * changes to this point.
7476 : : */
7477 : 42 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7478 : 42 : ControlFile->minRecoveryPoint = recptr;
1504 rhaas@postgresql.org 7479 : 42 : ControlFile->minRecoveryPointTLI = xlrec.ThisTimeLineID;
4704 simon@2ndQuadrant.co 7480 : 42 : UpdateControlFile();
7481 : 42 : LWLockRelease(ControlFileLock);
7482 : :
4706 7483 [ - + ]: 42 : END_CRIT_SECTION();
7484 : 42 : }
7485 : :
7486 : : /*
7487 : : * Write an OVERWRITE_CONTRECORD message.
7488 : : *
7489 : : * When on WAL replay we expect a continuation record at the start of a page
7490 : : * that is not there, recovery ends and WAL writing resumes at that point.
7491 : : * But it's wrong to resume writing new WAL back at the start of the record
7492 : : * that was broken, because downstream consumers of that WAL (physical
7493 : : * replicas) are not prepared to "rewind". So the first action after
7494 : : * finishing replay of all valid WAL must be to write a record of this type
7495 : : * at the point where the contrecord was missing; to support xlogreader
7496 : : * detecting the special case, XLP_FIRST_IS_OVERWRITE_CONTRECORD is also added
7497 : : * to the page header where the record occurs. xlogreader has an ad-hoc
7498 : : * mechanism to report metadata about the broken record, which is what we
7499 : : * use here.
7500 : : *
7501 : : * At replay time, XLP_FIRST_IS_OVERWRITE_CONTRECORD instructs xlogreader to
7502 : : * skip the record it was reading, and pass back the LSN of the skipped
7503 : : * record, so that its caller can verify (on "replay" of that record) that the
7504 : : * XLOG_OVERWRITE_CONTRECORD matches what was effectively overwritten.
7505 : : *
7506 : : * 'aborted_lsn' is the beginning position of the record that was incomplete.
7507 : : * It is included in the WAL record. 'pagePtr' and 'newTLI' point to the
7508 : : * beginning of the XLOG page where the record is to be inserted. They must
7509 : : * match the current WAL insert position, they're passed here just so that we
7510 : : * can verify that.
7511 : : */
7512 : : static XLogRecPtr
1401 heikki.linnakangas@i 7513 : 10 : CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn, XLogRecPtr pagePtr,
7514 : : TimeLineID newTLI)
7515 : : {
7516 : : xl_overwrite_contrecord xlrec;
7517 : : XLogRecPtr recptr;
7518 : : XLogPageHeader pagehdr;
7519 : : XLogRecPtr startPos;
7520 : :
7521 : : /* sanity checks */
1541 alvherre@alvh.no-ip. 7522 [ - + ]: 10 : if (!RecoveryInProgress())
1541 alvherre@alvh.no-ip. 7523 [ # # ]:UBC 0 : elog(ERROR, "can only be used at end of recovery");
1401 heikki.linnakangas@i 7524 [ - + ]:CBC 10 : if (pagePtr % XLOG_BLCKSZ != 0)
164 alvherre@kurilemu.de 7525 [ # # ]:UNC 0 : elog(ERROR, "invalid position for missing continuation record %X/%08X",
7526 : : LSN_FORMAT_ARGS(pagePtr));
7527 : :
7528 : : /* The current WAL insert position should be right after the page header */
1401 heikki.linnakangas@i 7529 :CBC 10 : startPos = pagePtr;
7530 [ + + ]: 10 : if (XLogSegmentOffset(startPos, wal_segment_size) == 0)
7531 : 1 : startPos += SizeOfXLogLongPHD;
7532 : : else
7533 : 9 : startPos += SizeOfXLogShortPHD;
7534 : 10 : recptr = GetXLogInsertRecPtr();
7535 [ - + ]: 10 : if (recptr != startPos)
164 alvherre@kurilemu.de 7536 [ # # ]:UNC 0 : elog(ERROR, "invalid WAL insert position %X/%08X for OVERWRITE_CONTRECORD",
7537 : : LSN_FORMAT_ARGS(recptr));
7538 : :
1541 alvherre@alvh.no-ip. 7539 :CBC 10 : START_CRIT_SECTION();
7540 : :
7541 : : /*
7542 : : * Initialize the XLOG page header (by GetXLogBuffer), and set the
7543 : : * XLP_FIRST_IS_OVERWRITE_CONTRECORD flag.
7544 : : *
7545 : : * No other backend is allowed to write WAL yet, so acquiring the WAL
7546 : : * insertion lock is just pro forma.
7547 : : */
1401 heikki.linnakangas@i 7548 : 10 : WALInsertLockAcquire();
7549 : 10 : pagehdr = (XLogPageHeader) GetXLogBuffer(pagePtr, newTLI);
7550 : 10 : pagehdr->xlp_info |= XLP_FIRST_IS_OVERWRITE_CONTRECORD;
7551 : 10 : WALInsertLockRelease();
7552 : :
7553 : : /*
7554 : : * Insert the XLOG_OVERWRITE_CONTRECORD record as the first record on the
7555 : : * page. We know it becomes the first record, because no other backend is
7556 : : * allowed to write WAL yet.
7557 : : */
1541 alvherre@alvh.no-ip. 7558 : 10 : XLogBeginInsert();
1401 heikki.linnakangas@i 7559 : 10 : xlrec.overwritten_lsn = aborted_lsn;
7560 : 10 : xlrec.overwrite_time = GetCurrentTimestamp();
310 peter@eisentraut.org 7561 : 10 : XLogRegisterData(&xlrec, sizeof(xl_overwrite_contrecord));
1541 alvherre@alvh.no-ip. 7562 : 10 : recptr = XLogInsert(RM_XLOG_ID, XLOG_OVERWRITE_CONTRECORD);
7563 : :
7564 : : /* check that the record was inserted to the right place */
1401 heikki.linnakangas@i 7565 [ - + ]: 10 : if (ProcLastRecPtr != startPos)
164 alvherre@kurilemu.de 7566 [ # # ]:UNC 0 : elog(ERROR, "OVERWRITE_CONTRECORD was inserted to unexpected position %X/%08X",
7567 : : LSN_FORMAT_ARGS(ProcLastRecPtr));
7568 : :
1541 alvherre@alvh.no-ip. 7569 :CBC 10 : XLogFlush(recptr);
7570 : :
7571 [ - + ]: 10 : END_CRIT_SECTION();
7572 : :
7573 : 10 : return recptr;
7574 : : }
7575 : :
7576 : : /*
7577 : : * Flush all data in shared memory to disk, and fsync
7578 : : *
7579 : : * This is the common code shared between regular checkpoints and
7580 : : * recovery restartpoints.
7581 : : */
7582 : : static void
6748 tgl@sss.pgh.pa.us 7583 : 1732 : CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
7584 : : {
5793 7585 : 1732 : CheckPointRelationMap();
826 akapila@postgresql.o 7586 : 1732 : CheckPointReplicationSlots(flags & CHECKPOINT_IS_SHUTDOWN);
4308 rhaas@postgresql.org 7587 : 1732 : CheckPointSnapBuild();
7588 : 1732 : CheckPointLogicalRewriteHeap();
3886 andres@anarazel.de 7589 : 1732 : CheckPointReplicationOrigin();
7590 : :
7591 : : /* Write out all dirty data in SLRUs and the main buffer pool */
7592 : : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_START(flags);
1910 tmunro@postgresql.or 7593 : 1732 : CheckpointStats.ckpt_write_t = GetCurrentTimestamp();
7594 : 1732 : CheckPointCLOG();
7595 : 1732 : CheckPointCommitTs();
7596 : 1732 : CheckPointSUBTRANS();
7597 : 1732 : CheckPointMultiXact();
7598 : 1732 : CheckPointPredicate();
7599 : 1732 : CheckPointBuffers(flags);
7600 : :
7601 : : /* Perform all queued up fsyncs */
7602 : : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_SYNC_START();
7603 : 1732 : CheckpointStats.ckpt_sync_t = GetCurrentTimestamp();
7604 : 1732 : ProcessSyncRequests();
7605 : 1732 : CheckpointStats.ckpt_sync_end_t = GetCurrentTimestamp();
7606 : : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_DONE();
7607 : :
7608 : : /* We deliberately delay 2PC checkpointing as long as possible */
7073 tgl@sss.pgh.pa.us 7609 : 1732 : CheckPointTwoPhase(checkPointRedo);
7610 : 1732 : }
7611 : :
7612 : : /*
7613 : : * Save a checkpoint for recovery restart if appropriate
7614 : : *
7615 : : * This function is called each time a checkpoint record is read from XLOG.
7616 : : * It must determine whether the checkpoint represents a safe restartpoint or
7617 : : * not. If so, the checkpoint record is stashed in shared memory so that
7618 : : * CreateRestartPoint can consult it. (Note that the latter function is
7619 : : * executed by the checkpointer, while this one will be executed by the
7620 : : * startup process.)
7621 : : */
7622 : : static void
1485 rhaas@postgresql.org 7623 : 702 : RecoveryRestartPoint(const CheckPoint *checkPoint, XLogReaderState *record)
7624 : : {
7625 : : /*
7626 : : * Also refrain from creating a restartpoint if we have seen any
7627 : : * references to non-existent pages. Restarting recovery from the
7628 : : * restartpoint would not see the references, so we would lose the
7629 : : * cross-check that the pages belonged to a relation that was dropped
7630 : : * later.
7631 : : */
5130 heikki.linnakangas@i 7632 [ - + ]: 702 : if (XLogHaveInvalidPages())
7633 : : {
738 michael@paquier.xyz 7634 [ # # ]:UBC 0 : elog(DEBUG2,
7635 : : "could not record restart point at %X/%08X because there are unresolved references to invalid pages",
7636 : : LSN_FORMAT_ARGS(checkPoint->redo));
5130 heikki.linnakangas@i 7637 : 0 : return;
7638 : : }
7639 : :
7640 : : /*
7641 : : * Copy the checkpoint record to shared memory, so that checkpointer can
7642 : : * work out the next time it wants to perform a restartpoint.
7643 : : */
4105 andres@anarazel.de 7644 [ - + ]:CBC 702 : SpinLockAcquire(&XLogCtl->info_lck);
1485 rhaas@postgresql.org 7645 : 702 : XLogCtl->lastCheckPointRecPtr = record->ReadRecPtr;
7646 : 702 : XLogCtl->lastCheckPointEndPtr = record->EndRecPtr;
4105 andres@anarazel.de 7647 : 702 : XLogCtl->lastCheckPoint = *checkPoint;
7648 : 702 : SpinLockRelease(&XLogCtl->info_lck);
7649 : : }
7650 : :
7651 : : /*
7652 : : * Establish a restartpoint if possible.
7653 : : *
7654 : : * This is similar to CreateCheckPoint, but is used during WAL recovery
7655 : : * to establish a point from which recovery can roll forward without
7656 : : * replaying the entire recovery log.
7657 : : *
7658 : : * Returns true if a new restartpoint was established. We can only establish
7659 : : * a restartpoint if we have replayed a safe checkpoint record since last
7660 : : * restartpoint.
7661 : : */
7662 : : bool
6147 heikki.linnakangas@i 7663 : 601 : CreateRestartPoint(int flags)
7664 : : {
7665 : : XLogRecPtr lastCheckPointRecPtr;
7666 : : XLogRecPtr lastCheckPointEndPtr;
7667 : : CheckPoint lastCheckPoint;
7668 : : XLogRecPtr PriorRedoPtr;
7669 : : XLogRecPtr receivePtr;
7670 : : XLogRecPtr replayPtr;
7671 : : TimeLineID replayTLI;
7672 : : XLogRecPtr endptr;
7673 : : XLogSegNo _logSegNo;
7674 : : TimestampTz xtime;
7675 : :
7676 : : /* Concurrent checkpoint/restartpoint cannot happen */
1319 michael@paquier.xyz 7677 [ + - - + ]: 601 : Assert(!IsUnderPostmaster || MyBackendType == B_CHECKPOINTER);
7678 : :
7679 : : /* Get a local copy of the last safe checkpoint record. */
4105 andres@anarazel.de 7680 [ - + ]: 601 : SpinLockAcquire(&XLogCtl->info_lck);
7681 : 601 : lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
3339 rhaas@postgresql.org 7682 : 601 : lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
4105 andres@anarazel.de 7683 : 601 : lastCheckPoint = XLogCtl->lastCheckPoint;
7684 : 601 : SpinLockRelease(&XLogCtl->info_lck);
7685 : :
7686 : : /*
7687 : : * Check that we're still in recovery mode. It's ok if we exit recovery
7688 : : * mode after this check, the restart point is valid anyway.
7689 : : */
6147 heikki.linnakangas@i 7690 [ - + ]: 601 : if (!RecoveryInProgress())
7691 : : {
6147 heikki.linnakangas@i 7692 [ # # ]:UBC 0 : ereport(DEBUG2,
7693 : : (errmsg_internal("skipping restartpoint, recovery has already ended")));
7694 : 0 : return false;
7695 : : }
7696 : :
7697 : : /*
7698 : : * If the last checkpoint record we've replayed is already our last
7699 : : * restartpoint, we can't perform a new restart point. We still update
7700 : : * minRecoveryPoint in that case, so that if this is a shutdown restart
7701 : : * point, we won't start up earlier than before. That's not strictly
7702 : : * necessary, but when hot standby is enabled, it would be rather weird if
7703 : : * the database opened up for read-only connections at a point-in-time
7704 : : * before the last shutdown. Such time travel is still possible in case of
7705 : : * immediate shutdown, though.
7706 : : *
7707 : : * We don't explicitly advance minRecoveryPoint when we do create a
7708 : : * restartpoint. It's assumed that flushing the buffers will do that as a
7709 : : * side-effect.
7710 : : */
42 alvherre@kurilemu.de 7711 [ + + ]:GNC 601 : if (!XLogRecPtrIsValid(lastCheckPointRecPtr) ||
4738 alvherre@alvh.no-ip. 7712 [ + + ]:CBC 270 : lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
7713 : : {
6147 heikki.linnakangas@i 7714 [ - + ]: 401 : ereport(DEBUG2,
7715 : : errmsg_internal("skipping restartpoint, already performed at %X/%08X",
7716 : : LSN_FORMAT_ARGS(lastCheckPoint.redo)));
7717 : :
7718 : 401 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5677 rhaas@postgresql.org 7719 [ + + ]: 401 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7720 : : {
7721 : 31 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7722 : 31 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7723 : 31 : UpdateControlFile();
7724 : 31 : LWLockRelease(ControlFileLock);
7725 : : }
6147 heikki.linnakangas@i 7726 : 401 : return false;
7727 : : }
7728 : :
7729 : : /*
7730 : : * Update the shared RedoRecPtr so that the startup process can calculate
7731 : : * the number of segments replayed since last restartpoint, and request a
7732 : : * restartpoint if it exceeds CheckPointSegments.
7733 : : *
7734 : : * Like in CreateCheckPoint(), hold off insertions to update it, although
7735 : : * during recovery this is just pro forma, because no WAL insertions are
7736 : : * happening.
7737 : : */
4290 7738 : 200 : WALInsertLockAcquireExclusive();
3951 7739 : 200 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
4290 7740 : 200 : WALInsertLockRelease();
7741 : :
7742 : : /* Also update the info_lck-protected copy */
4105 andres@anarazel.de 7743 [ - + ]: 200 : SpinLockAcquire(&XLogCtl->info_lck);
7744 : 200 : XLogCtl->RedoRecPtr = lastCheckPoint.redo;
7745 : 200 : SpinLockRelease(&XLogCtl->info_lck);
7746 : :
7747 : : /*
7748 : : * Prepare to accumulate statistics.
7749 : : *
7750 : : * Note: because it is possible for log_checkpoints to change while a
7751 : : * checkpoint proceeds, we always accumulate stats, even if
7752 : : * log_checkpoints is currently off.
7753 : : */
5433 rhaas@postgresql.org 7754 [ + - + - : 2200 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ - + - +
+ ]
7755 : 200 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
7756 : :
7757 [ + - ]: 200 : if (log_checkpoints)
6147 heikki.linnakangas@i 7758 : 200 : LogCheckpointStart(flags, true);
7759 : :
7760 : : /* Update the process title */
1830 michael@paquier.xyz 7761 : 200 : update_checkpoint_display(flags, true, false);
7762 : :
6147 heikki.linnakangas@i 7763 : 200 : CheckPointGuts(lastCheckPoint.redo, flags);
7764 : :
7765 : : /*
7766 : : * This location needs to be after CheckPointGuts() to ensure that some
7767 : : * work has already happened during this checkpoint.
7768 : : */
7769 : : INJECTION_POINT("create-restart-point", NULL);
7770 : :
7771 : : /*
7772 : : * Remember the prior checkpoint's redo ptr for
7773 : : * UpdateCheckPointDistanceEstimate()
7774 : : */
3951 7775 : 200 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7776 : :
7777 : : /*
7778 : : * Update pg_control, using current time. Check that it still shows an
7779 : : * older checkpoint, else do nothing; this is a quick hack to make sure
7780 : : * nothing really bad happens if somehow we get here after the
7781 : : * end-of-recovery checkpoint.
7782 : : */
6147 7783 : 200 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
1319 michael@paquier.xyz 7784 [ + - ]: 200 : if (ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
7785 : : {
7786 : : /*
7787 : : * Update the checkpoint information. We do this even if the cluster
7788 : : * does not show DB_IN_ARCHIVE_RECOVERY to match with the set of WAL
7789 : : * segments recycled below.
7790 : : */
6019 tgl@sss.pgh.pa.us 7791 : 200 : ControlFile->checkPoint = lastCheckPointRecPtr;
7792 : 200 : ControlFile->checkPointCopy = lastCheckPoint;
7793 : :
7794 : : /*
7795 : : * Ensure minRecoveryPoint is past the checkpoint record and update it
7796 : : * if the control file still shows DB_IN_ARCHIVE_RECOVERY. Normally,
7797 : : * this will have happened already while writing out dirty buffers,
7798 : : * but not necessarily - e.g. because no buffers were dirtied. We do
7799 : : * this because a backup performed in recovery uses minRecoveryPoint
7800 : : * to determine which WAL files must be included in the backup, and
7801 : : * the file (or files) containing the checkpoint record must be
7802 : : * included, at a minimum. Note that for an ordinary restart of
7803 : : * recovery there's no value in having the minimum recovery point any
7804 : : * earlier than this anyway, because redo will begin just after the
7805 : : * checkpoint record.
7806 : : */
1319 michael@paquier.xyz 7807 [ + - ]: 200 : if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY)
7808 : : {
7809 [ + + ]: 200 : if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
7810 : : {
7811 : 18 : ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
7812 : 18 : ControlFile->minRecoveryPointTLI = lastCheckPoint.ThisTimeLineID;
7813 : :
7814 : : /* update local copy */
7815 : 18 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
7816 : 18 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
7817 : : }
7818 [ + + ]: 200 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7819 : 21 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7820 : : }
6019 tgl@sss.pgh.pa.us 7821 : 200 : UpdateControlFile();
7822 : : }
6147 heikki.linnakangas@i 7823 : 200 : LWLockRelease(ControlFileLock);
7824 : :
7825 : : /*
7826 : : * Update the average distance between checkpoints/restartpoints if the
7827 : : * prior checkpoint exists.
7828 : : */
42 alvherre@kurilemu.de 7829 [ + - ]:GNC 200 : if (XLogRecPtrIsValid(PriorRedoPtr))
3951 heikki.linnakangas@i 7830 :CBC 200 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7831 : :
7832 : : /*
7833 : : * Delete old log files, those no longer needed for last restartpoint to
7834 : : * prevent the disk holding the xlog from growing full.
7835 : : */
2704 michael@paquier.xyz 7836 : 200 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7837 : :
7838 : : /*
7839 : : * Retreat _logSegNo using the current end of xlog replayed or received,
7840 : : * whichever is later.
7841 : : */
2080 tmunro@postgresql.or 7842 : 200 : receivePtr = GetWalRcvFlushRecPtr(NULL, NULL);
2704 michael@paquier.xyz 7843 : 200 : replayPtr = GetXLogReplayRecPtr(&replayTLI);
7844 : 200 : endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
7845 : 200 : KeepLogSeg(endptr, &_logSegNo);
302 akapila@postgresql.o 7846 [ + + ]: 200 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED | RS_INVAL_IDLE_TIMEOUT,
7847 : : _logSegNo, InvalidOid,
7848 : : InvalidTransactionId))
7849 : : {
7850 : : /*
7851 : : * Some slots have been invalidated; recalculate the old-segment
7852 : : * horizon, starting again from RedoRecPtr.
7853 : : */
1616 alvherre@alvh.no-ip. 7854 : 1 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7855 : 1 : KeepLogSeg(endptr, &_logSegNo);
7856 : : }
2704 michael@paquier.xyz 7857 : 200 : _logSegNo--;
7858 : :
7859 : : /*
7860 : : * Try to recycle segments on a useful timeline. If we've been promoted
7861 : : * since the beginning of this restartpoint, use the new timeline chosen
7862 : : * at end of recovery. If we're still in recovery, use the timeline we're
7863 : : * currently replaying.
7864 : : *
7865 : : * There is no guarantee that the WAL segments will be useful on the
7866 : : * current timeline; if recovery proceeds to a new timeline right after
7867 : : * this, the pre-allocated WAL segments on this timeline will not be used,
7868 : : * and will go wasted until recycled on the next restartpoint. We'll live
7869 : : * with that.
7870 : : */
1504 rhaas@postgresql.org 7871 [ - + ]: 200 : if (!RecoveryInProgress())
1499 rhaas@postgresql.org 7872 :UBC 0 : replayTLI = XLogCtl->InsertTimeLineID;
7873 : :
1504 rhaas@postgresql.org 7874 :CBC 200 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, endptr, replayTLI);
7875 : :
7876 : : /*
7877 : : * Make more log segments if needed. (Do this after recycling old log
7878 : : * segments, since that may supply some of the needed files.)
7879 : : */
7880 : 200 : PreallocXlogFiles(endptr, replayTLI);
7881 : :
7882 : : /*
7883 : : * Truncate pg_subtrans if possible. We can throw away all data before
7884 : : * the oldest XMIN of any running transaction. No future transaction will
7885 : : * attempt to reference any pg_subtrans entry older than that (see Asserts
7886 : : * in subtrans.c). When hot standby is disabled, though, we mustn't do
7887 : : * this because StartupSUBTRANS hasn't been called yet.
7888 : : */
5589 simon@2ndQuadrant.co 7889 [ + - ]: 200 : if (EnableHotStandby)
1954 andres@anarazel.de 7890 : 200 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7891 : :
7892 : : /* Real work is done; log and update stats. */
5005 rhaas@postgresql.org 7893 : 200 : LogCheckpointEnd(true);
7894 : :
7895 : : /* Reset the process title */
1830 michael@paquier.xyz 7896 : 200 : update_checkpoint_display(flags, true, true);
7897 : :
5647 tgl@sss.pgh.pa.us 7898 : 200 : xtime = GetLatestXTime();
6147 heikki.linnakangas@i 7899 [ + - + - : 200 : ereport((log_checkpoints ? LOG : DEBUG2),
+ + ]
7900 : : errmsg("recovery restart point at %X/%08X",
7901 : : LSN_FORMAT_ARGS(lastCheckPoint.redo)),
7902 : : xtime ? errdetail("Last completed transaction was at log time %s.",
7903 : : timestamptz_to_str(xtime)) : 0);
7904 : :
7905 : : /*
7906 : : * Finally, execute archive_cleanup_command, if any.
7907 : : */
2580 peter_e@gmx.net 7908 [ + - - + ]: 200 : if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
1046 michael@paquier.xyz 7909 :UBC 0 : ExecuteRecoveryCommand(archiveCleanupCommand,
7910 : : "archive_cleanup_command",
7911 : : false,
7912 : : WAIT_EVENT_ARCHIVE_CLEANUP_COMMAND);
7913 : :
6147 heikki.linnakangas@i 7914 :CBC 200 : return true;
7915 : : }
7916 : :
7917 : : /*
7918 : : * Report availability of WAL for the given target LSN
7919 : : * (typically a slot's restart_lsn)
7920 : : *
7921 : : * Returns one of the following enum values:
7922 : : *
7923 : : * * WALAVAIL_RESERVED means targetLSN is available and it is in the range of
7924 : : * max_wal_size.
7925 : : *
7926 : : * * WALAVAIL_EXTENDED means it is still available by preserving extra
7927 : : * segments beyond max_wal_size. If max_slot_wal_keep_size is smaller
7928 : : * than max_wal_size, this state is not returned.
7929 : : *
7930 : : * * WALAVAIL_UNRESERVED means it is being lost and the next checkpoint will
7931 : : * remove reserved segments. The walsender using this slot may return to the
7932 : : * above.
7933 : : *
7934 : : * * WALAVAIL_REMOVED means it has been removed. A replication stream on
7935 : : * a slot with this LSN cannot continue. (Any associated walsender
7936 : : * processes should have been terminated already.)
7937 : : *
7938 : : * * WALAVAIL_INVALID_LSN means the slot hasn't been set to reserve WAL.
7939 : : */
7940 : : WALAvailability
2081 alvherre@alvh.no-ip. 7941 : 391 : GetWALAvailability(XLogRecPtr targetLSN)
7942 : : {
7943 : : XLogRecPtr currpos; /* current write LSN */
7944 : : XLogSegNo currSeg; /* segid of currpos */
7945 : : XLogSegNo targetSeg; /* segid of targetLSN */
7946 : : XLogSegNo oldestSeg; /* actual oldest segid */
7947 : : XLogSegNo oldestSegMaxWalSize; /* oldest segid kept by max_wal_size */
7948 : : XLogSegNo oldestSlotSeg; /* oldest segid kept by slot */
7949 : : uint64 keepSegs;
7950 : :
7951 : : /*
7952 : : * slot does not reserve WAL. Either deactivated, or has never been active
7953 : : */
42 alvherre@kurilemu.de 7954 [ + + ]:GNC 391 : if (!XLogRecPtrIsValid(targetLSN))
2081 alvherre@alvh.no-ip. 7955 :CBC 23 : return WALAVAIL_INVALID_LSN;
7956 : :
7957 : : /*
7958 : : * Calculate the oldest segment currently reserved by all slots,
7959 : : * considering wal_keep_size and max_slot_wal_keep_size. Initialize
7960 : : * oldestSlotSeg to the current segment.
7961 : : */
1984 7962 : 368 : currpos = GetXLogWriteRecPtr();
7963 : 368 : XLByteToSeg(currpos, oldestSlotSeg, wal_segment_size);
2081 7964 : 368 : KeepLogSeg(currpos, &oldestSlotSeg);
7965 : :
7966 : : /*
7967 : : * Find the oldest extant segment file. We get 1 until checkpoint removes
7968 : : * the first WAL segment file since startup, which causes the status being
7969 : : * wrong under certain abnormal conditions but that doesn't actually harm.
7970 : : */
7971 : 368 : oldestSeg = XLogGetLastRemovedSegno() + 1;
7972 : :
7973 : : /* calculate oldest segment by max_wal_size */
7974 : 368 : XLByteToSeg(currpos, currSeg, wal_segment_size);
2003 7975 : 368 : keepSegs = ConvertToXSegs(max_wal_size_mb, wal_segment_size) + 1;
7976 : :
2081 7977 [ + + ]: 368 : if (currSeg > keepSegs)
7978 : 8 : oldestSegMaxWalSize = currSeg - keepSegs;
7979 : : else
7980 : 360 : oldestSegMaxWalSize = 1;
7981 : :
7982 : : /* the segment we care about */
1984 7983 : 368 : XLByteToSeg(targetLSN, targetSeg, wal_segment_size);
7984 : :
7985 : : /*
7986 : : * No point in returning reserved or extended status values if the
7987 : : * targetSeg is known to be lost.
7988 : : */
2003 7989 [ + + ]: 368 : if (targetSeg >= oldestSlotSeg)
7990 : : {
7991 : : /* show "reserved" when targetSeg is within max_wal_size */
7992 [ + + ]: 367 : if (targetSeg >= oldestSegMaxWalSize)
2081 7993 : 365 : return WALAVAIL_RESERVED;
7994 : :
7995 : : /* being retained by slots exceeding max_wal_size */
2003 7996 : 2 : return WALAVAIL_EXTENDED;
7997 : : }
7998 : :
7999 : : /* WAL segments are no longer retained but haven't been removed yet */
8000 [ + - ]: 1 : if (targetSeg >= oldestSeg)
8001 : 1 : return WALAVAIL_UNRESERVED;
8002 : :
8003 : : /* Definitely lost */
2081 alvherre@alvh.no-ip. 8004 :UBC 0 : return WALAVAIL_REMOVED;
8005 : : }
8006 : :
8007 : :
8008 : : /*
8009 : : * Retreat *logSegNo to the last segment that we need to retain because of
8010 : : * either wal_keep_size or replication slots.
8011 : : *
8012 : : * This is calculated by subtracting wal_keep_size from the given xlog
8013 : : * location, recptr and by making sure that that result is below the
8014 : : * requirement of replication slots. For the latter criterion we do consider
8015 : : * the effects of max_slot_wal_keep_size: reserve at most that much space back
8016 : : * from recptr.
8017 : : *
8018 : : * Note about replication slots: if this function calculates a value
8019 : : * that's further ahead than what slots need reserved, then affected
8020 : : * slots need to be invalidated and this function invoked again.
8021 : : * XXX it might be a good idea to rewrite this function so that
8022 : : * invalidation is optionally done here, instead.
8023 : : */
8024 : : static void
4925 heikki.linnakangas@i 8025 :CBC 2104 : KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
8026 : : {
8027 : : XLogSegNo currSegNo;
8028 : : XLogSegNo segno;
8029 : : XLogRecPtr keep;
8030 : :
2081 alvherre@alvh.no-ip. 8031 : 2104 : XLByteToSeg(recptr, currSegNo, wal_segment_size);
8032 : 2104 : segno = currSegNo;
8033 : :
8034 : : /* Calculate how many segments are kept by slots. */
8035 : 2104 : keep = XLogGetReplicationSlotMinimumLSN();
42 alvherre@kurilemu.de 8036 [ + + + + ]:GNC 2104 : if (XLogRecPtrIsValid(keep) && keep < recptr)
8037 : : {
2081 alvherre@alvh.no-ip. 8038 :CBC 627 : XLByteToSeg(keep, segno, wal_segment_size);
8039 : :
8040 : : /*
8041 : : * Account for max_slot_wal_keep_size to avoid keeping more than
8042 : : * configured. However, don't do that during a binary upgrade: if
8043 : : * slots were to be invalidated because of this, it would not be
8044 : : * possible to preserve logical ones during the upgrade.
8045 : : */
160 akapila@postgresql.o 8046 [ + + + - ]: 627 : if (max_slot_wal_keep_size_mb >= 0 && !IsBinaryUpgrade)
8047 : : {
8048 : : uint64 slot_keep_segs;
8049 : :
2081 alvherre@alvh.no-ip. 8050 : 20 : slot_keep_segs =
8051 : 20 : ConvertToXSegs(max_slot_wal_keep_size_mb, wal_segment_size);
8052 : :
8053 [ + + ]: 20 : if (currSegNo - segno > slot_keep_segs)
8054 : 5 : segno = currSegNo - slot_keep_segs;
8055 : : }
8056 : : }
8057 : :
8058 : : /*
8059 : : * If WAL summarization is in use, don't remove WAL that has yet to be
8060 : : * summarized.
8061 : : */
541 rhaas@postgresql.org 8062 : 2104 : keep = GetOldestUnsummarizedLSN(NULL, NULL);
42 alvherre@kurilemu.de 8063 [ + + ]:GNC 2104 : if (XLogRecPtrIsValid(keep))
8064 : : {
8065 : : XLogSegNo unsummarized_segno;
8066 : :
729 rhaas@postgresql.org 8067 :CBC 2 : XLByteToSeg(keep, unsummarized_segno, wal_segment_size);
8068 [ + + ]: 2 : if (unsummarized_segno < segno)
8069 : 1 : segno = unsummarized_segno;
8070 : : }
8071 : :
8072 : : /* but, keep at least wal_keep_size if that's set */
1977 fujii@postgresql.org 8073 [ + + ]: 2104 : if (wal_keep_size_mb > 0)
8074 : : {
8075 : : uint64 keep_segs;
8076 : :
8077 : 74 : keep_segs = ConvertToXSegs(wal_keep_size_mb, wal_segment_size);
8078 [ + - ]: 74 : if (currSegNo - segno < keep_segs)
8079 : : {
8080 : : /* avoid underflow, don't go below 1 */
8081 [ + + ]: 74 : if (currSegNo <= keep_segs)
8082 : 70 : segno = 1;
8083 : : else
8084 : 4 : segno = currSegNo - keep_segs;
8085 : : }
8086 : : }
8087 : :
8088 : : /* don't delete WAL segments newer than the calculated segment */
1984 alvherre@alvh.no-ip. 8089 [ + + ]: 2104 : if (segno < *logSegNo)
4925 heikki.linnakangas@i 8090 : 313 : *logSegNo = segno;
5266 simon@2ndQuadrant.co 8091 : 2104 : }
8092 : :
8093 : : /*
8094 : : * Write a NEXTOID log record
8095 : : */
8096 : : void
9176 vadim4o@yahoo.com 8097 : 606 : XLogPutNextOid(Oid nextOid)
8098 : : {
4046 heikki.linnakangas@i 8099 : 606 : XLogBeginInsert();
310 peter@eisentraut.org 8100 : 606 : XLogRegisterData(&nextOid, sizeof(Oid));
4046 heikki.linnakangas@i 8101 : 606 : (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
8102 : :
8103 : : /*
8104 : : * We need not flush the NEXTOID record immediately, because any of the
8105 : : * just-allocated OIDs could only reach disk as part of a tuple insert or
8106 : : * update that would have its own XLOG record that must follow the NEXTOID
8107 : : * record. Therefore, the standard buffer LSN interlock applied to those
8108 : : * records will ensure no such OID reaches disk before the NEXTOID record
8109 : : * does.
8110 : : *
8111 : : * Note, however, that the above statement only covers state "within" the
8112 : : * database. When we use a generated OID as a file or directory name, we
8113 : : * are in a sense violating the basic WAL rule, because that filesystem
8114 : : * change may reach disk before the NEXTOID WAL record does. The impact
8115 : : * of this is that if a database crash occurs immediately afterward, we
8116 : : * might after restart re-generate the same OID and find that it conflicts
8117 : : * with the leftover file or directory. But since for safety's sake we
8118 : : * always loop until finding a nonconflicting filename, this poses no real
8119 : : * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
8120 : : */
7539 tgl@sss.pgh.pa.us 8121 : 606 : }
8122 : :
8123 : : /*
8124 : : * Write an XLOG SWITCH record.
8125 : : *
8126 : : * Here we just blindly issue an XLogInsert request for the record.
8127 : : * All the magic happens inside XLogInsert.
8128 : : *
8129 : : * The return value is either the end+1 address of the switch record,
8130 : : * or the end+1 address of the prior segment if we did not need to
8131 : : * write a switch record because we are already at segment start.
8132 : : */
8133 : : XLogRecPtr
3283 andres@anarazel.de 8134 : 710 : RequestXLogSwitch(bool mark_unimportant)
8135 : : {
8136 : : XLogRecPtr RecPtr;
8137 : :
8138 : : /* XLOG SWITCH has no data */
4046 heikki.linnakangas@i 8139 : 710 : XLogBeginInsert();
8140 : :
3283 andres@anarazel.de 8141 [ - + ]: 710 : if (mark_unimportant)
3283 andres@anarazel.de 8142 :UBC 0 : XLogSetRecordFlags(XLOG_MARK_UNIMPORTANT);
4046 heikki.linnakangas@i 8143 :CBC 710 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH);
8144 : :
7074 tgl@sss.pgh.pa.us 8145 : 710 : return RecPtr;
8146 : : }
8147 : :
8148 : : /*
8149 : : * Write a RESTORE POINT record
8150 : : */
8151 : : XLogRecPtr
5427 simon@2ndQuadrant.co 8152 : 3 : XLogRestorePoint(const char *rpName)
8153 : : {
8154 : : XLogRecPtr RecPtr;
8155 : : xl_restore_point xlrec;
8156 : :
8157 : 3 : xlrec.rp_time = GetCurrentTimestamp();
4322 tgl@sss.pgh.pa.us 8158 : 3 : strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
8159 : :
4046 heikki.linnakangas@i 8160 : 3 : XLogBeginInsert();
310 peter@eisentraut.org 8161 : 3 : XLogRegisterData(&xlrec, sizeof(xl_restore_point));
8162 : :
4046 heikki.linnakangas@i 8163 : 3 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
8164 : :
5411 rhaas@postgresql.org 8165 [ + - ]: 3 : ereport(LOG,
8166 : : errmsg("restore point \"%s\" created at %X/%08X",
8167 : : rpName, LSN_FORMAT_ARGS(RecPtr)));
8168 : :
5427 simon@2ndQuadrant.co 8169 : 3 : return RecPtr;
8170 : : }
8171 : :
8172 : : /*
8173 : : * Check if any of the GUC parameters that are critical for hot standby
8174 : : * have changed, and update the value in pg_control file if necessary.
8175 : : */
8176 : : static void
5713 heikki.linnakangas@i 8177 : 872 : XLogReportParameters(void)
8178 : : {
8179 [ + + ]: 872 : if (wal_level != ControlFile->wal_level ||
4369 rhaas@postgresql.org 8180 [ + + ]: 647 : wal_log_hints != ControlFile->wal_log_hints ||
5713 heikki.linnakangas@i 8181 [ + + ]: 563 : MaxConnections != ControlFile->MaxConnections ||
4550 rhaas@postgresql.org 8182 [ + + ]: 562 : max_worker_processes != ControlFile->max_worker_processes ||
2501 michael@paquier.xyz 8183 [ + + ]: 561 : max_wal_senders != ControlFile->max_wal_senders ||
5713 heikki.linnakangas@i 8184 [ + + ]: 538 : max_prepared_xacts != ControlFile->max_prepared_xacts ||
4033 alvherre@alvh.no-ip. 8185 [ + - ]: 448 : max_locks_per_xact != ControlFile->max_locks_per_xact ||
8186 [ + + ]: 448 : track_commit_timestamp != ControlFile->track_commit_timestamp)
8187 : : {
8188 : : /*
8189 : : * The change in number of backend slots doesn't need to be WAL-logged
8190 : : * if archiving is not enabled, as you can't start archive recovery
8191 : : * with wal_level=minimal anyway. We don't really care about the
8192 : : * values in pg_control either if wal_level=minimal, but seems better
8193 : : * to keep them up-to-date to avoid confusion.
8194 : : */
5713 heikki.linnakangas@i 8195 [ + + + + ]: 435 : if (wal_level != ControlFile->wal_level || XLogIsNeeded())
8196 : : {
8197 : : xl_parameter_change xlrec;
8198 : : XLogRecPtr recptr;
8199 : :
8200 : 412 : xlrec.MaxConnections = MaxConnections;
4550 rhaas@postgresql.org 8201 : 412 : xlrec.max_worker_processes = max_worker_processes;
2501 michael@paquier.xyz 8202 : 412 : xlrec.max_wal_senders = max_wal_senders;
5713 heikki.linnakangas@i 8203 : 412 : xlrec.max_prepared_xacts = max_prepared_xacts;
8204 : 412 : xlrec.max_locks_per_xact = max_locks_per_xact;
8205 : 412 : xlrec.wal_level = wal_level;
4369 rhaas@postgresql.org 8206 : 412 : xlrec.wal_log_hints = wal_log_hints;
4033 alvherre@alvh.no-ip. 8207 : 412 : xlrec.track_commit_timestamp = track_commit_timestamp;
8208 : :
4046 heikki.linnakangas@i 8209 : 412 : XLogBeginInsert();
310 peter@eisentraut.org 8210 : 412 : XLogRegisterData(&xlrec, sizeof(xlrec));
8211 : :
4046 heikki.linnakangas@i 8212 : 412 : recptr = XLogInsert(RM_XLOG_ID, XLOG_PARAMETER_CHANGE);
4285 fujii@postgresql.org 8213 : 412 : XLogFlush(recptr);
8214 : : }
8215 : :
2019 tmunro@postgresql.or 8216 : 435 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8217 : :
5713 heikki.linnakangas@i 8218 : 435 : ControlFile->MaxConnections = MaxConnections;
4550 rhaas@postgresql.org 8219 : 435 : ControlFile->max_worker_processes = max_worker_processes;
2501 michael@paquier.xyz 8220 : 435 : ControlFile->max_wal_senders = max_wal_senders;
5713 heikki.linnakangas@i 8221 : 435 : ControlFile->max_prepared_xacts = max_prepared_xacts;
8222 : 435 : ControlFile->max_locks_per_xact = max_locks_per_xact;
8223 : 435 : ControlFile->wal_level = wal_level;
4369 rhaas@postgresql.org 8224 : 435 : ControlFile->wal_log_hints = wal_log_hints;
4033 alvherre@alvh.no-ip. 8225 : 435 : ControlFile->track_commit_timestamp = track_commit_timestamp;
5713 heikki.linnakangas@i 8226 : 435 : UpdateControlFile();
8227 : :
2019 tmunro@postgresql.or 8228 : 435 : LWLockRelease(ControlFileLock);
8229 : : }
5811 heikki.linnakangas@i 8230 : 872 : }
8231 : :
8232 : : /*
8233 : : * Update full_page_writes in shared memory, and write an
8234 : : * XLOG_FPW_CHANGE record if necessary.
8235 : : *
8236 : : * Note: this function assumes there is no other process running
8237 : : * concurrently that could update it.
8238 : : */
8239 : : void
5076 simon@2ndQuadrant.co 8240 : 1454 : UpdateFullPageWrites(void)
8241 : : {
8242 : 1454 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8243 : : bool recoveryInProgress;
8244 : :
8245 : : /*
8246 : : * Do nothing if full_page_writes has not been changed.
8247 : : *
8248 : : * It's safe to check the shared full_page_writes without the lock,
8249 : : * because we assume that there is no concurrently running process which
8250 : : * can update it.
8251 : : */
8252 [ + + ]: 1454 : if (fullPageWrites == Insert->fullPageWrites)
8253 : 998 : return;
8254 : :
8255 : : /*
8256 : : * Perform this outside critical section so that the WAL insert
8257 : : * initialization done by RecoveryInProgress() doesn't trigger an
8258 : : * assertion failure.
8259 : : */
2638 akapila@postgresql.o 8260 : 456 : recoveryInProgress = RecoveryInProgress();
8261 : :
5035 heikki.linnakangas@i 8262 : 456 : START_CRIT_SECTION();
8263 : :
8264 : : /*
8265 : : * It's always safe to take full page images, even when not strictly
8266 : : * required, but not the other round. So if we're setting full_page_writes
8267 : : * to true, first set it true and then write the WAL record. If we're
8268 : : * setting it to false, first write the WAL record and then set the global
8269 : : * flag.
8270 : : */
8271 [ + + ]: 456 : if (fullPageWrites)
8272 : : {
4290 8273 : 445 : WALInsertLockAcquireExclusive();
5035 8274 : 445 : Insert->fullPageWrites = true;
4290 8275 : 445 : WALInsertLockRelease();
8276 : : }
8277 : :
8278 : : /*
8279 : : * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
8280 : : * full_page_writes during archive recovery, if required.
8281 : : */
2638 akapila@postgresql.o 8282 [ + + - + ]: 456 : if (XLogStandbyInfoActive() && !recoveryInProgress)
8283 : : {
4046 heikki.linnakangas@i 8284 :UBC 0 : XLogBeginInsert();
310 peter@eisentraut.org 8285 : 0 : XLogRegisterData(&fullPageWrites, sizeof(bool));
8286 : :
4046 heikki.linnakangas@i 8287 : 0 : XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
8288 : : }
8289 : :
5035 heikki.linnakangas@i 8290 [ + + ]:CBC 456 : if (!fullPageWrites)
8291 : : {
4290 8292 : 11 : WALInsertLockAcquireExclusive();
5035 8293 : 11 : Insert->fullPageWrites = false;
4290 8294 : 11 : WALInsertLockRelease();
8295 : : }
5035 8296 [ - + ]: 456 : END_CRIT_SECTION();
8297 : : }
8298 : :
8299 : : /*
8300 : : * XLOG resource manager's routines
8301 : : *
8302 : : * Definitions of info values are in include/catalog/pg_control.h, though
8303 : : * not all record types are related to control file updates.
8304 : : *
8305 : : * NOTE: Some XLOG record types that are directly related to WAL recovery
8306 : : * are handled in xlogrecovery_redo().
8307 : : */
8308 : : void
4046 8309 : 41488 : xlog_redo(XLogReaderState *record)
8310 : : {
8311 : 41488 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
8312 : 41488 : XLogRecPtr lsn = record->EndRecPtr;
8313 : :
8314 : : /*
8315 : : * In XLOG rmgr, backup blocks are only used by XLOG_FPI and
8316 : : * XLOG_FPI_FOR_HINT records.
8317 : : */
4042 8318 [ + + + + : 41488 : Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
- + ]
8319 : : !XLogRecHasAnyBlockRefs(record));
8320 : :
9041 tgl@sss.pgh.pa.us 8321 [ + + ]: 41488 : if (info == XLOG_NEXTOID)
8322 : : {
8323 : : Oid nextOid;
8324 : :
8325 : : /*
8326 : : * We used to try to take the maximum of TransamVariables->nextOid and
8327 : : * the recorded nextOid, but that fails if the OID counter wraps
8328 : : * around. Since no OID allocation should be happening during replay
8329 : : * anyway, better to just believe the record exactly. We still take
8330 : : * OidGenLock while setting the variable, just in case.
8331 : : */
9176 vadim4o@yahoo.com 8332 : 91 : memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
5064 tgl@sss.pgh.pa.us 8333 : 91 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
741 heikki.linnakangas@i 8334 : 91 : TransamVariables->nextOid = nextOid;
8335 : 91 : TransamVariables->oidCount = 0;
5064 tgl@sss.pgh.pa.us 8336 : 91 : LWLockRelease(OidGenLock);
8337 : : }
9046 8338 [ + + ]: 41397 : else if (info == XLOG_CHECKPOINT_SHUTDOWN)
8339 : : {
8340 : : CheckPoint checkPoint;
8341 : : TimeLineID replayTLI;
8342 : :
8343 : 31 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8344 : : /* In a SHUTDOWN checkpoint, believe the counters exactly */
5064 8345 : 31 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
741 heikki.linnakangas@i 8346 : 31 : TransamVariables->nextXid = checkPoint.nextXid;
5064 tgl@sss.pgh.pa.us 8347 : 31 : LWLockRelease(XidGenLock);
8348 : 31 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
741 heikki.linnakangas@i 8349 : 31 : TransamVariables->nextOid = checkPoint.nextOid;
8350 : 31 : TransamVariables->oidCount = 0;
5064 tgl@sss.pgh.pa.us 8351 : 31 : LWLockRelease(OidGenLock);
7498 8352 : 31 : MultiXactSetNextMXact(checkPoint.nextMulti,
8353 : : checkPoint.nextMultiOffset);
8354 : :
3736 andres@anarazel.de 8355 : 31 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8356 : : checkPoint.oldestMultiDB);
8357 : :
8358 : : /*
8359 : : * No need to set oldestClogXid here as well; it'll be set when we
8360 : : * redo an xl_clog_truncate if it changed since initialization.
8361 : : */
5783 tgl@sss.pgh.pa.us 8362 : 31 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
8363 : :
8364 : : /*
8365 : : * If we see a shutdown checkpoint while waiting for an end-of-backup
8366 : : * record, the backup was canceled and the end-of-backup record will
8367 : : * never arrive.
8368 : : */
4682 heikki.linnakangas@i 8369 [ + - ]: 31 : if (ArchiveRecoveryRequested &&
42 alvherre@kurilemu.de 8370 [ - + ]:GNC 31 : XLogRecPtrIsValid(ControlFile->backupStartPoint) &&
42 alvherre@kurilemu.de 8371 [ # # ]:UNC 0 : !XLogRecPtrIsValid(ControlFile->backupEndPoint))
5064 tgl@sss.pgh.pa.us 8372 [ # # ]:UBC 0 : ereport(PANIC,
8373 : : (errmsg("online backup was canceled, recovery cannot continue")));
8374 : :
8375 : : /*
8376 : : * If we see a shutdown checkpoint, we know that nothing was running
8377 : : * on the primary at this point. So fake-up an empty running-xacts
8378 : : * record and use that here and now. Recover additional standby state
8379 : : * for prepared transactions.
8380 : : */
5843 simon@2ndQuadrant.co 8381 [ + + ]:CBC 31 : if (standbyState >= STANDBY_INITIALIZED)
8382 : : {
8383 : : TransactionId *xids;
8384 : : int nxids;
8385 : : TransactionId oldestActiveXID;
8386 : : TransactionId latestCompletedXid;
8387 : : RunningTransactionsData running;
8388 : :
5728 heikki.linnakangas@i 8389 : 29 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
8390 : :
8391 : : /* Update pg_subtrans entries for any prepared transactions */
539 8392 : 29 : StandbyRecoverPreparedTransactions();
8393 : :
8394 : : /*
8395 : : * Construct a RunningTransactions snapshot representing a shut
8396 : : * down server, with only prepared transactions still alive. We're
8397 : : * never overflowed at this point because all subxids are listed
8398 : : * with their parent prepared transactions.
8399 : : */
5728 8400 : 29 : running.xcnt = nxids;
4764 simon@2ndQuadrant.co 8401 : 29 : running.subxcnt = 0;
539 heikki.linnakangas@i 8402 : 29 : running.subxid_status = SUBXIDS_IN_SUBTRANS;
1955 andres@anarazel.de 8403 : 29 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5728 heikki.linnakangas@i 8404 : 29 : running.oldestRunningXid = oldestActiveXID;
1955 andres@anarazel.de 8405 : 29 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5698 simon@2ndQuadrant.co 8406 [ - + ]: 29 : TransactionIdRetreat(latestCompletedXid);
5697 8407 [ - + ]: 29 : Assert(TransactionIdIsNormal(latestCompletedXid));
5698 8408 : 29 : running.latestCompletedXid = latestCompletedXid;
5728 heikki.linnakangas@i 8409 : 29 : running.xids = xids;
8410 : :
8411 : 29 : ProcArrayApplyRecoveryInfo(&running);
8412 : : }
8413 : :
8414 : : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
2019 tmunro@postgresql.or 8415 : 31 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
1955 andres@anarazel.de 8416 : 31 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
2019 tmunro@postgresql.or 8417 : 31 : LWLockRelease(ControlFileLock);
8418 : :
8419 : : /*
8420 : : * We should've already switched to the new TLI before replaying this
8421 : : * record.
8422 : : */
1401 heikki.linnakangas@i 8423 : 31 : (void) GetCurrentReplayRecPtr(&replayTLI);
1504 rhaas@postgresql.org 8424 [ - + ]: 31 : if (checkPoint.ThisTimeLineID != replayTLI)
4759 heikki.linnakangas@i 8425 [ # # ]:UBC 0 : ereport(PANIC,
8426 : : (errmsg("unexpected timeline ID %u (should be %u) in shutdown checkpoint record",
8427 : : checkPoint.ThisTimeLineID, replayTLI)));
8428 : :
1485 rhaas@postgresql.org 8429 :CBC 31 : RecoveryRestartPoint(&checkPoint, record);
8430 : :
8431 : : /*
8432 : : * After replaying a checkpoint record, free all smgr objects.
8433 : : * Otherwise we would never do so for dropped relations, as the
8434 : : * startup does not process shared invalidation messages or call
8435 : : * AtEOXact_SMgr().
8436 : : */
99 michael@paquier.xyz 8437 : 31 : smgrdestroyall();
8438 : : }
9046 tgl@sss.pgh.pa.us 8439 [ + + ]: 41366 : else if (info == XLOG_CHECKPOINT_ONLINE)
8440 : : {
8441 : : CheckPoint checkPoint;
8442 : : TimeLineID replayTLI;
8443 : :
8444 : 671 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8445 : : /* In an ONLINE checkpoint, treat the XID counter as a minimum */
5064 8446 : 671 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
741 heikki.linnakangas@i 8447 [ - + ]: 671 : if (FullTransactionIdPrecedes(TransamVariables->nextXid,
8448 : : checkPoint.nextXid))
741 heikki.linnakangas@i 8449 :UBC 0 : TransamVariables->nextXid = checkPoint.nextXid;
5064 tgl@sss.pgh.pa.us 8450 :CBC 671 : LWLockRelease(XidGenLock);
8451 : :
8452 : : /*
8453 : : * We ignore the nextOid counter in an ONLINE checkpoint, preferring
8454 : : * to track OID assignment through XLOG_NEXTOID records. The nextOid
8455 : : * counter is from the start of the checkpoint and might well be stale
8456 : : * compared to later XLOG_NEXTOID records. We could try to take the
8457 : : * maximum of the nextOid counter and our latest value, but since
8458 : : * there's no particular guarantee about the speed with which the OID
8459 : : * counter wraps around, that's a risky thing to do. In any case,
8460 : : * users of the nextOid counter are required to avoid assignment of
8461 : : * duplicates, so that a somewhat out-of-date value should be safe.
8462 : : */
8463 : :
8464 : : /* Handle multixact */
7498 8465 : 671 : MultiXactAdvanceNextMXact(checkPoint.nextMulti,
8466 : : checkPoint.nextMultiOffset);
8467 : :
8468 : : /*
8469 : : * NB: This may perform multixact truncation when replaying WAL
8470 : : * generated by an older primary.
8471 : : */
3736 andres@anarazel.de 8472 : 671 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8473 : : checkPoint.oldestMultiDB);
741 heikki.linnakangas@i 8474 [ - + ]: 671 : if (TransactionIdPrecedes(TransamVariables->oldestXid,
8475 : : checkPoint.oldestXid))
5783 tgl@sss.pgh.pa.us 8476 :UBC 0 : SetTransactionIdLimit(checkPoint.oldestXid,
8477 : : checkPoint.oldestXidDB);
8478 : : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
2019 tmunro@postgresql.or 8479 :CBC 671 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
1955 andres@anarazel.de 8480 : 671 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
2019 tmunro@postgresql.or 8481 : 671 : LWLockRelease(ControlFileLock);
8482 : :
8483 : : /* TLI should not change in an on-line checkpoint */
1401 heikki.linnakangas@i 8484 : 671 : (void) GetCurrentReplayRecPtr(&replayTLI);
1504 rhaas@postgresql.org 8485 [ - + ]: 671 : if (checkPoint.ThisTimeLineID != replayTLI)
7981 tgl@sss.pgh.pa.us 8486 [ # # ]:UBC 0 : ereport(PANIC,
8487 : : (errmsg("unexpected timeline ID %u (should be %u) in online checkpoint record",
8488 : : checkPoint.ThisTimeLineID, replayTLI)));
8489 : :
1485 rhaas@postgresql.org 8490 :CBC 671 : RecoveryRestartPoint(&checkPoint, record);
8491 : :
8492 : : /*
8493 : : * After replaying a checkpoint record, free all smgr objects.
8494 : : * Otherwise we would never do so for dropped relations, as the
8495 : : * startup does not process shared invalidation messages or call
8496 : : * AtEOXact_SMgr().
8497 : : */
99 michael@paquier.xyz 8498 : 671 : smgrdestroyall();
8499 : : }
1541 alvherre@alvh.no-ip. 8500 [ + + ]: 40695 : else if (info == XLOG_OVERWRITE_CONTRECORD)
8501 : : {
8502 : : /* nothing to do here, handled in xlogrecovery_redo() */
8503 : : }
4706 simon@2ndQuadrant.co 8504 [ + + ]: 40694 : else if (info == XLOG_END_OF_RECOVERY)
8505 : : {
8506 : : xl_end_of_recovery xlrec;
8507 : : TimeLineID replayTLI;
8508 : :
8509 : 9 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
8510 : :
8511 : : /*
8512 : : * For Hot Standby, we could treat this like a Shutdown Checkpoint,
8513 : : * but this case is rarer and harder to test, so the benefit doesn't
8514 : : * outweigh the potential extra cost of maintenance.
8515 : : */
8516 : :
8517 : : /*
8518 : : * We should've already switched to the new TLI before replaying this
8519 : : * record.
8520 : : */
1401 heikki.linnakangas@i 8521 : 9 : (void) GetCurrentReplayRecPtr(&replayTLI);
1504 rhaas@postgresql.org 8522 [ - + ]: 9 : if (xlrec.ThisTimeLineID != replayTLI)
4706 simon@2ndQuadrant.co 8523 [ # # ]:UBC 0 : ereport(PANIC,
8524 : : (errmsg("unexpected timeline ID %u (should be %u) in end-of-recovery record",
8525 : : xlrec.ThisTimeLineID, replayTLI)));
8526 : : }
6787 tgl@sss.pgh.pa.us 8527 [ + - ]:CBC 40685 : else if (info == XLOG_NOOP)
8528 : : {
8529 : : /* nothing to do here */
8530 : : }
7074 8531 [ + + ]: 40685 : else if (info == XLOG_SWITCH)
8532 : : {
8533 : : /* nothing to do here */
8534 : : }
5427 simon@2ndQuadrant.co 8535 [ + + ]: 40247 : else if (info == XLOG_RESTORE_POINT)
8536 : : {
8537 : : /* nothing to do here, handled in xlogrecovery.c */
8538 : : }
4042 heikki.linnakangas@i 8539 [ + + + + ]: 40242 : else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
8540 : : {
8541 : : /*
8542 : : * XLOG_FPI records contain nothing else but one or more block
8543 : : * references. Every block reference must include a full-page image
8544 : : * even if full_page_writes was disabled when the record was generated
8545 : : * - otherwise there would be no point in this record.
8546 : : *
8547 : : * XLOG_FPI_FOR_HINT records are generated when a page needs to be
8548 : : * WAL-logged because of a hint bit update. They are only generated
8549 : : * when checksums and/or wal_log_hints are enabled. They may include
8550 : : * no full-page images if full_page_writes was disabled when they were
8551 : : * generated. In this case there is nothing to do here.
8552 : : *
8553 : : * No recovery conflicts are generated by these generic records - if a
8554 : : * resource manager needs to generate conflicts, it has to define a
8555 : : * separate WAL record type and redo routine.
8556 : : */
1371 tmunro@postgresql.or 8557 [ + + ]: 83729 : for (uint8 block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
8558 : : {
8559 : : Buffer buffer;
8560 : :
1611 fujii@postgresql.org 8561 [ + + ]: 44273 : if (!XLogRecHasBlockImage(record, block_id))
8562 : : {
8563 [ - + ]: 66 : if (info == XLOG_FPI)
1611 fujii@postgresql.org 8564 [ # # ]:UBC 0 : elog(ERROR, "XLOG_FPI record did not contain a full-page image");
1611 fujii@postgresql.org 8565 :CBC 66 : continue;
8566 : : }
8567 : :
2451 heikki.linnakangas@i 8568 [ - + ]: 44207 : if (XLogReadBufferForRedo(record, block_id, &buffer) != BLK_RESTORED)
2451 heikki.linnakangas@i 8569 [ # # ]:UBC 0 : elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
2451 heikki.linnakangas@i 8570 :CBC 44207 : UnlockReleaseBuffer(buffer);
8571 : : }
8572 : : }
5827 8573 [ + + ]: 786 : else if (info == XLOG_BACKUP_END)
8574 : : {
8575 : : /* nothing to do here, handled in xlogrecovery_redo() */
8576 : : }
5713 8577 [ + + ]: 703 : else if (info == XLOG_PARAMETER_CHANGE)
8578 : : {
8579 : : xl_parameter_change xlrec;
8580 : :
8581 : : /* Update our copy of the parameters in pg_control */
8582 : 31 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
8583 : :
8584 : : /*
8585 : : * Invalidate logical slots if we are in hot standby and the primary
8586 : : * does not have a WAL level sufficient for logical decoding. No need
8587 : : * to search for potentially conflicting logically slots if standby is
8588 : : * running with wal_level lower than logical, because in that case, we
8589 : : * would have either disallowed creation of logical slots or
8590 : : * invalidated existing ones.
8591 : : */
986 andres@anarazel.de 8592 [ + - + + ]: 31 : if (InRecovery && InHotStandby &&
8593 [ + + ]: 16 : xlrec.wal_level < WAL_LEVEL_LOGICAL &&
8594 [ + + ]: 6 : wal_level >= WAL_LEVEL_LOGICAL)
8595 : 3 : InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_LEVEL,
8596 : : 0, InvalidOid,
8597 : : InvalidTransactionId);
8598 : :
5708 heikki.linnakangas@i 8599 : 31 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
5713 8600 : 31 : ControlFile->MaxConnections = xlrec.MaxConnections;
4550 rhaas@postgresql.org 8601 : 31 : ControlFile->max_worker_processes = xlrec.max_worker_processes;
2501 michael@paquier.xyz 8602 : 31 : ControlFile->max_wal_senders = xlrec.max_wal_senders;
5713 heikki.linnakangas@i 8603 : 31 : ControlFile->max_prepared_xacts = xlrec.max_prepared_xacts;
8604 : 31 : ControlFile->max_locks_per_xact = xlrec.max_locks_per_xact;
8605 : 31 : ControlFile->wal_level = xlrec.wal_level;
3990 8606 : 31 : ControlFile->wal_log_hints = xlrec.wal_log_hints;
8607 : :
8608 : : /*
8609 : : * Update minRecoveryPoint to ensure that if recovery is aborted, we
8610 : : * recover back up to this point before allowing hot standby again.
8611 : : * This is important if the max_* settings are decreased, to ensure
8612 : : * you don't run queries against the WAL preceding the change. The
8613 : : * local copies cannot be updated as long as crash recovery is
8614 : : * happening and we expect all the WAL to be replayed.
8615 : : */
2723 michael@paquier.xyz 8616 [ + + ]: 31 : if (InArchiveRecovery)
8617 : : {
1401 heikki.linnakangas@i 8618 : 17 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
8619 : 17 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
8620 : : }
42 alvherre@kurilemu.de 8621 [ + + + + ]:GNC 31 : if (XLogRecPtrIsValid(LocalMinRecoveryPoint) && LocalMinRecoveryPoint < lsn)
8622 : : {
8623 : : TimeLineID replayTLI;
8624 : :
1401 heikki.linnakangas@i 8625 :CBC 6 : (void) GetCurrentReplayRecPtr(&replayTLI);
5708 8626 : 6 : ControlFile->minRecoveryPoint = lsn;
1504 rhaas@postgresql.org 8627 : 6 : ControlFile->minRecoveryPointTLI = replayTLI;
8628 : : }
8629 : :
3731 alvherre@alvh.no-ip. 8630 : 31 : CommitTsParameterChange(xlrec.track_commit_timestamp,
8631 : 31 : ControlFile->track_commit_timestamp);
8632 : 31 : ControlFile->track_commit_timestamp = xlrec.track_commit_timestamp;
8633 : :
5713 heikki.linnakangas@i 8634 : 31 : UpdateControlFile();
5708 8635 : 31 : LWLockRelease(ControlFileLock);
8636 : :
8637 : : /* Check to see if any parameter change gives a problem on recovery */
5713 8638 : 31 : CheckRequiredParameterValues();
8639 : : }
5076 simon@2ndQuadrant.co 8640 [ - + ]: 672 : else if (info == XLOG_FPW_CHANGE)
8641 : : {
8642 : : bool fpw;
8643 : :
5076 simon@2ndQuadrant.co 8644 :UBC 0 : memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
8645 : :
8646 : : /*
8647 : : * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
8648 : : * do_pg_backup_start() and do_pg_backup_stop() can check whether
8649 : : * full_page_writes has been disabled during online backup.
8650 : : */
8651 [ # # ]: 0 : if (!fpw)
8652 : : {
4105 andres@anarazel.de 8653 [ # # ]: 0 : SpinLockAcquire(&XLogCtl->info_lck);
1485 rhaas@postgresql.org 8654 [ # # ]: 0 : if (XLogCtl->lastFpwDisableRecPtr < record->ReadRecPtr)
8655 : 0 : XLogCtl->lastFpwDisableRecPtr = record->ReadRecPtr;
4105 andres@anarazel.de 8656 : 0 : SpinLockRelease(&XLogCtl->info_lck);
8657 : : }
8658 : :
8659 : : /* Keep track of full_page_writes */
5076 simon@2ndQuadrant.co 8660 : 0 : lastFullPageWrites = fpw;
8661 : : }
8662 : : else if (info == XLOG_CHECKPOINT_REDO)
8663 : : {
8664 : : /* nothing to do here, just for informational purposes */
8665 : : }
9189 vadim4o@yahoo.com 8666 :CBC 41486 : }
8667 : :
8668 : : /*
8669 : : * Return the extra open flags used for opening a file, depending on the
8670 : : * value of the GUCs wal_sync_method, fsync and debug_io_direct.
8671 : : */
8672 : : static int
6427 magnus@hagander.net 8673 : 15861 : get_sync_bit(int method)
8674 : : {
5774 bruce@momjian.us 8675 : 15861 : int o_direct_flag = 0;
8676 : :
8677 : : /*
8678 : : * Use O_DIRECT if requested, except in walreceiver process. The WAL
8679 : : * written by walreceiver is normally read by the startup process soon
8680 : : * after it's written. Also, walreceiver performs unaligned writes, which
8681 : : * don't work with O_DIRECT, so it is required for correctness too.
8682 : : */
985 tmunro@postgresql.or 8683 [ + + + - ]: 15861 : if ((io_direct_flags & IO_DIRECT_WAL) && !AmWalReceiverProcess())
5781 heikki.linnakangas@i 8684 : 8 : o_direct_flag = PG_O_DIRECT;
8685 : :
8686 : : /* If fsync is disabled, never open in sync mode */
985 tmunro@postgresql.or 8687 [ + - ]: 15861 : if (!enableFsync)
8688 : 15861 : return o_direct_flag;
8689 : :
6427 magnus@hagander.net 8690 [ # # # # ]:UBC 0 : switch (method)
8691 : : {
8692 : : /*
8693 : : * enum values for all sync options are defined even if they are
8694 : : * not supported on the current platform. But if not, they are
8695 : : * not included in the enum option array, and therefore will never
8696 : : * be seen here.
8697 : : */
797 nathan@postgresql.or 8698 : 0 : case WAL_SYNC_METHOD_FSYNC:
8699 : : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8700 : : case WAL_SYNC_METHOD_FDATASYNC:
985 tmunro@postgresql.or 8701 : 0 : return o_direct_flag;
8702 : : #ifdef O_SYNC
797 nathan@postgresql.or 8703 : 0 : case WAL_SYNC_METHOD_OPEN:
1245 tmunro@postgresql.or 8704 : 0 : return O_SYNC | o_direct_flag;
8705 : : #endif
8706 : : #ifdef O_DSYNC
797 nathan@postgresql.or 8707 : 0 : case WAL_SYNC_METHOD_OPEN_DSYNC:
1245 tmunro@postgresql.or 8708 : 0 : return O_DSYNC | o_direct_flag;
8709 : : #endif
6429 magnus@hagander.net 8710 : 0 : default:
8711 : : /* can't happen (unless we are out of sync with option array) */
580 peter@eisentraut.org 8712 [ # # ]: 0 : elog(ERROR, "unrecognized \"wal_sync_method\": %d", method);
8713 : : return 0; /* silence warning */
8714 : : }
8715 : : }
8716 : :
8717 : : /*
8718 : : * GUC support
8719 : : */
8720 : : void
797 nathan@postgresql.or 8721 :CBC 1109 : assign_wal_sync_method(int new_wal_sync_method, void *extra)
8722 : : {
8723 [ - + ]: 1109 : if (wal_sync_method != new_wal_sync_method)
8724 : : {
8725 : : /*
8726 : : * To ensure that no blocks escape unsynced, force an fsync on the
8727 : : * currently open log segment (if any). Also, if the open flag is
8728 : : * changing, close the log file so it will be reopened (with new flag
8729 : : * bit) at next use.
8730 : : */
9043 tgl@sss.pgh.pa.us 8731 [ # # ]:UBC 0 : if (openLogFile >= 0)
8732 : : {
3197 rhaas@postgresql.org 8733 : 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN);
9043 tgl@sss.pgh.pa.us 8734 [ # # ]: 0 : if (pg_fsync(openLogFile) != 0)
8735 : : {
8736 : : char xlogfname[MAXFNAMELEN];
8737 : : int save_errno;
8738 : :
2207 michael@paquier.xyz 8739 : 0 : save_errno = errno;
1504 rhaas@postgresql.org 8740 : 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo,
8741 : : wal_segment_size);
2207 michael@paquier.xyz 8742 : 0 : errno = save_errno;
8186 tgl@sss.pgh.pa.us 8743 [ # # ]: 0 : ereport(PANIC,
8744 : : (errcode_for_file_access(),
8745 : : errmsg("could not fsync file \"%s\": %m", xlogfname)));
8746 : : }
8747 : :
3197 rhaas@postgresql.org 8748 : 0 : pgstat_report_wait_end();
797 nathan@postgresql.or 8749 [ # # ]: 0 : if (get_sync_bit(wal_sync_method) != get_sync_bit(new_wal_sync_method))
7126 bruce@momjian.us 8750 : 0 : XLogFileClose();
8751 : : }
8752 : : }
9043 tgl@sss.pgh.pa.us 8753 :CBC 1109 : }
8754 : :
8755 : :
8756 : : /*
8757 : : * Issue appropriate kind of fsync (if any) for an XLOG output file.
8758 : : *
8759 : : * 'fd' is a file descriptor for the XLOG file to be fsync'd.
8760 : : * 'segno' is for error reporting purposes.
8761 : : */
8762 : : void
1504 rhaas@postgresql.org 8763 : 173110 : issue_xlog_fsync(int fd, XLogSegNo segno, TimeLineID tli)
8764 : : {
2207 michael@paquier.xyz 8765 : 173110 : char *msg = NULL;
8766 : : instr_time start;
8767 : :
1504 rhaas@postgresql.org 8768 [ - + ]: 173110 : Assert(tli != 0);
8769 : :
8770 : : /*
8771 : : * Quick exit if fsync is disabled or write() has already synced the WAL
8772 : : * file.
8773 : : */
1745 fujii@postgresql.org 8774 [ - + ]: 173110 : if (!enableFsync ||
797 nathan@postgresql.or 8775 [ # # ]:UBC 0 : wal_sync_method == WAL_SYNC_METHOD_OPEN ||
8776 [ # # ]: 0 : wal_sync_method == WAL_SYNC_METHOD_OPEN_DSYNC)
1745 fujii@postgresql.org 8777 :CBC 173110 : return;
8778 : :
8779 : : /*
8780 : : * Measure I/O timing to sync the WAL file for pg_stat_io.
8781 : : */
295 michael@paquier.xyz 8782 :UBC 0 : start = pgstat_prepare_io_time(track_wal_io_timing);
8783 : :
2726 8784 : 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC);
797 nathan@postgresql.or 8785 [ # # # # ]: 0 : switch (wal_sync_method)
8786 : : {
8787 : 0 : case WAL_SYNC_METHOD_FSYNC:
5816 heikki.linnakangas@i 8788 [ # # ]: 0 : if (pg_fsync_no_writethrough(fd) != 0)
2207 michael@paquier.xyz 8789 : 0 : msg = _("could not fsync file \"%s\": %m");
9043 tgl@sss.pgh.pa.us 8790 : 0 : break;
8791 : : #ifdef HAVE_FSYNC_WRITETHROUGH
8792 : : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8793 : : if (pg_fsync_writethrough(fd) != 0)
8794 : : msg = _("could not fsync write-through file \"%s\": %m");
8795 : : break;
8796 : : #endif
797 nathan@postgresql.or 8797 : 0 : case WAL_SYNC_METHOD_FDATASYNC:
5816 heikki.linnakangas@i 8798 [ # # ]: 0 : if (pg_fdatasync(fd) != 0)
2207 michael@paquier.xyz 8799 : 0 : msg = _("could not fdatasync file \"%s\": %m");
9043 tgl@sss.pgh.pa.us 8800 : 0 : break;
797 nathan@postgresql.or 8801 : 0 : case WAL_SYNC_METHOD_OPEN:
8802 : : case WAL_SYNC_METHOD_OPEN_DSYNC:
8803 : : /* not reachable */
1745 fujii@postgresql.org 8804 : 0 : Assert(false);
8805 : : break;
9043 tgl@sss.pgh.pa.us 8806 : 0 : default:
624 dgustafsson@postgres 8807 [ # # ]: 0 : ereport(PANIC,
8808 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8809 : : errmsg_internal("unrecognized \"wal_sync_method\": %d", wal_sync_method));
8810 : : break;
8811 : : }
8812 : :
8813 : : /* PANIC if failed to fsync */
2207 michael@paquier.xyz 8814 [ # # ]: 0 : if (msg)
8815 : : {
8816 : : char xlogfname[MAXFNAMELEN];
8817 : 0 : int save_errno = errno;
8818 : :
1504 rhaas@postgresql.org 8819 : 0 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
2207 michael@paquier.xyz 8820 : 0 : errno = save_errno;
8821 [ # # ]: 0 : ereport(PANIC,
8822 : : (errcode_for_file_access(),
8823 : : errmsg(msg, xlogfname)));
8824 : : }
8825 : :
8826 : 0 : pgstat_report_wait_end();
8827 : :
317 8828 : 0 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_FSYNC,
8829 : : start, 1, 0);
8830 : : }
8831 : :
8832 : : /*
8833 : : * do_pg_backup_start is the workhorse of the user-visible pg_backup_start()
8834 : : * function. It creates the necessary starting checkpoint and constructs the
8835 : : * backup state and tablespace map.
8836 : : *
8837 : : * Input parameters are "state" (the backup state), "fast" (if true, we do
8838 : : * the checkpoint in fast mode), and "tablespaces" (if non-NULL, indicates a
8839 : : * list of tablespaceinfo structs describing the cluster's tablespaces.).
8840 : : *
8841 : : * The tablespace map contents are appended to passed-in parameter
8842 : : * tablespace_map and the caller is responsible for including it in the backup
8843 : : * archive as 'tablespace_map'. The tablespace_map file is required mainly for
8844 : : * tar format in windows as native windows utilities are not able to create
8845 : : * symlinks while extracting files from tar. However for consistency and
8846 : : * platform-independence, we do it the same way everywhere.
8847 : : *
8848 : : * It fills in "state" with the information required for the backup, such
8849 : : * as the minimum WAL location that must be present to restore from this
8850 : : * backup (starttli) and the corresponding timeline ID (starttli).
8851 : : *
8852 : : * Every successfully started backup must be stopped by calling
8853 : : * do_pg_backup_stop() or do_pg_abort_backup(). There can be many
8854 : : * backups active at the same time.
8855 : : *
8856 : : * It is the responsibility of the caller of this function to verify the
8857 : : * permissions of the calling user!
8858 : : */
8859 : : void
1179 michael@paquier.xyz 8860 :CBC 168 : do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces,
8861 : : BackupState *state, StringInfo tblspcmapfile)
8862 : : {
8863 : : bool backup_started_in_recovery;
8864 : :
8865 [ - + ]: 168 : Assert(state != NULL);
5076 simon@2ndQuadrant.co 8866 : 168 : backup_started_in_recovery = RecoveryInProgress();
8867 : :
8868 : : /*
8869 : : * During recovery, we don't need to check WAL level. Because, if WAL
8870 : : * level is not sufficient, it's impossible to get here during recovery.
8871 : : */
8872 [ + + - + ]: 168 : if (!backup_started_in_recovery && !XLogIsNeeded())
6658 tgl@sss.pgh.pa.us 8873 [ # # ]:UBC 0 : ereport(ERROR,
8874 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8875 : : errmsg("WAL level not sufficient for making an online backup"),
8876 : : errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
8877 : :
5435 heikki.linnakangas@i 8878 [ + + ]:CBC 168 : if (strlen(backupidstr) > MAXPGPATH)
8879 [ + - ]: 1 : ereport(ERROR,
8880 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8881 : : errmsg("backup label too long (max %d bytes)",
8882 : : MAXPGPATH)));
8883 : :
534 dgustafsson@postgres 8884 : 167 : strlcpy(state->name, backupidstr, sizeof(state->name));
8885 : :
8886 : : /*
8887 : : * Mark backup active in shared memory. We must do full-page WAL writes
8888 : : * during an on-line backup even if not doing so at other times, because
8889 : : * it's quite possible for the backup dump to obtain a "torn" (partially
8890 : : * written) copy of a database page if it reads the page concurrently with
8891 : : * our write to the same page. This can be fixed as long as the first
8892 : : * write to the page in the WAL sequence is a full-page write. Hence, we
8893 : : * increment runningBackups then force a CHECKPOINT, to ensure there are
8894 : : * no dirty pages in shared memory that might get dumped while the backup
8895 : : * is in progress without having a corresponding WAL record. (Once the
8896 : : * backup is complete, we need not force full-page writes anymore, since
8897 : : * we expect that any pages not modified during the backup interval must
8898 : : * have been correctly captured by the backup.)
8899 : : *
8900 : : * Note that forcing full-page writes has no effect during an online
8901 : : * backup from the standby.
8902 : : *
8903 : : * We must hold all the insertion locks to change the value of
8904 : : * runningBackups, to ensure adequate interlocking against
8905 : : * XLogInsertRecord().
8906 : : */
4290 heikki.linnakangas@i 8907 : 167 : WALInsertLockAcquireExclusive();
1352 sfrost@snowman.net 8908 : 167 : XLogCtl->Insert.runningBackups++;
4290 heikki.linnakangas@i 8909 : 167 : WALInsertLockRelease();
8910 : :
8911 : : /*
8912 : : * Ensure we decrement runningBackups if we fail below. NB -- for this to
8913 : : * work correctly, it is critical that sessionBackupState is only updated
8914 : : * after this block is over.
8915 : : */
135 peter@eisentraut.org 8916 [ + - ]:GNC 167 : PG_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, BoolGetDatum(true));
8917 : : {
5366 bruce@momjian.us 8918 :CBC 167 : bool gotUniqueStartpoint = false;
8919 : : DIR *tblspcdir;
8920 : : struct dirent *de;
8921 : : tablespaceinfo *ti;
8922 : : int datadirpathlen;
8923 : :
8924 : : /*
8925 : : * Force an XLOG file switch before the checkpoint, to ensure that the
8926 : : * WAL segment the checkpoint is written to doesn't contain pages with
8927 : : * old timeline IDs. That would otherwise happen if you called
8928 : : * pg_backup_start() right after restoring from a PITR archive: the
8929 : : * first WAL segment containing the startup checkpoint has pages in
8930 : : * the beginning with the old timeline ID. That can cause trouble at
8931 : : * recovery: we won't have a history file covering the old timeline if
8932 : : * pg_wal directory was not included in the base backup and the WAL
8933 : : * archive was cleared too before starting the backup.
8934 : : *
8935 : : * This also ensures that we have emitted a WAL page header that has
8936 : : * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
8937 : : * Therefore, if a WAL archiver (such as pglesslog) is trying to
8938 : : * compress out removable backup blocks, it won't remove any that
8939 : : * occur after this point.
8940 : : *
8941 : : * During recovery, we skip forcing XLOG file switch, which means that
8942 : : * the backup taken during recovery is not available for the special
8943 : : * recovery case described above.
8944 : : */
5076 simon@2ndQuadrant.co 8945 [ + + ]: 167 : if (!backup_started_in_recovery)
3283 andres@anarazel.de 8946 : 161 : RequestXLogSwitch(false);
8947 : :
8948 : : do
8949 : : {
8950 : : bool checkpointfpw;
8951 : :
8952 : : /*
8953 : : * Force a CHECKPOINT. Aside from being necessary to prevent torn
8954 : : * page problems, this guarantees that two successive backup runs
8955 : : * will have different checkpoint positions and hence different
8956 : : * history file names, even if nothing happened in between.
8957 : : *
8958 : : * During recovery, establish a restartpoint if possible. We use
8959 : : * the last restartpoint as the backup starting checkpoint. This
8960 : : * means that two successive backup runs can have same checkpoint
8961 : : * positions.
8962 : : *
8963 : : * Since the fact that we are executing do_pg_backup_start()
8964 : : * during recovery means that checkpointer is running, we can use
8965 : : * RequestCheckpoint() to establish a restartpoint.
8966 : : *
8967 : : * We use CHECKPOINT_FAST only if requested by user (via passing
8968 : : * fast = true). Otherwise this can take awhile.
8969 : : */
5386 heikki.linnakangas@i 8970 [ + + ]: 167 : RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_WAIT |
8971 : : (fast ? CHECKPOINT_FAST : 0));
8972 : :
8973 : : /*
8974 : : * Now we need to fetch the checkpoint record location, and also
8975 : : * its REDO pointer. The oldest point in WAL that would be needed
8976 : : * to restore starting from the checkpoint is precisely the REDO
8977 : : * pointer.
8978 : : */
8979 : 167 : LWLockAcquire(ControlFileLock, LW_SHARED);
1179 michael@paquier.xyz 8980 : 167 : state->checkpointloc = ControlFile->checkPoint;
8981 : 167 : state->startpoint = ControlFile->checkPointCopy.redo;
8982 : 167 : state->starttli = ControlFile->checkPointCopy.ThisTimeLineID;
5076 simon@2ndQuadrant.co 8983 : 167 : checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
5386 heikki.linnakangas@i 8984 : 167 : LWLockRelease(ControlFileLock);
8985 : :
5076 simon@2ndQuadrant.co 8986 [ + + ]: 167 : if (backup_started_in_recovery)
8987 : : {
8988 : : XLogRecPtr recptr;
8989 : :
8990 : : /*
8991 : : * Check to see if all WAL replayed during online backup
8992 : : * (i.e., since last restartpoint used as backup starting
8993 : : * checkpoint) contain full-page writes.
8994 : : */
4105 andres@anarazel.de 8995 [ - + ]: 6 : SpinLockAcquire(&XLogCtl->info_lck);
8996 : 6 : recptr = XLogCtl->lastFpwDisableRecPtr;
8997 : 6 : SpinLockRelease(&XLogCtl->info_lck);
8998 : :
1179 michael@paquier.xyz 8999 [ + - - + ]: 6 : if (!checkpointfpw || state->startpoint <= recptr)
5076 simon@2ndQuadrant.co 9000 [ # # ]:UBC 0 : ereport(ERROR,
9001 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9002 : : errmsg("WAL generated with \"full_page_writes=off\" was replayed "
9003 : : "since last restartpoint"),
9004 : : errhint("This means that the backup being taken on the standby "
9005 : : "is corrupt and should not be used. "
9006 : : "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
9007 : : "and then try an online backup again.")));
9008 : :
9009 : : /*
9010 : : * During recovery, since we don't use the end-of-backup WAL
9011 : : * record and don't write the backup history file, the
9012 : : * starting WAL location doesn't need to be unique. This means
9013 : : * that two base backups started at the same time might use
9014 : : * the same checkpoint as starting locations.
9015 : : */
5076 simon@2ndQuadrant.co 9016 :CBC 6 : gotUniqueStartpoint = true;
9017 : : }
9018 : :
9019 : : /*
9020 : : * If two base backups are started at the same time (in WAL sender
9021 : : * processes), we need to make sure that they use different
9022 : : * checkpoints as starting locations, because we use the starting
9023 : : * WAL location as a unique identifier for the base backup in the
9024 : : * end-of-backup WAL record and when we write the backup history
9025 : : * file. Perhaps it would be better generate a separate unique ID
9026 : : * for each backup instead of forcing another checkpoint, but
9027 : : * taking a checkpoint right after another is not that expensive
9028 : : * either because only few buffers have been dirtied yet.
9029 : : */
4290 heikki.linnakangas@i 9030 : 167 : WALInsertLockAcquireExclusive();
1179 michael@paquier.xyz 9031 [ + - ]: 167 : if (XLogCtl->Insert.lastBackupStart < state->startpoint)
9032 : : {
9033 : 167 : XLogCtl->Insert.lastBackupStart = state->startpoint;
5386 heikki.linnakangas@i 9034 : 167 : gotUniqueStartpoint = true;
9035 : : }
4290 9036 : 167 : WALInsertLockRelease();
5366 bruce@momjian.us 9037 [ - + ]: 167 : } while (!gotUniqueStartpoint);
9038 : :
9039 : : /*
9040 : : * Construct tablespace_map file.
9041 : : */
3873 andrew@dunslane.net 9042 : 167 : datadirpathlen = strlen(DataDir);
9043 : :
9044 : : /* Collect information about all tablespaces */
471 michael@paquier.xyz 9045 : 167 : tblspcdir = AllocateDir(PG_TBLSPC_DIR);
9046 [ + + ]: 540 : while ((de = ReadDir(tblspcdir, PG_TBLSPC_DIR)) != NULL)
9047 : : {
9048 : : char fullpath[MAXPGPATH + sizeof(PG_TBLSPC_DIR)];
9049 : : char linkpath[MAXPGPATH];
3873 andrew@dunslane.net 9050 : 373 : char *relpath = NULL;
9051 : : char *s;
9052 : : PGFileType de_type;
9053 : : char *badp;
9054 : : Oid tsoid;
9055 : :
9056 : : /*
9057 : : * Try to parse the directory name as an unsigned integer.
9058 : : *
9059 : : * Tablespace directories should be positive integers that can be
9060 : : * represented in 32 bits, with no leading zeroes or trailing
9061 : : * garbage. If we come across a name that doesn't meet those
9062 : : * criteria, skip it.
9063 : : */
787 rhaas@postgresql.org 9064 [ + + - + ]: 373 : if (de->d_name[0] < '1' || de->d_name[1] > '9')
9065 : 334 : continue;
9066 : 39 : errno = 0;
9067 : 39 : tsoid = strtoul(de->d_name, &badp, 10);
9068 [ + - + - : 39 : if (*badp != '\0' || errno == EINVAL || errno == ERANGE)
- + ]
3873 andrew@dunslane.net 9069 :UBC 0 : continue;
9070 : :
471 michael@paquier.xyz 9071 :CBC 39 : snprintf(fullpath, sizeof(fullpath), "%s/%s", PG_TBLSPC_DIR, de->d_name);
9072 : :
975 rhaas@postgresql.org 9073 : 39 : de_type = get_dirent_type(fullpath, de, false, ERROR);
9074 : :
9075 [ + + ]: 39 : if (de_type == PGFILETYPE_LNK)
9076 : : {
9077 : : StringInfoData escapedpath;
9078 : : int rllen;
9079 : :
9080 : 25 : rllen = readlink(fullpath, linkpath, sizeof(linkpath));
9081 [ - + ]: 25 : if (rllen < 0)
9082 : : {
975 rhaas@postgresql.org 9083 [ # # ]:UBC 0 : ereport(WARNING,
9084 : : (errmsg("could not read symbolic link \"%s\": %m",
9085 : : fullpath)));
9086 : 0 : continue;
9087 : : }
975 rhaas@postgresql.org 9088 [ - + ]:CBC 25 : else if (rllen >= sizeof(linkpath))
9089 : : {
975 rhaas@postgresql.org 9090 [ # # ]:UBC 0 : ereport(WARNING,
9091 : : (errmsg("symbolic link \"%s\" target is too long",
9092 : : fullpath)));
9093 : 0 : continue;
9094 : : }
975 rhaas@postgresql.org 9095 :CBC 25 : linkpath[rllen] = '\0';
9096 : :
9097 : : /*
9098 : : * Relpath holds the relative path of the tablespace directory
9099 : : * when it's located within PGDATA, or NULL if it's located
9100 : : * elsewhere.
9101 : : */
9102 [ - + ]: 25 : if (rllen > datadirpathlen &&
975 rhaas@postgresql.org 9103 [ # # ]:UBC 0 : strncmp(linkpath, DataDir, datadirpathlen) == 0 &&
944 tgl@sss.pgh.pa.us 9104 [ # # ]: 0 : IS_DIR_SEP(linkpath[datadirpathlen]))
975 rhaas@postgresql.org 9105 : 0 : relpath = pstrdup(linkpath + datadirpathlen + 1);
9106 : :
9107 : : /*
9108 : : * Add a backslash-escaped version of the link path to the
9109 : : * tablespace map file.
9110 : : */
975 rhaas@postgresql.org 9111 :CBC 25 : initStringInfo(&escapedpath);
9112 [ + + ]: 594 : for (s = linkpath; *s; s++)
9113 : : {
9114 [ + - + - : 569 : if (*s == '\n' || *s == '\r' || *s == '\\')
- + ]
975 rhaas@postgresql.org 9115 :UBC 0 : appendStringInfoChar(&escapedpath, '\\');
975 rhaas@postgresql.org 9116 :CBC 569 : appendStringInfoChar(&escapedpath, *s);
9117 : : }
9118 : 25 : appendStringInfo(tblspcmapfile, "%s %s\n",
9119 : 25 : de->d_name, escapedpath.data);
9120 : 25 : pfree(escapedpath.data);
9121 : : }
9122 [ + - ]: 14 : else if (de_type == PGFILETYPE_DIR)
9123 : : {
9124 : : /*
9125 : : * It's possible to use allow_in_place_tablespaces to create
9126 : : * directories directly under pg_tblspc, for testing purposes
9127 : : * only.
9128 : : *
9129 : : * In this case, we store a relative path rather than an
9130 : : * absolute path into the tablespaceinfo.
9131 : : */
471 michael@paquier.xyz 9132 : 14 : snprintf(linkpath, sizeof(linkpath), "%s/%s",
9133 : 14 : PG_TBLSPC_DIR, de->d_name);
975 rhaas@postgresql.org 9134 : 14 : relpath = pstrdup(linkpath);
9135 : : }
9136 : : else
9137 : : {
9138 : : /* Skip any other file type that appears here. */
975 rhaas@postgresql.org 9139 :UBC 0 : continue;
9140 : : }
9141 : :
8 michael@paquier.xyz 9142 :GNC 39 : ti = palloc_object(tablespaceinfo);
787 rhaas@postgresql.org 9143 :CBC 39 : ti->oid = tsoid;
1737 tgl@sss.pgh.pa.us 9144 : 39 : ti->path = pstrdup(linkpath);
975 rhaas@postgresql.org 9145 : 39 : ti->rpath = relpath;
2010 9146 : 39 : ti->size = -1;
9147 : :
3862 bruce@momjian.us 9148 [ + - ]: 39 : if (tablespaces)
9149 : 39 : *tablespaces = lappend(*tablespaces, ti);
9150 : : }
2936 tgl@sss.pgh.pa.us 9151 : 167 : FreeDir(tblspcdir);
9152 : :
1179 michael@paquier.xyz 9153 : 167 : state->starttime = (pg_time_t) time(NULL);
9154 : : }
135 peter@eisentraut.org 9155 [ - + ]:GNC 167 : PG_END_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, BoolGetDatum(true));
9156 : :
1179 michael@paquier.xyz 9157 :CBC 167 : state->started_in_recovery = backup_started_in_recovery;
9158 : :
9159 : : /*
9160 : : * Mark that the start phase has correctly finished for the backup.
9161 : : */
1352 sfrost@snowman.net 9162 : 167 : sessionBackupState = SESSION_BACKUP_RUNNING;
7807 tgl@sss.pgh.pa.us 9163 : 167 : }
9164 : :
9165 : : /*
9166 : : * Utility routine to fetch the session-level status of a backup running.
9167 : : */
9168 : : SessionBackupState
3191 teodor@sigaev.ru 9169 : 188 : get_backup_status(void)
9170 : : {
9171 : 188 : return sessionBackupState;
9172 : : }
9173 : :
9174 : : /*
9175 : : * do_pg_backup_stop
9176 : : *
9177 : : * Utility function called at the end of an online backup. It creates history
9178 : : * file (if required), resets sessionBackupState and so on. It can optionally
9179 : : * wait for WAL segments to be archived.
9180 : : *
9181 : : * "state" is filled with the information necessary to restore from this
9182 : : * backup with its stop LSN (stoppoint), its timeline ID (stoptli), etc.
9183 : : *
9184 : : * It is the responsibility of the caller of this function to verify the
9185 : : * permissions of the calling user!
9186 : : */
9187 : : void
1179 michael@paquier.xyz 9188 : 161 : do_pg_backup_stop(BackupState *state, bool waitforarchive)
9189 : : {
9190 : 161 : bool backup_stopped_in_recovery = false;
9191 : : char histfilepath[MAXPGPATH];
9192 : : char lastxlogfilename[MAXFNAMELEN];
9193 : : char histfilename[MAXFNAMELEN];
9194 : : XLogSegNo _logSegNo;
9195 : : FILE *fp;
9196 : : int seconds_before_warning;
6466 bruce@momjian.us 9197 : 161 : int waits = 0;
5723 simon@2ndQuadrant.co 9198 : 161 : bool reported_waiting = false;
9199 : :
1179 michael@paquier.xyz 9200 [ - + ]: 161 : Assert(state != NULL);
9201 : :
9202 : 161 : backup_stopped_in_recovery = RecoveryInProgress();
9203 : :
9204 : : /*
9205 : : * During recovery, we don't need to check WAL level. Because, if WAL
9206 : : * level is not sufficient, it's impossible to get here during recovery.
9207 : : */
9208 [ + + - + ]: 161 : if (!backup_stopped_in_recovery && !XLogIsNeeded())
6310 tgl@sss.pgh.pa.us 9209 [ # # ]:UBC 0 : ereport(ERROR,
9210 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9211 : : errmsg("WAL level not sufficient for making an online backup"),
9212 : : errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
9213 : :
9214 : : /*
9215 : : * OK to update backup counter and session-level lock.
9216 : : *
9217 : : * Note that CHECK_FOR_INTERRUPTS() must not occur while updating them,
9218 : : * otherwise they can be updated inconsistently, which might cause
9219 : : * do_pg_abort_backup() to fail.
9220 : : */
3257 fujii@postgresql.org 9221 :CBC 161 : WALInsertLockAcquireExclusive();
9222 : :
9223 : : /*
9224 : : * It is expected that each do_pg_backup_start() call is matched by
9225 : : * exactly one do_pg_backup_stop() call.
9226 : : */
1352 sfrost@snowman.net 9227 [ - + ]: 161 : Assert(XLogCtl->Insert.runningBackups > 0);
9228 : 161 : XLogCtl->Insert.runningBackups--;
9229 : :
9230 : : /*
9231 : : * Clean up session-level lock.
9232 : : *
9233 : : * You might think that WALInsertLockRelease() can be called before
9234 : : * cleaning up session-level lock because session-level lock doesn't need
9235 : : * to be protected with WAL insertion lock. But since
9236 : : * CHECK_FOR_INTERRUPTS() can occur in it, session-level lock must be
9237 : : * cleaned up before it.
9238 : : */
3191 teodor@sigaev.ru 9239 : 161 : sessionBackupState = SESSION_BACKUP_NONE;
9240 : :
2921 fujii@postgresql.org 9241 : 161 : WALInsertLockRelease();
9242 : :
9243 : : /*
9244 : : * If we are taking an online backup from the standby, we confirm that the
9245 : : * standby has not been promoted during the backup.
9246 : : */
1179 michael@paquier.xyz 9247 [ + + - + ]: 161 : if (state->started_in_recovery && !backup_stopped_in_recovery)
5076 simon@2ndQuadrant.co 9248 [ # # ]:UBC 0 : ereport(ERROR,
9249 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9250 : : errmsg("the standby was promoted during online backup"),
9251 : : errhint("This means that the backup being taken is corrupt "
9252 : : "and should not be used. "
9253 : : "Try taking another online backup.")));
9254 : :
9255 : : /*
9256 : : * During recovery, we don't write an end-of-backup record. We assume that
9257 : : * pg_control was backed up last and its minimum recovery point can be
9258 : : * available as the backup end location. Since we don't have an
9259 : : * end-of-backup record, we use the pg_control value to check whether
9260 : : * we've reached the end of backup when starting recovery from this
9261 : : * backup. We have no way of checking if pg_control wasn't backed up last
9262 : : * however.
9263 : : *
9264 : : * We don't force a switch to new WAL file but it is still possible to
9265 : : * wait for all the required files to be archived if waitforarchive is
9266 : : * true. This is okay if we use the backup to start a standby and fetch
9267 : : * the missing WAL using streaming replication. But in the case of an
9268 : : * archive recovery, a user should set waitforarchive to true and wait for
9269 : : * them to be archived to ensure that all the required files are
9270 : : * available.
9271 : : *
9272 : : * We return the current minimum recovery point as the backup end
9273 : : * location. Note that it can be greater than the exact backup end
9274 : : * location if the minimum recovery point is updated after the backup of
9275 : : * pg_control. This is harmless for current uses.
9276 : : *
9277 : : * XXX currently a backup history file is for informational and debug
9278 : : * purposes only. It's not essential for an online backup. Furthermore,
9279 : : * even if it's created, it will not be archived during recovery because
9280 : : * an archiver is not invoked. So it doesn't seem worthwhile to write a
9281 : : * backup history file during recovery.
9282 : : */
1179 michael@paquier.xyz 9283 [ + + ]:CBC 161 : if (backup_stopped_in_recovery)
9284 : : {
9285 : : XLogRecPtr recptr;
9286 : :
9287 : : /*
9288 : : * Check to see if all WAL replayed during online backup contain
9289 : : * full-page writes.
9290 : : */
4105 andres@anarazel.de 9291 [ - + ]: 6 : SpinLockAcquire(&XLogCtl->info_lck);
9292 : 6 : recptr = XLogCtl->lastFpwDisableRecPtr;
9293 : 6 : SpinLockRelease(&XLogCtl->info_lck);
9294 : :
1179 michael@paquier.xyz 9295 [ - + ]: 6 : if (state->startpoint <= recptr)
5076 simon@2ndQuadrant.co 9296 [ # # ]:UBC 0 : ereport(ERROR,
9297 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9298 : : errmsg("WAL generated with \"full_page_writes=off\" was replayed "
9299 : : "during online backup"),
9300 : : errhint("This means that the backup being taken on the standby "
9301 : : "is corrupt and should not be used. "
9302 : : "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
9303 : : "and then try an online backup again.")));
9304 : :
9305 : :
5076 simon@2ndQuadrant.co 9306 :CBC 6 : LWLockAcquire(ControlFileLock, LW_SHARED);
1179 michael@paquier.xyz 9307 : 6 : state->stoppoint = ControlFile->minRecoveryPoint;
9308 : 6 : state->stoptli = ControlFile->minRecoveryPointTLI;
5076 simon@2ndQuadrant.co 9309 : 6 : LWLockRelease(ControlFileLock);
9310 : : }
9311 : : else
9312 : : {
9313 : : char *history_file;
9314 : :
9315 : : /*
9316 : : * Write the backup-end xlog record
9317 : : */
3057 rhaas@postgresql.org 9318 : 155 : XLogBeginInsert();
310 peter@eisentraut.org 9319 : 155 : XLogRegisterData(&state->startpoint,
9320 : : sizeof(state->startpoint));
1179 michael@paquier.xyz 9321 : 155 : state->stoppoint = XLogInsert(RM_XLOG_ID, XLOG_BACKUP_END);
9322 : :
9323 : : /*
9324 : : * Given that we're not in recovery, InsertTimeLineID is set and can't
9325 : : * change, so we can read it without a lock.
9326 : : */
9327 : 155 : state->stoptli = XLogCtl->InsertTimeLineID;
9328 : :
9329 : : /*
9330 : : * Force a switch to a new xlog segment file, so that the backup is
9331 : : * valid as soon as archiver moves out the current segment file.
9332 : : */
3057 rhaas@postgresql.org 9333 : 155 : RequestXLogSwitch(false);
9334 : :
1179 michael@paquier.xyz 9335 : 155 : state->stoptime = (pg_time_t) time(NULL);
9336 : :
9337 : : /*
9338 : : * Write the backup history file
9339 : : */
9340 : 155 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9341 : 155 : BackupHistoryFilePath(histfilepath, state->stoptli, _logSegNo,
9342 : : state->startpoint, wal_segment_size);
3057 rhaas@postgresql.org 9343 : 155 : fp = AllocateFile(histfilepath, "w");
9344 [ - + ]: 155 : if (!fp)
3057 rhaas@postgresql.org 9345 [ # # ]:UBC 0 : ereport(ERROR,
9346 : : (errcode_for_file_access(),
9347 : : errmsg("could not create file \"%s\": %m",
9348 : : histfilepath)));
9349 : :
9350 : : /* Build and save the contents of the backup history file */
1179 michael@paquier.xyz 9351 :CBC 155 : history_file = build_backup_content(state, true);
1178 9352 : 155 : fprintf(fp, "%s", history_file);
1179 9353 : 155 : pfree(history_file);
9354 : :
3057 rhaas@postgresql.org 9355 [ + - + - : 155 : if (fflush(fp) || ferror(fp) || FreeFile(fp))
- + ]
3057 rhaas@postgresql.org 9356 [ # # ]:UBC 0 : ereport(ERROR,
9357 : : (errcode_for_file_access(),
9358 : : errmsg("could not write file \"%s\": %m",
9359 : : histfilepath)));
9360 : :
9361 : : /*
9362 : : * Clean out any no-longer-needed history files. As a side effect,
9363 : : * this will post a .ready file for the newly created history file,
9364 : : * notifying the archiver that history file may be archived
9365 : : * immediately.
9366 : : */
3057 rhaas@postgresql.org 9367 :CBC 155 : CleanupBackupHistory();
9368 : : }
9369 : :
9370 : : /*
9371 : : * If archiving is enabled, wait for all the required WAL files to be
9372 : : * archived before returning. If archiving isn't enabled, the required WAL
9373 : : * needs to be transported via streaming replication (hopefully with
9374 : : * wal_keep_size set high enough), or some more exotic mechanism like
9375 : : * polling and copying files from pg_wal with script. We have no knowledge
9376 : : * of those mechanisms, so it's up to the user to ensure that he gets all
9377 : : * the required WAL.
9378 : : *
9379 : : * We wait until both the last WAL file filled during backup and the
9380 : : * history file have been archived, and assume that the alphabetic sorting
9381 : : * property of the WAL files ensures any earlier WAL files are safely
9382 : : * archived as well.
9383 : : *
9384 : : * We wait forever, since archive_command is supposed to work and we
9385 : : * assume the admin wanted his backup to work completely. If you don't
9386 : : * wish to wait, then either waitforarchive should be passed in as false,
9387 : : * or you can set statement_timeout. Also, some notices are issued to
9388 : : * clue in anyone who might be doing this interactively.
9389 : : */
9390 : :
9391 [ + + ]: 161 : if (waitforarchive &&
1179 michael@paquier.xyz 9392 [ + + + + : 10 : ((!backup_stopped_in_recovery && XLogArchivingActive()) ||
- + + + +
+ ]
9393 [ + - - + : 1 : (backup_stopped_in_recovery && XLogArchivingAlways())))
- + ]
9394 : : {
9395 : 4 : XLByteToPrevSeg(state->stoppoint, _logSegNo, wal_segment_size);
9396 : 4 : XLogFileName(lastxlogfilename, state->stoptli, _logSegNo,
9397 : : wal_segment_size);
9398 : :
9399 : 4 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9400 : 4 : BackupHistoryFileName(histfilename, state->stoptli, _logSegNo,
9401 : : state->startpoint, wal_segment_size);
9402 : :
5644 bruce@momjian.us 9403 : 4 : seconds_before_warning = 60;
9404 : 4 : waits = 0;
9405 : :
9406 [ + + - + ]: 12 : while (XLogArchiveIsBusy(lastxlogfilename) ||
9407 : 4 : XLogArchiveIsBusy(histfilename))
9408 : : {
9409 [ - + ]: 4 : CHECK_FOR_INTERRUPTS();
9410 : :
9411 [ + - - + ]: 4 : if (!reported_waiting && waits > 5)
9412 : : {
5644 bruce@momjian.us 9413 [ # # ]:UBC 0 : ereport(NOTICE,
9414 : : (errmsg("base backup done, waiting for required WAL segments to be archived")));
9415 : 0 : reported_waiting = true;
9416 : : }
9417 : :
1626 michael@paquier.xyz 9418 :CBC 4 : (void) WaitLatch(MyLatch,
9419 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
9420 : : 1000L,
9421 : : WAIT_EVENT_BACKUP_WAIT_WAL_ARCHIVE);
9422 : 4 : ResetLatch(MyLatch);
9423 : :
5644 bruce@momjian.us 9424 [ - + ]: 4 : if (++waits >= seconds_before_warning)
9425 : : {
5644 bruce@momjian.us 9426 :UBC 0 : seconds_before_warning *= 2; /* This wraps in >10 years... */
9427 [ # # ]: 0 : ereport(WARNING,
9428 : : (errmsg("still waiting for all required WAL segments to be archived (%d seconds elapsed)",
9429 : : waits),
9430 : : errhint("Check that your \"archive_command\" is executing properly. "
9431 : : "You can safely cancel this backup, "
9432 : : "but the database backup will not be usable without all the WAL segments.")));
9433 : : }
9434 : : }
9435 : :
5644 bruce@momjian.us 9436 [ + + ]:CBC 4 : ereport(NOTICE,
9437 : : (errmsg("all required WAL segments have been archived")));
9438 : : }
5426 magnus@hagander.net 9439 [ + + ]: 157 : else if (waitforarchive)
5712 tgl@sss.pgh.pa.us 9440 [ + - ]: 6 : ereport(NOTICE,
9441 : : (errmsg("WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup")));
5457 magnus@hagander.net 9442 : 161 : }
9443 : :
9444 : :
9445 : : /*
9446 : : * do_pg_abort_backup: abort a running backup
9447 : : *
9448 : : * This does just the most basic steps of do_pg_backup_stop(), by taking the
9449 : : * system out of backup mode, thus making it a lot more safe to call from
9450 : : * an error handler.
9451 : : *
9452 : : * 'arg' indicates that it's being called during backup setup; so
9453 : : * sessionBackupState has not been modified yet, but runningBackups has
9454 : : * already been incremented. When it's false, then it's invoked as a
9455 : : * before_shmem_exit handler, and therefore we must not change state
9456 : : * unless sessionBackupState indicates that a backup is actually running.
9457 : : *
9458 : : * NB: This gets used as a PG_ENSURE_ERROR_CLEANUP callback and
9459 : : * before_shmem_exit handler, hence the odd-looking signature.
9460 : : */
9461 : : void
2191 rhaas@postgresql.org 9462 : 8 : do_pg_abort_backup(int code, Datum arg)
9463 : : {
1156 alvherre@alvh.no-ip. 9464 : 8 : bool during_backup_start = DatumGetBool(arg);
9465 : :
9466 : : /* If called during backup start, there shouldn't be one already running */
1151 9467 [ - + - - ]: 8 : Assert(!during_backup_start || sessionBackupState == SESSION_BACKUP_NONE);
9468 : :
1156 9469 [ + - + + ]: 8 : if (during_backup_start || sessionBackupState != SESSION_BACKUP_NONE)
9470 : : {
9471 : 6 : WALInsertLockAcquireExclusive();
9472 [ - + ]: 6 : Assert(XLogCtl->Insert.runningBackups > 0);
9473 : 6 : XLogCtl->Insert.runningBackups--;
9474 : :
9475 : 6 : sessionBackupState = SESSION_BACKUP_NONE;
9476 : 6 : WALInsertLockRelease();
9477 : :
9478 [ + - ]: 6 : if (!during_backup_start)
9479 [ + - ]: 6 : ereport(WARNING,
9480 : : errmsg("aborting backup due to backend exiting before pg_backup_stop was called"));
9481 : : }
2191 rhaas@postgresql.org 9482 : 8 : }
9483 : :
9484 : : /*
9485 : : * Register a handler that will warn about unterminated backups at end of
9486 : : * session, unless this has already been done.
9487 : : */
9488 : : void
9489 : 4 : register_persistent_abort_backup_handler(void)
9490 : : {
9491 : : static bool already_done = false;
9492 : :
9493 [ + + ]: 4 : if (already_done)
9494 : 1 : return;
135 peter@eisentraut.org 9495 :GNC 3 : before_shmem_exit(do_pg_abort_backup, BoolGetDatum(false));
2191 rhaas@postgresql.org 9496 :CBC 3 : already_done = true;
9497 : : }
9498 : :
9499 : : /*
9500 : : * Get latest WAL insert pointer
9501 : : */
9502 : : XLogRecPtr
5090 heikki.linnakangas@i 9503 : 1951 : GetXLogInsertRecPtr(void)
9504 : : {
4105 andres@anarazel.de 9505 : 1951 : XLogCtlInsert *Insert = &XLogCtl->Insert;
9506 : : uint64 current_bytepos;
9507 : :
4546 heikki.linnakangas@i 9508 [ - + ]: 1951 : SpinLockAcquire(&Insert->insertpos_lck);
9509 : 1951 : current_bytepos = Insert->CurrBytePos;
9510 : 1951 : SpinLockRelease(&Insert->insertpos_lck);
9511 : :
9512 : 1951 : return XLogBytePosToRecPtr(current_bytepos);
9513 : : }
9514 : :
9515 : : /*
9516 : : * Get latest WAL write pointer
9517 : : */
9518 : : XLogRecPtr
1401 9519 : 1426 : GetXLogWriteRecPtr(void)
9520 : : {
624 alvherre@alvh.no-ip. 9521 : 1426 : RefreshXLogWriteResult(LogwrtResult);
9522 : :
1401 heikki.linnakangas@i 9523 : 1426 : return LogwrtResult.Write;
9524 : : }
9525 : :
9526 : : /*
9527 : : * Returns the redo pointer of the last checkpoint or restartpoint. This is
9528 : : * the oldest point in WAL that we still need, if we have to restart recovery.
9529 : : */
9530 : : void
9531 : 379 : GetOldestRestartPoint(XLogRecPtr *oldrecptr, TimeLineID *oldtli)
9532 : : {
9533 : 379 : LWLockAcquire(ControlFileLock, LW_SHARED);
9534 : 379 : *oldrecptr = ControlFile->checkPointCopy.redo;
9535 : 379 : *oldtli = ControlFile->checkPointCopy.ThisTimeLineID;
9536 : 379 : LWLockRelease(ControlFileLock);
7209 tgl@sss.pgh.pa.us 9537 : 379 : }
9538 : :
9539 : : /* Thin wrapper around ShutdownWalRcv(). */
9540 : : void
1634 noah@leadboat.com 9541 : 923 : XLogShutdownWalRcv(void)
9542 : : {
43 michael@paquier.xyz 9543 [ + + - + ]:GNC 923 : Assert(AmStartupProcess() || !IsUnderPostmaster);
9544 : :
1634 noah@leadboat.com 9545 :CBC 923 : ShutdownWalRcv();
44 michael@paquier.xyz 9546 : 923 : ResetInstallXLogFileSegmentActive();
1634 noah@leadboat.com 9547 : 923 : }
9548 : :
9549 : : /* Enable WAL file recycling and preallocation. */
9550 : : void
1401 heikki.linnakangas@i 9551 : 1076 : SetInstallXLogFileSegmentActive(void)
9552 : : {
9553 : 1076 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9554 : 1076 : XLogCtl->InstallXLogFileSegmentActive = true;
9555 : 1076 : LWLockRelease(ControlFileLock);
3753 fujii@postgresql.org 9556 : 1076 : }
9557 : :
9558 : : /* Disable WAL file recycling and preallocation. */
9559 : : void
44 michael@paquier.xyz 9560 : 1027 : ResetInstallXLogFileSegmentActive(void)
9561 : : {
9562 : 1027 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9563 : 1027 : XLogCtl->InstallXLogFileSegmentActive = false;
9564 : 1027 : LWLockRelease(ControlFileLock);
9565 : 1027 : }
9566 : :
9567 : : bool
1401 heikki.linnakangas@i 9568 : 357 : IsInstallXLogFileSegmentActive(void)
9569 : : {
9570 : : bool result;
9571 : :
9572 : 357 : LWLockAcquire(ControlFileLock, LW_SHARED);
9573 : 357 : result = XLogCtl->InstallXLogFileSegmentActive;
9574 : 357 : LWLockRelease(ControlFileLock);
9575 : :
9576 : 357 : return result;
9577 : : }
9578 : :
9579 : : /*
9580 : : * Update the WalWriterSleeping flag.
9581 : : */
9582 : : void
4972 tgl@sss.pgh.pa.us 9583 : 482 : SetWalWriterSleeping(bool sleeping)
9584 : : {
4105 andres@anarazel.de 9585 [ - + ]: 482 : SpinLockAcquire(&XLogCtl->info_lck);
9586 : 482 : XLogCtl->WalWriterSleeping = sleeping;
9587 : 482 : SpinLockRelease(&XLogCtl->info_lck);
4972 tgl@sss.pgh.pa.us 9588 : 482 : }
|
