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 "backup/basebackup.h"
66 : : #include "catalog/catversion.h"
67 : : #include "catalog/pg_control.h"
68 : : #include "catalog/pg_database.h"
69 : : #include "common/controldata_utils.h"
70 : : #include "common/file_utils.h"
71 : : #include "executor/instrument.h"
72 : : #include "miscadmin.h"
73 : : #include "pg_trace.h"
74 : : #include "pgstat.h"
75 : : #include "port/atomics.h"
76 : : #include "postmaster/bgwriter.h"
77 : : #include "postmaster/startup.h"
78 : : #include "postmaster/walsummarizer.h"
79 : : #include "postmaster/walwriter.h"
80 : : #include "replication/origin.h"
81 : : #include "replication/slot.h"
82 : : #include "replication/snapbuild.h"
83 : : #include "replication/walreceiver.h"
84 : : #include "replication/walsender.h"
85 : : #include "storage/bufmgr.h"
86 : : #include "storage/fd.h"
87 : : #include "storage/ipc.h"
88 : : #include "storage/large_object.h"
89 : : #include "storage/latch.h"
90 : : #include "storage/predicate.h"
91 : : #include "storage/proc.h"
92 : : #include "storage/procarray.h"
93 : : #include "storage/reinit.h"
94 : : #include "storage/spin.h"
95 : : #include "storage/sync.h"
96 : : #include "utils/guc_hooks.h"
97 : : #include "utils/guc_tables.h"
98 : : #include "utils/injection_point.h"
99 : : #include "utils/pgstat_internal.h"
100 : : #include "utils/ps_status.h"
101 : : #include "utils/relmapper.h"
102 : : #include "utils/snapmgr.h"
103 : : #include "utils/timeout.h"
104 : : #include "utils/timestamp.h"
105 : : #include "utils/varlena.h"
106 : :
107 : : #ifdef WAL_DEBUG
108 : : #include "utils/memutils.h"
109 : : #endif
110 : :
111 : : /* timeline ID to be used when bootstrapping */
112 : : #define BootstrapTimeLineID 1
113 : :
114 : : /* User-settable parameters */
115 : : int max_wal_size_mb = 1024; /* 1 GB */
116 : : int min_wal_size_mb = 80; /* 80 MB */
117 : : int wal_keep_size_mb = 0;
118 : : int XLOGbuffers = -1;
119 : : int XLogArchiveTimeout = 0;
120 : : int XLogArchiveMode = ARCHIVE_MODE_OFF;
121 : : char *XLogArchiveCommand = NULL;
122 : : bool EnableHotStandby = false;
123 : : bool fullPageWrites = true;
124 : : bool wal_log_hints = false;
125 : : int wal_compression = WAL_COMPRESSION_NONE;
126 : : char *wal_consistency_checking_string = NULL;
127 : : bool *wal_consistency_checking = NULL;
128 : : bool wal_init_zero = true;
129 : : bool wal_recycle = true;
130 : : bool log_checkpoints = true;
131 : : int wal_sync_method = DEFAULT_WAL_SYNC_METHOD;
132 : : int wal_level = WAL_LEVEL_REPLICA;
133 : : int CommitDelay = 0; /* precommit delay in microseconds */
134 : : int CommitSiblings = 5; /* # concurrent xacts needed to sleep */
135 : : int wal_retrieve_retry_interval = 5000;
136 : : int max_slot_wal_keep_size_mb = -1;
137 : : int wal_decode_buffer_size = 512 * 1024;
138 : : bool track_wal_io_timing = false;
139 : :
140 : : #ifdef WAL_DEBUG
141 : : bool XLOG_DEBUG = false;
142 : : #endif
143 : :
144 : : int wal_segment_size = DEFAULT_XLOG_SEG_SIZE;
145 : :
146 : : /*
147 : : * Number of WAL insertion locks to use. A higher value allows more insertions
148 : : * to happen concurrently, but adds some CPU overhead to flushing the WAL,
149 : : * which needs to iterate all the locks.
150 : : */
151 : : #define NUM_XLOGINSERT_LOCKS 8
152 : :
153 : : /*
154 : : * Max distance from last checkpoint, before triggering a new xlog-based
155 : : * checkpoint.
156 : : */
157 : : int CheckPointSegments;
158 : :
159 : : /* Estimated distance between checkpoints, in bytes */
160 : : static double CheckPointDistanceEstimate = 0;
161 : : static double PrevCheckPointDistance = 0;
162 : :
163 : : /*
164 : : * Track whether there were any deferred checks for custom resource managers
165 : : * specified in wal_consistency_checking.
166 : : */
167 : : static bool check_wal_consistency_checking_deferred = false;
168 : :
169 : : /*
170 : : * GUC support
171 : : */
172 : : const struct config_enum_entry wal_sync_method_options[] = {
173 : : {"fsync", WAL_SYNC_METHOD_FSYNC, false},
174 : : #ifdef HAVE_FSYNC_WRITETHROUGH
175 : : {"fsync_writethrough", WAL_SYNC_METHOD_FSYNC_WRITETHROUGH, false},
176 : : #endif
177 : : {"fdatasync", WAL_SYNC_METHOD_FDATASYNC, false},
178 : : #ifdef O_SYNC
179 : : {"open_sync", WAL_SYNC_METHOD_OPEN, false},
180 : : #endif
181 : : #ifdef O_DSYNC
182 : : {"open_datasync", WAL_SYNC_METHOD_OPEN_DSYNC, false},
183 : : #endif
184 : : {NULL, 0, false}
185 : : };
186 : :
187 : :
188 : : /*
189 : : * Although only "on", "off", and "always" are documented,
190 : : * we accept all the likely variants of "on" and "off".
191 : : */
192 : : const struct config_enum_entry archive_mode_options[] = {
193 : : {"always", ARCHIVE_MODE_ALWAYS, false},
194 : : {"on", ARCHIVE_MODE_ON, false},
195 : : {"off", ARCHIVE_MODE_OFF, false},
196 : : {"true", ARCHIVE_MODE_ON, true},
197 : : {"false", ARCHIVE_MODE_OFF, true},
198 : : {"yes", ARCHIVE_MODE_ON, true},
199 : : {"no", ARCHIVE_MODE_OFF, true},
200 : : {"1", ARCHIVE_MODE_ON, true},
201 : : {"0", ARCHIVE_MODE_OFF, true},
202 : : {NULL, 0, false}
203 : : };
204 : :
205 : : /*
206 : : * Statistics for current checkpoint are collected in this global struct.
207 : : * Because only the checkpointer or a stand-alone backend can perform
208 : : * checkpoints, this will be unused in normal backends.
209 : : */
210 : : CheckpointStatsData CheckpointStats;
211 : :
212 : : /*
213 : : * During recovery, lastFullPageWrites keeps track of full_page_writes that
214 : : * the replayed WAL records indicate. It's initialized with full_page_writes
215 : : * that the recovery starting checkpoint record indicates, and then updated
216 : : * each time XLOG_FPW_CHANGE record is replayed.
217 : : */
218 : : static bool lastFullPageWrites;
219 : :
220 : : /*
221 : : * Local copy of the state tracked by SharedRecoveryState in shared memory,
222 : : * It is false if SharedRecoveryState is RECOVERY_STATE_DONE. True actually
223 : : * means "not known, need to check the shared state".
224 : : */
225 : : static bool LocalRecoveryInProgress = true;
226 : :
227 : : /*
228 : : * Local state for XLogInsertAllowed():
229 : : * 1: unconditionally allowed to insert XLOG
230 : : * 0: unconditionally not allowed to insert XLOG
231 : : * -1: must check RecoveryInProgress(); disallow until it is false
232 : : * Most processes start with -1 and transition to 1 after seeing that recovery
233 : : * is not in progress. But we can also force the value for special cases.
234 : : * The coding in XLogInsertAllowed() depends on the first two of these states
235 : : * being numerically the same as bool true and false.
236 : : */
237 : : static int LocalXLogInsertAllowed = -1;
238 : :
239 : : /*
240 : : * ProcLastRecPtr points to the start of the last XLOG record inserted by the
241 : : * current backend. It is updated for all inserts. XactLastRecEnd points to
242 : : * end+1 of the last record, and is reset when we end a top-level transaction,
243 : : * or start a new one; so it can be used to tell if the current transaction has
244 : : * created any XLOG records.
245 : : *
246 : : * While in parallel mode, this may not be fully up to date. When committing,
247 : : * a transaction can assume this covers all xlog records written either by the
248 : : * user backend or by any parallel worker which was present at any point during
249 : : * the transaction. But when aborting, or when still in parallel mode, other
250 : : * parallel backends may have written WAL records at later LSNs than the value
251 : : * stored here. The parallel leader advances its own copy, when necessary,
252 : : * in WaitForParallelWorkersToFinish.
253 : : */
254 : : XLogRecPtr ProcLastRecPtr = InvalidXLogRecPtr;
255 : : XLogRecPtr XactLastRecEnd = InvalidXLogRecPtr;
256 : : XLogRecPtr XactLastCommitEnd = InvalidXLogRecPtr;
257 : :
258 : : /*
259 : : * RedoRecPtr is this backend's local copy of the REDO record pointer
260 : : * (which is almost but not quite the same as a pointer to the most recent
261 : : * CHECKPOINT record). We update this from the shared-memory copy,
262 : : * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
263 : : * hold an insertion lock). See XLogInsertRecord for details. We are also
264 : : * allowed to update from XLogCtl->RedoRecPtr if we hold the info_lck;
265 : : * see GetRedoRecPtr.
266 : : *
267 : : * NB: Code that uses this variable must be prepared not only for the
268 : : * possibility that it may be arbitrarily out of date, but also for the
269 : : * possibility that it might be set to InvalidXLogRecPtr. We used to
270 : : * initialize it as a side effect of the first call to RecoveryInProgress(),
271 : : * which meant that most code that might use it could assume that it had a
272 : : * real if perhaps stale value. That's no longer the case.
273 : : */
274 : : static XLogRecPtr RedoRecPtr;
275 : :
276 : : /*
277 : : * doPageWrites is this backend's local copy of (fullPageWrites ||
278 : : * runningBackups > 0). It is used together with RedoRecPtr to decide whether
279 : : * a full-page image of a page need to be taken.
280 : : *
281 : : * NB: Initially this is false, and there's no guarantee that it will be
282 : : * initialized to any other value before it is first used. Any code that
283 : : * makes use of it must recheck the value after obtaining a WALInsertLock,
284 : : * and respond appropriately if it turns out that the previous value wasn't
285 : : * accurate.
286 : : */
287 : : static bool doPageWrites;
288 : :
289 : : /*----------
290 : : * Shared-memory data structures for XLOG control
291 : : *
292 : : * LogwrtRqst indicates a byte position that we need to write and/or fsync
293 : : * the log up to (all records before that point must be written or fsynced).
294 : : * The positions already written/fsynced are maintained in logWriteResult
295 : : * and logFlushResult using atomic access.
296 : : * In addition to the shared variable, each backend has a private copy of
297 : : * both in LogwrtResult, which is updated when convenient.
298 : : *
299 : : * The request bookkeeping is simpler: there is a shared XLogCtl->LogwrtRqst
300 : : * (protected by info_lck), but we don't need to cache any copies of it.
301 : : *
302 : : * info_lck is only held long enough to read/update the protected variables,
303 : : * so it's a plain spinlock. The other locks are held longer (potentially
304 : : * over I/O operations), so we use LWLocks for them. These locks are:
305 : : *
306 : : * WALBufMappingLock: must be held to replace a page in the WAL buffer cache.
307 : : * It is only held while initializing and changing the mapping. If the
308 : : * contents of the buffer being replaced haven't been written yet, the mapping
309 : : * lock is released while the write is done, and reacquired afterwards.
310 : : *
311 : : * WALWriteLock: must be held to write WAL buffers to disk (XLogWrite or
312 : : * XLogFlush).
313 : : *
314 : : * ControlFileLock: must be held to read/update control file or create
315 : : * new log file.
316 : : *
317 : : *----------
318 : : */
319 : :
320 : : typedef struct XLogwrtRqst
321 : : {
322 : : XLogRecPtr Write; /* last byte + 1 to write out */
323 : : XLogRecPtr Flush; /* last byte + 1 to flush */
324 : : } XLogwrtRqst;
325 : :
326 : : typedef struct XLogwrtResult
327 : : {
328 : : XLogRecPtr Write; /* last byte + 1 written out */
329 : : XLogRecPtr Flush; /* last byte + 1 flushed */
330 : : } XLogwrtResult;
331 : :
332 : : /*
333 : : * Inserting to WAL is protected by a small fixed number of WAL insertion
334 : : * locks. To insert to the WAL, you must hold one of the locks - it doesn't
335 : : * matter which one. To lock out other concurrent insertions, you must hold
336 : : * of them. Each WAL insertion lock consists of a lightweight lock, plus an
337 : : * indicator of how far the insertion has progressed (insertingAt).
338 : : *
339 : : * The insertingAt values are read when a process wants to flush WAL from
340 : : * the in-memory buffers to disk, to check that all the insertions to the
341 : : * region the process is about to write out have finished. You could simply
342 : : * wait for all currently in-progress insertions to finish, but the
343 : : * insertingAt indicator allows you to ignore insertions to later in the WAL,
344 : : * so that you only wait for the insertions that are modifying the buffers
345 : : * you're about to write out.
346 : : *
347 : : * This isn't just an optimization. If all the WAL buffers are dirty, an
348 : : * inserter that's holding a WAL insert lock might need to evict an old WAL
349 : : * buffer, which requires flushing the WAL. If it's possible for an inserter
350 : : * to block on another inserter unnecessarily, deadlock can arise when two
351 : : * inserters holding a WAL insert lock wait for each other to finish their
352 : : * insertion.
353 : : *
354 : : * Small WAL records that don't cross a page boundary never update the value,
355 : : * the WAL record is just copied to the page and the lock is released. But
356 : : * to avoid the deadlock-scenario explained above, the indicator is always
357 : : * updated before sleeping while holding an insertion lock.
358 : : *
359 : : * lastImportantAt contains the LSN of the last important WAL record inserted
360 : : * using a given lock. This value is used to detect if there has been
361 : : * important WAL activity since the last time some action, like a checkpoint,
362 : : * was performed - allowing to not repeat the action if not. The LSN is
363 : : * updated for all insertions, unless the XLOG_MARK_UNIMPORTANT flag was
364 : : * set. lastImportantAt is never cleared, only overwritten by the LSN of newer
365 : : * records. Tracking the WAL activity directly in WALInsertLock has the
366 : : * advantage of not needing any additional locks to update the value.
367 : : */
368 : : typedef struct
369 : : {
370 : : LWLock lock;
371 : : pg_atomic_uint64 insertingAt;
372 : : XLogRecPtr lastImportantAt;
373 : : } WALInsertLock;
374 : :
375 : : /*
376 : : * All the WAL insertion locks are allocated as an array in shared memory. We
377 : : * force the array stride to be a power of 2, which saves a few cycles in
378 : : * indexing, but more importantly also ensures that individual slots don't
379 : : * cross cache line boundaries. (Of course, we have to also ensure that the
380 : : * array start address is suitably aligned.)
381 : : */
382 : : typedef union WALInsertLockPadded
383 : : {
384 : : WALInsertLock l;
385 : : char pad[PG_CACHE_LINE_SIZE];
386 : : } WALInsertLockPadded;
387 : :
388 : : /*
389 : : * Session status of running backup, used for sanity checks in SQL-callable
390 : : * functions to start and stop backups.
391 : : */
392 : : static SessionBackupState sessionBackupState = SESSION_BACKUP_NONE;
393 : :
394 : : /*
395 : : * Shared state data for WAL insertion.
396 : : */
397 : : typedef struct XLogCtlInsert
398 : : {
399 : : slock_t insertpos_lck; /* protects CurrBytePos and PrevBytePos */
400 : :
401 : : /*
402 : : * CurrBytePos is the end of reserved WAL. The next record will be
403 : : * inserted at that position. PrevBytePos is the start position of the
404 : : * previously inserted (or rather, reserved) record - it is copied to the
405 : : * prev-link of the next record. These are stored as "usable byte
406 : : * positions" rather than XLogRecPtrs (see XLogBytePosToRecPtr()).
407 : : */
408 : : uint64 CurrBytePos;
409 : : uint64 PrevBytePos;
410 : :
411 : : /*
412 : : * Make sure the above heavily-contended spinlock and byte positions are
413 : : * on their own cache line. In particular, the RedoRecPtr and full page
414 : : * write variables below should be on a different cache line. They are
415 : : * read on every WAL insertion, but updated rarely, and we don't want
416 : : * those reads to steal the cache line containing Curr/PrevBytePos.
417 : : */
418 : : char pad[PG_CACHE_LINE_SIZE];
419 : :
420 : : /*
421 : : * fullPageWrites is the authoritative value used by all backends to
422 : : * determine whether to write full-page image to WAL. This shared value,
423 : : * instead of the process-local fullPageWrites, is required because, when
424 : : * full_page_writes is changed by SIGHUP, we must WAL-log it before it
425 : : * actually affects WAL-logging by backends. Checkpointer sets at startup
426 : : * or after SIGHUP.
427 : : *
428 : : * To read these fields, you must hold an insertion lock. To modify them,
429 : : * you must hold ALL the locks.
430 : : */
431 : : XLogRecPtr RedoRecPtr; /* current redo point for insertions */
432 : : bool fullPageWrites;
433 : :
434 : : /*
435 : : * runningBackups is a counter indicating the number of backups currently
436 : : * in progress. lastBackupStart is the latest checkpoint redo location
437 : : * used as a starting point for an online backup.
438 : : */
439 : : int runningBackups;
440 : : XLogRecPtr lastBackupStart;
441 : :
442 : : /*
443 : : * WAL insertion locks.
444 : : */
445 : : WALInsertLockPadded *WALInsertLocks;
446 : : } XLogCtlInsert;
447 : :
448 : : /*
449 : : * Total shared-memory state for XLOG.
450 : : */
451 : : typedef struct XLogCtlData
452 : : {
453 : : XLogCtlInsert Insert;
454 : :
455 : : /* Protected by info_lck: */
456 : : XLogwrtRqst LogwrtRqst;
457 : : XLogRecPtr RedoRecPtr; /* a recent copy of Insert->RedoRecPtr */
458 : : XLogRecPtr asyncXactLSN; /* LSN of newest async commit/abort */
459 : : XLogRecPtr replicationSlotMinLSN; /* oldest LSN needed by any slot */
460 : :
461 : : XLogSegNo lastRemovedSegNo; /* latest removed/recycled XLOG segment */
462 : :
463 : : /* Fake LSN counter, for unlogged relations. */
464 : : pg_atomic_uint64 unloggedLSN;
465 : :
466 : : /* Time and LSN of last xlog segment switch. Protected by WALWriteLock. */
467 : : pg_time_t lastSegSwitchTime;
468 : : XLogRecPtr lastSegSwitchLSN;
469 : :
470 : : /* These are accessed using atomics -- info_lck not needed */
471 : : pg_atomic_uint64 logInsertResult; /* last byte + 1 inserted to buffers */
472 : : pg_atomic_uint64 logWriteResult; /* last byte + 1 written out */
473 : : pg_atomic_uint64 logFlushResult; /* last byte + 1 flushed */
474 : :
475 : : /*
476 : : * Latest initialized page in the cache (last byte position + 1).
477 : : *
478 : : * To change the identity of a buffer (and InitializedUpTo), you need to
479 : : * hold WALBufMappingLock. To change the identity of a buffer that's
480 : : * still dirty, the old page needs to be written out first, and for that
481 : : * you need WALWriteLock, and you need to ensure that there are no
482 : : * in-progress insertions to the page by calling
483 : : * WaitXLogInsertionsToFinish().
484 : : */
485 : : XLogRecPtr InitializedUpTo;
486 : :
487 : : /*
488 : : * These values do not change after startup, although the pointed-to pages
489 : : * and xlblocks values certainly do. xlblocks values are protected by
490 : : * WALBufMappingLock.
491 : : */
492 : : char *pages; /* buffers for unwritten XLOG pages */
493 : : pg_atomic_uint64 *xlblocks; /* 1st byte ptr-s + XLOG_BLCKSZ */
494 : : int XLogCacheBlck; /* highest allocated xlog buffer index */
495 : :
496 : : /*
497 : : * InsertTimeLineID is the timeline into which new WAL is being inserted
498 : : * and flushed. It is zero during recovery, and does not change once set.
499 : : *
500 : : * If we create a new timeline when the system was started up,
501 : : * PrevTimeLineID is the old timeline's ID that we forked off from.
502 : : * Otherwise it's equal to InsertTimeLineID.
503 : : *
504 : : * We set these fields while holding info_lck. Most that reads these
505 : : * values knows that recovery is no longer in progress and so can safely
506 : : * read the value without a lock, but code that could be run either during
507 : : * or after recovery can take info_lck while reading these values.
508 : : */
509 : : TimeLineID InsertTimeLineID;
510 : : TimeLineID PrevTimeLineID;
511 : :
512 : : /*
513 : : * SharedRecoveryState indicates if we're still in crash or archive
514 : : * recovery. Protected by info_lck.
515 : : */
516 : : RecoveryState SharedRecoveryState;
517 : :
518 : : /*
519 : : * InstallXLogFileSegmentActive indicates whether the checkpointer should
520 : : * arrange for future segments by recycling and/or PreallocXlogFiles().
521 : : * Protected by ControlFileLock. Only the startup process changes it. If
522 : : * true, anyone can use InstallXLogFileSegment(). If false, the startup
523 : : * process owns the exclusive right to install segments, by reading from
524 : : * the archive and possibly replacing existing files.
525 : : */
526 : : bool InstallXLogFileSegmentActive;
527 : :
528 : : /*
529 : : * WalWriterSleeping indicates whether the WAL writer is currently in
530 : : * low-power mode (and hence should be nudged if an async commit occurs).
531 : : * Protected by info_lck.
532 : : */
533 : : bool WalWriterSleeping;
534 : :
535 : : /*
536 : : * During recovery, we keep a copy of the latest checkpoint record here.
537 : : * lastCheckPointRecPtr points to start of checkpoint record and
538 : : * lastCheckPointEndPtr points to end+1 of checkpoint record. Used by the
539 : : * checkpointer when it wants to create a restartpoint.
540 : : *
541 : : * Protected by info_lck.
542 : : */
543 : : XLogRecPtr lastCheckPointRecPtr;
544 : : XLogRecPtr lastCheckPointEndPtr;
545 : : CheckPoint lastCheckPoint;
546 : :
547 : : /*
548 : : * lastFpwDisableRecPtr points to the start of the last replayed
549 : : * XLOG_FPW_CHANGE record that instructs full_page_writes is disabled.
550 : : */
551 : : XLogRecPtr lastFpwDisableRecPtr;
552 : :
553 : : slock_t info_lck; /* locks shared variables shown above */
554 : : } XLogCtlData;
555 : :
556 : : /*
557 : : * Classification of XLogInsertRecord operations.
558 : : */
559 : : typedef enum
560 : : {
561 : : WALINSERT_NORMAL,
562 : : WALINSERT_SPECIAL_SWITCH,
563 : : WALINSERT_SPECIAL_CHECKPOINT
564 : : } WalInsertClass;
565 : :
566 : : static XLogCtlData *XLogCtl = NULL;
567 : :
568 : : /* a private copy of XLogCtl->Insert.WALInsertLocks, for convenience */
569 : : static WALInsertLockPadded *WALInsertLocks = NULL;
570 : :
571 : : /*
572 : : * We maintain an image of pg_control in shared memory.
573 : : */
574 : : static ControlFileData *ControlFile = NULL;
575 : :
576 : : /*
577 : : * Calculate the amount of space left on the page after 'endptr'. Beware
578 : : * multiple evaluation!
579 : : */
580 : : #define INSERT_FREESPACE(endptr) \
581 : : (((endptr) % XLOG_BLCKSZ == 0) ? 0 : (XLOG_BLCKSZ - (endptr) % XLOG_BLCKSZ))
582 : :
583 : : /* Macro to advance to next buffer index. */
584 : : #define NextBufIdx(idx) \
585 : : (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))
586 : :
587 : : /*
588 : : * XLogRecPtrToBufIdx returns the index of the WAL buffer that holds, or
589 : : * would hold if it was in cache, the page containing 'recptr'.
590 : : */
591 : : #define XLogRecPtrToBufIdx(recptr) \
592 : : (((recptr) / XLOG_BLCKSZ) % (XLogCtl->XLogCacheBlck + 1))
593 : :
594 : : /*
595 : : * These are the number of bytes in a WAL page usable for WAL data.
596 : : */
597 : : #define UsableBytesInPage (XLOG_BLCKSZ - SizeOfXLogShortPHD)
598 : :
599 : : /*
600 : : * Convert values of GUCs measured in megabytes to equiv. segment count.
601 : : * Rounds down.
602 : : */
603 : : #define ConvertToXSegs(x, segsize) XLogMBVarToSegs((x), (segsize))
604 : :
605 : : /* The number of bytes in a WAL segment usable for WAL data. */
606 : : static int UsableBytesInSegment;
607 : :
608 : : /*
609 : : * Private, possibly out-of-date copy of shared LogwrtResult.
610 : : * See discussion above.
611 : : */
612 : : static XLogwrtResult LogwrtResult = {0, 0};
613 : :
614 : : /*
615 : : * Update local copy of shared XLogCtl->log{Write,Flush}Result
616 : : *
617 : : * It's critical that Flush always trails Write, so the order of the reads is
618 : : * important, as is the barrier. See also XLogWrite.
619 : : */
620 : : #define RefreshXLogWriteResult(_target) \
621 : : do { \
622 : : _target.Flush = pg_atomic_read_u64(&XLogCtl->logFlushResult); \
623 : : pg_read_barrier(); \
624 : : _target.Write = pg_atomic_read_u64(&XLogCtl->logWriteResult); \
625 : : } while (0)
626 : :
627 : : /*
628 : : * openLogFile is -1 or a kernel FD for an open log file segment.
629 : : * openLogSegNo identifies the segment, and openLogTLI the corresponding TLI.
630 : : * These variables are only used to write the XLOG, and so will normally refer
631 : : * to the active segment.
632 : : *
633 : : * Note: call Reserve/ReleaseExternalFD to track consumption of this FD.
634 : : */
635 : : static int openLogFile = -1;
636 : : static XLogSegNo openLogSegNo = 0;
637 : : static TimeLineID openLogTLI = 0;
638 : :
639 : : /*
640 : : * Local copies of equivalent fields in the control file. When running
641 : : * crash recovery, LocalMinRecoveryPoint is set to InvalidXLogRecPtr as we
642 : : * expect to replay all the WAL available, and updateMinRecoveryPoint is
643 : : * switched to false to prevent any updates while replaying records.
644 : : * Those values are kept consistent as long as crash recovery runs.
645 : : */
646 : : static XLogRecPtr LocalMinRecoveryPoint;
647 : : static TimeLineID LocalMinRecoveryPointTLI;
648 : : static bool updateMinRecoveryPoint = true;
649 : :
650 : : /* For WALInsertLockAcquire/Release functions */
651 : : static int MyLockNo = 0;
652 : : static bool holdingAllLocks = false;
653 : :
654 : : #ifdef WAL_DEBUG
655 : : static MemoryContext walDebugCxt = NULL;
656 : : #endif
657 : :
658 : : static void CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI,
659 : : XLogRecPtr EndOfLog,
660 : : TimeLineID newTLI);
661 : : static void CheckRequiredParameterValues(void);
662 : : static void XLogReportParameters(void);
663 : : static int LocalSetXLogInsertAllowed(void);
664 : : static void CreateEndOfRecoveryRecord(void);
665 : : static XLogRecPtr CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn,
666 : : XLogRecPtr pagePtr,
667 : : TimeLineID newTLI);
668 : : static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
669 : : static void KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo);
670 : : static XLogRecPtr XLogGetReplicationSlotMinimumLSN(void);
671 : :
672 : : static void AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli,
673 : : bool opportunistic);
674 : : static void XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible);
675 : : static bool InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
676 : : bool find_free, XLogSegNo max_segno,
677 : : TimeLineID tli);
678 : : static void XLogFileClose(void);
679 : : static void PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli);
680 : : static void RemoveTempXlogFiles(void);
681 : : static void RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr,
682 : : XLogRecPtr endptr, TimeLineID insertTLI);
683 : : static void RemoveXlogFile(const struct dirent *segment_de,
684 : : XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo,
685 : : TimeLineID insertTLI);
686 : : static void UpdateLastRemovedPtr(char *filename);
687 : : static void ValidateXLOGDirectoryStructure(void);
688 : : static void CleanupBackupHistory(void);
689 : : static void UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force);
690 : : static bool PerformRecoveryXLogAction(void);
691 : : static void InitControlFile(uint64 sysidentifier, uint32 data_checksum_version);
692 : : static void WriteControlFile(void);
693 : : static void ReadControlFile(void);
694 : : static void UpdateControlFile(void);
695 : : static char *str_time(pg_time_t tnow, char *buf, size_t bufsize);
696 : :
697 : : static int get_sync_bit(int method);
698 : :
699 : : static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch,
700 : : XLogRecData *rdata,
701 : : XLogRecPtr StartPos, XLogRecPtr EndPos,
702 : : TimeLineID tli);
703 : : static void ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos,
704 : : XLogRecPtr *EndPos, XLogRecPtr *PrevPtr);
705 : : static bool ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos,
706 : : XLogRecPtr *PrevPtr);
707 : : static XLogRecPtr WaitXLogInsertionsToFinish(XLogRecPtr upto);
708 : : static char *GetXLogBuffer(XLogRecPtr ptr, TimeLineID tli);
709 : : static XLogRecPtr XLogBytePosToRecPtr(uint64 bytepos);
710 : : static XLogRecPtr XLogBytePosToEndRecPtr(uint64 bytepos);
711 : : static uint64 XLogRecPtrToBytePos(XLogRecPtr ptr);
712 : :
713 : : static void WALInsertLockAcquire(void);
714 : : static void WALInsertLockAcquireExclusive(void);
715 : : static void WALInsertLockRelease(void);
716 : : static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
717 : :
718 : : /*
719 : : * Insert an XLOG record represented by an already-constructed chain of data
720 : : * chunks. This is a low-level routine; to construct the WAL record header
721 : : * and data, use the higher-level routines in xloginsert.c.
722 : : *
723 : : * If 'fpw_lsn' is valid, it is the oldest LSN among the pages that this
724 : : * WAL record applies to, that were not included in the record as full page
725 : : * images. If fpw_lsn <= RedoRecPtr, the function does not perform the
726 : : * insertion and returns InvalidXLogRecPtr. The caller can then recalculate
727 : : * which pages need a full-page image, and retry. If fpw_lsn is invalid, the
728 : : * record is always inserted.
729 : : *
730 : : * 'flags' gives more in-depth control on the record being inserted. See
731 : : * XLogSetRecordFlags() for details.
732 : : *
733 : : * 'topxid_included' tells whether the top-transaction id is logged along with
734 : : * current subtransaction. See XLogRecordAssemble().
735 : : *
736 : : * The first XLogRecData in the chain must be for the record header, and its
737 : : * data must be MAXALIGNed. XLogInsertRecord fills in the xl_prev and
738 : : * xl_crc fields in the header, the rest of the header must already be filled
739 : : * by the caller.
740 : : *
741 : : * Returns XLOG pointer to end of record (beginning of next record).
742 : : * This can be used as LSN for data pages affected by the logged action.
743 : : * (LSN is the XLOG point up to which the XLOG must be flushed to disk
744 : : * before the data page can be written out. This implements the basic
745 : : * WAL rule "write the log before the data".)
746 : : */
747 : : XLogRecPtr
3231 andres@anarazel.de 748 :CBC 13898692 : XLogInsertRecord(XLogRecData *rdata,
749 : : XLogRecPtr fpw_lsn,
750 : : uint8 flags,
751 : : int num_fpi,
752 : : bool topxid_included)
753 : : {
8985 bruce@momjian.us 754 : 13898692 : XLogCtlInsert *Insert = &XLogCtl->Insert;
755 : : pg_crc32c rdata_crc;
756 : : bool inserted;
4008 heikki.linnakangas@i 757 : 13898692 : XLogRecord *rechdr = (XLogRecord *) rdata->data;
3279 tgl@sss.pgh.pa.us 758 : 13898692 : uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
739 rhaas@postgresql.org 759 : 13898692 : WalInsertClass class = WALINSERT_NORMAL;
760 : : XLogRecPtr StartPos;
761 : : XLogRecPtr EndPos;
2601 akapila@postgresql.o 762 : 13898692 : bool prevDoPageWrites = doPageWrites;
763 : : TimeLineID insertTLI;
764 : :
765 : : /* Does this record type require special handling? */
739 rhaas@postgresql.org 766 [ + + ]: 13898692 : if (unlikely(rechdr->xl_rmid == RM_XLOG_ID))
767 : : {
768 [ + + ]: 217049 : if (info == XLOG_SWITCH)
769 : 707 : class = WALINSERT_SPECIAL_SWITCH;
770 [ + + ]: 216342 : else if (info == XLOG_CHECKPOINT_REDO)
771 : 903 : class = WALINSERT_SPECIAL_CHECKPOINT;
772 : : }
773 : :
774 : : /* we assume that all of the record header is in the first chunk */
3994 heikki.linnakangas@i 775 [ - + ]: 13898692 : Assert(rdata->len >= SizeOfXLogRecord);
776 : :
777 : : /* cross-check on whether we should be here or not */
5967 tgl@sss.pgh.pa.us 778 [ - + ]: 13898692 : if (!XLogInsertAllowed())
5967 tgl@sss.pgh.pa.us 779 [ # # ]:UBC 0 : elog(ERROR, "cannot make new WAL entries during recovery");
780 : :
781 : : /*
782 : : * Given that we're not in recovery, InsertTimeLineID is set and can't
783 : : * change, so we can read it without a lock.
784 : : */
1447 rhaas@postgresql.org 785 :CBC 13898692 : insertTLI = XLogCtl->InsertTimeLineID;
786 : :
787 : : /*----------
788 : : *
789 : : * We have now done all the preparatory work we can without holding a
790 : : * lock or modifying shared state. From here on, inserting the new WAL
791 : : * record to the shared WAL buffer cache is a two-step process:
792 : : *
793 : : * 1. Reserve the right amount of space from the WAL. The current head of
794 : : * reserved space is kept in Insert->CurrBytePos, and is protected by
795 : : * insertpos_lck.
796 : : *
797 : : * 2. Copy the record to the reserved WAL space. This involves finding the
798 : : * correct WAL buffer containing the reserved space, and copying the
799 : : * record in place. This can be done concurrently in multiple processes.
800 : : *
801 : : * To keep track of which insertions are still in-progress, each concurrent
802 : : * inserter acquires an insertion lock. In addition to just indicating that
803 : : * an insertion is in progress, the lock tells others how far the inserter
804 : : * has progressed. There is a small fixed number of insertion locks,
805 : : * determined by NUM_XLOGINSERT_LOCKS. When an inserter crosses a page
806 : : * boundary, it updates the value stored in the lock to the how far it has
807 : : * inserted, to allow the previous buffer to be flushed.
808 : : *
809 : : * Holding onto an insertion lock also protects RedoRecPtr and
810 : : * fullPageWrites from changing until the insertion is finished.
811 : : *
812 : : * Step 2 can usually be done completely in parallel. If the required WAL
813 : : * page is not initialized yet, you have to grab WALBufMappingLock to
814 : : * initialize it, but the WAL writer tries to do that ahead of insertions
815 : : * to avoid that from happening in the critical path.
816 : : *
817 : : *----------
818 : : */
5038 heikki.linnakangas@i 819 : 13898692 : START_CRIT_SECTION();
820 : :
739 rhaas@postgresql.org 821 [ + + ]: 13898692 : if (likely(class == WALINSERT_NORMAL))
822 : : {
748 823 : 13897082 : WALInsertLockAcquire();
824 : :
825 : : /*
826 : : * Check to see if my copy of RedoRecPtr is out of date. If so, may
827 : : * have to go back and have the caller recompute everything. This can
828 : : * only happen just after a checkpoint, so it's better to be slow in
829 : : * this case and fast otherwise.
830 : : *
831 : : * Also check to see if fullPageWrites was just turned on or there's a
832 : : * running backup (which forces full-page writes); if we weren't
833 : : * already doing full-page writes then go back and recompute.
834 : : *
835 : : * If we aren't doing full-page writes then RedoRecPtr doesn't
836 : : * actually affect the contents of the XLOG record, so we'll update
837 : : * our local copy but not force a recomputation. (If doPageWrites was
838 : : * just turned off, we could recompute the record without full pages,
839 : : * but we choose not to bother.)
840 : : */
841 [ + + ]: 13897082 : if (RedoRecPtr != Insert->RedoRecPtr)
842 : : {
843 [ - + ]: 6183 : Assert(RedoRecPtr < Insert->RedoRecPtr);
844 : 6183 : RedoRecPtr = Insert->RedoRecPtr;
845 : : }
846 [ + + + + ]: 13897082 : doPageWrites = (Insert->fullPageWrites || Insert->runningBackups > 0);
847 : :
848 [ + + ]: 13897082 : if (doPageWrites &&
849 [ + + + + ]: 13654179 : (!prevDoPageWrites ||
850 [ + + ]: 12927156 : (fpw_lsn != InvalidXLogRecPtr && fpw_lsn <= RedoRecPtr)))
851 : : {
852 : : /*
853 : : * Oops, some buffer now needs to be backed up that the caller
854 : : * didn't back up. Start over.
855 : : */
856 : 6913 : WALInsertLockRelease();
857 [ - + ]: 6913 : END_CRIT_SECTION();
858 : 6913 : return InvalidXLogRecPtr;
859 : : }
860 : :
861 : : /*
862 : : * Reserve space for the record in the WAL. This also sets the xl_prev
863 : : * pointer.
864 : : */
4008 heikki.linnakangas@i 865 : 13890169 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
866 : : &rechdr->xl_prev);
867 : :
868 : : /* Normal records are always inserted. */
4494 869 : 13890169 : inserted = true;
870 : : }
739 rhaas@postgresql.org 871 [ + + ]: 1610 : else if (class == WALINSERT_SPECIAL_SWITCH)
872 : : {
873 : : /*
874 : : * In order to insert an XLOG_SWITCH record, we need to hold all of
875 : : * the WAL insertion locks, not just one, so that no one else can
876 : : * begin inserting a record until we've figured out how much space
877 : : * remains in the current WAL segment and claimed all of it.
878 : : *
879 : : * Nonetheless, this case is simpler than the normal cases handled
880 : : * below, which must check for changes in doPageWrites and RedoRecPtr.
881 : : * Those checks are only needed for records that can contain buffer
882 : : * references, and an XLOG_SWITCH record never does.
883 : : */
748 884 [ - + ]: 707 : Assert(fpw_lsn == InvalidXLogRecPtr);
885 : 707 : WALInsertLockAcquireExclusive();
886 : 707 : inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
887 : : }
888 : : else
889 : : {
739 890 [ - + ]: 903 : Assert(class == WALINSERT_SPECIAL_CHECKPOINT);
891 : :
892 : : /*
893 : : * We need to update both the local and shared copies of RedoRecPtr,
894 : : * which means that we need to hold all the WAL insertion locks.
895 : : * However, there can't be any buffer references, so as above, we need
896 : : * not check RedoRecPtr before inserting the record; we just need to
897 : : * update it afterwards.
898 : : */
899 [ - + ]: 903 : Assert(fpw_lsn == InvalidXLogRecPtr);
900 : 903 : WALInsertLockAcquireExclusive();
901 : 903 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
902 : : &rechdr->xl_prev);
903 : 903 : RedoRecPtr = Insert->RedoRecPtr = StartPos;
904 : 903 : inserted = true;
905 : : }
906 : :
4494 heikki.linnakangas@i 907 [ + + ]: 13891779 : if (inserted)
908 : : {
909 : : /*
910 : : * Now that xl_prev has been filled in, calculate CRC of the record
911 : : * header.
912 : : */
3994 913 : 13891714 : rdata_crc = rechdr->xl_crc;
914 : 13891714 : COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
4010 915 : 13891714 : FIN_CRC32C(rdata_crc);
4494 916 : 13891714 : rechdr->xl_crc = rdata_crc;
917 : :
918 : : /*
919 : : * All the record data, including the header, is now ready to be
920 : : * inserted. Copy the record in the space reserved.
921 : : */
739 rhaas@postgresql.org 922 : 13891714 : CopyXLogRecordToWAL(rechdr->xl_tot_len,
923 : : class == WALINSERT_SPECIAL_SWITCH, rdata,
924 : : StartPos, EndPos, insertTLI);
925 : :
926 : : /*
927 : : * Unless record is flagged as not important, update LSN of last
928 : : * important record in the current slot. When holding all locks, just
929 : : * update the first one.
930 : : */
3231 andres@anarazel.de 931 [ + + ]: 13891714 : if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
932 : : {
3085 bruce@momjian.us 933 [ + + ]: 13804827 : int lockno = holdingAllLocks ? 0 : MyLockNo;
934 : :
3231 andres@anarazel.de 935 : 13804827 : WALInsertLocks[lockno].l.lastImportantAt = StartPos;
936 : : }
937 : : }
938 : : else
939 : : {
940 : : /*
941 : : * This was an xlog-switch record, but the current insert location was
942 : : * already exactly at the beginning of a segment, so there was no need
943 : : * to do anything.
944 : : */
945 : : }
946 : :
947 : : /*
948 : : * Done! Let others know that we're finished.
949 : : */
4238 heikki.linnakangas@i 950 : 13891779 : WALInsertLockRelease();
951 : :
4494 952 [ - + ]: 13891779 : END_CRIT_SECTION();
953 : :
1455 akapila@postgresql.o 954 : 13891779 : MarkCurrentTransactionIdLoggedIfAny();
955 : :
956 : : /*
957 : : * Mark top transaction id is logged (if needed) so that we should not try
958 : : * to log it again with the next WAL record in the current subtransaction.
959 : : */
960 [ + + ]: 13891779 : if (topxid_included)
961 : 219 : MarkSubxactTopXidLogged();
962 : :
963 : : /*
964 : : * Update shared LogwrtRqst.Write, if we crossed page boundary.
965 : : */
4494 heikki.linnakangas@i 966 [ + + ]: 13891779 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
967 : : {
4053 andres@anarazel.de 968 [ + + ]: 1654635 : SpinLockAcquire(&XLogCtl->info_lck);
969 : : /* advance global request to include new block(s) */
970 [ + + ]: 1654635 : if (XLogCtl->LogwrtRqst.Write < EndPos)
971 : 1588834 : XLogCtl->LogwrtRqst.Write = EndPos;
972 : 1654635 : SpinLockRelease(&XLogCtl->info_lck);
570 alvherre@alvh.no-ip. 973 : 1654635 : RefreshXLogWriteResult(LogwrtResult);
974 : : }
975 : :
976 : : /*
977 : : * If this was an XLOG_SWITCH record, flush the record and the empty
978 : : * padding space that fills the rest of the segment, and perform
979 : : * end-of-segment actions (eg, notifying archiver).
980 : : */
739 rhaas@postgresql.org 981 [ + + ]: 13891779 : if (class == WALINSERT_SPECIAL_SWITCH)
982 : : {
983 : : TRACE_POSTGRESQL_WAL_SWITCH();
4494 heikki.linnakangas@i 984 : 707 : XLogFlush(EndPos);
985 : :
986 : : /*
987 : : * Even though we reserved the rest of the segment for us, which is
988 : : * reflected in EndPos, we return a pointer to just the end of the
989 : : * xlog-switch record.
990 : : */
991 [ + + ]: 707 : if (inserted)
992 : : {
993 : 642 : EndPos = StartPos + SizeOfXLogRecord;
994 [ - + ]: 642 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
995 : : {
2960 andres@anarazel.de 996 :UBC 0 : uint64 offset = XLogSegmentOffset(EndPos, wal_segment_size);
997 : :
998 [ # # ]: 0 : if (offset == EndPos % XLOG_BLCKSZ)
4494 heikki.linnakangas@i 999 : 0 : EndPos += SizeOfXLogLongPHD;
1000 : : else
1001 : 0 : EndPos += SizeOfXLogShortPHD;
1002 : : }
1003 : : }
1004 : : }
1005 : :
1006 : : #ifdef WAL_DEBUG
1007 : : if (XLOG_DEBUG)
1008 : : {
1009 : : static XLogReaderState *debug_reader = NULL;
1010 : : XLogRecord *record;
1011 : : DecodedXLogRecord *decoded;
1012 : : StringInfoData buf;
1013 : : StringInfoData recordBuf;
1014 : : char *errormsg = NULL;
1015 : : MemoryContext oldCxt;
1016 : :
1017 : : oldCxt = MemoryContextSwitchTo(walDebugCxt);
1018 : :
1019 : : initStringInfo(&buf);
1020 : : appendStringInfo(&buf, "INSERT @ %X/%08X: ", LSN_FORMAT_ARGS(EndPos));
1021 : :
1022 : : /*
1023 : : * We have to piece together the WAL record data from the XLogRecData
1024 : : * entries, so that we can pass it to the rm_desc function as one
1025 : : * contiguous chunk.
1026 : : */
1027 : : initStringInfo(&recordBuf);
1028 : : for (; rdata != NULL; rdata = rdata->next)
1029 : : appendBinaryStringInfo(&recordBuf, rdata->data, rdata->len);
1030 : :
1031 : : /* We also need temporary space to decode the record. */
1032 : : record = (XLogRecord *) recordBuf.data;
1033 : : decoded = (DecodedXLogRecord *)
1034 : : palloc(DecodeXLogRecordRequiredSpace(record->xl_tot_len));
1035 : :
1036 : : if (!debug_reader)
1037 : : debug_reader = XLogReaderAllocate(wal_segment_size, NULL,
1038 : : XL_ROUTINE(.page_read = NULL,
1039 : : .segment_open = NULL,
1040 : : .segment_close = NULL),
1041 : : NULL);
1042 : : if (!debug_reader)
1043 : : {
1044 : : appendStringInfoString(&buf, "error decoding record: out of memory while allocating a WAL reading processor");
1045 : : }
1046 : : else if (!DecodeXLogRecord(debug_reader,
1047 : : decoded,
1048 : : record,
1049 : : EndPos,
1050 : : &errormsg))
1051 : : {
1052 : : appendStringInfo(&buf, "error decoding record: %s",
1053 : : errormsg ? errormsg : "no error message");
1054 : : }
1055 : : else
1056 : : {
1057 : : appendStringInfoString(&buf, " - ");
1058 : :
1059 : : debug_reader->record = decoded;
1060 : : xlog_outdesc(&buf, debug_reader);
1061 : : debug_reader->record = NULL;
1062 : : }
1063 : : elog(LOG, "%s", buf.data);
1064 : :
1065 : : pfree(decoded);
1066 : : pfree(buf.data);
1067 : : pfree(recordBuf.data);
1068 : : MemoryContextSwitchTo(oldCxt);
1069 : : }
1070 : : #endif
1071 : :
1072 : : /*
1073 : : * Update our global variables
1074 : : */
4494 heikki.linnakangas@i 1075 :CBC 13891779 : ProcLastRecPtr = StartPos;
1076 : 13891779 : XactLastRecEnd = EndPos;
1077 : :
1078 : : /* Report WAL traffic to the instrumentation. */
2032 akapila@postgresql.o 1079 [ + + ]: 13891779 : if (inserted)
1080 : : {
1081 : 13891714 : pgWalUsage.wal_bytes += rechdr->xl_tot_len;
1082 : 13891714 : pgWalUsage.wal_records++;
2001 1083 : 13891714 : pgWalUsage.wal_fpi += num_fpi;
1084 : :
1085 : : /* Required for the flush of pending stats WAL data */
91 michael@paquier.xyz 1086 : 13891714 : pgstat_report_fixed = true;
1087 : : }
1088 : :
4494 heikki.linnakangas@i 1089 : 13891779 : return EndPos;
1090 : : }
1091 : :
1092 : : /*
1093 : : * Reserves the right amount of space for a record of given size from the WAL.
1094 : : * *StartPos is set to the beginning of the reserved section, *EndPos to
1095 : : * its end+1. *PrevPtr is set to the beginning of the previous record; it is
1096 : : * used to set the xl_prev of this record.
1097 : : *
1098 : : * This is the performance critical part of XLogInsert that must be serialized
1099 : : * across backends. The rest can happen mostly in parallel. Try to keep this
1100 : : * section as short as possible, insertpos_lck can be heavily contended on a
1101 : : * busy system.
1102 : : *
1103 : : * NB: The space calculation here must match the code in CopyXLogRecordToWAL,
1104 : : * where we actually copy the record to the reserved space.
1105 : : *
1106 : : * NB: Testing shows that XLogInsertRecord runs faster if this code is inlined;
1107 : : * however, because there are two call sites, the compiler is reluctant to
1108 : : * inline. We use pg_attribute_always_inline here to try to convince it.
1109 : : */
1110 : : static pg_attribute_always_inline void
1111 : 13891072 : ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos,
1112 : : XLogRecPtr *PrevPtr)
1113 : : {
4053 andres@anarazel.de 1114 : 13891072 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1115 : : uint64 startbytepos;
1116 : : uint64 endbytepos;
1117 : : uint64 prevbytepos;
1118 : :
4494 heikki.linnakangas@i 1119 : 13891072 : size = MAXALIGN(size);
1120 : :
1121 : : /* All (non xlog-switch) records should contain data. */
1122 [ - + ]: 13891072 : Assert(size > SizeOfXLogRecord);
1123 : :
1124 : : /*
1125 : : * The duration the spinlock needs to be held is minimized by minimizing
1126 : : * the calculations that have to be done while holding the lock. The
1127 : : * current tip of reserved WAL is kept in CurrBytePos, as a byte position
1128 : : * that only counts "usable" bytes in WAL, that is, it excludes all WAL
1129 : : * page headers. The mapping between "usable" byte positions and physical
1130 : : * positions (XLogRecPtrs) can be done outside the locked region, and
1131 : : * because the usable byte position doesn't include any headers, reserving
1132 : : * X bytes from WAL is almost as simple as "CurrBytePos += X".
1133 : : */
1134 [ + + ]: 13891072 : SpinLockAcquire(&Insert->insertpos_lck);
1135 : :
1136 : 13891072 : startbytepos = Insert->CurrBytePos;
1137 : 13891072 : endbytepos = startbytepos + size;
1138 : 13891072 : prevbytepos = Insert->PrevBytePos;
1139 : 13891072 : Insert->CurrBytePos = endbytepos;
1140 : 13891072 : Insert->PrevBytePos = startbytepos;
1141 : :
1142 : 13891072 : SpinLockRelease(&Insert->insertpos_lck);
1143 : :
1144 : 13891072 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1145 : 13891072 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1146 : 13891072 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1147 : :
1148 : : /*
1149 : : * Check that the conversions between "usable byte positions" and
1150 : : * XLogRecPtrs work consistently in both directions.
1151 : : */
1152 [ - + ]: 13891072 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1153 [ - + ]: 13891072 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1154 [ - + ]: 13891072 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1155 : 13891072 : }
1156 : :
1157 : : /*
1158 : : * Like ReserveXLogInsertLocation(), but for an xlog-switch record.
1159 : : *
1160 : : * A log-switch record is handled slightly differently. The rest of the
1161 : : * segment will be reserved for this insertion, as indicated by the returned
1162 : : * *EndPos value. However, if we are already at the beginning of the current
1163 : : * segment, *StartPos and *EndPos are set to the current location without
1164 : : * reserving any space, and the function returns false.
1165 : : */
1166 : : static bool
1167 : 707 : ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
1168 : : {
4053 andres@anarazel.de 1169 : 707 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1170 : : uint64 startbytepos;
1171 : : uint64 endbytepos;
1172 : : uint64 prevbytepos;
3994 heikki.linnakangas@i 1173 : 707 : uint32 size = MAXALIGN(SizeOfXLogRecord);
1174 : : XLogRecPtr ptr;
1175 : : uint32 segleft;
1176 : :
1177 : : /*
1178 : : * These calculations are a bit heavy-weight to be done while holding a
1179 : : * spinlock, but since we're holding all the WAL insertion locks, there
1180 : : * are no other inserters competing for it. GetXLogInsertRecPtr() does
1181 : : * compete for it, but that's not called very frequently.
1182 : : */
4494 1183 [ - + ]: 707 : SpinLockAcquire(&Insert->insertpos_lck);
1184 : :
1185 : 707 : startbytepos = Insert->CurrBytePos;
1186 : :
1187 : 707 : ptr = XLogBytePosToEndRecPtr(startbytepos);
2960 andres@anarazel.de 1188 [ + + ]: 707 : if (XLogSegmentOffset(ptr, wal_segment_size) == 0)
1189 : : {
4494 heikki.linnakangas@i 1190 : 65 : SpinLockRelease(&Insert->insertpos_lck);
1191 : 65 : *EndPos = *StartPos = ptr;
1192 : 65 : return false;
1193 : : }
1194 : :
1195 : 642 : endbytepos = startbytepos + size;
1196 : 642 : prevbytepos = Insert->PrevBytePos;
1197 : :
1198 : 642 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1199 : 642 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1200 : :
2960 andres@anarazel.de 1201 : 642 : segleft = wal_segment_size - XLogSegmentOffset(*EndPos, wal_segment_size);
1202 [ + - ]: 642 : if (segleft != wal_segment_size)
1203 : : {
1204 : : /* consume the rest of the segment */
4494 heikki.linnakangas@i 1205 : 642 : *EndPos += segleft;
1206 : 642 : endbytepos = XLogRecPtrToBytePos(*EndPos);
1207 : : }
1208 : 642 : Insert->CurrBytePos = endbytepos;
1209 : 642 : Insert->PrevBytePos = startbytepos;
1210 : :
1211 : 642 : SpinLockRelease(&Insert->insertpos_lck);
1212 : :
1213 : 642 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1214 : :
2960 andres@anarazel.de 1215 [ - + ]: 642 : Assert(XLogSegmentOffset(*EndPos, wal_segment_size) == 0);
4494 heikki.linnakangas@i 1216 [ - + ]: 642 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1217 [ - + ]: 642 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1218 [ - + ]: 642 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1219 : :
1220 : 642 : return true;
1221 : : }
1222 : :
1223 : : /*
1224 : : * Subroutine of XLogInsertRecord. Copies a WAL record to an already-reserved
1225 : : * area in the WAL.
1226 : : */
1227 : : static void
1228 : 13891714 : CopyXLogRecordToWAL(int write_len, bool isLogSwitch, XLogRecData *rdata,
1229 : : XLogRecPtr StartPos, XLogRecPtr EndPos, TimeLineID tli)
1230 : : {
1231 : : char *currpos;
1232 : : int freespace;
1233 : : int written;
1234 : : XLogRecPtr CurrPos;
1235 : : XLogPageHeader pagehdr;
1236 : :
1237 : : /*
1238 : : * Get a pointer to the right place in the right WAL buffer to start
1239 : : * inserting to.
1240 : : */
1241 : 13891714 : CurrPos = StartPos;
1452 rhaas@postgresql.org 1242 : 13891714 : currpos = GetXLogBuffer(CurrPos, tli);
4494 heikki.linnakangas@i 1243 [ + - ]: 13891714 : freespace = INSERT_FREESPACE(CurrPos);
1244 : :
1245 : : /*
1246 : : * there should be enough space for at least the first field (xl_tot_len)
1247 : : * on this page.
1248 : : */
1249 [ - + ]: 13891714 : Assert(freespace >= sizeof(uint32));
1250 : :
1251 : : /* Copy record data */
1252 : 13891714 : written = 0;
1253 [ + + ]: 66541535 : while (rdata != NULL)
1254 : : {
419 peter@eisentraut.org 1255 : 52649821 : const char *rdata_data = rdata->data;
4494 heikki.linnakangas@i 1256 : 52649821 : int rdata_len = rdata->len;
1257 : :
1258 [ + + ]: 54423963 : while (rdata_len > freespace)
1259 : : {
1260 : : /*
1261 : : * Write what fits on this page, and continue on the next page.
1262 : : */
1263 [ + + - + ]: 1774142 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || freespace == 0);
1264 : 1774142 : memcpy(currpos, rdata_data, freespace);
1265 : 1774142 : rdata_data += freespace;
1266 : 1774142 : rdata_len -= freespace;
1267 : 1774142 : written += freespace;
1268 : 1774142 : CurrPos += freespace;
1269 : :
1270 : : /*
1271 : : * Get pointer to beginning of next page, and set the xlp_rem_len
1272 : : * in the page header. Set XLP_FIRST_IS_CONTRECORD.
1273 : : *
1274 : : * It's safe to set the contrecord flag and xlp_rem_len without a
1275 : : * lock on the page. All the other flags were already set when the
1276 : : * page was initialized, in AdvanceXLInsertBuffer, and we're the
1277 : : * only backend that needs to set the contrecord flag.
1278 : : */
1452 rhaas@postgresql.org 1279 : 1774142 : currpos = GetXLogBuffer(CurrPos, tli);
4494 heikki.linnakangas@i 1280 : 1774142 : pagehdr = (XLogPageHeader) currpos;
1281 : 1774142 : pagehdr->xlp_rem_len = write_len - written;
1282 : 1774142 : pagehdr->xlp_info |= XLP_FIRST_IS_CONTRECORD;
1283 : :
1284 : : /* skip over the page header */
2960 andres@anarazel.de 1285 [ + + ]: 1774142 : if (XLogSegmentOffset(CurrPos, wal_segment_size) == 0)
1286 : : {
4494 heikki.linnakangas@i 1287 : 1195 : CurrPos += SizeOfXLogLongPHD;
1288 : 1195 : currpos += SizeOfXLogLongPHD;
1289 : : }
1290 : : else
1291 : : {
1292 : 1772947 : CurrPos += SizeOfXLogShortPHD;
1293 : 1772947 : currpos += SizeOfXLogShortPHD;
1294 : : }
1295 [ + - ]: 1774142 : freespace = INSERT_FREESPACE(CurrPos);
1296 : : }
1297 : :
1298 [ + + - + ]: 52649821 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || rdata_len == 0);
1299 : 52649821 : memcpy(currpos, rdata_data, rdata_len);
1300 : 52649821 : currpos += rdata_len;
1301 : 52649821 : CurrPos += rdata_len;
1302 : 52649821 : freespace -= rdata_len;
1303 : 52649821 : written += rdata_len;
1304 : :
1305 : 52649821 : rdata = rdata->next;
1306 : : }
1307 [ - + ]: 13891714 : Assert(written == write_len);
1308 : :
1309 : : /*
1310 : : * If this was an xlog-switch, it's not enough to write the switch record,
1311 : : * we also have to consume all the remaining space in the WAL segment. We
1312 : : * have already reserved that space, but we need to actually fill it.
1313 : : */
2960 andres@anarazel.de 1314 [ + + + - ]: 13891714 : if (isLogSwitch && XLogSegmentOffset(CurrPos, wal_segment_size) != 0)
1315 : : {
1316 : : /* An xlog-switch record doesn't contain any data besides the header */
4494 heikki.linnakangas@i 1317 [ - + ]: 642 : Assert(write_len == SizeOfXLogRecord);
1318 : :
1319 : : /* Assert that we did reserve the right amount of space */
2960 andres@anarazel.de 1320 [ - + ]: 642 : Assert(XLogSegmentOffset(EndPos, wal_segment_size) == 0);
1321 : :
1322 : : /* Use up all the remaining space on the current page */
4494 heikki.linnakangas@i 1323 : 642 : CurrPos += freespace;
1324 : :
1325 : : /*
1326 : : * Cause all remaining pages in the segment to be flushed, leaving the
1327 : : * XLog position where it should be, at the start of the next segment.
1328 : : * We do this one page at a time, to make sure we don't deadlock
1329 : : * against ourselves if wal_buffers < wal_segment_size.
1330 : : */
1331 [ + + ]: 532922 : while (CurrPos < EndPos)
1332 : : {
1333 : : /*
1334 : : * The minimal action to flush the page would be to call
1335 : : * WALInsertLockUpdateInsertingAt(CurrPos) followed by
1336 : : * AdvanceXLInsertBuffer(...). The page would be left initialized
1337 : : * mostly to zeros, except for the page header (always the short
1338 : : * variant, as this is never a segment's first page).
1339 : : *
1340 : : * The large vistas of zeros are good for compressibility, but the
1341 : : * headers interrupting them every XLOG_BLCKSZ (with values that
1342 : : * differ from page to page) are not. The effect varies with
1343 : : * compression tool, but bzip2 for instance compresses about an
1344 : : * order of magnitude worse if those headers are left in place.
1345 : : *
1346 : : * Rather than complicating AdvanceXLInsertBuffer itself (which is
1347 : : * called in heavily-loaded circumstances as well as this lightly-
1348 : : * loaded one) with variant behavior, we just use GetXLogBuffer
1349 : : * (which itself calls the two methods we need) to get the pointer
1350 : : * and zero most of the page. Then we just zero the page header.
1351 : : */
1452 rhaas@postgresql.org 1352 : 532280 : currpos = GetXLogBuffer(CurrPos, tli);
2768 tgl@sss.pgh.pa.us 1353 [ + - + - : 2129120 : MemSet(currpos, 0, SizeOfXLogShortPHD);
+ - + - +
+ ]
1354 : :
4494 heikki.linnakangas@i 1355 : 532280 : CurrPos += XLOG_BLCKSZ;
1356 : : }
1357 : : }
1358 : : else
1359 : : {
1360 : : /* Align the end position, so that the next record starts aligned */
3994 1361 : 13891072 : CurrPos = MAXALIGN64(CurrPos);
1362 : : }
1363 : :
4494 1364 [ - + ]: 13891714 : if (CurrPos != EndPos)
572 dgustafsson@postgres 1365 [ # # ]:UBC 0 : ereport(PANIC,
1366 : : errcode(ERRCODE_DATA_CORRUPTED),
1367 : : errmsg_internal("space reserved for WAL record does not match what was written"));
4494 heikki.linnakangas@i 1368 :CBC 13891714 : }
1369 : :
1370 : : /*
1371 : : * Acquire a WAL insertion lock, for inserting to WAL.
1372 : : */
1373 : : static void
4238 1374 : 13897093 : WALInsertLockAcquire(void)
1375 : : {
1376 : : bool immed;
1377 : :
1378 : : /*
1379 : : * It doesn't matter which of the WAL insertion locks we acquire, so try
1380 : : * the one we used last time. If the system isn't particularly busy, it's
1381 : : * a good bet that it's still available, and it's good to have some
1382 : : * affinity to a particular lock so that you don't unnecessarily bounce
1383 : : * cache lines between processes when there's no contention.
1384 : : *
1385 : : * If this is the first time through in this backend, pick a lock
1386 : : * (semi-)randomly. This allows the locks to be used evenly if you have a
1387 : : * lot of very short connections.
1388 : : */
1389 : : static int lockToTry = -1;
1390 : :
1391 [ + + ]: 13897093 : if (lockToTry == -1)
613 1392 : 7305 : lockToTry = MyProcNumber % NUM_XLOGINSERT_LOCKS;
4238 1393 : 13897093 : MyLockNo = lockToTry;
1394 : :
1395 : : /*
1396 : : * The insertingAt value is initially set to 0, as we don't know our
1397 : : * insert location yet.
1398 : : */
3741 andres@anarazel.de 1399 : 13897093 : immed = LWLockAcquire(&WALInsertLocks[MyLockNo].l.lock, LW_EXCLUSIVE);
4238 heikki.linnakangas@i 1400 [ + + ]: 13897093 : if (!immed)
1401 : : {
1402 : : /*
1403 : : * If we couldn't get the lock immediately, try another lock next
1404 : : * time. On a system with more insertion locks than concurrent
1405 : : * inserters, this causes all the inserters to eventually migrate to a
1406 : : * lock that no-one else is using. On a system with more inserters
1407 : : * than locks, it still helps to distribute the inserters evenly
1408 : : * across the locks.
1409 : : */
4044 1410 : 19056 : lockToTry = (lockToTry + 1) % NUM_XLOGINSERT_LOCKS;
1411 : : }
4494 1412 : 13897093 : }
1413 : :
1414 : : /*
1415 : : * Acquire all WAL insertion locks, to prevent other backends from inserting
1416 : : * to WAL.
1417 : : */
1418 : : static void
4238 1419 : 4295 : WALInsertLockAcquireExclusive(void)
1420 : : {
1421 : : int i;
1422 : :
1423 : : /*
1424 : : * When holding all the locks, all but the last lock's insertingAt
1425 : : * indicator is set to 0xFFFFFFFFFFFFFFFF, which is higher than any real
1426 : : * XLogRecPtr value, to make sure that no-one blocks waiting on those.
1427 : : */
4044 1428 [ + + ]: 34360 : for (i = 0; i < NUM_XLOGINSERT_LOCKS - 1; i++)
1429 : : {
3741 andres@anarazel.de 1430 : 30065 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1431 : 30065 : LWLockUpdateVar(&WALInsertLocks[i].l.lock,
1432 : 30065 : &WALInsertLocks[i].l.insertingAt,
1433 : : PG_UINT64_MAX);
1434 : : }
1435 : : /* Variable value reset to 0 at release */
1436 : 4295 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1437 : :
4238 heikki.linnakangas@i 1438 : 4295 : holdingAllLocks = true;
4494 1439 : 4295 : }
1440 : :
1441 : : /*
1442 : : * Release our insertion lock (or locks, if we're holding them all).
1443 : : *
1444 : : * NB: Reset all variables to 0, so they cause LWLockWaitForVar to block the
1445 : : * next time the lock is acquired.
1446 : : */
1447 : : static void
4238 1448 : 13901388 : WALInsertLockRelease(void)
1449 : : {
1450 [ + + ]: 13901388 : if (holdingAllLocks)
1451 : : {
1452 : : int i;
1453 : :
4044 1454 [ + + ]: 38655 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
3741 andres@anarazel.de 1455 : 34360 : LWLockReleaseClearVar(&WALInsertLocks[i].l.lock,
1456 : 34360 : &WALInsertLocks[i].l.insertingAt,
1457 : : 0);
1458 : :
4238 heikki.linnakangas@i 1459 : 4295 : holdingAllLocks = false;
1460 : : }
1461 : : else
1462 : : {
3741 andres@anarazel.de 1463 : 13897093 : LWLockReleaseClearVar(&WALInsertLocks[MyLockNo].l.lock,
1464 : 13897093 : &WALInsertLocks[MyLockNo].l.insertingAt,
1465 : : 0);
1466 : : }
4494 heikki.linnakangas@i 1467 : 13901388 : }
1468 : :
1469 : : /*
1470 : : * Update our insertingAt value, to let others know that we've finished
1471 : : * inserting up to that point.
1472 : : */
1473 : : static void
4238 1474 : 2239320 : WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt)
1475 : : {
1476 [ + + ]: 2239320 : if (holdingAllLocks)
1477 : : {
1478 : : /*
1479 : : * We use the last lock to mark our actual position, see comments in
1480 : : * WALInsertLockAcquireExclusive.
1481 : : */
4044 1482 : 532048 : LWLockUpdateVar(&WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.lock,
3050 tgl@sss.pgh.pa.us 1483 : 532048 : &WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.insertingAt,
1484 : : insertingAt);
1485 : : }
1486 : : else
4238 heikki.linnakangas@i 1487 : 1707272 : LWLockUpdateVar(&WALInsertLocks[MyLockNo].l.lock,
1488 : 1707272 : &WALInsertLocks[MyLockNo].l.insertingAt,
1489 : : insertingAt);
4494 1490 : 2239320 : }
1491 : :
1492 : : /*
1493 : : * Wait for any WAL insertions < upto to finish.
1494 : : *
1495 : : * Returns the location of the oldest insertion that is still in-progress.
1496 : : * Any WAL prior to that point has been fully copied into WAL buffers, and
1497 : : * can be flushed out to disk. Because this waits for any insertions older
1498 : : * than 'upto' to finish, the return value is always >= 'upto'.
1499 : : *
1500 : : * Note: When you are about to write out WAL, you must call this function
1501 : : * *before* acquiring WALWriteLock, to avoid deadlocks. This function might
1502 : : * need to wait for an insertion to finish (or at least advance to next
1503 : : * uninitialized page), and the inserter might need to evict an old WAL buffer
1504 : : * to make room for a new one, which in turn requires WALWriteLock.
1505 : : */
1506 : : static XLogRecPtr
1507 : 2126174 : WaitXLogInsertionsToFinish(XLogRecPtr upto)
1508 : : {
1509 : : uint64 bytepos;
1510 : : XLogRecPtr inserted;
1511 : : XLogRecPtr reservedUpto;
1512 : : XLogRecPtr finishedUpto;
4053 andres@anarazel.de 1513 : 2126174 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1514 : : int i;
1515 : :
4494 heikki.linnakangas@i 1516 [ - + ]: 2126174 : if (MyProc == NULL)
4494 heikki.linnakangas@i 1517 [ # # ]:UBC 0 : elog(PANIC, "cannot wait without a PGPROC structure");
1518 : :
1519 : : /*
1520 : : * Check if there's any work to do. Use a barrier to ensure we get the
1521 : : * freshest value.
1522 : : */
568 alvherre@alvh.no-ip. 1523 :CBC 2126174 : inserted = pg_atomic_read_membarrier_u64(&XLogCtl->logInsertResult);
1524 [ + + ]: 2126174 : if (upto <= inserted)
1525 : 1672739 : return inserted;
1526 : :
1527 : : /* Read the current insert position */
4494 heikki.linnakangas@i 1528 [ + + ]: 453435 : SpinLockAcquire(&Insert->insertpos_lck);
1529 : 453435 : bytepos = Insert->CurrBytePos;
1530 : 453435 : SpinLockRelease(&Insert->insertpos_lck);
1531 : 453435 : reservedUpto = XLogBytePosToEndRecPtr(bytepos);
1532 : :
1533 : : /*
1534 : : * No-one should request to flush a piece of WAL that hasn't even been
1535 : : * reserved yet. However, it can happen if there is a block with a bogus
1536 : : * LSN on disk, for example. XLogFlush checks for that situation and
1537 : : * complains, but only after the flush. Here we just assume that to mean
1538 : : * that all WAL that has been reserved needs to be finished. In this
1539 : : * corner-case, the return value can be smaller than 'upto' argument.
1540 : : */
1541 [ - + ]: 453435 : if (upto > reservedUpto)
1542 : : {
1788 peter@eisentraut.org 1543 [ # # ]:UBC 0 : ereport(LOG,
1544 : : errmsg("request to flush past end of generated WAL; request %X/%08X, current position %X/%08X",
1545 : : LSN_FORMAT_ARGS(upto), LSN_FORMAT_ARGS(reservedUpto)));
4494 heikki.linnakangas@i 1546 : 0 : upto = reservedUpto;
1547 : : }
1548 : :
1549 : : /*
1550 : : * Loop through all the locks, sleeping on any in-progress insert older
1551 : : * than 'upto'.
1552 : : *
1553 : : * finishedUpto is our return value, indicating the point upto which all
1554 : : * the WAL insertions have been finished. Initialize it to the head of
1555 : : * reserved WAL, and as we iterate through the insertion locks, back it
1556 : : * out for any insertion that's still in progress.
1557 : : */
4494 heikki.linnakangas@i 1558 :CBC 453435 : finishedUpto = reservedUpto;
4044 1559 [ + + ]: 4080915 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1560 : : {
4192 bruce@momjian.us 1561 : 3627480 : XLogRecPtr insertingat = InvalidXLogRecPtr;
1562 : :
1563 : : do
1564 : : {
1565 : : /*
1566 : : * See if this insertion is in progress. LWLockWaitForVar will
1567 : : * wait for the lock to be released, or for the 'value' to be set
1568 : : * by a LWLockUpdateVar call. When a lock is initially acquired,
1569 : : * its value is 0 (InvalidXLogRecPtr), which means that we don't
1570 : : * know where it's inserting yet. We will have to wait for it. If
1571 : : * it's a small insertion, the record will most likely fit on the
1572 : : * same page and the inserter will release the lock without ever
1573 : : * calling LWLockUpdateVar. But if it has to sleep, it will
1574 : : * advertise the insertion point with LWLockUpdateVar before
1575 : : * sleeping.
1576 : : *
1577 : : * In this loop we are only waiting for insertions that started
1578 : : * before WaitXLogInsertionsToFinish was called. The lack of
1579 : : * memory barriers in the loop means that we might see locks as
1580 : : * "unused" that have since become used. This is fine because
1581 : : * they only can be used for later insertions that we would not
1582 : : * want to wait on anyway. Not taking a lock to acquire the
1583 : : * current insertingAt value means that we might see older
1584 : : * insertingAt values. This is also fine, because if we read a
1585 : : * value too old, we will add ourselves to the wait queue, which
1586 : : * contains atomic operations.
1587 : : */
4238 heikki.linnakangas@i 1588 [ + + ]: 3752987 : if (LWLockWaitForVar(&WALInsertLocks[i].l.lock,
1589 : 3752987 : &WALInsertLocks[i].l.insertingAt,
1590 : : insertingat, &insertingat))
1591 : : {
1592 : : /* the lock was free, so no insertion in progress */
1593 : 2569190 : insertingat = InvalidXLogRecPtr;
1594 : 2569190 : break;
1595 : : }
1596 : :
1597 : : /*
1598 : : * This insertion is still in progress. Have to wait, unless the
1599 : : * inserter has proceeded past 'upto'.
1600 : : */
1601 [ + + ]: 1183797 : } while (insertingat < upto);
1602 : :
1603 [ + + + + ]: 3627480 : if (insertingat != InvalidXLogRecPtr && insertingat < finishedUpto)
1604 : 400400 : finishedUpto = insertingat;
1605 : : }
1606 : :
1607 : : /*
1608 : : * Advance the limit we know to have been inserted and return the freshest
1609 : : * value we know of, which might be beyond what we requested if somebody
1610 : : * is concurrently doing this with an 'upto' pointer ahead of us.
1611 : : */
568 alvherre@alvh.no-ip. 1612 : 453435 : finishedUpto = pg_atomic_monotonic_advance_u64(&XLogCtl->logInsertResult,
1613 : : finishedUpto);
1614 : :
4494 heikki.linnakangas@i 1615 : 453435 : return finishedUpto;
1616 : : }
1617 : :
1618 : : /*
1619 : : * Get a pointer to the right location in the WAL buffer containing the
1620 : : * given XLogRecPtr.
1621 : : *
1622 : : * If the page is not initialized yet, it is initialized. That might require
1623 : : * evicting an old dirty buffer from the buffer cache, which means I/O.
1624 : : *
1625 : : * The caller must ensure that the page containing the requested location
1626 : : * isn't evicted yet, and won't be evicted. The way to ensure that is to
1627 : : * hold onto a WAL insertion lock with the insertingAt position set to
1628 : : * something <= ptr. GetXLogBuffer() will update insertingAt if it needs
1629 : : * to evict an old page from the buffer. (This means that once you call
1630 : : * GetXLogBuffer() with a given 'ptr', you must not access anything before
1631 : : * that point anymore, and must not call GetXLogBuffer() with an older 'ptr'
1632 : : * later, because older buffers might be recycled already)
1633 : : */
1634 : : static char *
1452 rhaas@postgresql.org 1635 : 16198147 : GetXLogBuffer(XLogRecPtr ptr, TimeLineID tli)
1636 : : {
1637 : : int idx;
1638 : : XLogRecPtr endptr;
1639 : : static uint64 cachedPage = 0;
1640 : : static char *cachedPos = NULL;
1641 : : XLogRecPtr expectedEndPtr;
1642 : :
1643 : : /*
1644 : : * Fast path for the common case that we need to access again the same
1645 : : * page as last time.
1646 : : */
4494 heikki.linnakangas@i 1647 [ + + ]: 16198147 : if (ptr / XLOG_BLCKSZ == cachedPage)
1648 : : {
1649 [ - + ]: 13549244 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1650 [ - + ]: 13549244 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1651 : 13549244 : return cachedPos + ptr % XLOG_BLCKSZ;
1652 : : }
1653 : :
1654 : : /*
1655 : : * The XLog buffer cache is organized so that a page is always loaded to a
1656 : : * particular buffer. That way we can easily calculate the buffer a given
1657 : : * page must be loaded into, from the XLogRecPtr alone.
1658 : : */
1659 : 2648903 : idx = XLogRecPtrToBufIdx(ptr);
1660 : :
1661 : : /*
1662 : : * See what page is loaded in the buffer at the moment. It could be the
1663 : : * page we're looking for, or something older. It can't be anything newer
1664 : : * - that would imply the page we're looking for has already been written
1665 : : * out to disk and evicted, and the caller is responsible for making sure
1666 : : * that doesn't happen.
1667 : : *
1668 : : * We don't hold a lock while we read the value. If someone is just about
1669 : : * to initialize or has just initialized the page, it's possible that we
1670 : : * get InvalidXLogRecPtr. That's ok, we'll grab the mapping lock (in
1671 : : * AdvanceXLInsertBuffer) and retry if we see anything other than the page
1672 : : * we're looking for.
1673 : : */
1674 : 2648903 : expectedEndPtr = ptr;
1675 : 2648903 : expectedEndPtr += XLOG_BLCKSZ - ptr % XLOG_BLCKSZ;
1676 : :
678 jdavis@postgresql.or 1677 : 2648903 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
4494 heikki.linnakangas@i 1678 [ + + ]: 2648903 : if (expectedEndPtr != endptr)
1679 : : {
1680 : : XLogRecPtr initializedUpto;
1681 : :
1682 : : /*
1683 : : * Before calling AdvanceXLInsertBuffer(), which can block, let others
1684 : : * know how far we're finished with inserting the record.
1685 : : *
1686 : : * NB: If 'ptr' points to just after the page header, advertise a
1687 : : * position at the beginning of the page rather than 'ptr' itself. If
1688 : : * there are no other insertions running, someone might try to flush
1689 : : * up to our advertised location. If we advertised a position after
1690 : : * the page header, someone might try to flush the page header, even
1691 : : * though page might actually not be initialized yet. As the first
1692 : : * inserter on the page, we are effectively responsible for making
1693 : : * sure that it's initialized, before we let insertingAt to move past
1694 : : * the page header.
1695 : : */
3739 1696 [ + + ]: 2239320 : if (ptr % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
2960 andres@anarazel.de 1697 [ + - ]: 5331 : XLogSegmentOffset(ptr, wal_segment_size) > XLOG_BLCKSZ)
3739 heikki.linnakangas@i 1698 : 5331 : initializedUpto = ptr - SizeOfXLogShortPHD;
1699 [ + + ]: 2233989 : else if (ptr % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
2960 andres@anarazel.de 1700 [ + + ]: 761 : XLogSegmentOffset(ptr, wal_segment_size) < XLOG_BLCKSZ)
3739 heikki.linnakangas@i 1701 : 517 : initializedUpto = ptr - SizeOfXLogLongPHD;
1702 : : else
1703 : 2233472 : initializedUpto = ptr;
1704 : :
1705 : 2239320 : WALInsertLockUpdateInsertingAt(initializedUpto);
1706 : :
1452 rhaas@postgresql.org 1707 : 2239320 : AdvanceXLInsertBuffer(ptr, tli, false);
678 jdavis@postgresql.or 1708 : 2239320 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1709 : :
4494 heikki.linnakangas@i 1710 [ - + ]: 2239320 : if (expectedEndPtr != endptr)
112 alvherre@kurilemu.de 1711 [ # # ]:UNC 0 : elog(PANIC, "could not find WAL buffer for %X/%08X",
1712 : : LSN_FORMAT_ARGS(ptr));
1713 : : }
1714 : : else
1715 : : {
1716 : : /*
1717 : : * Make sure the initialization of the page is visible to us, and
1718 : : * won't arrive later to overwrite the WAL data we write on the page.
1719 : : */
4494 heikki.linnakangas@i 1720 :CBC 409583 : pg_memory_barrier();
1721 : : }
1722 : :
1723 : : /*
1724 : : * Found the buffer holding this page. Return a pointer to the right
1725 : : * offset within the page.
1726 : : */
1727 : 2648903 : cachedPage = ptr / XLOG_BLCKSZ;
1728 : 2648903 : cachedPos = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1729 : :
1730 [ - + ]: 2648903 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1731 [ - + ]: 2648903 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1732 : :
1733 : 2648903 : return cachedPos + ptr % XLOG_BLCKSZ;
1734 : : }
1735 : :
1736 : : /*
1737 : : * Read WAL data directly from WAL buffers, if available. Returns the number
1738 : : * of bytes read successfully.
1739 : : *
1740 : : * Fewer than 'count' bytes may be read if some of the requested WAL data has
1741 : : * already been evicted.
1742 : : *
1743 : : * No locks are taken.
1744 : : *
1745 : : * Caller should ensure that it reads no further than LogwrtResult.Write
1746 : : * (which should have been updated by the caller when determining how far to
1747 : : * read). The 'tli' argument is only used as a convenient safety check so that
1748 : : * callers do not read from WAL buffers on a historical timeline.
1749 : : */
1750 : : Size
623 jdavis@postgresql.or 1751 : 100440 : WALReadFromBuffers(char *dstbuf, XLogRecPtr startptr, Size count,
1752 : : TimeLineID tli)
1753 : : {
1754 : 100440 : char *pdst = dstbuf;
1755 : 100440 : XLogRecPtr recptr = startptr;
1756 : : XLogRecPtr inserted;
619 1757 : 100440 : Size nbytes = count;
1758 : :
623 1759 [ + + + + ]: 100440 : if (RecoveryInProgress() || tli != GetWALInsertionTimeLine())
1760 : 847 : return 0;
1761 : :
1762 [ - + ]: 99593 : Assert(!XLogRecPtrIsInvalid(startptr));
1763 : :
1764 : : /*
1765 : : * Caller should ensure that the requested data has been inserted into WAL
1766 : : * buffers before we try to read it.
1767 : : */
568 alvherre@alvh.no-ip. 1768 : 99593 : inserted = pg_atomic_read_u64(&XLogCtl->logInsertResult);
1769 [ - + ]: 99593 : if (startptr + count > inserted)
568 alvherre@alvh.no-ip. 1770 [ # # ]:UBC 0 : ereport(ERROR,
1771 : : errmsg("cannot read past end of generated WAL: requested %X/%08X, current position %X/%08X",
1772 : : LSN_FORMAT_ARGS(startptr + count),
1773 : : LSN_FORMAT_ARGS(inserted)));
1774 : :
1775 : : /*
1776 : : * Loop through the buffers without a lock. For each buffer, atomically
1777 : : * read and verify the end pointer, then copy the data out, and finally
1778 : : * re-read and re-verify the end pointer.
1779 : : *
1780 : : * Once a page is evicted, it never returns to the WAL buffers, so if the
1781 : : * end pointer matches the expected end pointer before and after we copy
1782 : : * the data, then the right page must have been present during the data
1783 : : * copy. Read barriers are necessary to ensure that the data copy actually
1784 : : * happens between the two verification steps.
1785 : : *
1786 : : * If either verification fails, we simply terminate the loop and return
1787 : : * with the data that had been already copied out successfully.
1788 : : */
623 jdavis@postgresql.or 1789 [ + + ]:CBC 124687 : while (nbytes > 0)
1790 : : {
1791 : 117490 : uint32 offset = recptr % XLOG_BLCKSZ;
1792 : 117490 : int idx = XLogRecPtrToBufIdx(recptr);
1793 : : XLogRecPtr expectedEndPtr;
1794 : : XLogRecPtr endptr;
1795 : : const char *page;
1796 : : const char *psrc;
1797 : : Size npagebytes;
1798 : :
1799 : : /*
1800 : : * Calculate the end pointer we expect in the xlblocks array if the
1801 : : * correct page is present.
1802 : : */
1803 : 117490 : expectedEndPtr = recptr + (XLOG_BLCKSZ - offset);
1804 : :
1805 : : /*
1806 : : * First verification step: check that the correct page is present in
1807 : : * the WAL buffers.
1808 : : */
1809 : 117490 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1810 [ + + ]: 117490 : if (expectedEndPtr != endptr)
1811 : 92394 : break;
1812 : :
1813 : : /*
1814 : : * The correct page is present (or was at the time the endptr was
1815 : : * read; must re-verify later). Calculate pointer to source data and
1816 : : * determine how much data to read from this page.
1817 : : */
1818 : 25096 : page = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1819 : 25096 : psrc = page + offset;
1820 : 25096 : npagebytes = Min(nbytes, XLOG_BLCKSZ - offset);
1821 : :
1822 : : /*
1823 : : * Ensure that the data copy and the first verification step are not
1824 : : * reordered.
1825 : : */
1826 : 25096 : pg_read_barrier();
1827 : :
1828 : : /* data copy */
1829 : 25096 : memcpy(pdst, psrc, npagebytes);
1830 : :
1831 : : /*
1832 : : * Ensure that the data copy and the second verification step are not
1833 : : * reordered.
1834 : : */
1835 : 25096 : pg_read_barrier();
1836 : :
1837 : : /*
1838 : : * Second verification step: check that the page we read from wasn't
1839 : : * evicted while we were copying the data.
1840 : : */
1841 : 25096 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1842 [ + + ]: 25096 : if (expectedEndPtr != endptr)
1843 : 2 : break;
1844 : :
1845 : 25094 : pdst += npagebytes;
1846 : 25094 : recptr += npagebytes;
1847 : 25094 : nbytes -= npagebytes;
1848 : : }
1849 : :
1850 [ - + ]: 99593 : Assert(pdst - dstbuf <= count);
1851 : :
1852 : 99593 : return pdst - dstbuf;
1853 : : }
1854 : :
1855 : : /*
1856 : : * Converts a "usable byte position" to XLogRecPtr. A usable byte position
1857 : : * is the position starting from the beginning of WAL, excluding all WAL
1858 : : * page headers.
1859 : : */
1860 : : static XLogRecPtr
4494 heikki.linnakangas@i 1861 : 27785958 : XLogBytePosToRecPtr(uint64 bytepos)
1862 : : {
1863 : : uint64 fullsegs;
1864 : : uint64 fullpages;
1865 : : uint64 bytesleft;
1866 : : uint32 seg_offset;
1867 : : XLogRecPtr result;
1868 : :
1869 : 27785958 : fullsegs = bytepos / UsableBytesInSegment;
1870 : 27785958 : bytesleft = bytepos % UsableBytesInSegment;
1871 : :
1872 [ + + ]: 27785958 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1873 : : {
1874 : : /* fits on first page of segment */
1875 : 49991 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1876 : : }
1877 : : else
1878 : : {
1879 : : /* account for the first page on segment with long header */
1880 : 27735967 : seg_offset = XLOG_BLCKSZ;
1881 : 27735967 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1882 : :
1883 : 27735967 : fullpages = bytesleft / UsableBytesInPage;
1884 : 27735967 : bytesleft = bytesleft % UsableBytesInPage;
1885 : :
1886 : 27735967 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1887 : : }
1888 : :
2667 alvherre@alvh.no-ip. 1889 : 27785958 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1890 : :
4494 heikki.linnakangas@i 1891 : 27785958 : return result;
1892 : : }
1893 : :
1894 : : /*
1895 : : * Like XLogBytePosToRecPtr, but if the position is at a page boundary,
1896 : : * returns a pointer to the beginning of the page (ie. before page header),
1897 : : * not to where the first xlog record on that page would go to. This is used
1898 : : * when converting a pointer to the end of a record.
1899 : : */
1900 : : static XLogRecPtr
1901 : 14345856 : XLogBytePosToEndRecPtr(uint64 bytepos)
1902 : : {
1903 : : uint64 fullsegs;
1904 : : uint64 fullpages;
1905 : : uint64 bytesleft;
1906 : : uint32 seg_offset;
1907 : : XLogRecPtr result;
1908 : :
1909 : 14345856 : fullsegs = bytepos / UsableBytesInSegment;
1910 : 14345856 : bytesleft = bytepos % UsableBytesInSegment;
1911 : :
1912 [ + + ]: 14345856 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1913 : : {
1914 : : /* fits on first page of segment */
1915 [ + + ]: 80824 : if (bytesleft == 0)
1916 : 54576 : seg_offset = 0;
1917 : : else
1918 : 26248 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1919 : : }
1920 : : else
1921 : : {
1922 : : /* account for the first page on segment with long header */
1923 : 14265032 : seg_offset = XLOG_BLCKSZ;
1924 : 14265032 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1925 : :
1926 : 14265032 : fullpages = bytesleft / UsableBytesInPage;
1927 : 14265032 : bytesleft = bytesleft % UsableBytesInPage;
1928 : :
1929 [ + + ]: 14265032 : if (bytesleft == 0)
1930 : 13961 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft;
1931 : : else
1932 : 14251071 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1933 : : }
1934 : :
2667 alvherre@alvh.no-ip. 1935 : 14345856 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1936 : :
4494 heikki.linnakangas@i 1937 : 14345856 : return result;
1938 : : }
1939 : :
1940 : : /*
1941 : : * Convert an XLogRecPtr to a "usable byte position".
1942 : : */
1943 : : static uint64
1944 : 41677488 : XLogRecPtrToBytePos(XLogRecPtr ptr)
1945 : : {
1946 : : uint64 fullsegs;
1947 : : uint32 fullpages;
1948 : : uint32 offset;
1949 : : uint64 result;
1950 : :
2960 andres@anarazel.de 1951 : 41677488 : XLByteToSeg(ptr, fullsegs, wal_segment_size);
1952 : :
1953 : 41677488 : fullpages = (XLogSegmentOffset(ptr, wal_segment_size)) / XLOG_BLCKSZ;
4494 heikki.linnakangas@i 1954 : 41677488 : offset = ptr % XLOG_BLCKSZ;
1955 : :
1956 [ + + ]: 41677488 : if (fullpages == 0)
1957 : : {
1958 : 75699 : result = fullsegs * UsableBytesInSegment;
1959 [ + + ]: 75699 : if (offset > 0)
1960 : : {
1961 [ - + ]: 74384 : Assert(offset >= SizeOfXLogLongPHD);
1962 : 74384 : result += offset - SizeOfXLogLongPHD;
1963 : : }
1964 : : }
1965 : : else
1966 : : {
1967 : 41601789 : result = fullsegs * UsableBytesInSegment +
4192 bruce@momjian.us 1968 : 41601789 : (XLOG_BLCKSZ - SizeOfXLogLongPHD) + /* account for first page */
3050 tgl@sss.pgh.pa.us 1969 : 41601789 : (fullpages - 1) * UsableBytesInPage; /* full pages */
4494 heikki.linnakangas@i 1970 [ + + ]: 41601789 : if (offset > 0)
1971 : : {
1972 [ - + ]: 41588095 : Assert(offset >= SizeOfXLogShortPHD);
1973 : 41588095 : result += offset - SizeOfXLogShortPHD;
1974 : : }
1975 : : }
1976 : :
1977 : 41677488 : return result;
1978 : : }
1979 : :
1980 : : /*
1981 : : * Initialize XLOG buffers, writing out old buffers if they still contain
1982 : : * unwritten data, upto the page containing 'upto'. Or if 'opportunistic' is
1983 : : * true, initialize as many pages as we can without having to write out
1984 : : * unwritten data. Any new pages are initialized to zeros, with pages headers
1985 : : * initialized properly.
1986 : : */
1987 : : static void
1452 rhaas@postgresql.org 1988 : 2242561 : AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
1989 : : {
8994 tgl@sss.pgh.pa.us 1990 : 2242561 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1991 : : int nextidx;
1992 : : XLogRecPtr OldPageRqstPtr;
1993 : : XLogwrtRqst WriteRqst;
4494 heikki.linnakangas@i 1994 : 2242561 : XLogRecPtr NewPageEndPtr = InvalidXLogRecPtr;
1995 : : XLogRecPtr NewPageBeginPtr;
1996 : : XLogPageHeader NewPage;
1133 tgl@sss.pgh.pa.us 1997 : 2242561 : int npages pg_attribute_unused() = 0;
1998 : :
66 akorotkov@postgresql 1999 : 2242561 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2000 : :
2001 : : /*
2002 : : * Now that we have the lock, check if someone initialized the page
2003 : : * already.
2004 : : */
2005 [ + + + + ]: 6603576 : while (upto >= XLogCtl->InitializedUpTo || opportunistic)
2006 : : {
2007 : 4364256 : nextidx = XLogRecPtrToBufIdx(XLogCtl->InitializedUpTo);
2008 : :
2009 : : /*
2010 : : * Get ending-offset of the buffer page we need to replace (this may
2011 : : * be zero if the buffer hasn't been used yet). Fall through if it's
2012 : : * already written out.
2013 : : */
2014 : 4364256 : OldPageRqstPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[nextidx]);
2015 [ + + ]: 4364256 : if (LogwrtResult.Write < OldPageRqstPtr)
2016 : : {
2017 : : /*
2018 : : * Nope, got work to do. If we just want to pre-initialize as much
2019 : : * as we can without flushing, give up now.
2020 : : */
2021 [ + + ]: 2010674 : if (opportunistic)
2022 : 3241 : break;
2023 : :
2024 : : /* Advance shared memory write request position */
4053 andres@anarazel.de 2025 [ + + ]: 2007433 : SpinLockAcquire(&XLogCtl->info_lck);
2026 [ + + ]: 2007433 : if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
2027 : 504172 : XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
2028 : 2007433 : SpinLockRelease(&XLogCtl->info_lck);
2029 : :
2030 : : /*
2031 : : * Acquire an up-to-date LogwrtResult value and see if we still
2032 : : * need to write it or if someone else already did.
2033 : : */
570 alvherre@alvh.no-ip. 2034 : 2007433 : RefreshXLogWriteResult(LogwrtResult);
4494 heikki.linnakangas@i 2035 [ + + ]: 2007433 : if (LogwrtResult.Write < OldPageRqstPtr)
2036 : : {
2037 : : /*
2038 : : * Must acquire write lock. Release WALBufMappingLock first,
2039 : : * to make sure that all insertions that we need to wait for
2040 : : * can finish (up to this same position). Otherwise we risk
2041 : : * deadlock.
2042 : : */
66 akorotkov@postgresql 2043 : 1993133 : LWLockRelease(WALBufMappingLock);
2044 : :
4494 heikki.linnakangas@i 2045 : 1993133 : WaitXLogInsertionsToFinish(OldPageRqstPtr);
2046 : :
2047 : 1993133 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
2048 : :
572 alvherre@alvh.no-ip. 2049 : 1993133 : RefreshXLogWriteResult(LogwrtResult);
4494 heikki.linnakangas@i 2050 [ + + ]: 1993133 : if (LogwrtResult.Write >= OldPageRqstPtr)
2051 : : {
2052 : : /* OK, someone wrote it already */
2053 : 122032 : LWLockRelease(WALWriteLock);
2054 : : }
2055 : : else
2056 : : {
2057 : : /* Have to write it ourselves */
2058 : : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_START();
2059 : 1871101 : WriteRqst.Write = OldPageRqstPtr;
2060 : 1871101 : WriteRqst.Flush = 0;
1452 rhaas@postgresql.org 2061 : 1871101 : XLogWrite(WriteRqst, tli, false);
4494 heikki.linnakangas@i 2062 : 1871101 : LWLockRelease(WALWriteLock);
252 michael@paquier.xyz 2063 : 1871101 : pgWalUsage.wal_buffers_full++;
2064 : : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_DONE();
2065 : :
2066 : : /*
2067 : : * Required for the flush of pending stats WAL data, per
2068 : : * update of pgWalUsage.
2069 : : */
91 2070 : 1871101 : pgstat_report_fixed = true;
2071 : : }
2072 : : /* Re-acquire WALBufMappingLock and retry */
66 akorotkov@postgresql 2073 : 1993133 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2074 : 1993133 : continue;
2075 : : }
2076 : : }
2077 : :
2078 : : /*
2079 : : * Now the next buffer slot is free and we can set it up to be the
2080 : : * next output page.
2081 : : */
2082 : 2367882 : NewPageBeginPtr = XLogCtl->InitializedUpTo;
4494 heikki.linnakangas@i 2083 : 2367882 : NewPageEndPtr = NewPageBeginPtr + XLOG_BLCKSZ;
2084 : :
66 akorotkov@postgresql 2085 [ - + ]: 2367882 : Assert(XLogRecPtrToBufIdx(NewPageBeginPtr) == nextidx);
2086 : :
4494 heikki.linnakangas@i 2087 : 2367882 : NewPage = (XLogPageHeader) (XLogCtl->pages + nextidx * (Size) XLOG_BLCKSZ);
2088 : :
2089 : : /*
2090 : : * Mark the xlblock with InvalidXLogRecPtr and issue a write barrier
2091 : : * before initializing. Otherwise, the old page may be partially
2092 : : * zeroed but look valid.
2093 : : */
678 jdavis@postgresql.or 2094 : 2367882 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], InvalidXLogRecPtr);
2095 : 2367882 : pg_write_barrier();
2096 : :
2097 : : /*
2098 : : * Be sure to re-zero the buffer so that bytes beyond what we've
2099 : : * written will look like zeroes and not valid XLOG records...
2100 : : */
257 peter@eisentraut.org 2101 [ + - + - : 2367882 : MemSet(NewPage, 0, XLOG_BLCKSZ);
+ - - + -
- ]
2102 : :
2103 : : /*
2104 : : * Fill the new page's header
2105 : : */
3810 bruce@momjian.us 2106 : 2367882 : NewPage->xlp_magic = XLOG_PAGE_MAGIC;
2107 : :
2108 : : /* NewPage->xlp_info = 0; */ /* done by memset */
1452 rhaas@postgresql.org 2109 : 2367882 : NewPage->xlp_tli = tli;
3810 bruce@momjian.us 2110 : 2367882 : NewPage->xlp_pageaddr = NewPageBeginPtr;
2111 : :
2112 : : /* NewPage->xlp_rem_len = 0; */ /* done by memset */
2113 : :
2114 : : /*
2115 : : * If online backup is not in progress, mark the header to indicate
2116 : : * that WAL records beginning in this page have removable backup
2117 : : * blocks. This allows the WAL archiver to know whether it is safe to
2118 : : * compress archived WAL data by transforming full-block records into
2119 : : * the non-full-block format. It is sufficient to record this at the
2120 : : * page level because we force a page switch (in fact a segment
2121 : : * switch) when starting a backup, so the flag will be off before any
2122 : : * records can be written during the backup. At the end of a backup,
2123 : : * the last page will be marked as all unsafe when perhaps only part
2124 : : * is unsafe, but at worst the archiver would miss the opportunity to
2125 : : * compress a few records.
2126 : : */
1104 alvherre@alvh.no-ip. 2127 [ + + ]: 2367882 : if (Insert->runningBackups == 0)
3810 bruce@momjian.us 2128 : 2240688 : NewPage->xlp_info |= XLP_BKP_REMOVABLE;
2129 : :
2130 : : /*
2131 : : * If first page of an XLOG segment file, make it a long header.
2132 : : */
2960 andres@anarazel.de 2133 [ + + ]: 2367882 : if ((XLogSegmentOffset(NewPage->xlp_pageaddr, wal_segment_size)) == 0)
2134 : : {
4494 heikki.linnakangas@i 2135 : 1753 : XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
2136 : :
2137 : 1753 : NewLongPage->xlp_sysid = ControlFile->system_identifier;
2960 andres@anarazel.de 2138 : 1753 : NewLongPage->xlp_seg_size = wal_segment_size;
4494 heikki.linnakangas@i 2139 : 1753 : NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
3810 bruce@momjian.us 2140 : 1753 : NewPage->xlp_info |= XLP_LONG_HEADER;
2141 : : }
2142 : :
2143 : : /*
2144 : : * Make sure the initialization of the page becomes visible to others
2145 : : * before the xlblocks update. GetXLogBuffer() reads xlblocks without
2146 : : * holding a lock.
2147 : : */
4494 heikki.linnakangas@i 2148 : 2367882 : pg_write_barrier();
2149 : :
678 jdavis@postgresql.or 2150 : 2367882 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], NewPageEndPtr);
66 akorotkov@postgresql 2151 : 2367882 : XLogCtl->InitializedUpTo = NewPageEndPtr;
2152 : :
4494 heikki.linnakangas@i 2153 : 2367882 : npages++;
2154 : : }
66 akorotkov@postgresql 2155 : 2242561 : LWLockRelease(WALBufMappingLock);
2156 : :
2157 : : #ifdef WAL_DEBUG
2158 : : if (XLOG_DEBUG && npages > 0)
2159 : : {
2160 : : elog(DEBUG1, "initialized %d pages, up to %X/%08X",
2161 : : npages, LSN_FORMAT_ARGS(NewPageEndPtr));
2162 : : }
2163 : : #endif
9527 vadim4o@yahoo.com 2164 : 2242561 : }
2165 : :
2166 : : /*
2167 : : * Calculate CheckPointSegments based on max_wal_size_mb and
2168 : : * checkpoint_completion_target.
2169 : : */
2170 : : static void
3899 heikki.linnakangas@i 2171 : 7508 : CalculateCheckpointSegments(void)
2172 : : {
2173 : : double target;
2174 : :
2175 : : /*-------
2176 : : * Calculate the distance at which to trigger a checkpoint, to avoid
2177 : : * exceeding max_wal_size_mb. This is based on two assumptions:
2178 : : *
2179 : : * a) we keep WAL for only one checkpoint cycle (prior to PG11 we kept
2180 : : * WAL for two checkpoint cycles to allow us to recover from the
2181 : : * secondary checkpoint if the first checkpoint failed, though we
2182 : : * only did this on the primary anyway, not on standby. Keeping just
2183 : : * one checkpoint simplifies processing and reduces disk space in
2184 : : * many smaller databases.)
2185 : : * b) during checkpoint, we consume checkpoint_completion_target *
2186 : : * number of segments consumed between checkpoints.
2187 : : *-------
2188 : : */
2960 andres@anarazel.de 2189 : 7508 : target = (double) ConvertToXSegs(max_wal_size_mb, wal_segment_size) /
2911 simon@2ndQuadrant.co 2190 : 7508 : (1.0 + CheckPointCompletionTarget);
2191 : :
2192 : : /* round down */
3899 heikki.linnakangas@i 2193 : 7508 : CheckPointSegments = (int) target;
2194 : :
2195 [ + + ]: 7508 : if (CheckPointSegments < 1)
2196 : 10 : CheckPointSegments = 1;
2197 : 7508 : }
2198 : :
2199 : : void
2200 : 5464 : assign_max_wal_size(int newval, void *extra)
2201 : : {
3128 simon@2ndQuadrant.co 2202 : 5464 : max_wal_size_mb = newval;
3899 heikki.linnakangas@i 2203 : 5464 : CalculateCheckpointSegments();
2204 : 5464 : }
2205 : :
2206 : : void
2207 : 1087 : assign_checkpoint_completion_target(double newval, void *extra)
2208 : : {
2209 : 1087 : CheckPointCompletionTarget = newval;
2210 : 1087 : CalculateCheckpointSegments();
2211 : 1087 : }
2212 : :
2213 : : bool
791 peter@eisentraut.org 2214 : 2095 : check_wal_segment_size(int *newval, void **extra, GucSource source)
2215 : : {
2216 [ + - + - : 2095 : if (!IsValidWalSegSize(*newval))
+ - - + ]
2217 : : {
791 peter@eisentraut.org 2218 :UBC 0 : GUC_check_errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.");
2219 : 0 : return false;
2220 : : }
2221 : :
791 peter@eisentraut.org 2222 :CBC 2095 : return true;
2223 : : }
2224 : :
2225 : : /*
2226 : : * At a checkpoint, how many WAL segments to recycle as preallocated future
2227 : : * XLOG segments? Returns the highest segment that should be preallocated.
2228 : : */
2229 : : static XLogSegNo
2140 michael@paquier.xyz 2230 : 1701 : XLOGfileslop(XLogRecPtr lastredoptr)
2231 : : {
2232 : : XLogSegNo minSegNo;
2233 : : XLogSegNo maxSegNo;
2234 : : double distance;
2235 : : XLogSegNo recycleSegNo;
2236 : :
2237 : : /*
2238 : : * Calculate the segment numbers that min_wal_size_mb and max_wal_size_mb
2239 : : * correspond to. Always recycle enough segments to meet the minimum, and
2240 : : * remove enough segments to stay below the maximum.
2241 : : */
2242 : 1701 : minSegNo = lastredoptr / wal_segment_size +
2960 andres@anarazel.de 2243 : 1701 : ConvertToXSegs(min_wal_size_mb, wal_segment_size) - 1;
2140 michael@paquier.xyz 2244 : 1701 : maxSegNo = lastredoptr / wal_segment_size +
2960 andres@anarazel.de 2245 : 1701 : ConvertToXSegs(max_wal_size_mb, wal_segment_size) - 1;
2246 : :
2247 : : /*
2248 : : * Between those limits, recycle enough segments to get us through to the
2249 : : * estimated end of next checkpoint.
2250 : : *
2251 : : * To estimate where the next checkpoint will finish, assume that the
2252 : : * system runs steadily consuming CheckPointDistanceEstimate bytes between
2253 : : * every checkpoint.
2254 : : */
2911 simon@2ndQuadrant.co 2255 : 1701 : distance = (1.0 + CheckPointCompletionTarget) * CheckPointDistanceEstimate;
2256 : : /* add 10% for good measure. */
3899 heikki.linnakangas@i 2257 : 1701 : distance *= 1.10;
2258 : :
2140 michael@paquier.xyz 2259 : 1701 : recycleSegNo = (XLogSegNo) ceil(((double) lastredoptr + distance) /
2260 : : wal_segment_size);
2261 : :
3899 heikki.linnakangas@i 2262 [ + + ]: 1701 : if (recycleSegNo < minSegNo)
2263 : 1192 : recycleSegNo = minSegNo;
2264 [ + + ]: 1701 : if (recycleSegNo > maxSegNo)
2265 : 391 : recycleSegNo = maxSegNo;
2266 : :
2267 : 1701 : return recycleSegNo;
2268 : : }
2269 : :
2270 : : /*
2271 : : * Check whether we've consumed enough xlog space that a checkpoint is needed.
2272 : : *
2273 : : * new_segno indicates a log file that has just been filled up (or read
2274 : : * during recovery). We measure the distance from RedoRecPtr to new_segno
2275 : : * and see if that exceeds CheckPointSegments.
2276 : : *
2277 : : * Note: it is caller's responsibility that RedoRecPtr is up-to-date.
2278 : : */
2279 : : bool
4873 2280 : 4622 : XLogCheckpointNeeded(XLogSegNo new_segno)
2281 : : {
2282 : : XLogSegNo old_segno;
2283 : :
2960 andres@anarazel.de 2284 : 4622 : XLByteToSeg(RedoRecPtr, old_segno, wal_segment_size);
2285 : :
4873 heikki.linnakangas@i 2286 [ + + ]: 4622 : if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
6590 tgl@sss.pgh.pa.us 2287 : 2907 : return true;
2288 : 1715 : return false;
2289 : : }
2290 : :
2291 : : /*
2292 : : * Write and/or fsync the log at least as far as WriteRqst indicates.
2293 : : *
2294 : : * If flexible == true, we don't have to write as far as WriteRqst, but
2295 : : * may stop at any convenient boundary (such as a cache or logfile boundary).
2296 : : * This option allows us to avoid uselessly issuing multiple writes when a
2297 : : * single one would do.
2298 : : *
2299 : : * Must be called with WALWriteLock held. WaitXLogInsertionsToFinish(WriteRqst)
2300 : : * must be called before grabbing the lock, to make sure the data is ready to
2301 : : * write.
2302 : : */
2303 : : static void
1452 rhaas@postgresql.org 2304 : 1998369 : XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
2305 : : {
2306 : : bool ispartialpage;
2307 : : bool last_iteration;
2308 : : bool finishing_seg;
2309 : : int curridx;
2310 : : int npages;
2311 : : int startidx;
2312 : : uint32 startoffset;
2313 : :
2314 : : /* We should always be inside a critical section here */
7500 tgl@sss.pgh.pa.us 2315 [ - + ]: 1998369 : Assert(CritSectionCount > 0);
2316 : :
2317 : : /*
2318 : : * Update local LogwrtResult (caller probably did this already, but...)
2319 : : */
572 alvherre@alvh.no-ip. 2320 : 1998369 : RefreshXLogWriteResult(LogwrtResult);
2321 : :
2322 : : /*
2323 : : * Since successive pages in the xlog cache are consecutively allocated,
2324 : : * we can usually gather multiple pages together and issue just one
2325 : : * write() call. npages is the number of pages we have determined can be
2326 : : * written together; startidx is the cache block index of the first one,
2327 : : * and startoffset is the file offset at which it should go. The latter
2328 : : * two variables are only valid when npages > 0, but we must initialize
2329 : : * all of them to keep the compiler quiet.
2330 : : */
7371 tgl@sss.pgh.pa.us 2331 : 1998369 : npages = 0;
2332 : 1998369 : startidx = 0;
2333 : 1998369 : startoffset = 0;
2334 : :
2335 : : /*
2336 : : * Within the loop, curridx is the cache block index of the page to
2337 : : * consider writing. Begin at the buffer containing the next unwritten
2338 : : * page, or last partially written page.
2339 : : */
4485 heikki.linnakangas@i 2340 : 1998369 : curridx = XLogRecPtrToBufIdx(LogwrtResult.Write);
2341 : :
4686 alvherre@alvh.no-ip. 2342 [ + + ]: 4310915 : while (LogwrtResult.Write < WriteRqst.Write)
2343 : : {
2344 : : /*
2345 : : * Make sure we're not ahead of the insert process. This could happen
2346 : : * if we're passed a bogus WriteRqst.Write that is past the end of the
2347 : : * last page that's been initialized by AdvanceXLInsertBuffer.
2348 : : */
678 jdavis@postgresql.or 2349 : 2434549 : XLogRecPtr EndPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[curridx]);
2350 : :
4494 heikki.linnakangas@i 2351 [ - + ]: 2434549 : if (LogwrtResult.Write >= EndPtr)
112 alvherre@kurilemu.de 2352 [ # # ]:UNC 0 : elog(PANIC, "xlog write request %X/%08X is past end of log %X/%08X",
2353 : : LSN_FORMAT_ARGS(LogwrtResult.Write),
2354 : : LSN_FORMAT_ARGS(EndPtr));
2355 : :
2356 : : /* Advance LogwrtResult.Write to end of current buffer page */
4494 heikki.linnakangas@i 2357 :CBC 2434549 : LogwrtResult.Write = EndPtr;
4686 alvherre@alvh.no-ip. 2358 : 2434549 : ispartialpage = WriteRqst.Write < LogwrtResult.Write;
2359 : :
2960 andres@anarazel.de 2360 [ + + ]: 2434549 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2361 : : wal_segment_size))
2362 : : {
2363 : : /*
2364 : : * Switch to new logfile segment. We cannot have any pending
2365 : : * pages here (since we dump what we have at segment end).
2366 : : */
7371 tgl@sss.pgh.pa.us 2367 [ - + ]: 12854 : Assert(npages == 0);
8994 2368 [ + + ]: 12854 : if (openLogFile >= 0)
7074 bruce@momjian.us 2369 : 6091 : XLogFileClose();
2960 andres@anarazel.de 2370 : 12854 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2371 : : wal_segment_size);
1452 rhaas@postgresql.org 2372 : 12854 : openLogTLI = tli;
2373 : :
2374 : : /* create/use new log file */
2375 : 12854 : openLogFile = XLogFileInit(openLogSegNo, tli);
2072 tgl@sss.pgh.pa.us 2376 : 12854 : ReserveExternalFD();
2377 : : }
2378 : :
2379 : : /* Make sure we have the current logfile open */
8994 2380 [ - + ]: 2434549 : if (openLogFile < 0)
2381 : : {
2960 andres@anarazel.de 2382 :UBC 0 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2383 : : wal_segment_size);
1452 rhaas@postgresql.org 2384 : 0 : openLogTLI = tli;
2385 : 0 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2072 tgl@sss.pgh.pa.us 2386 : 0 : ReserveExternalFD();
2387 : : }
2388 : :
2389 : : /* Add current page to the set of pending pages-to-dump */
7371 tgl@sss.pgh.pa.us 2390 [ + + ]:CBC 2434549 : if (npages == 0)
2391 : : {
2392 : : /* first of group */
2393 : 2010272 : startidx = curridx;
2960 andres@anarazel.de 2394 : 2010272 : startoffset = XLogSegmentOffset(LogwrtResult.Write - XLOG_BLCKSZ,
2395 : : wal_segment_size);
2396 : : }
7371 tgl@sss.pgh.pa.us 2397 : 2434549 : npages++;
2398 : :
2399 : : /*
2400 : : * Dump the set if this will be the last loop iteration, or if we are
2401 : : * at the last page of the cache area (since the next page won't be
2402 : : * contiguous in memory), or if we are at the end of the logfile
2403 : : * segment.
2404 : : */
4686 alvherre@alvh.no-ip. 2405 : 2434549 : last_iteration = WriteRqst.Write <= LogwrtResult.Write;
2406 : :
7371 tgl@sss.pgh.pa.us 2407 [ + + ]: 4749488 : finishing_seg = !ispartialpage &&
2960 andres@anarazel.de 2408 [ + + ]: 2314939 : (startoffset + npages * XLOG_BLCKSZ) >= wal_segment_size;
2409 : :
7022 tgl@sss.pgh.pa.us 2410 [ + + ]: 2434549 : if (last_iteration ||
7371 2411 [ + + - + ]: 438581 : curridx == XLogCtl->XLogCacheBlck ||
2412 : : finishing_seg)
2413 : : {
2414 : : char *from;
2415 : : Size nbytes;
2416 : : Size nleft;
2417 : : ssize_t written;
2418 : : instr_time start;
2419 : :
2420 : : /* OK to write the page(s) */
7147 2421 : 2010272 : from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
2422 : 2010272 : nbytes = npages * (Size) XLOG_BLCKSZ;
4501 heikki.linnakangas@i 2423 : 2010272 : nleft = nbytes;
2424 : : do
2425 : : {
2426 : 2010272 : errno = 0;
2427 : :
2428 : : /*
2429 : : * Measure I/O timing to write WAL data, for pg_stat_io.
2430 : : */
243 michael@paquier.xyz 2431 : 2010272 : start = pgstat_prepare_io_time(track_wal_io_timing);
2432 : :
3145 rhaas@postgresql.org 2433 : 2010272 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
1124 tmunro@postgresql.or 2434 : 2010272 : written = pg_pwrite(openLogFile, from, nleft, startoffset);
3145 rhaas@postgresql.org 2435 : 2010272 : pgstat_report_wait_end();
2436 : :
265 michael@paquier.xyz 2437 : 2010272 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL,
2438 : : IOOP_WRITE, start, 1, written);
2439 : :
4501 heikki.linnakangas@i 2440 [ - + ]: 2010272 : if (written <= 0)
2441 : : {
2442 : : char xlogfname[MAXFNAMELEN];
2443 : : int save_errno;
2444 : :
4501 heikki.linnakangas@i 2445 [ # # ]:UBC 0 : if (errno == EINTR)
2446 : 0 : continue;
2447 : :
2155 michael@paquier.xyz 2448 : 0 : save_errno = errno;
1452 rhaas@postgresql.org 2449 : 0 : XLogFileName(xlogfname, tli, openLogSegNo,
2450 : : wal_segment_size);
2155 michael@paquier.xyz 2451 : 0 : errno = save_errno;
4501 heikki.linnakangas@i 2452 [ # # ]: 0 : ereport(PANIC,
2453 : : (errcode_for_file_access(),
2454 : : errmsg("could not write to log file \"%s\" at offset %u, length %zu: %m",
2455 : : xlogfname, startoffset, nleft)));
2456 : : }
4501 heikki.linnakangas@i 2457 :CBC 2010272 : nleft -= written;
2458 : 2010272 : from += written;
2546 tmunro@postgresql.or 2459 : 2010272 : startoffset += written;
4501 heikki.linnakangas@i 2460 [ - + ]: 2010272 : } while (nleft > 0);
2461 : :
7371 tgl@sss.pgh.pa.us 2462 : 2010272 : npages = 0;
2463 : :
2464 : : /*
2465 : : * If we just wrote the whole last page of a logfile segment,
2466 : : * fsync the segment immediately. This avoids having to go back
2467 : : * and re-open prior segments when an fsync request comes along
2468 : : * later. Doing it here ensures that one and only one backend will
2469 : : * perform this fsync.
2470 : : *
2471 : : * This is also the right place to notify the Archiver that the
2472 : : * segment is ready to copy to archival storage, and to update the
2473 : : * timer for archive_timeout, and to signal for a checkpoint if
2474 : : * too many logfile segments have been used since the last
2475 : : * checkpoint.
2476 : : */
4494 heikki.linnakangas@i 2477 [ + + ]: 2010272 : if (finishing_seg)
2478 : : {
1452 rhaas@postgresql.org 2479 : 1848 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2480 : :
2481 : : /* signal that we need to wakeup walsenders later */
4865 2482 : 1848 : WalSndWakeupRequest();
2483 : :
3050 tgl@sss.pgh.pa.us 2484 : 1848 : LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
2485 : :
7371 2486 [ + + - + : 1848 : if (XLogArchivingActive())
+ + ]
1452 rhaas@postgresql.org 2487 : 401 : XLogArchiveNotifySeg(openLogSegNo, tli);
2488 : :
4485 heikki.linnakangas@i 2489 : 1848 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
3231 andres@anarazel.de 2490 : 1848 : XLogCtl->lastSegSwitchLSN = LogwrtResult.Flush;
2491 : :
2492 : : /*
2493 : : * Request a checkpoint if we've consumed too much xlog since
2494 : : * the last one. For speed, we first check using the local
2495 : : * copy of RedoRecPtr, which might be out of date; if it looks
2496 : : * like a checkpoint is needed, forcibly update RedoRecPtr and
2497 : : * recheck.
2498 : : */
4873 heikki.linnakangas@i 2499 [ + + + + ]: 1848 : if (IsUnderPostmaster && XLogCheckpointNeeded(openLogSegNo))
2500 : : {
6590 tgl@sss.pgh.pa.us 2501 : 241 : (void) GetRedoRecPtr();
4873 heikki.linnakangas@i 2502 [ + + ]: 241 : if (XLogCheckpointNeeded(openLogSegNo))
6694 tgl@sss.pgh.pa.us 2503 : 197 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
2504 : : }
2505 : : }
2506 : : }
2507 : :
8994 2508 [ + + ]: 2434549 : if (ispartialpage)
2509 : : {
2510 : : /* Only asked to write a partial page */
2511 : 119610 : LogwrtResult.Write = WriteRqst.Write;
2512 : 119610 : break;
2513 : : }
7371 2514 [ + + ]: 2314939 : curridx = NextBufIdx(curridx);
2515 : :
2516 : : /* If flexible, break out of loop as soon as we wrote something */
2517 [ + + + + ]: 2314939 : if (flexible && npages == 0)
2518 : 2393 : break;
2519 : : }
2520 : :
2521 [ - + ]: 1998369 : Assert(npages == 0);
2522 : :
2523 : : /*
2524 : : * If asked to flush, do so
2525 : : */
4686 alvherre@alvh.no-ip. 2526 [ + + ]: 1998369 : if (LogwrtResult.Flush < WriteRqst.Flush &&
2527 [ + + ]: 126609 : LogwrtResult.Flush < LogwrtResult.Write)
2528 : : {
2529 : : /*
2530 : : * Could get here without iterating above loop, in which case we might
2531 : : * have no open file or the wrong one. However, we do not need to
2532 : : * fsync more than one file.
2533 : : */
745 nathan@postgresql.or 2534 [ + - ]: 126545 : if (wal_sync_method != WAL_SYNC_METHOD_OPEN &&
2535 [ + - ]: 126545 : wal_sync_method != WAL_SYNC_METHOD_OPEN_DSYNC)
2536 : : {
8991 tgl@sss.pgh.pa.us 2537 [ + + ]: 126545 : if (openLogFile >= 0 &&
2960 andres@anarazel.de 2538 [ + + ]: 126522 : !XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2539 : : wal_segment_size))
7074 bruce@momjian.us 2540 : 147 : XLogFileClose();
8991 tgl@sss.pgh.pa.us 2541 [ + + ]: 126545 : if (openLogFile < 0)
2542 : : {
2960 andres@anarazel.de 2543 : 170 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2544 : : wal_segment_size);
1452 rhaas@postgresql.org 2545 : 170 : openLogTLI = tli;
2546 : 170 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2072 tgl@sss.pgh.pa.us 2547 : 170 : ReserveExternalFD();
2548 : : }
2549 : :
1452 rhaas@postgresql.org 2550 : 126545 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2551 : : }
2552 : :
2553 : : /* signal that we need to wakeup walsenders later */
4865 2554 : 126545 : WalSndWakeupRequest();
2555 : :
8994 tgl@sss.pgh.pa.us 2556 : 126545 : LogwrtResult.Flush = LogwrtResult.Write;
2557 : : }
2558 : :
2559 : : /*
2560 : : * Update shared-memory status
2561 : : *
2562 : : * We make sure that the shared 'request' values do not fall behind the
2563 : : * 'result' values. This is not absolutely essential, but it saves some
2564 : : * code in a couple of places.
2565 : : */
570 alvherre@alvh.no-ip. 2566 [ + + ]: 1998369 : SpinLockAcquire(&XLogCtl->info_lck);
2567 [ + + ]: 1998369 : if (XLogCtl->LogwrtRqst.Write < LogwrtResult.Write)
2568 : 111926 : XLogCtl->LogwrtRqst.Write = LogwrtResult.Write;
2569 [ + + ]: 1998369 : if (XLogCtl->LogwrtRqst.Flush < LogwrtResult.Flush)
2570 : 128002 : XLogCtl->LogwrtRqst.Flush = LogwrtResult.Flush;
2571 : 1998369 : SpinLockRelease(&XLogCtl->info_lck);
2572 : :
2573 : : /*
2574 : : * We write Write first, bar, then Flush. When reading, the opposite must
2575 : : * be done (with a matching barrier in between), so that we always see a
2576 : : * Flush value that trails behind the Write value seen.
2577 : : */
2578 : 1998369 : pg_atomic_write_u64(&XLogCtl->logWriteResult, LogwrtResult.Write);
2579 : 1998369 : pg_write_barrier();
2580 : 1998369 : pg_atomic_write_u64(&XLogCtl->logFlushResult, LogwrtResult.Flush);
2581 : :
2582 : : #ifdef USE_ASSERT_CHECKING
2583 : : {
2584 : : XLogRecPtr Flush;
2585 : : XLogRecPtr Write;
2586 : : XLogRecPtr Insert;
2587 : :
2588 : 1998369 : Flush = pg_atomic_read_u64(&XLogCtl->logFlushResult);
2589 : 1998369 : pg_read_barrier();
2590 : 1998369 : Write = pg_atomic_read_u64(&XLogCtl->logWriteResult);
568 2591 : 1998369 : pg_read_barrier();
2592 : 1998369 : Insert = pg_atomic_read_u64(&XLogCtl->logInsertResult);
2593 : :
2594 : : /* WAL written to disk is always ahead of WAL flushed */
570 2595 [ - + ]: 1998369 : Assert(Write >= Flush);
2596 : :
2597 : : /* WAL inserted to buffers is always ahead of WAL written */
568 2598 [ - + ]: 1998369 : Assert(Insert >= Write);
2599 : : }
2600 : : #endif
8994 tgl@sss.pgh.pa.us 2601 : 1998369 : }
2602 : :
2603 : : /*
2604 : : * Record the LSN for an asynchronous transaction commit/abort
2605 : : * and nudge the WALWriter if there is work for it to do.
2606 : : * (This should not be called for synchronous commits.)
2607 : : */
2608 : : void
5569 simon@2ndQuadrant.co 2609 : 29324 : XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN)
2610 : : {
5097 2611 : 29324 : XLogRecPtr WriteRqstPtr = asyncXactLSN;
2612 : : bool sleeping;
700 heikki.linnakangas@i 2613 : 29324 : bool wakeup = false;
2614 : : XLogRecPtr prevAsyncXactLSN;
2615 : :
4053 andres@anarazel.de 2616 [ + + ]: 29324 : SpinLockAcquire(&XLogCtl->info_lck);
2617 : 29324 : sleeping = XLogCtl->WalWriterSleeping;
700 heikki.linnakangas@i 2618 : 29324 : prevAsyncXactLSN = XLogCtl->asyncXactLSN;
4053 andres@anarazel.de 2619 [ + + ]: 29324 : if (XLogCtl->asyncXactLSN < asyncXactLSN)
2620 : 28869 : XLogCtl->asyncXactLSN = asyncXactLSN;
2621 : 29324 : SpinLockRelease(&XLogCtl->info_lck);
2622 : :
2623 : : /*
2624 : : * If somebody else already called this function with a more aggressive
2625 : : * LSN, they will have done what we needed (and perhaps more).
2626 : : */
700 heikki.linnakangas@i 2627 [ + + ]: 29324 : if (asyncXactLSN <= prevAsyncXactLSN)
2628 : 455 : return;
2629 : :
2630 : : /*
2631 : : * If the WALWriter is sleeping, kick it to make it come out of low-power
2632 : : * mode, so that this async commit will reach disk within the expected
2633 : : * amount of time. Otherwise, determine whether it has enough WAL
2634 : : * available to flush, the same way that XLogBackgroundFlush() does.
2635 : : */
2636 [ + + ]: 28869 : if (sleeping)
2637 : 12 : wakeup = true;
2638 : : else
2639 : : {
2640 : : int flushblocks;
2641 : :
570 alvherre@alvh.no-ip. 2642 : 28857 : RefreshXLogWriteResult(LogwrtResult);
2643 : :
700 heikki.linnakangas@i 2644 : 28857 : flushblocks =
2645 : 28857 : WriteRqstPtr / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
2646 : :
2647 [ + - + + ]: 28857 : if (WalWriterFlushAfter == 0 || flushblocks >= WalWriterFlushAfter)
2648 : 3715 : wakeup = true;
2649 : : }
2650 : :
360 2651 [ + + ]: 28869 : if (wakeup)
2652 : : {
2653 : 3727 : volatile PROC_HDR *procglobal = ProcGlobal;
2654 : 3727 : ProcNumber walwriterProc = procglobal->walwriterProc;
2655 : :
2656 [ + + ]: 3727 : if (walwriterProc != INVALID_PROC_NUMBER)
2657 : 199 : SetLatch(&GetPGProcByNumber(walwriterProc)->procLatch);
2658 : : }
2659 : : }
2660 : :
2661 : : /*
2662 : : * Record the LSN up to which we can remove WAL because it's not required by
2663 : : * any replication slot.
2664 : : */
2665 : : void
4287 rhaas@postgresql.org 2666 : 41060 : XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn)
2667 : : {
4053 andres@anarazel.de 2668 [ + + ]: 41060 : SpinLockAcquire(&XLogCtl->info_lck);
2669 : 41060 : XLogCtl->replicationSlotMinLSN = lsn;
2670 : 41060 : SpinLockRelease(&XLogCtl->info_lck);
4287 rhaas@postgresql.org 2671 : 41060 : }
2672 : :
2673 : :
2674 : : /*
2675 : : * Return the oldest LSN we must retain to satisfy the needs of some
2676 : : * replication slot.
2677 : : */
2678 : : static XLogRecPtr
2679 : 2088 : XLogGetReplicationSlotMinimumLSN(void)
2680 : : {
2681 : : XLogRecPtr retval;
2682 : :
4053 andres@anarazel.de 2683 [ - + ]: 2088 : SpinLockAcquire(&XLogCtl->info_lck);
2684 : 2088 : retval = XLogCtl->replicationSlotMinLSN;
2685 : 2088 : SpinLockRelease(&XLogCtl->info_lck);
2686 : :
4287 rhaas@postgresql.org 2687 : 2088 : return retval;
2688 : : }
2689 : :
2690 : : /*
2691 : : * Advance minRecoveryPoint in control file.
2692 : : *
2693 : : * If we crash during recovery, we must reach this point again before the
2694 : : * database is consistent.
2695 : : *
2696 : : * If 'force' is true, 'lsn' argument is ignored. Otherwise, minRecoveryPoint
2697 : : * is only updated if it's not already greater than or equal to 'lsn'.
2698 : : */
2699 : : static void
6095 heikki.linnakangas@i 2700 : 103532 : UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
2701 : : {
2702 : : /* Quick check using our local copy of the variable */
1349 2703 [ + + + + : 103532 : if (!updateMinRecoveryPoint || (!force && lsn <= LocalMinRecoveryPoint))
+ + ]
6095 2704 : 96749 : return;
2705 : :
2706 : : /*
2707 : : * An invalid minRecoveryPoint means that we need to recover all the WAL,
2708 : : * i.e., we're doing crash recovery. We never modify the control file's
2709 : : * value in that case, so we can short-circuit future checks here too. The
2710 : : * local values of minRecoveryPoint and minRecoveryPointTLI should not be
2711 : : * updated until crash recovery finishes. We only do this for the startup
2712 : : * process as it should not update its own reference of minRecoveryPoint
2713 : : * until it has finished crash recovery to make sure that all WAL
2714 : : * available is replayed in this case. This also saves from extra locks
2715 : : * taken on the control file from the startup process.
2716 : : */
1349 2717 [ + + + + ]: 6783 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint) && InRecovery)
2718 : : {
2671 michael@paquier.xyz 2719 : 31 : updateMinRecoveryPoint = false;
2720 : 31 : return;
2721 : : }
2722 : :
6095 heikki.linnakangas@i 2723 : 6752 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2724 : :
2725 : : /* update local copy */
1349 2726 : 6752 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
2727 : 6752 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
2728 : :
2729 [ + + ]: 6752 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint))
2614 michael@paquier.xyz 2730 : 1 : updateMinRecoveryPoint = false;
1349 heikki.linnakangas@i 2731 [ + + + + ]: 6751 : else if (force || LocalMinRecoveryPoint < lsn)
2732 : : {
2733 : : XLogRecPtr newMinRecoveryPoint;
2734 : : TimeLineID newMinRecoveryPointTLI;
2735 : :
2736 : : /*
2737 : : * To avoid having to update the control file too often, we update it
2738 : : * all the way to the last record being replayed, even though 'lsn'
2739 : : * would suffice for correctness. This also allows the 'force' case
2740 : : * to not need a valid 'lsn' value.
2741 : : *
2742 : : * Another important reason for doing it this way is that the passed
2743 : : * 'lsn' value could be bogus, i.e., past the end of available WAL, if
2744 : : * the caller got it from a corrupted heap page. Accepting such a
2745 : : * value as the min recovery point would prevent us from coming up at
2746 : : * all. Instead, we just log a warning and continue with recovery.
2747 : : * (See also the comments about corrupt LSNs in XLogFlush.)
2748 : : */
2749 : 5346 : newMinRecoveryPoint = GetCurrentReplayRecPtr(&newMinRecoveryPointTLI);
4686 alvherre@alvh.no-ip. 2750 [ + + - + ]: 5346 : if (!force && newMinRecoveryPoint < lsn)
5967 tgl@sss.pgh.pa.us 2751 [ # # ]:UBC 0 : elog(WARNING,
2752 : : "xlog min recovery request %X/%08X is past current point %X/%08X",
2753 : : LSN_FORMAT_ARGS(lsn), LSN_FORMAT_ARGS(newMinRecoveryPoint));
2754 : :
2755 : : /* update control file */
4686 alvherre@alvh.no-ip. 2756 [ + + ]:CBC 5346 : if (ControlFile->minRecoveryPoint < newMinRecoveryPoint)
2757 : : {
6095 heikki.linnakangas@i 2758 : 4998 : ControlFile->minRecoveryPoint = newMinRecoveryPoint;
4710 2759 : 4998 : ControlFile->minRecoveryPointTLI = newMinRecoveryPointTLI;
6095 2760 : 4998 : UpdateControlFile();
1349 2761 : 4998 : LocalMinRecoveryPoint = newMinRecoveryPoint;
2762 : 4998 : LocalMinRecoveryPointTLI = newMinRecoveryPointTLI;
2763 : :
6095 2764 [ + + ]: 4998 : ereport(DEBUG2,
2765 : : errmsg_internal("updated min recovery point to %X/%08X on timeline %u",
2766 : : LSN_FORMAT_ARGS(newMinRecoveryPoint),
2767 : : newMinRecoveryPointTLI));
2768 : : }
2769 : : }
2770 : 6752 : LWLockRelease(ControlFileLock);
2771 : : }
2772 : :
2773 : : /*
2774 : : * Ensure that all XLOG data through the given position is flushed to disk.
2775 : : *
2776 : : * NOTE: this differs from XLogWrite mainly in that the WALWriteLock is not
2777 : : * already held, and we try to avoid acquiring it if possible.
2778 : : */
2779 : : void
8994 tgl@sss.pgh.pa.us 2780 : 681575 : XLogFlush(XLogRecPtr record)
2781 : : {
2782 : : XLogRecPtr WriteRqstPtr;
2783 : : XLogwrtRqst WriteRqst;
1447 rhaas@postgresql.org 2784 : 681575 : TimeLineID insertTLI = XLogCtl->InsertTimeLineID;
2785 : :
2786 : : /*
2787 : : * During REDO, we are reading not writing WAL. Therefore, instead of
2788 : : * trying to flush the WAL, we should update minRecoveryPoint instead. We
2789 : : * test XLogInsertAllowed(), not InRecovery, because we need checkpointer
2790 : : * to act this way too, and because when it tries to write the
2791 : : * end-of-recovery checkpoint, it should indeed flush.
2792 : : */
5967 tgl@sss.pgh.pa.us 2793 [ + + ]: 681575 : if (!XLogInsertAllowed())
2794 : : {
6095 heikki.linnakangas@i 2795 : 103080 : UpdateMinRecoveryPoint(record, false);
8994 tgl@sss.pgh.pa.us 2796 : 543726 : return;
2797 : : }
2798 : :
2799 : : /* Quick exit if already known flushed */
4686 alvherre@alvh.no-ip. 2800 [ + + ]: 578495 : if (record <= LogwrtResult.Flush)
8994 tgl@sss.pgh.pa.us 2801 : 440646 : return;
2802 : :
2803 : : #ifdef WAL_DEBUG
2804 : : if (XLOG_DEBUG)
2805 : : elog(LOG, "xlog flush request %X/%08X; write %X/%08X; flush %X/%08X",
2806 : : LSN_FORMAT_ARGS(record),
2807 : : LSN_FORMAT_ARGS(LogwrtResult.Write),
2808 : : LSN_FORMAT_ARGS(LogwrtResult.Flush));
2809 : : #endif
2810 : :
2811 : 137849 : START_CRIT_SECTION();
2812 : :
2813 : : /*
2814 : : * Since fsync is usually a horribly expensive operation, we try to
2815 : : * piggyback as much data as we can on each fsync: if we see any more data
2816 : : * entered into the xlog buffer, we'll write and fsync that too, so that
2817 : : * the final value of LogwrtResult.Flush is as large as possible. This
2818 : : * gives us some chance of avoiding another fsync immediately after.
2819 : : */
2820 : :
2821 : : /* initialize to given target; may increase below */
2822 : 137849 : WriteRqstPtr = record;
2823 : :
2824 : : /*
2825 : : * Now wait until we get the write lock, or someone else does the flush
2826 : : * for us.
2827 : : */
2828 : : for (;;)
8704 2829 : 2392 : {
2830 : : XLogRecPtr insertpos;
2831 : :
2832 : : /* done already? */
570 alvherre@alvh.no-ip. 2833 : 140241 : RefreshXLogWriteResult(LogwrtResult);
4686 2834 [ + + ]: 140241 : if (record <= LogwrtResult.Flush)
5019 heikki.linnakangas@i 2835 : 10441 : break;
2836 : :
2837 : : /*
2838 : : * Before actually performing the write, wait for all in-flight
2839 : : * insertions to the pages we're about to write to finish.
2840 : : */
570 alvherre@alvh.no-ip. 2841 [ + + ]: 129800 : SpinLockAcquire(&XLogCtl->info_lck);
2842 [ + + ]: 129800 : if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2843 : 8759 : WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2844 : 129800 : SpinLockRelease(&XLogCtl->info_lck);
4494 heikki.linnakangas@i 2845 : 129800 : insertpos = WaitXLogInsertionsToFinish(WriteRqstPtr);
2846 : :
2847 : : /*
2848 : : * Try to get the write lock. If we can't get it immediately, wait
2849 : : * until it's released, and recheck if we still need to do the flush
2850 : : * or if the backend that held the lock did it for us already. This
2851 : : * helps to maintain a good rate of group committing when the system
2852 : : * is bottlenecked by the speed of fsyncing.
2853 : : */
5010 2854 [ + + ]: 129800 : if (!LWLockAcquireOrWait(WALWriteLock, LW_EXCLUSIVE))
2855 : : {
2856 : : /*
2857 : : * The lock is now free, but we didn't acquire it yet. Before we
2858 : : * do, loop back to check if someone else flushed the record for
2859 : : * us already.
2860 : : */
5019 2861 : 2392 : continue;
2862 : : }
2863 : :
2864 : : /* Got the lock; recheck whether request is satisfied */
572 alvherre@alvh.no-ip. 2865 : 127408 : RefreshXLogWriteResult(LogwrtResult);
4686 2866 [ + + ]: 127408 : if (record <= LogwrtResult.Flush)
2867 : : {
4865 rhaas@postgresql.org 2868 : 3302 : LWLockRelease(WALWriteLock);
2869 : 3302 : break;
2870 : : }
2871 : :
2872 : : /*
2873 : : * Sleep before flush! By adding a delay here, we may give further
2874 : : * backends the opportunity to join the backlog of group commit
2875 : : * followers; this can significantly improve transaction throughput,
2876 : : * at the risk of increasing transaction latency.
2877 : : *
2878 : : * We do not sleep if enableFsync is not turned on, nor if there are
2879 : : * fewer than CommitSiblings other backends with active transactions.
2880 : : */
2881 [ - + - - : 124106 : if (CommitDelay > 0 && enableFsync &&
- - ]
4865 rhaas@postgresql.org 2882 :UBC 0 : MinimumActiveBackends(CommitSiblings))
2883 : : {
2884 : 0 : pg_usleep(CommitDelay);
2885 : :
2886 : : /*
2887 : : * Re-check how far we can now flush the WAL. It's generally not
2888 : : * safe to call WaitXLogInsertionsToFinish while holding
2889 : : * WALWriteLock, because an in-progress insertion might need to
2890 : : * also grab WALWriteLock to make progress. But we know that all
2891 : : * the insertions up to insertpos have already finished, because
2892 : : * that's what the earlier WaitXLogInsertionsToFinish() returned.
2893 : : * We're only calling it again to allow insertpos to be moved
2894 : : * further forward, not to actually wait for anyone.
2895 : : */
4494 heikki.linnakangas@i 2896 : 0 : insertpos = WaitXLogInsertionsToFinish(insertpos);
2897 : : }
2898 : :
2899 : : /* try to write/flush later additions to XLOG as well */
4494 heikki.linnakangas@i 2900 :CBC 124106 : WriteRqst.Write = insertpos;
2901 : 124106 : WriteRqst.Flush = insertpos;
2902 : :
1452 rhaas@postgresql.org 2903 : 124106 : XLogWrite(WriteRqst, insertTLI, false);
2904 : :
8794 tgl@sss.pgh.pa.us 2905 : 124106 : LWLockRelease(WALWriteLock);
2906 : : /* done */
5019 heikki.linnakangas@i 2907 : 124106 : break;
2908 : : }
2909 : :
8994 tgl@sss.pgh.pa.us 2910 [ - + ]: 137849 : END_CRIT_SECTION();
2911 : :
2912 : : /* wake up walsenders now that we've released heavily contended locks */
933 andres@anarazel.de 2913 : 137849 : WalSndWakeupProcessRequests(true, !RecoveryInProgress());
2914 : :
2915 : : /*
2916 : : * If we still haven't flushed to the request point then we have a
2917 : : * problem; most likely, the requested flush point is past end of XLOG.
2918 : : * This has been seen to occur when a disk page has a corrupted LSN.
2919 : : *
2920 : : * Formerly we treated this as a PANIC condition, but that hurts the
2921 : : * system's robustness rather than helping it: we do not want to take down
2922 : : * the whole system due to corruption on one data page. In particular, if
2923 : : * the bad page is encountered again during recovery then we would be
2924 : : * unable to restart the database at all! (This scenario actually
2925 : : * happened in the field several times with 7.1 releases.) As of 8.4, bad
2926 : : * LSNs encountered during recovery are UpdateMinRecoveryPoint's problem;
2927 : : * the only time we can reach here during recovery is while flushing the
2928 : : * end-of-recovery checkpoint record, and we don't expect that to have a
2929 : : * bad LSN.
2930 : : *
2931 : : * Note that for calls from xact.c, the ERROR will be promoted to PANIC
2932 : : * since xact.c calls this routine inside a critical section. However,
2933 : : * calls from bufmgr.c are not within critical sections and so we will not
2934 : : * force a restart for a bad LSN on a data page.
2935 : : */
4686 alvherre@alvh.no-ip. 2936 [ - + ]: 137849 : if (LogwrtResult.Flush < record)
5967 tgl@sss.pgh.pa.us 2937 [ # # ]:UBC 0 : elog(ERROR,
2938 : : "xlog flush request %X/%08X is not satisfied --- flushed only to %X/%08X",
2939 : : LSN_FORMAT_ARGS(record),
2940 : : LSN_FORMAT_ARGS(LogwrtResult.Flush));
2941 : :
2942 : : /*
2943 : : * Cross-check XLogNeedsFlush(). Some of the checks of XLogFlush() and
2944 : : * XLogNeedsFlush() are duplicated, and this assertion ensures that these
2945 : : * remain consistent.
2946 : : */
38 michael@paquier.xyz 2947 [ - + ]:GNC 137849 : Assert(!XLogNeedsFlush(record));
2948 : : }
2949 : :
2950 : : /*
2951 : : * Write & flush xlog, but without specifying exactly where to.
2952 : : *
2953 : : * We normally write only completed blocks; but if there is nothing to do on
2954 : : * that basis, we check for unwritten async commits in the current incomplete
2955 : : * block, and write through the latest one of those. Thus, if async commits
2956 : : * are not being used, we will write complete blocks only.
2957 : : *
2958 : : * If, based on the above, there's anything to write we do so immediately. But
2959 : : * to avoid calling fsync, fdatasync et. al. at a rate that'd impact
2960 : : * concurrent IO, we only flush WAL every wal_writer_delay ms, or if there's
2961 : : * more than wal_writer_flush_after unflushed blocks.
2962 : : *
2963 : : * We can guarantee that async commits reach disk after at most three
2964 : : * wal_writer_delay cycles. (When flushing complete blocks, we allow XLogWrite
2965 : : * to write "flexibly", meaning it can stop at the end of the buffer ring;
2966 : : * this makes a difference only with very high load or long wal_writer_delay,
2967 : : * but imposes one extra cycle for the worst case for async commits.)
2968 : : *
2969 : : * This routine is invoked periodically by the background walwriter process.
2970 : : *
2971 : : * Returns true if there was any work to do, even if we skipped flushing due
2972 : : * to wal_writer_delay/wal_writer_flush_after.
2973 : : */
2974 : : bool
6670 tgl@sss.pgh.pa.us 2975 :CBC 10620 : XLogBackgroundFlush(void)
2976 : : {
2977 : : XLogwrtRqst WriteRqst;
2978 : 10620 : bool flexible = true;
2979 : : static TimestampTz lastflush;
2980 : : TimestampTz now;
2981 : : int flushblocks;
2982 : : TimeLineID insertTLI;
2983 : :
2984 : : /* XLOG doesn't need flushing during recovery */
6095 heikki.linnakangas@i 2985 [ - + ]: 10620 : if (RecoveryInProgress())
4920 tgl@sss.pgh.pa.us 2986 :UBC 0 : return false;
2987 : :
2988 : : /*
2989 : : * Since we're not in recovery, InsertTimeLineID is set and can't change,
2990 : : * so we can read it without a lock.
2991 : : */
1447 rhaas@postgresql.org 2992 :CBC 10620 : insertTLI = XLogCtl->InsertTimeLineID;
2993 : :
2994 : : /* read updated LogwrtRqst */
4053 andres@anarazel.de 2995 [ + + ]: 10620 : SpinLockAcquire(&XLogCtl->info_lck);
3542 2996 : 10620 : WriteRqst = XLogCtl->LogwrtRqst;
4053 2997 : 10620 : SpinLockRelease(&XLogCtl->info_lck);
2998 : :
2999 : : /* back off to last completed page boundary */
3542 3000 : 10620 : WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
3001 : :
3002 : : /* if we have already flushed that far, consider async commit records */
570 alvherre@alvh.no-ip. 3003 : 10620 : RefreshXLogWriteResult(LogwrtResult);
3542 andres@anarazel.de 3004 [ + + ]: 10620 : if (WriteRqst.Write <= LogwrtResult.Flush)
3005 : : {
4053 3006 [ - + ]: 8041 : SpinLockAcquire(&XLogCtl->info_lck);
3542 3007 : 8041 : WriteRqst.Write = XLogCtl->asyncXactLSN;
4053 3008 : 8041 : SpinLockRelease(&XLogCtl->info_lck);
6670 tgl@sss.pgh.pa.us 3009 : 8041 : flexible = false; /* ensure it all gets written */
3010 : : }
3011 : :
3012 : : /*
3013 : : * If already known flushed, we're done. Just need to check if we are
3014 : : * holding an open file handle to a logfile that's no longer in use,
3015 : : * preventing the file from being deleted.
3016 : : */
3542 andres@anarazel.de 3017 [ + + ]: 10620 : if (WriteRqst.Write <= LogwrtResult.Flush)
3018 : : {
5592 bruce@momjian.us 3019 [ + + ]: 7379 : if (openLogFile >= 0)
3020 : : {
2960 andres@anarazel.de 3021 [ + + ]: 4723 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
3022 : : wal_segment_size))
3023 : : {
5619 magnus@hagander.net 3024 : 201 : XLogFileClose();
3025 : : }
3026 : : }
4920 tgl@sss.pgh.pa.us 3027 : 7379 : return false;
3028 : : }
3029 : :
3030 : : /*
3031 : : * Determine how far to flush WAL, based on the wal_writer_delay and
3032 : : * wal_writer_flush_after GUCs.
3033 : : *
3034 : : * Note that XLogSetAsyncXactLSN() performs similar calculation based on
3035 : : * wal_writer_flush_after, to decide when to wake us up. Make sure the
3036 : : * logic is the same in both places if you change this.
3037 : : */
3542 andres@anarazel.de 3038 : 3241 : now = GetCurrentTimestamp();
700 heikki.linnakangas@i 3039 : 3241 : flushblocks =
3542 andres@anarazel.de 3040 : 3241 : WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
3041 : :
3042 [ + - + + ]: 3241 : if (WalWriterFlushAfter == 0 || lastflush == 0)
3043 : : {
3044 : : /* first call, or block based limits disabled */
3045 : 267 : WriteRqst.Flush = WriteRqst.Write;
3046 : 267 : lastflush = now;
3047 : : }
3048 [ + + ]: 2974 : else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
3049 : : {
3050 : : /*
3051 : : * Flush the writes at least every WalWriterDelay ms. This is
3052 : : * important to bound the amount of time it takes for an asynchronous
3053 : : * commit to hit disk.
3054 : : */
3055 : 2817 : WriteRqst.Flush = WriteRqst.Write;
3056 : 2817 : lastflush = now;
3057 : : }
700 heikki.linnakangas@i 3058 [ + + ]: 157 : else if (flushblocks >= WalWriterFlushAfter)
3059 : : {
3060 : : /* exceeded wal_writer_flush_after blocks, flush */
3542 andres@anarazel.de 3061 : 143 : WriteRqst.Flush = WriteRqst.Write;
3062 : 143 : lastflush = now;
3063 : : }
3064 : : else
3065 : : {
3066 : : /* no flushing, this time round */
3067 : 14 : WriteRqst.Flush = 0;
3068 : : }
3069 : :
3070 : : #ifdef WAL_DEBUG
3071 : : if (XLOG_DEBUG)
3072 : : elog(LOG, "xlog bg flush request write %X/%08X; flush: %X/%08X, current is write %X/%08X; flush %X/%08X",
3073 : : LSN_FORMAT_ARGS(WriteRqst.Write),
3074 : : LSN_FORMAT_ARGS(WriteRqst.Flush),
3075 : : LSN_FORMAT_ARGS(LogwrtResult.Write),
3076 : : LSN_FORMAT_ARGS(LogwrtResult.Flush));
3077 : : #endif
3078 : :
6670 tgl@sss.pgh.pa.us 3079 : 3241 : START_CRIT_SECTION();
3080 : :
3081 : : /* now wait for any in-progress insertions to finish and get write lock */
3542 andres@anarazel.de 3082 : 3241 : WaitXLogInsertionsToFinish(WriteRqst.Write);
6670 tgl@sss.pgh.pa.us 3083 : 3241 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
572 alvherre@alvh.no-ip. 3084 : 3241 : RefreshXLogWriteResult(LogwrtResult);
3542 andres@anarazel.de 3085 [ + + ]: 3241 : if (WriteRqst.Write > LogwrtResult.Write ||
3086 [ + + ]: 189 : WriteRqst.Flush > LogwrtResult.Flush)
3087 : : {
1452 rhaas@postgresql.org 3088 : 3162 : XLogWrite(WriteRqst, insertTLI, flexible);
3089 : : }
6670 tgl@sss.pgh.pa.us 3090 : 3241 : LWLockRelease(WALWriteLock);
3091 : :
3092 [ - + ]: 3241 : END_CRIT_SECTION();
3093 : :
3094 : : /* wake up walsenders now that we've released heavily contended locks */
933 andres@anarazel.de 3095 : 3241 : WalSndWakeupProcessRequests(true, !RecoveryInProgress());
3096 : :
3097 : : /*
3098 : : * Great, done. To take some work off the critical path, try to initialize
3099 : : * as many of the no-longer-needed WAL buffers for future use as we can.
3100 : : */
1452 rhaas@postgresql.org 3101 : 3241 : AdvanceXLInsertBuffer(InvalidXLogRecPtr, insertTLI, true);
3102 : :
3103 : : /*
3104 : : * If we determined that we need to write data, but somebody else
3105 : : * wrote/flushed already, it should be considered as being active, to
3106 : : * avoid hibernating too early.
3107 : : */
3542 andres@anarazel.de 3108 : 3241 : return true;
3109 : : }
3110 : :
3111 : : /*
3112 : : * Test whether XLOG data has been flushed up to (at least) the given
3113 : : * position, or whether the minimum recovery point has been updated past
3114 : : * the given position.
3115 : : *
3116 : : * Returns true if a flush is still needed, or if the minimum recovery point
3117 : : * must be updated.
3118 : : *
3119 : : * It is possible that someone else is already in the process of flushing
3120 : : * that far, or has updated the minimum recovery point up to the given
3121 : : * position.
3122 : : */
3123 : : bool
6725 tgl@sss.pgh.pa.us 3124 : 8792622 : XLogNeedsFlush(XLogRecPtr record)
3125 : : {
3126 : : /*
3127 : : * During recovery, we don't flush WAL but update minRecoveryPoint
3128 : : * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3129 : : * would need to be updated.
3130 : : *
3131 : : * Using XLogInsertAllowed() rather than RecoveryInProgress() matters for
3132 : : * the case of an end-of-recovery checkpoint, where WAL data is flushed.
3133 : : * This check should be consistent with the one in XLogFlush().
3134 : : */
38 michael@paquier.xyz 3135 [ + + ]:GNC 8792622 : if (!XLogInsertAllowed())
3136 : : {
3137 : : /* Quick exit if already known to be updated or cannot be updated */
27 3138 [ + - + + ]: 625165 : if (!updateMinRecoveryPoint || record <= LocalMinRecoveryPoint)
3139 : 616815 : return false;
3140 : :
3141 : : /*
3142 : : * An invalid minRecoveryPoint means that we need to recover all the
3143 : : * WAL, i.e., we're doing crash recovery. We never modify the control
3144 : : * file's value in that case, so we can short-circuit future checks
3145 : : * here too. This triggers a quick exit path for the startup process,
3146 : : * which cannot update its local copy of minRecoveryPoint as long as
3147 : : * it has not replayed all WAL available when doing crash recovery.
3148 : : */
1349 heikki.linnakangas@i 3149 [ + + - + ]:CBC 8350 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint) && InRecovery)
3150 : : {
2671 michael@paquier.xyz 3151 :UBC 0 : updateMinRecoveryPoint = false;
5791 simon@2ndQuadrant.co 3152 :LBC (622910) : return false;
3153 : : }
3154 : :
3155 : : /*
3156 : : * Update local copy of minRecoveryPoint. But if the lock is busy,
3157 : : * just return a conservative guess.
3158 : : */
5791 simon@2ndQuadrant.co 3159 [ - + ]:CBC 8350 : if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
5791 simon@2ndQuadrant.co 3160 :UBC 0 : return true;
1349 heikki.linnakangas@i 3161 :CBC 8350 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
3162 : 8350 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
5791 simon@2ndQuadrant.co 3163 : 8350 : LWLockRelease(ControlFileLock);
3164 : :
3165 : : /*
3166 : : * Check minRecoveryPoint for any other process than the startup
3167 : : * process doing crash recovery, which should not update the control
3168 : : * file value if crash recovery is still running.
3169 : : */
1349 heikki.linnakangas@i 3170 [ - + ]: 8350 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint))
2614 michael@paquier.xyz 3171 :UBC 0 : updateMinRecoveryPoint = false;
3172 : :
3173 : : /* check again */
1349 heikki.linnakangas@i 3174 [ + + - + ]:CBC 8350 : if (record <= LocalMinRecoveryPoint || !updateMinRecoveryPoint)
2614 michael@paquier.xyz 3175 : 73 : return false;
3176 : : else
3177 : 8277 : return true;
3178 : : }
3179 : :
3180 : : /* Quick exit if already known flushed */
4686 alvherre@alvh.no-ip. 3181 [ + + ]: 8167457 : if (record <= LogwrtResult.Flush)
6725 tgl@sss.pgh.pa.us 3182 : 7978574 : return false;
3183 : :
3184 : : /* read LogwrtResult and update local state */
572 alvherre@alvh.no-ip. 3185 : 188883 : RefreshXLogWriteResult(LogwrtResult);
3186 : :
3187 : : /* check again */
4686 3188 [ + + ]: 188883 : if (record <= LogwrtResult.Flush)
6725 tgl@sss.pgh.pa.us 3189 : 3643 : return false;
3190 : :
3191 : 185240 : return true;
3192 : : }
3193 : :
3194 : : /*
3195 : : * Try to make a given XLOG file segment exist.
3196 : : *
3197 : : * logsegno: identify segment.
3198 : : *
3199 : : * *added: on return, true if this call raised the number of extant segments.
3200 : : *
3201 : : * path: on return, this char[MAXPGPATH] has the path to the logsegno file.
3202 : : *
3203 : : * Returns -1 or FD of opened file. A -1 here is not an error; a caller
3204 : : * wanting an open segment should attempt to open "path", which usually will
3205 : : * succeed. (This is weird, but it's efficient for the callers.)
3206 : : */
3207 : : static int
1452 rhaas@postgresql.org 3208 : 13951 : XLogFileInitInternal(XLogSegNo logsegno, TimeLineID logtli,
3209 : : bool *added, char *path)
3210 : : {
3211 : : char tmppath[MAXPGPATH];
3212 : : XLogSegNo installed_segno;
3213 : : XLogSegNo max_segno;
3214 : : int fd;
3215 : : int save_errno;
933 tmunro@postgresql.or 3216 : 13951 : int open_flags = O_RDWR | O_CREAT | O_EXCL | PG_BINARY;
3217 : : instr_time io_start;
3218 : :
1452 rhaas@postgresql.org 3219 [ - + ]: 13951 : Assert(logtli != 0);
3220 : :
3221 : 13951 : XLogFilePath(path, logtli, logsegno, wal_segment_size);
3222 : :
3223 : : /*
3224 : : * Try to use existent file (checkpoint maker may have created it already)
3225 : : */
1582 noah@leadboat.com 3226 : 13951 : *added = false;
969 tmunro@postgresql.or 3227 : 13951 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
745 nathan@postgresql.or 3228 : 13951 : get_sync_bit(wal_sync_method));
1582 noah@leadboat.com 3229 [ + + ]: 13951 : if (fd < 0)
3230 : : {
3231 [ - + ]: 1326 : if (errno != ENOENT)
1582 noah@leadboat.com 3232 [ # # ]:UBC 0 : ereport(ERROR,
3233 : : (errcode_for_file_access(),
3234 : : errmsg("could not open file \"%s\": %m", path)));
3235 : : }
3236 : : else
1582 noah@leadboat.com 3237 :CBC 12625 : return fd;
3238 : :
3239 : : /*
3240 : : * Initialize an empty (all zeroes) segment. NOTE: it is possible that
3241 : : * another process is doing the same thing. If so, we will end up
3242 : : * pre-creating an extra log segment. That seems OK, and better than
3243 : : * holding the lock throughout this lengthy process.
3244 : : */
6694 tgl@sss.pgh.pa.us 3245 [ + + ]: 1326 : elog(DEBUG2, "creating and filling new WAL file");
3246 : :
7420 3247 : 1326 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3248 : :
8990 3249 : 1326 : unlink(tmppath);
3250 : :
933 tmunro@postgresql.or 3251 [ - + ]: 1326 : if (io_direct_flags & IO_DIRECT_WAL_INIT)
933 tmunro@postgresql.or 3252 :UBC 0 : open_flags |= PG_O_DIRECT;
3253 : :
3254 : : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
933 tmunro@postgresql.or 3255 :CBC 1326 : fd = BasicOpenFile(tmppath, open_flags);
9527 vadim4o@yahoo.com 3256 [ - + ]: 1326 : if (fd < 0)
7500 tgl@sss.pgh.pa.us 3257 [ # # ]:UBC 0 : ereport(ERROR,
3258 : : (errcode_for_file_access(),
3259 : : errmsg("could not create file \"%s\": %m", tmppath)));
3260 : :
3261 : : /* Measure I/O timing when initializing segment */
243 michael@paquier.xyz 3262 :CBC 1326 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3263 : :
2400 tmunro@postgresql.or 3264 : 1326 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_WRITE);
3265 : 1326 : save_errno = 0;
3266 [ + - ]: 1326 : if (wal_init_zero)
3267 : : {
3268 : : ssize_t rc;
3269 : :
3270 : : /*
3271 : : * Zero-fill the file. With this setting, we do this the hard way to
3272 : : * ensure that all the file space has really been allocated. On
3273 : : * platforms that allow "holes" in files, just seeking to the end
3274 : : * doesn't allocate intermediate space. This way, we know that we
3275 : : * have all the space and (after the fsync below) that all the
3276 : : * indirect blocks are down on disk. Therefore, fdatasync(2) or
3277 : : * O_DSYNC will be sufficient to sync future writes to the log file.
3278 : : */
966 michael@paquier.xyz 3279 : 1326 : rc = pg_pwrite_zeros(fd, wal_segment_size, 0);
3280 : :
1084 3281 [ - + ]: 1326 : if (rc < 0)
1084 michael@paquier.xyz 3282 :UBC 0 : save_errno = errno;
3283 : : }
3284 : : else
3285 : : {
3286 : : /*
3287 : : * Otherwise, seeking to the end and writing a solitary byte is
3288 : : * enough.
3289 : : */
4436 jdavis@postgresql.or 3290 : 0 : errno = 0;
1084 michael@paquier.xyz 3291 [ # # ]: 0 : if (pg_pwrite(fd, "\0", 1, wal_segment_size - 1) != 1)
3292 : : {
3293 : : /* if write didn't set errno, assume no disk space */
2400 tmunro@postgresql.or 3294 [ # # ]: 0 : save_errno = errno ? errno : ENOSPC;
3295 : : }
3296 : : }
2400 tmunro@postgresql.or 3297 :CBC 1326 : pgstat_report_wait_end();
3298 : :
3299 : : /*
3300 : : * A full segment worth of data is written when using wal_init_zero. One
3301 : : * byte is written when not using it.
3302 : : */
265 michael@paquier.xyz 3303 : 1326 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_INIT, IOOP_WRITE,
3304 : : io_start, 1,
3305 [ + - ]: 1326 : wal_init_zero ? wal_segment_size : 1);
3306 : :
2400 tmunro@postgresql.or 3307 [ - + ]: 1326 : if (save_errno)
3308 : : {
3309 : : /*
3310 : : * If we fail to make the file, delete it to release disk space
3311 : : */
2400 tmunro@postgresql.or 3312 :UBC 0 : unlink(tmppath);
3313 : :
3314 : 0 : close(fd);
3315 : :
3316 : 0 : errno = save_errno;
3317 : :
3318 [ # # ]: 0 : ereport(ERROR,
3319 : : (errcode_for_file_access(),
3320 : : errmsg("could not write to file \"%s\": %m", tmppath)));
3321 : : }
3322 : :
3323 : : /* Measure I/O timing when flushing segment */
243 michael@paquier.xyz 3324 :CBC 1326 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3325 : :
3145 rhaas@postgresql.org 3326 : 1326 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_SYNC);
9089 tgl@sss.pgh.pa.us 3327 [ - + ]: 1326 : if (pg_fsync(fd) != 0)
3328 : : {
1158 drowley@postgresql.o 3329 :UBC 0 : save_errno = errno;
4717 heikki.linnakangas@i 3330 : 0 : close(fd);
2681 michael@paquier.xyz 3331 : 0 : errno = save_errno;
7500 tgl@sss.pgh.pa.us 3332 [ # # ]: 0 : ereport(ERROR,
3333 : : (errcode_for_file_access(),
3334 : : errmsg("could not fsync file \"%s\": %m", tmppath)));
3335 : : }
3145 rhaas@postgresql.org 3336 :CBC 1326 : pgstat_report_wait_end();
3337 : :
265 michael@paquier.xyz 3338 : 1326 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_INIT,
3339 : : IOOP_FSYNC, io_start, 1, 0);
3340 : :
2305 peter@eisentraut.org 3341 [ - + ]: 1326 : if (close(fd) != 0)
7500 tgl@sss.pgh.pa.us 3342 [ # # ]:UBC 0 : ereport(ERROR,
3343 : : (errcode_for_file_access(),
3344 : : errmsg("could not close file \"%s\": %m", tmppath)));
3345 : :
3346 : : /*
3347 : : * Now move the segment into place with its final name. Cope with
3348 : : * possibility that someone else has created the file while we were
3349 : : * filling ours: if so, use ours to pre-create a future log segment.
3350 : : */
4873 heikki.linnakangas@i 3351 :CBC 1326 : installed_segno = logsegno;
3352 : :
3353 : : /*
3354 : : * XXX: What should we use as max_segno? We used to use XLOGfileslop when
3355 : : * that was a constant, but that was always a bit dubious: normally, at a
3356 : : * checkpoint, XLOGfileslop was the offset from the checkpoint record, but
3357 : : * here, it was the offset from the insert location. We can't do the
3358 : : * normal XLOGfileslop calculation here because we don't have access to
3359 : : * the prior checkpoint's redo location. So somewhat arbitrarily, just use
3360 : : * CheckPointSegments.
3361 : : */
3899 3362 : 1326 : max_segno = logsegno + CheckPointSegments;
1452 rhaas@postgresql.org 3363 [ + - ]: 1326 : if (InstallXLogFileSegment(&installed_segno, tmppath, true, max_segno,
3364 : : logtli))
3365 : : {
1582 noah@leadboat.com 3366 : 1326 : *added = true;
3367 [ + + ]: 1326 : elog(DEBUG2, "done creating and filling new WAL file");
3368 : : }
3369 : : else
3370 : : {
3371 : : /*
3372 : : * No need for any more future segments, or InstallXLogFileSegment()
3373 : : * failed to rename the file into place. If the rename failed, a
3374 : : * caller opening the file may fail.
3375 : : */
8866 tgl@sss.pgh.pa.us 3376 :UBC 0 : unlink(tmppath);
1582 noah@leadboat.com 3377 [ # # ]: 0 : elog(DEBUG2, "abandoned new WAL file");
3378 : : }
3379 : :
1582 noah@leadboat.com 3380 :CBC 1326 : return -1;
3381 : : }
3382 : :
3383 : : /*
3384 : : * Create a new XLOG file segment, or open a pre-existing one.
3385 : : *
3386 : : * logsegno: identify segment to be created/opened.
3387 : : *
3388 : : * Returns FD of opened file.
3389 : : *
3390 : : * Note: errors here are ERROR not PANIC because we might or might not be
3391 : : * inside a critical section (eg, during checkpoint there is no reason to
3392 : : * take down the system on failure). They will promote to PANIC if we are
3393 : : * in a critical section.
3394 : : */
3395 : : int
1452 rhaas@postgresql.org 3396 : 13734 : XLogFileInit(XLogSegNo logsegno, TimeLineID logtli)
3397 : : {
3398 : : bool ignore_added;
3399 : : char path[MAXPGPATH];
3400 : : int fd;
3401 : :
3402 [ - + ]: 13734 : Assert(logtli != 0);
3403 : :
3404 : 13734 : fd = XLogFileInitInternal(logsegno, logtli, &ignore_added, path);
1582 noah@leadboat.com 3405 [ + + ]: 13734 : if (fd >= 0)
3406 : 12461 : return fd;
3407 : :
3408 : : /* Now open original target segment (might not be file I just made) */
969 tmunro@postgresql.or 3409 : 1273 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
745 nathan@postgresql.or 3410 : 1273 : get_sync_bit(wal_sync_method));
8866 tgl@sss.pgh.pa.us 3411 [ - + ]: 1273 : if (fd < 0)
7500 tgl@sss.pgh.pa.us 3412 [ # # ]:UBC 0 : ereport(ERROR,
3413 : : (errcode_for_file_access(),
3414 : : errmsg("could not open file \"%s\": %m", path)));
7229 neilc@samurai.com 3415 :CBC 1273 : return fd;
3416 : : }
3417 : :
3418 : : /*
3419 : : * Create a new XLOG file segment by copying a pre-existing one.
3420 : : *
3421 : : * destsegno: identify segment to be created.
3422 : : *
3423 : : * srcTLI, srcsegno: identify segment to be copied (could be from
3424 : : * a different timeline)
3425 : : *
3426 : : * upto: how much of the source file to copy (the rest is filled with
3427 : : * zeros)
3428 : : *
3429 : : * Currently this is only used during recovery, and so there are no locking
3430 : : * considerations. But we should be just as tense as XLogFileInit to avoid
3431 : : * emplacing a bogus file.
3432 : : */
3433 : : static void
1452 rhaas@postgresql.org 3434 : 40 : XLogFileCopy(TimeLineID destTLI, XLogSegNo destsegno,
3435 : : TimeLineID srcTLI, XLogSegNo srcsegno,
3436 : : int upto)
3437 : : {
3438 : : char path[MAXPGPATH];
3439 : : char tmppath[MAXPGPATH];
3440 : : PGAlignedXLogBlock buffer;
3441 : : int srcfd;
3442 : : int fd;
3443 : : int nbytes;
3444 : :
3445 : : /*
3446 : : * Open the source file
3447 : : */
2960 andres@anarazel.de 3448 : 40 : XLogFilePath(path, srcTLI, srcsegno, wal_segment_size);
2956 peter_e@gmx.net 3449 : 40 : srcfd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
7768 tgl@sss.pgh.pa.us 3450 [ - + ]: 40 : if (srcfd < 0)
7500 tgl@sss.pgh.pa.us 3451 [ # # ]:UBC 0 : ereport(ERROR,
3452 : : (errcode_for_file_access(),
3453 : : errmsg("could not open file \"%s\": %m", path)));
3454 : :
3455 : : /*
3456 : : * Copy into a temp file name.
3457 : : */
7420 tgl@sss.pgh.pa.us 3458 :CBC 40 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3459 : :
7768 3460 : 40 : unlink(tmppath);
3461 : :
3462 : : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
2956 peter_e@gmx.net 3463 : 40 : fd = OpenTransientFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
7768 tgl@sss.pgh.pa.us 3464 [ - + ]: 40 : if (fd < 0)
7500 tgl@sss.pgh.pa.us 3465 [ # # ]:UBC 0 : ereport(ERROR,
3466 : : (errcode_for_file_access(),
3467 : : errmsg("could not create file \"%s\": %m", tmppath)));
3468 : :
3469 : : /*
3470 : : * Do the data copying.
3471 : : */
2960 andres@anarazel.de 3472 [ + + ]:CBC 81960 : for (nbytes = 0; nbytes < wal_segment_size; nbytes += sizeof(buffer))
3473 : : {
3474 : : int nread;
3475 : :
3966 heikki.linnakangas@i 3476 : 81920 : nread = upto - nbytes;
3477 : :
3478 : : /*
3479 : : * The part that is not read from the source file is filled with
3480 : : * zeros.
3481 : : */
3482 [ + + ]: 81920 : if (nread < sizeof(buffer))
2613 tgl@sss.pgh.pa.us 3483 : 40 : memset(buffer.data, 0, sizeof(buffer));
3484 : :
3966 heikki.linnakangas@i 3485 [ + + ]: 81920 : if (nread > 0)
3486 : : {
3487 : : int r;
3488 : :
3489 [ + + ]: 2184 : if (nread > sizeof(buffer))
3490 : 2144 : nread = sizeof(buffer);
3145 rhaas@postgresql.org 3491 : 2184 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_READ);
2613 tgl@sss.pgh.pa.us 3492 : 2184 : r = read(srcfd, buffer.data, nread);
2658 michael@paquier.xyz 3493 [ - + ]: 2184 : if (r != nread)
3494 : : {
2658 michael@paquier.xyz 3495 [ # # ]:UBC 0 : if (r < 0)
3966 heikki.linnakangas@i 3496 [ # # ]: 0 : ereport(ERROR,
3497 : : (errcode_for_file_access(),
3498 : : errmsg("could not read file \"%s\": %m",
3499 : : path)));
3500 : : else
3501 [ # # ]: 0 : ereport(ERROR,
3502 : : (errcode(ERRCODE_DATA_CORRUPTED),
3503 : : errmsg("could not read file \"%s\": read %d of %zu",
3504 : : path, r, (Size) nread)));
3505 : : }
3145 rhaas@postgresql.org 3506 :CBC 2184 : pgstat_report_wait_end();
3507 : : }
7768 tgl@sss.pgh.pa.us 3508 : 81920 : errno = 0;
3145 rhaas@postgresql.org 3509 : 81920 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_WRITE);
2613 tgl@sss.pgh.pa.us 3510 [ - + ]: 81920 : if ((int) write(fd, buffer.data, sizeof(buffer)) != (int) sizeof(buffer))
3511 : : {
7768 tgl@sss.pgh.pa.us 3512 :UBC 0 : int save_errno = errno;
3513 : :
3514 : : /*
3515 : : * If we fail to make the file, delete it to release disk space
3516 : : */
3517 : 0 : unlink(tmppath);
3518 : : /* if write didn't set errno, assume problem is no disk space */
3519 [ # # ]: 0 : errno = save_errno ? save_errno : ENOSPC;
3520 : :
7500 3521 [ # # ]: 0 : ereport(ERROR,
3522 : : (errcode_for_file_access(),
3523 : : errmsg("could not write to file \"%s\": %m", tmppath)));
3524 : : }
3145 rhaas@postgresql.org 3525 :CBC 81920 : pgstat_report_wait_end();
3526 : : }
3527 : :
3528 : 40 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_SYNC);
7768 tgl@sss.pgh.pa.us 3529 [ - + ]: 40 : if (pg_fsync(fd) != 0)
2534 tmunro@postgresql.or 3530 [ # # ]:UBC 0 : ereport(data_sync_elevel(ERROR),
3531 : : (errcode_for_file_access(),
3532 : : errmsg("could not fsync file \"%s\": %m", tmppath)));
3145 rhaas@postgresql.org 3533 :CBC 40 : pgstat_report_wait_end();
3534 : :
2305 peter@eisentraut.org 3535 [ - + ]: 40 : if (CloseTransientFile(fd) != 0)
7500 tgl@sss.pgh.pa.us 3536 [ # # ]:UBC 0 : ereport(ERROR,
3537 : : (errcode_for_file_access(),
3538 : : errmsg("could not close file \"%s\": %m", tmppath)));
3539 : :
2305 peter@eisentraut.org 3540 [ - + ]:CBC 40 : if (CloseTransientFile(srcfd) != 0)
2424 michael@paquier.xyz 3541 [ # # ]:UBC 0 : ereport(ERROR,
3542 : : (errcode_for_file_access(),
3543 : : errmsg("could not close file \"%s\": %m", path)));
3544 : :
3545 : : /*
3546 : : * Now move the segment into place with its final name.
3547 : : */
1452 rhaas@postgresql.org 3548 [ - + ]:CBC 40 : if (!InstallXLogFileSegment(&destsegno, tmppath, false, 0, destTLI))
3771 fujii@postgresql.org 3549 [ # # ]:UBC 0 : elog(ERROR, "InstallXLogFileSegment should not have failed");
7768 tgl@sss.pgh.pa.us 3550 :CBC 40 : }
3551 : :
3552 : : /*
3553 : : * Install a new XLOG segment file as a current or future log segment.
3554 : : *
3555 : : * This is used both to install a newly-created segment (which has a temp
3556 : : * filename while it's being created) and to recycle an old segment.
3557 : : *
3558 : : * *segno: identify segment to install as (or first possible target).
3559 : : * When find_free is true, this is modified on return to indicate the
3560 : : * actual installation location or last segment searched.
3561 : : *
3562 : : * tmppath: initial name of file to install. It will be renamed into place.
3563 : : *
3564 : : * find_free: if true, install the new segment at the first empty segno
3565 : : * number at or after the passed numbers. If false, install the new segment
3566 : : * exactly where specified, deleting any existing segment file there.
3567 : : *
3568 : : * max_segno: maximum segment number to install the new file as. Fail if no
3569 : : * free slot is found between *segno and max_segno. (Ignored when find_free
3570 : : * is false.)
3571 : : *
3572 : : * tli: The timeline on which the new segment should be installed.
3573 : : *
3574 : : * Returns true if the file was installed successfully. false indicates that
3575 : : * max_segno limit was exceeded, the startup process has disabled this
3576 : : * function for now, or an error occurred while renaming the file into place.
3577 : : */
3578 : : static bool
4873 heikki.linnakangas@i 3579 : 3002 : InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
3580 : : bool find_free, XLogSegNo max_segno, TimeLineID tli)
3581 : : {
3582 : : char path[MAXPGPATH];
3583 : : struct stat stat_buf;
3584 : :
1452 rhaas@postgresql.org 3585 [ - + ]: 3002 : Assert(tli != 0);
3586 : :
3587 : 3002 : XLogFilePath(path, tli, *segno, wal_segment_size);
3588 : :
1582 noah@leadboat.com 3589 : 3002 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
3590 [ - + ]: 3002 : if (!XLogCtl->InstallXLogFileSegmentActive)
3591 : : {
1582 noah@leadboat.com 3592 :UBC 0 : LWLockRelease(ControlFileLock);
3593 : 0 : return false;
3594 : : }
3595 : :
8866 tgl@sss.pgh.pa.us 3596 [ + + ]:CBC 3002 : if (!find_free)
3597 : : {
3598 : : /* Force installation: get rid of any pre-existing segment file */
3136 teodor@sigaev.ru 3599 : 40 : durable_unlink(path, DEBUG1);
3600 : : }
3601 : : else
3602 : : {
3603 : : /* Find a free slot to put it in */
8311 tgl@sss.pgh.pa.us 3604 [ + + ]: 4130 : while (stat(path, &stat_buf) == 0)
3605 : : {
3899 heikki.linnakangas@i 3606 [ + + ]: 1391 : if ((*segno) >= max_segno)
3607 : : {
3608 : : /* Failed to find a free slot within specified range */
1582 noah@leadboat.com 3609 : 223 : LWLockRelease(ControlFileLock);
8866 tgl@sss.pgh.pa.us 3610 : 223 : return false;
3611 : : }
4873 heikki.linnakangas@i 3612 : 1168 : (*segno)++;
1452 rhaas@postgresql.org 3613 : 1168 : XLogFilePath(path, tli, *segno, wal_segment_size);
3614 : : }
3615 : : }
3616 : :
1210 michael@paquier.xyz 3617 [ + - - + ]: 2779 : Assert(access(path, F_OK) != 0 && errno == ENOENT);
3618 [ - + ]: 2779 : if (durable_rename(tmppath, path, LOG) != 0)
3619 : : {
1582 noah@leadboat.com 3620 :UBC 0 : LWLockRelease(ControlFileLock);
3621 : : /* durable_rename already emitted log message */
5888 heikki.linnakangas@i 3622 : 0 : return false;
3623 : : }
3624 : :
1582 noah@leadboat.com 3625 :CBC 2779 : LWLockRelease(ControlFileLock);
3626 : :
8866 tgl@sss.pgh.pa.us 3627 : 2779 : return true;
3628 : : }
3629 : :
3630 : : /*
3631 : : * Open a pre-existing logfile segment for writing.
3632 : : */
3633 : : int
1452 rhaas@postgresql.org 3634 : 170 : XLogFileOpen(XLogSegNo segno, TimeLineID tli)
3635 : : {
3636 : : char path[MAXPGPATH];
3637 : : int fd;
3638 : :
3639 : 170 : XLogFilePath(path, tli, segno, wal_segment_size);
3640 : :
969 tmunro@postgresql.or 3641 : 170 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
745 nathan@postgresql.or 3642 : 170 : get_sync_bit(wal_sync_method));
9527 vadim4o@yahoo.com 3643 [ - + ]: 170 : if (fd < 0)
8134 tgl@sss.pgh.pa.us 3644 [ # # ]:UBC 0 : ereport(PANIC,
3645 : : (errcode_for_file_access(),
3646 : : errmsg("could not open file \"%s\": %m", path)));
3647 : :
7768 tgl@sss.pgh.pa.us 3648 :CBC 170 : return fd;
3649 : : }
3650 : :
3651 : : /*
3652 : : * Close the current logfile segment for writing.
3653 : : */
3654 : : static void
7074 bruce@momjian.us 3655 : 6439 : XLogFileClose(void)
3656 : : {
3657 [ - + ]: 6439 : Assert(openLogFile >= 0);
3658 : :
3659 : : /*
3660 : : * WAL segment files will not be re-read in normal operation, so we advise
3661 : : * the OS to release any cached pages. But do not do so if WAL archiving
3662 : : * or streaming is active, because archiver and walsender process could
3663 : : * use the cache to read the WAL segment.
3664 : : */
3665 : : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
933 tmunro@postgresql.or 3666 [ + + + - ]: 6439 : if (!XLogIsNeeded() && (io_direct_flags & IO_DIRECT_WAL) == 0)
6133 tgl@sss.pgh.pa.us 3667 : 1597 : (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
3668 : : #endif
3669 : :
2305 peter@eisentraut.org 3670 [ - + ]: 6439 : if (close(openLogFile) != 0)
3671 : : {
3672 : : char xlogfname[MAXFNAMELEN];
2155 michael@paquier.xyz 3673 :UBC 0 : int save_errno = errno;
3674 : :
1452 rhaas@postgresql.org 3675 : 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo, wal_segment_size);
2155 michael@paquier.xyz 3676 : 0 : errno = save_errno;
7074 bruce@momjian.us 3677 [ # # ]: 0 : ereport(PANIC,
3678 : : (errcode_for_file_access(),
3679 : : errmsg("could not close file \"%s\": %m", xlogfname)));
3680 : : }
3681 : :
7074 bruce@momjian.us 3682 :CBC 6439 : openLogFile = -1;
2072 tgl@sss.pgh.pa.us 3683 : 6439 : ReleaseExternalFD();
7074 bruce@momjian.us 3684 : 6439 : }
3685 : :
3686 : : /*
3687 : : * Preallocate log files beyond the specified log endpoint.
3688 : : *
3689 : : * XXX this is currently extremely conservative, since it forces only one
3690 : : * future log segment to exist, and even that only if we are 75% done with
3691 : : * the current one. This is only appropriate for very low-WAL-volume systems.
3692 : : * High-volume systems will be OK once they've built up a sufficient set of
3693 : : * recycled log segments, but the startup transient is likely to include
3694 : : * a lot of segment creations by foreground processes, which is not so good.
3695 : : *
3696 : : * XLogFileInitInternal() can ereport(ERROR). All known causes indicate big
3697 : : * trouble; for example, a full filesystem is one cause. The checkpoint WAL
3698 : : * and/or ControlFile updates already completed. If a RequestCheckpoint()
3699 : : * initiated the present checkpoint and an ERROR ends this function, the
3700 : : * command that called RequestCheckpoint() fails. That's not ideal, but it's
3701 : : * not worth contorting more functions to use caller-specified elevel values.
3702 : : * (With or without RequestCheckpoint(), an ERROR forestalls some inessential
3703 : : * reporting and resource reclamation.)
3704 : : */
3705 : : static void
1452 rhaas@postgresql.org 3706 : 1952 : PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli)
3707 : : {
3708 : : XLogSegNo _logSegNo;
3709 : : int lf;
3710 : : bool added;
3711 : : char path[MAXPGPATH];
3712 : : uint64 offset;
3713 : :
1582 noah@leadboat.com 3714 [ + + ]: 1952 : if (!XLogCtl->InstallXLogFileSegmentActive)
3715 : 10 : return; /* unlocked check says no */
3716 : :
2960 andres@anarazel.de 3717 : 1942 : XLByteToPrevSeg(endptr, _logSegNo, wal_segment_size);
3718 : 1942 : offset = XLogSegmentOffset(endptr - 1, wal_segment_size);
3719 [ + + ]: 1942 : if (offset >= (uint32) (0.75 * wal_segment_size))
3720 : : {
4873 heikki.linnakangas@i 3721 : 217 : _logSegNo++;
1452 rhaas@postgresql.org 3722 : 217 : lf = XLogFileInitInternal(_logSegNo, tli, &added, path);
1582 noah@leadboat.com 3723 [ + + ]: 217 : if (lf >= 0)
3724 : 164 : close(lf);
3725 [ + + ]: 217 : if (added)
6694 tgl@sss.pgh.pa.us 3726 : 53 : CheckpointStats.ckpt_segs_added++;
3727 : : }
3728 : : }
3729 : :
3730 : : /*
3731 : : * Throws an error if the given log segment has already been removed or
3732 : : * recycled. The caller should only pass a segment that it knows to have
3733 : : * existed while the server has been running, as this function always
3734 : : * succeeds if no WAL segments have been removed since startup.
3735 : : * 'tli' is only used in the error message.
3736 : : *
3737 : : * Note: this function guarantees to keep errno unchanged on return.
3738 : : * This supports callers that use this to possibly deliver a better
3739 : : * error message about a missing file, while still being able to throw
3740 : : * a normal file-access error afterwards, if this does return.
3741 : : */
3742 : : void
4680 heikki.linnakangas@i 3743 : 123464 : CheckXLogRemoved(XLogSegNo segno, TimeLineID tli)
3744 : : {
2884 tgl@sss.pgh.pa.us 3745 : 123464 : int save_errno = errno;
3746 : : XLogSegNo lastRemovedSegNo;
3747 : :
4053 andres@anarazel.de 3748 [ + + ]: 123464 : SpinLockAcquire(&XLogCtl->info_lck);
3749 : 123464 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3750 : 123464 : SpinLockRelease(&XLogCtl->info_lck);
3751 : :
4680 heikki.linnakangas@i 3752 [ - + ]: 123464 : if (segno <= lastRemovedSegNo)
3753 : : {
3754 : : char filename[MAXFNAMELEN];
3755 : :
2960 andres@anarazel.de 3756 :UBC 0 : XLogFileName(filename, tli, segno, wal_segment_size);
2884 tgl@sss.pgh.pa.us 3757 : 0 : errno = save_errno;
4680 heikki.linnakangas@i 3758 [ # # ]: 0 : ereport(ERROR,
3759 : : (errcode_for_file_access(),
3760 : : errmsg("requested WAL segment %s has already been removed",
3761 : : filename)));
3762 : : }
2884 tgl@sss.pgh.pa.us 3763 :CBC 123464 : errno = save_errno;
5677 heikki.linnakangas@i 3764 : 123464 : }
3765 : :
3766 : : /*
3767 : : * Return the last WAL segment removed, or 0 if no segment has been removed
3768 : : * since startup.
3769 : : *
3770 : : * NB: the result can be out of date arbitrarily fast, the caller has to deal
3771 : : * with that.
3772 : : */
3773 : : XLogSegNo
4256 rhaas@postgresql.org 3774 : 970 : XLogGetLastRemovedSegno(void)
3775 : : {
3776 : : XLogSegNo lastRemovedSegNo;
3777 : :
4053 andres@anarazel.de 3778 [ - + ]: 970 : SpinLockAcquire(&XLogCtl->info_lck);
3779 : 970 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3780 : 970 : SpinLockRelease(&XLogCtl->info_lck);
3781 : :
4256 rhaas@postgresql.org 3782 : 970 : return lastRemovedSegNo;
3783 : : }
3784 : :
3785 : : /*
3786 : : * Return the oldest WAL segment on the given TLI that still exists in
3787 : : * XLOGDIR, or 0 if none.
3788 : : */
3789 : : XLogSegNo
677 3790 : 5 : XLogGetOldestSegno(TimeLineID tli)
3791 : : {
3792 : : DIR *xldir;
3793 : : struct dirent *xlde;
3794 : 5 : XLogSegNo oldest_segno = 0;
3795 : :
3796 : 5 : xldir = AllocateDir(XLOGDIR);
3797 [ + + ]: 34 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3798 : : {
3799 : : TimeLineID file_tli;
3800 : : XLogSegNo file_segno;
3801 : :
3802 : : /* Ignore files that are not XLOG segments. */
3803 [ + + ]: 29 : if (!IsXLogFileName(xlde->d_name))
3804 : 20 : continue;
3805 : :
3806 : : /* Parse filename to get TLI and segno. */
3807 : 9 : XLogFromFileName(xlde->d_name, &file_tli, &file_segno,
3808 : : wal_segment_size);
3809 : :
3810 : : /* Ignore anything that's not from the TLI of interest. */
3811 [ - + ]: 9 : if (tli != file_tli)
677 rhaas@postgresql.org 3812 :UBC 0 : continue;
3813 : :
3814 : : /* If it's the oldest so far, update oldest_segno. */
677 rhaas@postgresql.org 3815 [ + + + + ]:CBC 9 : if (oldest_segno == 0 || file_segno < oldest_segno)
3816 : 6 : oldest_segno = file_segno;
3817 : : }
3818 : :
3819 : 5 : FreeDir(xldir);
3820 : 5 : return oldest_segno;
3821 : : }
3822 : :
3823 : : /*
3824 : : * Update the last removed segno pointer in shared memory, to reflect that the
3825 : : * given XLOG file has been removed.
3826 : : */
3827 : : static void
5677 heikki.linnakangas@i 3828 : 2573 : UpdateLastRemovedPtr(char *filename)
3829 : : {
3830 : : uint32 tli;
3831 : : XLogSegNo segno;
3832 : :
2960 andres@anarazel.de 3833 : 2573 : XLogFromFileName(filename, &tli, &segno, wal_segment_size);
3834 : :
4053 3835 [ - + ]: 2573 : SpinLockAcquire(&XLogCtl->info_lck);
3836 [ + + ]: 2573 : if (segno > XLogCtl->lastRemovedSegNo)
3837 : 1155 : XLogCtl->lastRemovedSegNo = segno;
3838 : 2573 : SpinLockRelease(&XLogCtl->info_lck);
5677 heikki.linnakangas@i 3839 : 2573 : }
3840 : :
3841 : : /*
3842 : : * Remove all temporary log files in pg_wal
3843 : : *
3844 : : * This is called at the beginning of recovery after a previous crash,
3845 : : * at a point where no other processes write fresh WAL data.
3846 : : */
3847 : : static void
2663 michael@paquier.xyz 3848 : 170 : RemoveTempXlogFiles(void)
3849 : : {
3850 : : DIR *xldir;
3851 : : struct dirent *xlde;
3852 : :
3853 [ + + ]: 170 : elog(DEBUG2, "removing all temporary WAL segments");
3854 : :
3855 : 170 : xldir = AllocateDir(XLOGDIR);
3856 [ + + ]: 1099 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3857 : : {
3858 : : char path[MAXPGPATH];
3859 : :
3860 [ + - ]: 929 : if (strncmp(xlde->d_name, "xlogtemp.", 9) != 0)
3861 : 929 : continue;
3862 : :
2663 michael@paquier.xyz 3863 :UBC 0 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
3864 : 0 : unlink(path);
3865 [ # # ]: 0 : elog(DEBUG2, "removed temporary WAL segment \"%s\"", path);
3866 : : }
2663 michael@paquier.xyz 3867 :CBC 170 : FreeDir(xldir);
3868 : 170 : }
3869 : :
3870 : : /*
3871 : : * Recycle or remove all log files older or equal to passed segno.
3872 : : *
3873 : : * endptr is current (or recent) end of xlog, and lastredoptr is the
3874 : : * redo pointer of the last checkpoint. These are used to determine
3875 : : * whether we want to recycle rather than delete no-longer-wanted log files.
3876 : : *
3877 : : * insertTLI is the current timeline for XLOG insertion. Any recycled
3878 : : * segments should be reused for this timeline.
3879 : : */
3880 : : static void
1452 rhaas@postgresql.org 3881 : 1701 : RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr,
3882 : : TimeLineID insertTLI)
3883 : : {
3884 : : DIR *xldir;
3885 : : struct dirent *xlde;
3886 : : char lastoff[MAXFNAMELEN];
3887 : : XLogSegNo endlogSegNo;
3888 : : XLogSegNo recycleSegNo;
3889 : :
3890 : : /* Initialize info about where to try to recycle to */
1746 michael@paquier.xyz 3891 : 1701 : XLByteToSeg(endptr, endlogSegNo, wal_segment_size);
3892 : 1701 : recycleSegNo = XLOGfileslop(lastredoptr);
3893 : :
3894 : : /*
3895 : : * Construct a filename of the last segment to be kept. The timeline ID
3896 : : * doesn't matter, we ignore that in the comparison. (During recovery,
3897 : : * InsertTimeLineID isn't set, so we can't use that.)
3898 : : */
2960 andres@anarazel.de 3899 : 1701 : XLogFileName(lastoff, 0, segno, wal_segment_size);
3900 : :
5537 simon@2ndQuadrant.co 3901 [ + + ]: 1701 : elog(DEBUG2, "attempting to remove WAL segments older than log file %s",
3902 : : lastoff);
3903 : :
2884 tgl@sss.pgh.pa.us 3904 : 1701 : xldir = AllocateDir(XLOGDIR);
3905 : :
7420 3906 [ + + ]: 38698 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3907 : : {
3908 : : /* Ignore files that are not XLOG segments */
3825 heikki.linnakangas@i 3909 [ + + ]: 36997 : if (!IsXLogFileName(xlde->d_name) &&
3910 [ + + ]: 7242 : !IsPartialXLogFileName(xlde->d_name))
3850 3911 : 7238 : continue;
3912 : :
3913 : : /*
3914 : : * We ignore the timeline part of the XLOG segment identifiers in
3915 : : * deciding whether a segment is still needed. This ensures that we
3916 : : * won't prematurely remove a segment from a parent timeline. We could
3917 : : * probably be a little more proactive about removing segments of
3918 : : * non-parent timelines, but that would be a whole lot more
3919 : : * complicated.
3920 : : *
3921 : : * We use the alphanumeric sorting property of the filenames to decide
3922 : : * which ones are earlier than the lastoff segment.
3923 : : */
3924 [ + + ]: 29759 : if (strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
3925 : : {
4637 3926 [ + + ]: 23400 : if (XLogArchiveCheckDone(xlde->d_name))
3927 : : {
3928 : : /* Update the last removed location in shared memory first */
5677 3929 : 2573 : UpdateLastRemovedPtr(xlde->d_name);
3930 : :
1151 michael@paquier.xyz 3931 : 2573 : RemoveXlogFile(xlde, recycleSegNo, &endlogSegNo, insertTLI);
3932 : : }
3933 : : }
3934 : : }
3935 : :
3850 heikki.linnakangas@i 3936 : 1701 : FreeDir(xldir);
3937 : 1701 : }
3938 : :
3939 : : /*
3940 : : * Recycle or remove WAL files that are not part of the given timeline's
3941 : : * history.
3942 : : *
3943 : : * This is called during recovery, whenever we switch to follow a new
3944 : : * timeline, and at the end of recovery when we create a new timeline. We
3945 : : * wouldn't otherwise care about extra WAL files lying in pg_wal, but they
3946 : : * might be leftover pre-allocated or recycled WAL segments on the old timeline
3947 : : * that we haven't used yet, and contain garbage. If we just leave them in
3948 : : * pg_wal, they will eventually be archived, and we can't let that happen.
3949 : : * Files that belong to our timeline history are valid, because we have
3950 : : * successfully replayed them, but from others we can't be sure.
3951 : : *
3952 : : * 'switchpoint' is the current point in WAL where we switch to new timeline,
3953 : : * and 'newTLI' is the new timeline we switch to.
3954 : : */
3955 : : void
3956 : 58 : RemoveNonParentXlogFiles(XLogRecPtr switchpoint, TimeLineID newTLI)
3957 : : {
3958 : : DIR *xldir;
3959 : : struct dirent *xlde;
3960 : : char switchseg[MAXFNAMELEN];
3961 : : XLogSegNo endLogSegNo;
3962 : : XLogSegNo switchLogSegNo;
3963 : : XLogSegNo recycleSegNo;
3964 : :
3965 : : /*
3966 : : * Initialize info about where to begin the work. This will recycle,
3967 : : * somewhat arbitrarily, 10 future segments.
3968 : : */
1746 michael@paquier.xyz 3969 : 58 : XLByteToPrevSeg(switchpoint, switchLogSegNo, wal_segment_size);
3970 : 58 : XLByteToSeg(switchpoint, endLogSegNo, wal_segment_size);
3971 : 58 : recycleSegNo = endLogSegNo + 10;
3972 : :
3973 : : /*
3974 : : * Construct a filename of the last segment to be kept.
3975 : : */
3976 : 58 : XLogFileName(switchseg, newTLI, switchLogSegNo, wal_segment_size);
3977 : :
3850 heikki.linnakangas@i 3978 [ + + ]: 58 : elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
3979 : : switchseg);
3980 : :
2884 tgl@sss.pgh.pa.us 3981 : 58 : xldir = AllocateDir(XLOGDIR);
3982 : :
3850 heikki.linnakangas@i 3983 [ + + ]: 555 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3984 : : {
3985 : : /* Ignore files that are not XLOG segments */
3825 3986 [ + + ]: 497 : if (!IsXLogFileName(xlde->d_name))
3850 3987 : 308 : continue;
3988 : :
3989 : : /*
3990 : : * Remove files that are on a timeline older than the new one we're
3991 : : * switching to, but with a segment number >= the first segment on the
3992 : : * new timeline.
3993 : : */
3994 [ + + ]: 189 : if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
3995 [ + + ]: 123 : strcmp(xlde->d_name + 8, switchseg + 8) > 0)
3996 : : {
3997 : : /*
3998 : : * If the file has already been marked as .ready, however, don't
3999 : : * remove it yet. It should be OK to remove it - files that are
4000 : : * not part of our timeline history are not required for recovery
4001 : : * - but seems safer to let them be archived and removed later.
4002 : : */
4003 [ + - ]: 15 : if (!XLogArchiveIsReady(xlde->d_name))
1151 michael@paquier.xyz 4004 : 15 : RemoveXlogFile(xlde, recycleSegNo, &endLogSegNo, newTLI);
4005 : : }
4006 : : }
4007 : :
3850 heikki.linnakangas@i 4008 : 58 : FreeDir(xldir);
4009 : 58 : }
4010 : :
4011 : : /*
4012 : : * Recycle or remove a log file that's no longer needed.
4013 : : *
4014 : : * segment_de is the dirent structure of the segment to recycle or remove.
4015 : : * recycleSegNo is the segment number to recycle up to. endlogSegNo is
4016 : : * the segment number of the current (or recent) end of WAL.
4017 : : *
4018 : : * endlogSegNo gets incremented if the segment is recycled so as it is not
4019 : : * checked again with future callers of this function.
4020 : : *
4021 : : * insertTLI is the current timeline for XLOG insertion. Any recycled segments
4022 : : * should be used for this timeline.
4023 : : */
4024 : : static void
1151 michael@paquier.xyz 4025 : 2588 : RemoveXlogFile(const struct dirent *segment_de,
4026 : : XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo,
4027 : : TimeLineID insertTLI)
4028 : : {
4029 : : char path[MAXPGPATH];
4030 : : #ifdef WIN32
4031 : : char newpath[MAXPGPATH];
4032 : : #endif
4033 : 2588 : const char *segname = segment_de->d_name;
4034 : :
3850 heikki.linnakangas@i 4035 : 2588 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", segname);
4036 : :
4037 : : /*
4038 : : * Before deleting the file, see if it can be recycled as a future log
4039 : : * segment. Only recycle normal files, because we don't want to recycle
4040 : : * symbolic links pointing to a separate archive directory.
4041 : : */
2400 tmunro@postgresql.or 4042 [ + - ]: 2588 : if (wal_recycle &&
1746 michael@paquier.xyz 4043 [ + + ]: 2588 : *endlogSegNo <= recycleSegNo &&
1582 noah@leadboat.com 4044 [ + + + - ]: 3598 : XLogCtl->InstallXLogFileSegmentActive && /* callee rechecks this */
1151 michael@paquier.xyz 4045 [ + + ]: 3272 : get_dirent_type(path, segment_de, false, DEBUG2) == PGFILETYPE_REG &&
1746 4046 : 1636 : InstallXLogFileSegment(endlogSegNo, path,
4047 : : true, recycleSegNo, insertTLI))
4048 : : {
3850 heikki.linnakangas@i 4049 [ + + ]: 1413 : ereport(DEBUG2,
4050 : : (errmsg_internal("recycled write-ahead log file \"%s\"",
4051 : : segname)));
4052 : 1413 : CheckpointStats.ckpt_segs_recycled++;
4053 : : /* Needn't recheck that slot on future iterations */
1746 michael@paquier.xyz 4054 : 1413 : (*endlogSegNo)++;
4055 : : }
4056 : : else
4057 : : {
4058 : : /* No need for any more future segments, or recycling failed ... */
4059 : : int rc;
4060 : :
3850 heikki.linnakangas@i 4061 [ + + ]: 1175 : ereport(DEBUG2,
4062 : : (errmsg_internal("removing write-ahead log file \"%s\"",
4063 : : segname)));
4064 : :
4065 : : #ifdef WIN32
4066 : :
4067 : : /*
4068 : : * On Windows, if another process (e.g another backend) holds the file
4069 : : * open in FILE_SHARE_DELETE mode, unlink will succeed, but the file
4070 : : * will still show up in directory listing until the last handle is
4071 : : * closed. To avoid confusing the lingering deleted file for a live
4072 : : * WAL file that needs to be archived, rename it before deleting it.
4073 : : *
4074 : : * If another process holds the file open without FILE_SHARE_DELETE
4075 : : * flag, rename will fail. We'll try again at the next checkpoint.
4076 : : */
4077 : : snprintf(newpath, MAXPGPATH, "%s.deleted", path);
4078 : : if (rename(path, newpath) != 0)
4079 : : {
4080 : : ereport(LOG,
4081 : : (errcode_for_file_access(),
4082 : : errmsg("could not rename file \"%s\": %m",
4083 : : path)));
4084 : : return;
4085 : : }
4086 : : rc = durable_unlink(newpath, LOG);
4087 : : #else
3136 teodor@sigaev.ru 4088 : 1175 : rc = durable_unlink(path, LOG);
4089 : : #endif
3850 heikki.linnakangas@i 4090 [ - + ]: 1175 : if (rc != 0)
4091 : : {
4092 : : /* Message already logged by durable_unlink() */
3850 heikki.linnakangas@i 4093 :UBC 0 : return;
4094 : : }
3850 heikki.linnakangas@i 4095 :CBC 1175 : CheckpointStats.ckpt_segs_removed++;
4096 : : }
4097 : :
4098 : 2588 : XLogArchiveCleanup(segname);
4099 : : }
4100 : :
4101 : : /*
4102 : : * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
4103 : : * If the latter do not exist, recreate them.
4104 : : *
4105 : : * It is not the goal of this function to verify the contents of these
4106 : : * directories, but to help in cases where someone has performed a cluster
4107 : : * copy for PITR purposes but omitted pg_wal from the copy.
4108 : : *
4109 : : * We could also recreate pg_wal if it doesn't exist, but a deliberate
4110 : : * policy decision was made not to. It is fairly common for pg_wal to be
4111 : : * a symlink, and if that was the DBA's intent then automatically making a
4112 : : * plain directory would result in degraded performance with no notice.
4113 : : */
4114 : : static void
6196 tgl@sss.pgh.pa.us 4115 : 907 : ValidateXLOGDirectoryStructure(void)
4116 : : {
4117 : : char path[MAXPGPATH];
4118 : : struct stat stat_buf;
4119 : :
4120 : : /* Check for pg_wal; if it doesn't exist, error out */
4121 [ + - ]: 907 : if (stat(XLOGDIR, &stat_buf) != 0 ||
4122 [ - + ]: 907 : !S_ISDIR(stat_buf.st_mode))
5982 bruce@momjian.us 4123 [ # # ]:UBC 0 : ereport(FATAL,
4124 : : (errcode_for_file_access(),
4125 : : errmsg("required WAL directory \"%s\" does not exist",
4126 : : XLOGDIR)));
4127 : :
4128 : : /* Check for archive_status */
6196 tgl@sss.pgh.pa.us 4129 :CBC 907 : snprintf(path, MAXPGPATH, XLOGDIR "/archive_status");
4130 [ + + ]: 907 : if (stat(path, &stat_buf) == 0)
4131 : : {
4132 : : /* Check for weird cases where it exists but isn't a directory */
4133 [ - + ]: 906 : if (!S_ISDIR(stat_buf.st_mode))
5982 bruce@momjian.us 4134 [ # # ]:UBC 0 : ereport(FATAL,
4135 : : (errcode_for_file_access(),
4136 : : errmsg("required WAL directory \"%s\" does not exist",
4137 : : path)));
4138 : : }
4139 : : else
4140 : : {
6196 tgl@sss.pgh.pa.us 4141 [ + - ]:CBC 1 : ereport(LOG,
4142 : : (errmsg("creating missing WAL directory \"%s\"", path)));
2760 sfrost@snowman.net 4143 [ - + ]: 1 : if (MakePGDirectory(path) < 0)
5982 bruce@momjian.us 4144 [ # # ]:UBC 0 : ereport(FATAL,
4145 : : (errcode_for_file_access(),
4146 : : errmsg("could not create missing directory \"%s\": %m",
4147 : : path)));
4148 : : }
4149 : :
4150 : : /* Check for summaries */
677 rhaas@postgresql.org 4151 :CBC 907 : snprintf(path, MAXPGPATH, XLOGDIR "/summaries");
4152 [ + + ]: 907 : if (stat(path, &stat_buf) == 0)
4153 : : {
4154 : : /* Check for weird cases where it exists but isn't a directory */
4155 [ - + ]: 906 : if (!S_ISDIR(stat_buf.st_mode))
677 rhaas@postgresql.org 4156 [ # # ]:UBC 0 : ereport(FATAL,
4157 : : (errmsg("required WAL directory \"%s\" does not exist",
4158 : : path)));
4159 : : }
4160 : : else
4161 : : {
677 rhaas@postgresql.org 4162 [ + - ]:CBC 1 : ereport(LOG,
4163 : : (errmsg("creating missing WAL directory \"%s\"", path)));
4164 [ - + ]: 1 : if (MakePGDirectory(path) < 0)
677 rhaas@postgresql.org 4165 [ # # ]:UBC 0 : ereport(FATAL,
4166 : : (errmsg("could not create missing directory \"%s\": %m",
4167 : : path)));
4168 : : }
6196 tgl@sss.pgh.pa.us 4169 :CBC 907 : }
4170 : :
4171 : : /*
4172 : : * Remove previous backup history files. This also retries creation of
4173 : : * .ready files for any backup history files for which XLogArchiveNotify
4174 : : * failed earlier.
4175 : : */
4176 : : static void
7067 4177 : 154 : CleanupBackupHistory(void)
4178 : : {
4179 : : DIR *xldir;
4180 : : struct dirent *xlde;
4181 : : char path[MAXPGPATH + sizeof(XLOGDIR)];
4182 : :
7420 4183 : 154 : xldir = AllocateDir(XLOGDIR);
4184 : :
4185 [ + + ]: 1542 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4186 : : {
3825 heikki.linnakangas@i 4187 [ + + ]: 1234 : if (IsBackupHistoryFileName(xlde->d_name))
4188 : : {
6258 tgl@sss.pgh.pa.us 4189 [ + + ]: 162 : if (XLogArchiveCheckDone(xlde->d_name))
4190 : : {
3090 peter_e@gmx.net 4191 [ + + ]: 130 : elog(DEBUG2, "removing WAL backup history file \"%s\"",
4192 : : xlde->d_name);
3121 4193 : 130 : snprintf(path, sizeof(path), XLOGDIR "/%s", xlde->d_name);
7439 bruce@momjian.us 4194 : 130 : unlink(path);
4195 : 130 : XLogArchiveCleanup(xlde->d_name);
4196 : : }
4197 : : }
4198 : : }
4199 : :
4200 : 154 : FreeDir(xldir);
4201 : 154 : }
4202 : :
4203 : : /*
4204 : : * I/O routines for pg_control
4205 : : *
4206 : : * *ControlFile is a buffer in shared memory that holds an image of the
4207 : : * contents of pg_control. WriteControlFile() initializes pg_control
4208 : : * given a preloaded buffer, ReadControlFile() loads the buffer from
4209 : : * the pg_control file (during postmaster or standalone-backend startup),
4210 : : * and UpdateControlFile() rewrites pg_control after we modify xlog state.
4211 : : * InitControlFile() fills the buffer with initial values.
4212 : : *
4213 : : * For simplicity, WriteControlFile() initializes the fields of pg_control
4214 : : * that are related to checking backend/database compatibility, and
4215 : : * ReadControlFile() verifies they are correct. We could split out the
4216 : : * I/O and compatibility-check functions, but there seems no need currently.
4217 : : */
4218 : :
4219 : : static void
461 peter@eisentraut.org 4220 : 50 : InitControlFile(uint64 sysidentifier, uint32 data_checksum_version)
4221 : : {
4222 : : char mock_auth_nonce[MOCK_AUTH_NONCE_LEN];
4223 : :
4224 : : /*
4225 : : * Generate a random nonce. This is used for authentication requests that
4226 : : * will fail because the user does not exist. The nonce is used to create
4227 : : * a genuine-looking password challenge for the non-existent user, in lieu
4228 : : * of an actual stored password.
4229 : : */
1349 heikki.linnakangas@i 4230 [ - + ]: 50 : if (!pg_strong_random(mock_auth_nonce, MOCK_AUTH_NONCE_LEN))
1349 heikki.linnakangas@i 4231 [ # # ]:UBC 0 : ereport(PANIC,
4232 : : (errcode(ERRCODE_INTERNAL_ERROR),
4233 : : errmsg("could not generate secret authorization token")));
4234 : :
1349 heikki.linnakangas@i 4235 :CBC 50 : memset(ControlFile, 0, sizeof(ControlFileData));
4236 : : /* Initialize pg_control status fields */
4237 : 50 : ControlFile->system_identifier = sysidentifier;
4238 : 50 : memcpy(ControlFile->mock_authentication_nonce, mock_auth_nonce, MOCK_AUTH_NONCE_LEN);
4239 : 50 : ControlFile->state = DB_SHUTDOWNED;
4240 : 50 : ControlFile->unloggedLSN = FirstNormalUnloggedLSN;
4241 : :
4242 : : /* Set important parameter values for use when replaying WAL */
2079 peter@eisentraut.org 4243 : 50 : ControlFile->MaxConnections = MaxConnections;
4244 : 50 : ControlFile->max_worker_processes = max_worker_processes;
4245 : 50 : ControlFile->max_wal_senders = max_wal_senders;
4246 : 50 : ControlFile->max_prepared_xacts = max_prepared_xacts;
4247 : 50 : ControlFile->max_locks_per_xact = max_locks_per_xact;
4248 : 50 : ControlFile->wal_level = wal_level;
4249 : 50 : ControlFile->wal_log_hints = wal_log_hints;
4250 : 50 : ControlFile->track_commit_timestamp = track_commit_timestamp;
461 4251 : 50 : ControlFile->data_checksum_version = data_checksum_version;
2079 4252 : 50 : }
4253 : :
4254 : : static void
9102 tgl@sss.pgh.pa.us 4255 : 50 : WriteControlFile(void)
4256 : : {
4257 : : int fd;
4258 : : char buffer[PG_CONTROL_FILE_SIZE]; /* need not be aligned */
4259 : :
4260 : : /*
4261 : : * Initialize version and compatibility-check fields
4262 : : */
8994 4263 : 50 : ControlFile->pg_control_version = PG_CONTROL_VERSION;
4264 : 50 : ControlFile->catalog_version_no = CATALOG_VERSION_NO;
4265 : :
7329 4266 : 50 : ControlFile->maxAlign = MAXIMUM_ALIGNOF;
4267 : 50 : ControlFile->floatFormat = FLOATFORMAT_VALUE;
4268 : :
9102 4269 : 50 : ControlFile->blcksz = BLCKSZ;
4270 : 50 : ControlFile->relseg_size = RELSEG_SIZE;
7147 4271 : 50 : ControlFile->xlog_blcksz = XLOG_BLCKSZ;
2960 andres@anarazel.de 4272 : 50 : ControlFile->xlog_seg_size = wal_segment_size;
4273 : :
8590 lockhart@fourpalms.o 4274 : 50 : ControlFile->nameDataLen = NAMEDATALEN;
7517 tgl@sss.pgh.pa.us 4275 : 50 : ControlFile->indexMaxKeys = INDEX_MAX_KEYS;
4276 : :
6782 4277 : 50 : ControlFile->toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
4162 4278 : 50 : ControlFile->loblksize = LOBLKSIZE;
4279 : :
75 tgl@sss.pgh.pa.us 4280 :GNC 50 : ControlFile->float8ByVal = true; /* vestigial */
4281 : :
4282 : : /*
4283 : : * Initialize the default 'char' signedness.
4284 : : *
4285 : : * The signedness of the char type is implementation-defined. For instance
4286 : : * on x86 architecture CPUs, the char data type is typically treated as
4287 : : * signed by default, whereas on aarch architecture CPUs, it is typically
4288 : : * treated as unsigned by default. In v17 or earlier, we accidentally let
4289 : : * C implementation signedness affect persistent data. This led to
4290 : : * inconsistent results when comparing char data across different
4291 : : * platforms.
4292 : : *
4293 : : * This flag can be used as a hint to ensure consistent behavior for
4294 : : * pre-v18 data files that store data sorted by the 'char' type on disk,
4295 : : * especially in cross-platform replication scenarios.
4296 : : *
4297 : : * Newly created database clusters unconditionally set the default char
4298 : : * signedness to true. pg_upgrade changes this flag for clusters that were
4299 : : * initialized on signedness=false platforms. As a result,
4300 : : * signedness=false setting will become rare over time. If we had known
4301 : : * about this problem during the last development cycle that forced initdb
4302 : : * (v8.3), we would have made all clusters signed or all clusters
4303 : : * unsigned. Making pg_upgrade the only source of signedness=false will
4304 : : * cause the population of database clusters to converge toward that
4305 : : * retrospective ideal.
4306 : : */
248 msawada@postgresql.o 4307 :CBC 50 : ControlFile->default_char_signedness = true;
4308 : :
4309 : : /* Contents are protected with a CRC */
4010 heikki.linnakangas@i 4310 : 50 : INIT_CRC32C(ControlFile->crc);
4311 : 50 : COMP_CRC32C(ControlFile->crc,
4312 : : ControlFile,
4313 : : offsetof(ControlFileData, crc));
4314 : 50 : FIN_CRC32C(ControlFile->crc);
4315 : :
4316 : : /*
4317 : : * We write out PG_CONTROL_FILE_SIZE bytes into pg_control, zero-padding
4318 : : * the excess over sizeof(ControlFileData). This reduces the odds of
4319 : : * premature-EOF errors when reading pg_control. We'll still fail when we
4320 : : * check the contents of the file, but hopefully with a more specific
4321 : : * error than "couldn't read pg_control".
4322 : : */
3022 tgl@sss.pgh.pa.us 4323 : 50 : memset(buffer, 0, PG_CONTROL_FILE_SIZE);
9102 4324 : 50 : memcpy(buffer, ControlFile, sizeof(ControlFileData));
4325 : :
7420 4326 : 50 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4327 : : O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
9102 4328 [ - + ]: 50 : if (fd < 0)
8134 tgl@sss.pgh.pa.us 4329 [ # # ]:UBC 0 : ereport(PANIC,
4330 : : (errcode_for_file_access(),
4331 : : errmsg("could not create file \"%s\": %m",
4332 : : XLOG_CONTROL_FILE)));
4333 : :
8909 tgl@sss.pgh.pa.us 4334 :CBC 50 : errno = 0;
3145 rhaas@postgresql.org 4335 : 50 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE);
3022 tgl@sss.pgh.pa.us 4336 [ - + ]: 50 : if (write(fd, buffer, PG_CONTROL_FILE_SIZE) != PG_CONTROL_FILE_SIZE)
4337 : : {
4338 : : /* if write didn't set errno, assume problem is no disk space */
8909 tgl@sss.pgh.pa.us 4339 [ # # ]:UBC 0 : if (errno == 0)
4340 : 0 : errno = ENOSPC;
8134 4341 [ # # ]: 0 : ereport(PANIC,
4342 : : (errcode_for_file_access(),
4343 : : errmsg("could not write to file \"%s\": %m",
4344 : : XLOG_CONTROL_FILE)));
4345 : : }
3145 rhaas@postgresql.org 4346 :CBC 50 : pgstat_report_wait_end();
4347 : :
4348 : 50 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC);
9089 tgl@sss.pgh.pa.us 4349 [ - + ]: 50 : if (pg_fsync(fd) != 0)
8134 tgl@sss.pgh.pa.us 4350 [ # # ]:UBC 0 : ereport(PANIC,
4351 : : (errcode_for_file_access(),
4352 : : errmsg("could not fsync file \"%s\": %m",
4353 : : XLOG_CONTROL_FILE)));
3145 rhaas@postgresql.org 4354 :CBC 50 : pgstat_report_wait_end();
4355 : :
2305 peter@eisentraut.org 4356 [ - + ]: 50 : if (close(fd) != 0)
7945 tgl@sss.pgh.pa.us 4357 [ # # ]:UBC 0 : ereport(PANIC,
4358 : : (errcode_for_file_access(),
4359 : : errmsg("could not close file \"%s\": %m",
4360 : : XLOG_CONTROL_FILE)));
9102 tgl@sss.pgh.pa.us 4361 :CBC 50 : }
4362 : :
4363 : : static void
4364 : 957 : ReadControlFile(void)
4365 : : {
4366 : : pg_crc32c crc;
4367 : : int fd;
4368 : : char wal_segsz_str[20];
4369 : : int r;
4370 : :
4371 : : /*
4372 : : * Read data...
4373 : : */
7420 4374 : 957 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4375 : : O_RDWR | PG_BINARY);
9102 4376 [ - + ]: 957 : if (fd < 0)
8134 tgl@sss.pgh.pa.us 4377 [ # # ]:UBC 0 : ereport(PANIC,
4378 : : (errcode_for_file_access(),
4379 : : errmsg("could not open file \"%s\": %m",
4380 : : XLOG_CONTROL_FILE)));
4381 : :
3145 rhaas@postgresql.org 4382 :CBC 957 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_READ);
2719 magnus@hagander.net 4383 : 957 : r = read(fd, ControlFile, sizeof(ControlFileData));
4384 [ - + ]: 957 : if (r != sizeof(ControlFileData))
4385 : : {
2719 magnus@hagander.net 4386 [ # # ]:UBC 0 : if (r < 0)
4387 [ # # ]: 0 : ereport(PANIC,
4388 : : (errcode_for_file_access(),
4389 : : errmsg("could not read file \"%s\": %m",
4390 : : XLOG_CONTROL_FILE)));
4391 : : else
4392 [ # # ]: 0 : ereport(PANIC,
4393 : : (errcode(ERRCODE_DATA_CORRUPTED),
4394 : : errmsg("could not read file \"%s\": read %d of %zu",
4395 : : XLOG_CONTROL_FILE, r, sizeof(ControlFileData))));
4396 : : }
3145 rhaas@postgresql.org 4397 :CBC 957 : pgstat_report_wait_end();
4398 : :
9102 tgl@sss.pgh.pa.us 4399 : 957 : close(fd);
4400 : :
4401 : : /*
4402 : : * Check for expected pg_control format version. If this is wrong, the
4403 : : * CRC check will likely fail because we'll be checking the wrong number
4404 : : * of bytes. Complaining about wrong version will probably be more
4405 : : * enlightening than complaining about wrong CRC.
4406 : : */
4407 : :
6489 peter_e@gmx.net 4408 [ - + - - : 957 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION && ControlFile->pg_control_version % 65536 == 0 && ControlFile->pg_control_version / 65536 != 0)
- - ]
6489 peter_e@gmx.net 4409 [ # # ]:UBC 0 : ereport(FATAL,
4410 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4411 : : errmsg("database files are incompatible with server"),
4412 : : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
4413 : : " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
4414 : : ControlFile->pg_control_version, ControlFile->pg_control_version,
4415 : : PG_CONTROL_VERSION, PG_CONTROL_VERSION),
4416 : : errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
4417 : :
8994 tgl@sss.pgh.pa.us 4418 [ - + ]:CBC 957 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION)
8134 tgl@sss.pgh.pa.us 4419 [ # # ]:UBC 0 : ereport(FATAL,
4420 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4421 : : errmsg("database files are incompatible with server"),
4422 : : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
4423 : : " but the server was compiled with PG_CONTROL_VERSION %d.",
4424 : : ControlFile->pg_control_version, PG_CONTROL_VERSION),
4425 : : errhint("It looks like you need to initdb.")));
4426 : :
4427 : : /* Now check the CRC. */
4010 heikki.linnakangas@i 4428 :CBC 957 : INIT_CRC32C(crc);
4429 : 957 : COMP_CRC32C(crc,
4430 : : ControlFile,
4431 : : offsetof(ControlFileData, crc));
4432 : 957 : FIN_CRC32C(crc);
4433 : :
4434 [ - + ]: 957 : if (!EQ_CRC32C(crc, ControlFile->crc))
8134 tgl@sss.pgh.pa.us 4435 [ # # ]:UBC 0 : ereport(FATAL,
4436 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4437 : : errmsg("incorrect checksum in control file")));
4438 : :
4439 : : /*
4440 : : * Do compatibility checking immediately. If the database isn't
4441 : : * compatible with the backend executable, we want to abort before we can
4442 : : * possibly do any damage.
4443 : : */
8994 tgl@sss.pgh.pa.us 4444 [ - + ]:CBC 957 : if (ControlFile->catalog_version_no != CATALOG_VERSION_NO)
8134 tgl@sss.pgh.pa.us 4445 [ # # ]:UBC 0 : ereport(FATAL,
4446 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4447 : : errmsg("database files are incompatible with server"),
4448 : : /* translator: %s is a variable name and %d is its value */
4449 : : errdetail("The database cluster was initialized with %s %d,"
4450 : : " but the server was compiled with %s %d.",
4451 : : "CATALOG_VERSION_NO", ControlFile->catalog_version_no,
4452 : : "CATALOG_VERSION_NO", CATALOG_VERSION_NO),
4453 : : errhint("It looks like you need to initdb.")));
7329 tgl@sss.pgh.pa.us 4454 [ - + ]:CBC 957 : if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
7329 tgl@sss.pgh.pa.us 4455 [ # # ]:UBC 0 : ereport(FATAL,
4456 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4457 : : errmsg("database files are incompatible with server"),
4458 : : /* translator: %s is a variable name and %d is its value */
4459 : : errdetail("The database cluster was initialized with %s %d,"
4460 : : " but the server was compiled with %s %d.",
4461 : : "MAXALIGN", ControlFile->maxAlign,
4462 : : "MAXALIGN", MAXIMUM_ALIGNOF),
4463 : : errhint("It looks like you need to initdb.")));
7329 tgl@sss.pgh.pa.us 4464 [ - + ]:CBC 957 : if (ControlFile->floatFormat != FLOATFORMAT_VALUE)
7329 tgl@sss.pgh.pa.us 4465 [ # # ]:UBC 0 : ereport(FATAL,
4466 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4467 : : errmsg("database files are incompatible with server"),
4468 : : errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
4469 : : errhint("It looks like you need to initdb.")));
9102 tgl@sss.pgh.pa.us 4470 [ - + ]:CBC 957 : if (ControlFile->blcksz != BLCKSZ)
8134 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 : : /* translator: %s is a variable name and %d is its value */
4475 : : errdetail("The database cluster was initialized with %s %d,"
4476 : : " but the server was compiled with %s %d.",
4477 : : "BLCKSZ", ControlFile->blcksz,
4478 : : "BLCKSZ", BLCKSZ),
4479 : : errhint("It looks like you need to recompile or initdb.")));
9102 tgl@sss.pgh.pa.us 4480 [ - + ]:CBC 957 : if (ControlFile->relseg_size != RELSEG_SIZE)
8134 tgl@sss.pgh.pa.us 4481 [ # # ]:UBC 0 : ereport(FATAL,
4482 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4483 : : errmsg("database files are incompatible with server"),
4484 : : /* translator: %s is a variable name and %d is its value */
4485 : : errdetail("The database cluster was initialized with %s %d,"
4486 : : " but the server was compiled with %s %d.",
4487 : : "RELSEG_SIZE", ControlFile->relseg_size,
4488 : : "RELSEG_SIZE", RELSEG_SIZE),
4489 : : errhint("It looks like you need to recompile or initdb.")));
7147 tgl@sss.pgh.pa.us 4490 [ - + ]:CBC 957 : if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
7147 tgl@sss.pgh.pa.us 4491 [ # # ]:UBC 0 : ereport(FATAL,
4492 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4493 : : errmsg("database files are incompatible with server"),
4494 : : /* translator: %s is a variable name and %d is its value */
4495 : : errdetail("The database cluster was initialized with %s %d,"
4496 : : " but the server was compiled with %s %d.",
4497 : : "XLOG_BLCKSZ", ControlFile->xlog_blcksz,
4498 : : "XLOG_BLCKSZ", XLOG_BLCKSZ),
4499 : : errhint("It looks like you need to recompile or initdb.")));
8590 lockhart@fourpalms.o 4500 [ - + ]:CBC 957 : if (ControlFile->nameDataLen != NAMEDATALEN)
8134 tgl@sss.pgh.pa.us 4501 [ # # ]:UBC 0 : ereport(FATAL,
4502 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4503 : : errmsg("database files are incompatible with server"),
4504 : : /* translator: %s is a variable name and %d is its value */
4505 : : errdetail("The database cluster was initialized with %s %d,"
4506 : : " but the server was compiled with %s %d.",
4507 : : "NAMEDATALEN", ControlFile->nameDataLen,
4508 : : "NAMEDATALEN", NAMEDATALEN),
4509 : : errhint("It looks like you need to recompile or initdb.")));
7517 tgl@sss.pgh.pa.us 4510 [ - + ]:CBC 957 : if (ControlFile->indexMaxKeys != INDEX_MAX_KEYS)
8134 tgl@sss.pgh.pa.us 4511 [ # # ]:UBC 0 : ereport(FATAL,
4512 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4513 : : errmsg("database files are incompatible with server"),
4514 : : /* translator: %s is a variable name and %d is its value */
4515 : : errdetail("The database cluster was initialized with %s %d,"
4516 : : " but the server was compiled with %s %d.",
4517 : : "INDEX_MAX_KEYS", ControlFile->indexMaxKeys,
4518 : : "INDEX_MAX_KEYS", INDEX_MAX_KEYS),
4519 : : errhint("It looks like you need to recompile or initdb.")));
6782 tgl@sss.pgh.pa.us 4520 [ - + ]:CBC 957 : if (ControlFile->toast_max_chunk_size != TOAST_MAX_CHUNK_SIZE)
6782 tgl@sss.pgh.pa.us 4521 [ # # ]:UBC 0 : ereport(FATAL,
4522 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4523 : : errmsg("database files are incompatible with server"),
4524 : : /* translator: %s is a variable name and %d is its value */
4525 : : errdetail("The database cluster was initialized with %s %d,"
4526 : : " but the server was compiled with %s %d.",
4527 : : "TOAST_MAX_CHUNK_SIZE", ControlFile->toast_max_chunk_size,
4528 : : "TOAST_MAX_CHUNK_SIZE", (int) TOAST_MAX_CHUNK_SIZE),
4529 : : errhint("It looks like you need to recompile or initdb.")));
4162 tgl@sss.pgh.pa.us 4530 [ - + ]:CBC 957 : if (ControlFile->loblksize != LOBLKSIZE)
4162 tgl@sss.pgh.pa.us 4531 [ # # ]:UBC 0 : ereport(FATAL,
4532 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4533 : : errmsg("database files are incompatible with server"),
4534 : : /* translator: %s is a variable name and %d is its value */
4535 : : errdetail("The database cluster was initialized with %s %d,"
4536 : : " but the server was compiled with %s %d.",
4537 : : "LOBLKSIZE", ControlFile->loblksize,
4538 : : "LOBLKSIZE", (int) LOBLKSIZE),
4539 : : errhint("It looks like you need to recompile or initdb.")));
4540 : :
75 tgl@sss.pgh.pa.us 4541 [ - + ]:GNC 957 : Assert(ControlFile->float8ByVal); /* vestigial, not worth an error msg */
4542 : :
2960 andres@anarazel.de 4543 :CBC 957 : wal_segment_size = ControlFile->xlog_seg_size;
4544 : :
4545 [ + - + - : 957 : if (!IsValidWalSegSize(wal_segment_size))
+ - - + ]
2960 andres@anarazel.de 4546 [ # # ]:UBC 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4547 : : errmsg_plural("invalid WAL segment size in control file (%d byte)",
4548 : : "invalid WAL segment size in control file (%d bytes)",
4549 : : wal_segment_size,
4550 : : wal_segment_size),
4551 : : errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.")));
4552 : :
2960 andres@anarazel.de 4553 :CBC 957 : snprintf(wal_segsz_str, sizeof(wal_segsz_str), "%d", wal_segment_size);
4554 : 957 : SetConfigOption("wal_segment_size", wal_segsz_str, PGC_INTERNAL,
4555 : : PGC_S_DYNAMIC_DEFAULT);
4556 : :
4557 : : /* check and update variables dependent on wal_segment_size */
4558 [ - + ]: 957 : if (ConvertToXSegs(min_wal_size_mb, wal_segment_size) < 2)
2960 andres@anarazel.de 4559 [ # # ]:UBC 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4560 : : /* translator: both %s are GUC names */
4561 : : errmsg("\"%s\" must be at least twice \"%s\"",
4562 : : "min_wal_size", "wal_segment_size")));
4563 : :
2960 andres@anarazel.de 4564 [ - + ]:CBC 957 : if (ConvertToXSegs(max_wal_size_mb, wal_segment_size) < 2)
2960 andres@anarazel.de 4565 [ # # ]:UBC 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4566 : : /* translator: both %s are GUC names */
4567 : : errmsg("\"%s\" must be at least twice \"%s\"",
4568 : : "max_wal_size", "wal_segment_size")));
4569 : :
2960 andres@anarazel.de 4570 :CBC 957 : UsableBytesInSegment =
4571 : 957 : (wal_segment_size / XLOG_BLCKSZ * UsableBytesInPage) -
4572 : : (SizeOfXLogLongPHD - SizeOfXLogShortPHD);
4573 : :
4574 : 957 : CalculateCheckpointSegments();
4575 : :
4576 : : /* Make the initdb settings visible as GUC variables, too */
2758 magnus@hagander.net 4577 [ + + ]: 957 : SetConfigOption("data_checksums", DataChecksumsEnabled() ? "yes" : "no",
4578 : : PGC_INTERNAL, PGC_S_DYNAMIC_DEFAULT);
9102 tgl@sss.pgh.pa.us 4579 : 957 : }
4580 : :
4581 : : /*
4582 : : * Utility wrapper to update the control file. Note that the control
4583 : : * file gets flushed.
4584 : : */
4585 : : static void
4586 : 8961 : UpdateControlFile(void)
4587 : : {
2401 peter@eisentraut.org 4588 : 8961 : update_controlfile(DataDir, ControlFile, true);
9518 vadim4o@yahoo.com 4589 : 8961 : }
4590 : :
4591 : : /*
4592 : : * Returns the unique system identifier from control file.
4593 : : */
4594 : : uint64
5764 heikki.linnakangas@i 4595 : 1353 : GetSystemIdentifier(void)
4596 : : {
4597 [ - + ]: 1353 : Assert(ControlFile != NULL);
4598 : 1353 : return ControlFile->system_identifier;
4599 : : }
4600 : :
4601 : : /*
4602 : : * Returns the random nonce from control file.
4603 : : */
4604 : : char *
3156 4605 : 1 : GetMockAuthenticationNonce(void)
4606 : : {
4607 [ - + ]: 1 : Assert(ControlFile != NULL);
4608 : 1 : return ControlFile->mock_authentication_nonce;
4609 : : }
4610 : :
4611 : : /*
4612 : : * Are checksums enabled for data pages?
4613 : : */
4614 : : bool
2758 magnus@hagander.net 4615 : 9046961 : DataChecksumsEnabled(void)
4616 : : {
4602 simon@2ndQuadrant.co 4617 [ - + ]: 9046961 : Assert(ControlFile != NULL);
4563 4618 : 9046961 : return (ControlFile->data_checksum_version > 0);
4619 : : }
4620 : :
4621 : : /*
4622 : : * Return true if the cluster was initialized on a platform where the
4623 : : * default signedness of char is "signed". This function exists for code
4624 : : * that deals with pre-v18 data files that store data sorted by the 'char'
4625 : : * type on disk (e.g., GIN and GiST indexes). See the comments in
4626 : : * WriteControlFile() for details.
4627 : : */
4628 : : bool
248 msawada@postgresql.o 4629 : 3 : GetDefaultCharSignedness(void)
4630 : : {
4631 : 3 : return ControlFile->default_char_signedness;
4632 : : }
4633 : :
4634 : : /*
4635 : : * Returns a fake LSN for unlogged relations.
4636 : : *
4637 : : * Each call generates an LSN that is greater than any previous value
4638 : : * returned. The current counter value is saved and restored across clean
4639 : : * shutdowns, but like unlogged relations, does not survive a crash. This can
4640 : : * be used in lieu of real LSN values returned by XLogInsert, if you need an
4641 : : * LSN-like increasing sequence of numbers without writing any WAL.
4642 : : */
4643 : : XLogRecPtr
4641 heikki.linnakangas@i 4644 : 33 : GetFakeLSNForUnloggedRel(void)
4645 : : {
606 nathan@postgresql.or 4646 : 33 : return pg_atomic_fetch_add_u64(&XLogCtl->unloggedLSN, 1);
4647 : : }
4648 : :
4649 : : /*
4650 : : * Auto-tune the number of XLOG buffers.
4651 : : *
4652 : : * The preferred setting for wal_buffers is about 3% of shared_buffers, with
4653 : : * a maximum of one XLOG segment (there is little reason to think that more
4654 : : * is helpful, at least so long as we force an fsync when switching log files)
4655 : : * and a minimum of 8 blocks (which was the default value prior to PostgreSQL
4656 : : * 9.1, when auto-tuning was added).
4657 : : *
4658 : : * This should not be called until NBuffers has received its final value.
4659 : : */
4660 : : static int
5317 tgl@sss.pgh.pa.us 4661 : 1048 : XLOGChooseNumBuffers(void)
4662 : : {
4663 : : int xbuffers;
4664 : :
4665 : 1048 : xbuffers = NBuffers / 32;
2960 andres@anarazel.de 4666 [ + + ]: 1048 : if (xbuffers > (wal_segment_size / XLOG_BLCKSZ))
4667 : 24 : xbuffers = (wal_segment_size / XLOG_BLCKSZ);
5317 tgl@sss.pgh.pa.us 4668 [ + + ]: 1048 : if (xbuffers < 8)
4669 : 412 : xbuffers = 8;
4670 : 1048 : return xbuffers;
4671 : : }
4672 : :
4673 : : /*
4674 : : * GUC check_hook for wal_buffers
4675 : : */
4676 : : bool
4677 : 2135 : check_wal_buffers(int *newval, void **extra, GucSource source)
4678 : : {
4679 : : /*
4680 : : * -1 indicates a request for auto-tune.
4681 : : */
4682 [ + + ]: 2135 : if (*newval == -1)
4683 : : {
4684 : : /*
4685 : : * If we haven't yet changed the boot_val default of -1, just let it
4686 : : * be. We'll fix it when XLOGShmemSize is called.
4687 : : */
4688 [ + - ]: 1087 : if (XLOGbuffers == -1)
4689 : 1087 : return true;
4690 : :
4691 : : /* Otherwise, substitute the auto-tune value */
5317 tgl@sss.pgh.pa.us 4692 :UBC 0 : *newval = XLOGChooseNumBuffers();
4693 : : }
4694 : :
4695 : : /*
4696 : : * We clamp manually-set values to at least 4 blocks. Prior to PostgreSQL
4697 : : * 9.1, a minimum of 4 was enforced by guc.c, but since that is no longer
4698 : : * the case, we just silently treat such values as a request for the
4699 : : * minimum. (We could throw an error instead, but that doesn't seem very
4700 : : * helpful.)
4701 : : */
5317 tgl@sss.pgh.pa.us 4702 [ - + ]:CBC 1048 : if (*newval < 4)
5317 tgl@sss.pgh.pa.us 4703 :UBC 0 : *newval = 4;
4704 : :
5317 tgl@sss.pgh.pa.us 4705 :CBC 1048 : return true;
4706 : : }
4707 : :
4708 : : /*
4709 : : * GUC check_hook for wal_consistency_checking
4710 : : */
4711 : : bool
1140 4712 : 1987 : check_wal_consistency_checking(char **newval, void **extra, GucSource source)
4713 : : {
4714 : : char *rawstring;
4715 : : List *elemlist;
4716 : : ListCell *l;
4717 : : bool newwalconsistency[RM_MAX_ID + 1];
4718 : :
4719 : : /* Initialize the array */
4720 [ + - + - : 65571 : MemSet(newwalconsistency, 0, (RM_MAX_ID + 1) * sizeof(bool));
+ - + - +
+ ]
4721 : :
4722 : : /* Need a modifiable copy of string */
4723 : 1987 : rawstring = pstrdup(*newval);
4724 : :
4725 : : /* Parse string into list of identifiers */
4726 [ - + ]: 1987 : if (!SplitIdentifierString(rawstring, ',', &elemlist))
4727 : : {
4728 : : /* syntax error in list */
1140 tgl@sss.pgh.pa.us 4729 :UBC 0 : GUC_check_errdetail("List syntax is invalid.");
4730 : 0 : pfree(rawstring);
4731 : 0 : list_free(elemlist);
4732 : 0 : return false;
4733 : : }
4734 : :
1140 tgl@sss.pgh.pa.us 4735 [ + + + + :CBC 2440 : foreach(l, elemlist)
+ + ]
4736 : : {
4737 : 453 : char *tok = (char *) lfirst(l);
4738 : : int rmid;
4739 : :
4740 : : /* Check for 'all'. */
4741 [ + + ]: 453 : if (pg_strcasecmp(tok, "all") == 0)
4742 : : {
4743 [ + + ]: 115907 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4744 [ + + + + ]: 115456 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL)
4745 : 4510 : newwalconsistency[rmid] = true;
4746 : : }
4747 : : else
4748 : : {
4749 : : /* Check if the token matches any known resource manager. */
4750 : 2 : bool found = false;
4751 : :
4752 [ + - ]: 36 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4753 : : {
4754 [ + - + + : 54 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL &&
+ + ]
4755 : 18 : pg_strcasecmp(tok, GetRmgr(rmid).rm_name) == 0)
4756 : : {
4757 : 2 : newwalconsistency[rmid] = true;
4758 : 2 : found = true;
4759 : 2 : break;
4760 : : }
4761 : : }
4762 [ - + ]: 2 : if (!found)
4763 : : {
4764 : : /*
4765 : : * During startup, it might be a not-yet-loaded custom
4766 : : * resource manager. Defer checking until
4767 : : * InitializeWalConsistencyChecking().
4768 : : */
1140 tgl@sss.pgh.pa.us 4769 [ # # ]:UBC 0 : if (!process_shared_preload_libraries_done)
4770 : : {
4771 : 0 : check_wal_consistency_checking_deferred = true;
4772 : : }
4773 : : else
4774 : : {
4775 : 0 : GUC_check_errdetail("Unrecognized key word: \"%s\".", tok);
4776 : 0 : pfree(rawstring);
4777 : 0 : list_free(elemlist);
4778 : 0 : return false;
4779 : : }
4780 : : }
4781 : : }
4782 : : }
4783 : :
1140 tgl@sss.pgh.pa.us 4784 :CBC 1987 : pfree(rawstring);
4785 : 1987 : list_free(elemlist);
4786 : :
4787 : : /* assign new value */
214 dgustafsson@postgres 4788 : 1987 : *extra = guc_malloc(LOG, (RM_MAX_ID + 1) * sizeof(bool));
4789 [ - + ]: 1987 : if (!*extra)
214 dgustafsson@postgres 4790 :UBC 0 : return false;
1140 tgl@sss.pgh.pa.us 4791 :CBC 1987 : memcpy(*extra, newwalconsistency, (RM_MAX_ID + 1) * sizeof(bool));
4792 : 1987 : return true;
4793 : : }
4794 : :
4795 : : /*
4796 : : * GUC assign_hook for wal_consistency_checking
4797 : : */
4798 : : void
4799 : 1986 : assign_wal_consistency_checking(const char *newval, void *extra)
4800 : : {
4801 : : /*
4802 : : * If some checks were deferred, it's possible that the checks will fail
4803 : : * later during InitializeWalConsistencyChecking(). But in that case, the
4804 : : * postmaster will exit anyway, so it's safe to proceed with the
4805 : : * assignment.
4806 : : *
4807 : : * Any built-in resource managers specified are assigned immediately,
4808 : : * which affects WAL created before shared_preload_libraries are
4809 : : * processed. Any custom resource managers specified won't be assigned
4810 : : * until after shared_preload_libraries are processed, but that's OK
4811 : : * because WAL for a custom resource manager can't be written before the
4812 : : * module is loaded anyway.
4813 : : */
4814 : 1986 : wal_consistency_checking = extra;
4815 : 1986 : }
4816 : :
4817 : : /*
4818 : : * InitializeWalConsistencyChecking: run after loading custom resource managers
4819 : : *
4820 : : * If any unknown resource managers were specified in the
4821 : : * wal_consistency_checking GUC, processing was deferred. Now that
4822 : : * shared_preload_libraries have been loaded, process wal_consistency_checking
4823 : : * again.
4824 : : */
4825 : : void
4826 : 898 : InitializeWalConsistencyChecking(void)
4827 : : {
4828 [ - + ]: 898 : Assert(process_shared_preload_libraries_done);
4829 : :
4830 [ - + ]: 898 : if (check_wal_consistency_checking_deferred)
4831 : : {
4832 : : struct config_generic *guc;
4833 : :
1140 tgl@sss.pgh.pa.us 4834 :UBC 0 : guc = find_option("wal_consistency_checking", false, false, ERROR);
4835 : :
4836 : 0 : check_wal_consistency_checking_deferred = false;
4837 : :
4838 : 0 : set_config_option_ext("wal_consistency_checking",
4839 : : wal_consistency_checking_string,
4840 : : guc->scontext, guc->source, guc->srole,
4841 : : GUC_ACTION_SET, true, ERROR, false);
4842 : :
4843 : : /* checking should not be deferred again */
4844 [ # # ]: 0 : Assert(!check_wal_consistency_checking_deferred);
4845 : : }
1140 tgl@sss.pgh.pa.us 4846 :CBC 898 : }
4847 : :
4848 : : /*
4849 : : * GUC show_hook for archive_command
4850 : : */
4851 : : const char *
4852 : 1684 : show_archive_command(void)
4853 : : {
4854 [ + + - + : 1684 : if (XLogArchivingActive())
+ + ]
4855 : 2 : return XLogArchiveCommand;
4856 : : else
4857 : 1682 : return "(disabled)";
4858 : : }
4859 : :
4860 : : /*
4861 : : * GUC show_hook for in_hot_standby
4862 : : */
4863 : : const char *
4864 : 13900 : show_in_hot_standby(void)
4865 : : {
4866 : : /*
4867 : : * We display the actual state based on shared memory, so that this GUC
4868 : : * reports up-to-date state if examined intra-query. The underlying
4869 : : * variable (in_hot_standby_guc) changes only when we transmit a new value
4870 : : * to the client.
4871 : : */
4872 [ + + ]: 13900 : return RecoveryInProgress() ? "on" : "off";
4873 : : }
4874 : :
4875 : : /*
4876 : : * Read the control file, set respective GUCs.
4877 : : *
4878 : : * This is to be called during startup, including a crash recovery cycle,
4879 : : * unless in bootstrap mode, where no control file yet exists. As there's no
4880 : : * usable shared memory yet (its sizing can depend on the contents of the
4881 : : * control file!), first store the contents in local memory. XLOGShmemInit()
4882 : : * will then copy it to shared memory later.
4883 : : *
4884 : : * reset just controls whether previous contents are to be expected (in the
4885 : : * reset case, there's a dangling pointer into old shared memory), or not.
4886 : : */
4887 : : void
2962 andres@anarazel.de 4888 : 907 : LocalProcessControlFile(bool reset)
4889 : : {
4890 [ + + - + ]: 907 : Assert(reset || ControlFile == NULL);
2966 4891 : 907 : ControlFile = palloc(sizeof(ControlFileData));
4892 : 907 : ReadControlFile();
4893 : 907 : }
4894 : :
4895 : : /*
4896 : : * Get the wal_level from the control file. For a standby, this value should be
4897 : : * considered as its active wal_level, because it may be different from what
4898 : : * was originally configured on standby.
4899 : : */
4900 : : WalLevel
933 4901 : 1 : GetActiveWalLevelOnStandby(void)
4902 : : {
4903 : 1 : return ControlFile->wal_level;
4904 : : }
4905 : :
4906 : : /*
4907 : : * Initialization of shared memory for XLOG
4908 : : */
4909 : : Size
9106 peter_e@gmx.net 4910 : 2998 : XLOGShmemSize(void)
4911 : : {
4912 : : Size size;
4913 : :
4914 : : /*
4915 : : * If the value of wal_buffers is -1, use the preferred auto-tune value.
4916 : : * This isn't an amazingly clean place to do this, but we must wait till
4917 : : * NBuffers has received its final value, and must do it before using the
4918 : : * value of XLOGbuffers to do anything important.
4919 : : *
4920 : : * We prefer to report this value's source as PGC_S_DYNAMIC_DEFAULT.
4921 : : * However, if the DBA explicitly set wal_buffers = -1 in the config file,
4922 : : * then PGC_S_DYNAMIC_DEFAULT will fail to override that and we must force
4923 : : * the matter with PGC_S_OVERRIDE.
4924 : : */
5317 tgl@sss.pgh.pa.us 4925 [ + + ]: 2998 : if (XLOGbuffers == -1)
4926 : : {
4927 : : char buf[32];
4928 : :
4929 : 1048 : snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
1237 4930 : 1048 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4931 : : PGC_S_DYNAMIC_DEFAULT);
4932 [ - + ]: 1048 : if (XLOGbuffers == -1) /* failed to apply it? */
1237 tgl@sss.pgh.pa.us 4933 :UBC 0 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4934 : : PGC_S_OVERRIDE);
4935 : : }
5392 tgl@sss.pgh.pa.us 4936 [ - + ]:CBC 2998 : Assert(XLOGbuffers > 0);
4937 : :
4938 : : /* XLogCtl */
7373 4939 : 2998 : size = sizeof(XLogCtlData);
4940 : :
4941 : : /* WAL insertion locks, plus alignment */
4044 heikki.linnakangas@i 4942 : 2998 : size = add_size(size, mul_size(sizeof(WALInsertLockPadded), NUM_XLOGINSERT_LOCKS + 1));
4943 : : /* xlblocks array */
678 jdavis@postgresql.or 4944 : 2998 : size = add_size(size, mul_size(sizeof(pg_atomic_uint64), XLOGbuffers));
4945 : : /* extra alignment padding for XLOG I/O buffers */
933 tmunro@postgresql.or 4946 : 2998 : size = add_size(size, Max(XLOG_BLCKSZ, PG_IO_ALIGN_SIZE));
4947 : : /* and the buffers themselves */
7147 tgl@sss.pgh.pa.us 4948 : 2998 : size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4949 : :
4950 : : /*
4951 : : * Note: we don't count ControlFileData, it comes out of the "slop factor"
4952 : : * added by CreateSharedMemoryAndSemaphores. This lets us use this
4953 : : * routine again below to compute the actual allocation size.
4954 : : */
4955 : :
7373 4956 : 2998 : return size;
4957 : : }
4958 : :
4959 : : void
9518 vadim4o@yahoo.com 4960 : 1049 : XLOGShmemInit(void)
4961 : : {
4962 : : bool foundCFile,
4963 : : foundXLog;
4964 : : char *allocptr;
4965 : : int i;
4966 : : ControlFileData *localControlFile;
4967 : :
4968 : : #ifdef WAL_DEBUG
4969 : :
4970 : : /*
4971 : : * Create a memory context for WAL debugging that's exempt from the normal
4972 : : * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
4973 : : * an allocation fails, but wal_debug is not for production use anyway.
4974 : : */
4975 : : if (walDebugCxt == NULL)
4976 : : {
4977 : : walDebugCxt = AllocSetContextCreate(TopMemoryContext,
4978 : : "WAL Debug",
4979 : : ALLOCSET_DEFAULT_SIZES);
4980 : : MemoryContextAllowInCriticalSection(walDebugCxt, true);
4981 : : }
4982 : : #endif
4983 : :
4984 : :
2962 andres@anarazel.de 4985 : 1049 : XLogCtl = (XLogCtlData *)
4986 : 1049 : ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
4987 : :
2966 4988 : 1049 : localControlFile = ControlFile;
9102 tgl@sss.pgh.pa.us 4989 : 1049 : ControlFile = (ControlFileData *)
7982 bruce@momjian.us 4990 : 1049 : ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
4991 : :
7371 tgl@sss.pgh.pa.us 4992 [ + - - + ]: 1049 : if (foundCFile || foundXLog)
4993 : : {
4994 : : /* both should be present or neither */
7371 tgl@sss.pgh.pa.us 4995 [ # # # # ]:UBC 0 : Assert(foundCFile && foundXLog);
4996 : :
4997 : : /* Initialize local copy of WALInsertLocks */
4113 rhaas@postgresql.org 4998 : 0 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks;
4999 : :
2962 andres@anarazel.de 5000 [ # # ]: 0 : if (localControlFile)
5001 : 0 : pfree(localControlFile);
7982 bruce@momjian.us 5002 : 0 : return;
5003 : : }
8994 tgl@sss.pgh.pa.us 5004 :CBC 1049 : memset(XLogCtl, 0, sizeof(XLogCtlData));
5005 : :
5006 : : /*
5007 : : * Already have read control file locally, unless in bootstrap mode. Move
5008 : : * contents into shared memory.
5009 : : */
2962 andres@anarazel.de 5010 [ + + ]: 1049 : if (localControlFile)
5011 : : {
5012 : 899 : memcpy(ControlFile, localControlFile, sizeof(ControlFileData));
5013 : 899 : pfree(localControlFile);
5014 : : }
5015 : :
5016 : : /*
5017 : : * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
5018 : : * multiple of the alignment for same, so no extra alignment padding is
5019 : : * needed here.
5020 : : */
4494 heikki.linnakangas@i 5021 : 1049 : allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
678 jdavis@postgresql.or 5022 : 1049 : XLogCtl->xlblocks = (pg_atomic_uint64 *) allocptr;
5023 : 1049 : allocptr += sizeof(pg_atomic_uint64) * XLOGbuffers;
5024 : :
5025 [ + + ]: 296215 : for (i = 0; i < XLOGbuffers; i++)
5026 : : {
5027 : 295166 : pg_atomic_init_u64(&XLogCtl->xlblocks[i], InvalidXLogRecPtr);
5028 : : }
5029 : :
5030 : : /* WAL insertion locks. Ensure they're aligned to the full padded size */
4238 heikki.linnakangas@i 5031 : 1049 : allocptr += sizeof(WALInsertLockPadded) -
3050 tgl@sss.pgh.pa.us 5032 : 1049 : ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
4238 heikki.linnakangas@i 5033 : 1049 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks =
5034 : : (WALInsertLockPadded *) allocptr;
4044 5035 : 1049 : allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
5036 : :
5037 [ + + ]: 9441 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
5038 : : {
3604 rhaas@postgresql.org 5039 : 8392 : LWLockInitialize(&WALInsertLocks[i].l.lock, LWTRANCHE_WAL_INSERT);
825 michael@paquier.xyz 5040 : 8392 : pg_atomic_init_u64(&WALInsertLocks[i].l.insertingAt, InvalidXLogRecPtr);
3231 andres@anarazel.de 5041 : 8392 : WALInsertLocks[i].l.lastImportantAt = InvalidXLogRecPtr;
5042 : : }
5043 : :
5044 : : /*
5045 : : * Align the start of the page buffers to a full xlog block size boundary.
5046 : : * This simplifies some calculations in XLOG insertion. It is also
5047 : : * required for O_DIRECT.
5048 : : */
4494 heikki.linnakangas@i 5049 : 1049 : allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
7373 tgl@sss.pgh.pa.us 5050 : 1049 : XLogCtl->pages = allocptr;
7147 5051 : 1049 : memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
5052 : :
5053 : : /*
5054 : : * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
5055 : : * in additional info.)
5056 : : */
8994 5057 : 1049 : XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
2012 michael@paquier.xyz 5058 : 1049 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
1582 noah@leadboat.com 5059 : 1049 : XLogCtl->InstallXLogFileSegmentActive = false;
4920 tgl@sss.pgh.pa.us 5060 : 1049 : XLogCtl->WalWriterSleeping = false;
5061 : :
4494 heikki.linnakangas@i 5062 : 1049 : SpinLockInit(&XLogCtl->Insert.insertpos_lck);
8794 tgl@sss.pgh.pa.us 5063 : 1049 : SpinLockInit(&XLogCtl->info_lck);
568 alvherre@alvh.no-ip. 5064 : 1049 : pg_atomic_init_u64(&XLogCtl->logInsertResult, InvalidXLogRecPtr);
570 5065 : 1049 : pg_atomic_init_u64(&XLogCtl->logWriteResult, InvalidXLogRecPtr);
5066 : 1049 : pg_atomic_init_u64(&XLogCtl->logFlushResult, InvalidXLogRecPtr);
606 nathan@postgresql.or 5067 : 1049 : pg_atomic_init_u64(&XLogCtl->unloggedLSN, InvalidXLogRecPtr);
5068 : : }
5069 : :
5070 : : /*
5071 : : * This func must be called ONCE on system install. It creates pg_control
5072 : : * and the initial XLOG segment.
5073 : : */
5074 : : void
461 peter@eisentraut.org 5075 : 50 : BootStrapXLOG(uint32 data_checksum_version)
5076 : : {
5077 : : CheckPoint checkPoint;
5078 : : char *buffer;
5079 : : XLogPageHeader page;
5080 : : XLogLongPageHeader longpage;
5081 : : XLogRecord *record;
5082 : : char *recptr;
5083 : : uint64 sysidentifier;
5084 : : struct timeval tv;
5085 : : pg_crc32c crc;
5086 : :
5087 : : /* allow ordinary WAL segment creation, like StartupXLOG() would */
1167 michael@paquier.xyz 5088 : 50 : SetInstallXLogFileSegmentActive();
5089 : :
5090 : : /*
5091 : : * Select a hopefully-unique system identifier code for this installation.
5092 : : * We use the result of gettimeofday(), including the fractional seconds
5093 : : * field, as being about as unique as we can easily get. (Think not to
5094 : : * use random(), since it hasn't been seeded and there's no portable way
5095 : : * to seed it other than the system clock value...) The upper half of the
5096 : : * uint64 value is just the tv_sec part, while the lower half contains the
5097 : : * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
5098 : : * PID for a little extra uniqueness. A person knowing this encoding can
5099 : : * determine the initialization time of the installation, which could
5100 : : * perhaps be useful sometimes.
5101 : : */
7929 tgl@sss.pgh.pa.us 5102 : 50 : gettimeofday(&tv, NULL);
5103 : 50 : sysidentifier = ((uint64) tv.tv_sec) << 32;
4202 5104 : 50 : sysidentifier |= ((uint64) tv.tv_usec) << 12;
5105 : 50 : sysidentifier |= getpid() & 0xFFF;
5106 : :
5107 : : /* page buffer must be aligned suitably for O_DIRECT */
4494 heikki.linnakangas@i 5108 : 50 : buffer = (char *) palloc(XLOG_BLCKSZ + XLOG_BLCKSZ);
5109 : 50 : page = (XLogPageHeader) TYPEALIGN(XLOG_BLCKSZ, buffer);
7147 tgl@sss.pgh.pa.us 5110 : 50 : memset(page, 0, XLOG_BLCKSZ);
5111 : :
5112 : : /*
5113 : : * Set up information for the initial checkpoint record
5114 : : *
5115 : : * The initial checkpoint record is written to the beginning of the WAL
5116 : : * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
5117 : : * used, so that we can use 0/0 to mean "before any valid WAL segment".
5118 : : */
2960 andres@anarazel.de 5119 : 50 : checkPoint.redo = wal_segment_size + SizeOfXLogLongPHD;
1452 rhaas@postgresql.org 5120 : 50 : checkPoint.ThisTimeLineID = BootstrapTimeLineID;
5121 : 50 : checkPoint.PrevTimeLineID = BootstrapTimeLineID;
5024 simon@2ndQuadrant.co 5122 : 50 : checkPoint.fullPageWrites = fullPageWrites;
462 rhaas@postgresql.org 5123 : 50 : checkPoint.wal_level = wal_level;
5124 : : checkPoint.nextXid =
2405 tmunro@postgresql.or 5125 : 50 : FullTransactionIdFromEpochAndXid(0, FirstNormalTransactionId);
1565 tgl@sss.pgh.pa.us 5126 : 50 : checkPoint.nextOid = FirstGenbkiObjectId;
7487 5127 : 50 : checkPoint.nextMulti = FirstMultiXactId;
7446 5128 : 50 : checkPoint.nextMultiOffset = 0;
5901 5129 : 50 : checkPoint.oldestXid = FirstNormalTransactionId;
1285 5130 : 50 : checkPoint.oldestXidDB = Template1DbOid;
4660 alvherre@alvh.no-ip. 5131 : 50 : checkPoint.oldestMulti = FirstMultiXactId;
1285 tgl@sss.pgh.pa.us 5132 : 50 : checkPoint.oldestMultiDB = Template1DbOid;
3591 mail@joeconway.com 5133 : 50 : checkPoint.oldestCommitTsXid = InvalidTransactionId;
5134 : 50 : checkPoint.newestCommitTsXid = InvalidTransactionId;
6462 tgl@sss.pgh.pa.us 5135 : 50 : checkPoint.time = (pg_time_t) time(NULL);
5791 simon@2ndQuadrant.co 5136 : 50 : checkPoint.oldestActiveXid = InvalidTransactionId;
5137 : :
689 heikki.linnakangas@i 5138 : 50 : TransamVariables->nextXid = checkPoint.nextXid;
5139 : 50 : TransamVariables->nextOid = checkPoint.nextOid;
5140 : 50 : TransamVariables->oidCount = 0;
7446 tgl@sss.pgh.pa.us 5141 : 50 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
3140 rhaas@postgresql.org 5142 : 50 : AdvanceOldestClogXid(checkPoint.oldestXid);
5731 tgl@sss.pgh.pa.us 5143 : 50 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
3149 5144 : 50 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
3981 alvherre@alvh.no-ip. 5145 : 50 : SetCommitTsLimit(InvalidTransactionId, InvalidTransactionId);
5146 : :
5147 : : /* Set up the XLOG page header */
9518 vadim4o@yahoo.com 5148 : 50 : page->xlp_magic = XLOG_PAGE_MAGIC;
7768 tgl@sss.pgh.pa.us 5149 : 50 : page->xlp_info = XLP_LONG_HEADER;
1452 rhaas@postgresql.org 5150 : 50 : page->xlp_tli = BootstrapTimeLineID;
2960 andres@anarazel.de 5151 : 50 : page->xlp_pageaddr = wal_segment_size;
7768 tgl@sss.pgh.pa.us 5152 : 50 : longpage = (XLogLongPageHeader) page;
5153 : 50 : longpage->xlp_sysid = sysidentifier;
2960 andres@anarazel.de 5154 : 50 : longpage->xlp_seg_size = wal_segment_size;
7145 tgl@sss.pgh.pa.us 5155 : 50 : longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5156 : :
5157 : : /* Insert the initial checkpoint record */
3994 heikki.linnakangas@i 5158 : 50 : recptr = ((char *) page + SizeOfXLogLongPHD);
5159 : 50 : record = (XLogRecord *) recptr;
4873 5160 : 50 : record->xl_prev = 0;
9518 vadim4o@yahoo.com 5161 : 50 : record->xl_xid = InvalidTransactionId;
3994 heikki.linnakangas@i 5162 : 50 : record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
8994 tgl@sss.pgh.pa.us 5163 : 50 : record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
9518 vadim4o@yahoo.com 5164 : 50 : record->xl_rmid = RM_XLOG_ID;
3994 heikki.linnakangas@i 5165 : 50 : recptr += SizeOfXLogRecord;
5166 : : /* fill the XLogRecordDataHeaderShort struct */
3135 tgl@sss.pgh.pa.us 5167 : 50 : *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
3994 heikki.linnakangas@i 5168 : 50 : *(recptr++) = sizeof(checkPoint);
5169 : 50 : memcpy(recptr, &checkPoint, sizeof(checkPoint));
5170 : 50 : recptr += sizeof(checkPoint);
5171 [ - + ]: 50 : Assert(recptr - (char *) record == record->xl_tot_len);
5172 : :
4010 5173 : 50 : INIT_CRC32C(crc);
3994 5174 : 50 : COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
4010 5175 : 50 : COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5176 : 50 : FIN_CRC32C(crc);
9069 vadim4o@yahoo.com 5177 : 50 : record->xl_crc = crc;
5178 : :
5179 : : /* Create first XLOG segment file */
1452 rhaas@postgresql.org 5180 : 50 : openLogTLI = BootstrapTimeLineID;
5181 : 50 : openLogFile = XLogFileInit(1, BootstrapTimeLineID);
5182 : :
5183 : : /*
5184 : : * We needn't bother with Reserve/ReleaseExternalFD here, since we'll
5185 : : * close the file again in a moment.
5186 : : */
5187 : :
5188 : : /* Write the first page with the initial record */
8909 tgl@sss.pgh.pa.us 5189 : 50 : errno = 0;
3145 rhaas@postgresql.org 5190 : 50 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
7147 tgl@sss.pgh.pa.us 5191 [ - + ]: 50 : if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5192 : : {
5193 : : /* if write didn't set errno, assume problem is no disk space */
8909 tgl@sss.pgh.pa.us 5194 [ # # ]:UBC 0 : if (errno == 0)
5195 : 0 : errno = ENOSPC;
8134 5196 [ # # ]: 0 : ereport(PANIC,
5197 : : (errcode_for_file_access(),
5198 : : errmsg("could not write bootstrap write-ahead log file: %m")));
5199 : : }
3145 rhaas@postgresql.org 5200 :CBC 50 : pgstat_report_wait_end();
5201 : :
5202 : 50 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
8994 tgl@sss.pgh.pa.us 5203 [ - + ]: 50 : if (pg_fsync(openLogFile) != 0)
8134 tgl@sss.pgh.pa.us 5204 [ # # ]:UBC 0 : ereport(PANIC,
5205 : : (errcode_for_file_access(),
5206 : : errmsg("could not fsync bootstrap write-ahead log file: %m")));
3145 rhaas@postgresql.org 5207 :CBC 50 : pgstat_report_wait_end();
5208 : :
2305 peter@eisentraut.org 5209 [ - + ]: 50 : if (close(openLogFile) != 0)
7945 tgl@sss.pgh.pa.us 5210 [ # # ]:UBC 0 : ereport(PANIC,
5211 : : (errcode_for_file_access(),
5212 : : errmsg("could not close bootstrap write-ahead log file: %m")));
5213 : :
8994 tgl@sss.pgh.pa.us 5214 :CBC 50 : openLogFile = -1;
5215 : :
5216 : : /* Now create pg_control */
461 peter@eisentraut.org 5217 : 50 : InitControlFile(sysidentifier, data_checksum_version);
8994 tgl@sss.pgh.pa.us 5218 : 50 : ControlFile->time = checkPoint.time;
9518 vadim4o@yahoo.com 5219 : 50 : ControlFile->checkPoint = checkPoint.redo;
8994 tgl@sss.pgh.pa.us 5220 : 50 : ControlFile->checkPointCopy = checkPoint;
5221 : :
5222 : : /* some additional ControlFile fields are set in WriteControlFile() */
9102 5223 : 50 : WriteControlFile();
5224 : :
5225 : : /* Bootstrap the commit log, too */
8829 5226 : 50 : BootStrapCLOG();
3981 alvherre@alvh.no-ip. 5227 : 50 : BootStrapCommitTs();
7788 tgl@sss.pgh.pa.us 5228 : 50 : BootStrapSUBTRANS();
7487 5229 : 50 : BootStrapMultiXact();
5230 : :
7373 5231 : 50 : pfree(buffer);
5232 : :
5233 : : /*
5234 : : * Force control file to be read - in contrast to normal processing we'd
5235 : : * otherwise never run the checks and GUC related initializations therein.
5236 : : */
2966 andres@anarazel.de 5237 : 50 : ReadControlFile();
9518 vadim4o@yahoo.com 5238 : 50 : }
5239 : :
5240 : : static char *
86 tgl@sss.pgh.pa.us 5241 :GNC 797 : str_time(pg_time_t tnow, char *buf, size_t bufsize)
5242 : : {
5243 : 797 : pg_strftime(buf, bufsize,
5244 : : "%Y-%m-%d %H:%M:%S %Z",
6659 tgl@sss.pgh.pa.us 5245 :CBC 797 : pg_localtime(&tnow, log_timezone));
5246 : :
9106 peter_e@gmx.net 5247 : 797 : return buf;
5248 : : }
5249 : :
5250 : : /*
5251 : : * Initialize the first WAL segment on new timeline.
5252 : : */
5253 : : static void
1349 heikki.linnakangas@i 5254 : 48 : XLogInitNewTimeline(TimeLineID endTLI, XLogRecPtr endOfLog, TimeLineID newTLI)
5255 : : {
5256 : : char xlogfname[MAXFNAMELEN];
5257 : : XLogSegNo endLogSegNo;
5258 : : XLogSegNo startLogSegNo;
5259 : :
5260 : : /* we always switch to a new timeline after archive recovery */
1452 rhaas@postgresql.org 5261 [ - + ]: 48 : Assert(endTLI != newTLI);
5262 : :
5263 : : /*
5264 : : * Update min recovery point one last time.
5265 : : */
5968 heikki.linnakangas@i 5266 : 48 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5267 : :
5268 : : /*
5269 : : * Calculate the last segment on the old timeline, and the first segment
5270 : : * on the new timeline. If the switch happens in the middle of a segment,
5271 : : * they are the same, but if the switch happens exactly at a segment
5272 : : * boundary, startLogSegNo will be endLogSegNo + 1.
5273 : : */
2960 andres@anarazel.de 5274 : 48 : XLByteToPrevSeg(endOfLog, endLogSegNo, wal_segment_size);
5275 : 48 : XLByteToSeg(endOfLog, startLogSegNo, wal_segment_size);
5276 : :
5277 : : /*
5278 : : * Initialize the starting WAL segment for the new timeline. If the switch
5279 : : * happens in the middle of a segment, copy data from the last WAL segment
5280 : : * of the old timeline up to the switch point, to the starting WAL segment
5281 : : * on the new timeline.
5282 : : */
3966 heikki.linnakangas@i 5283 [ + + ]: 48 : if (endLogSegNo == startLogSegNo)
5284 : : {
5285 : : /*
5286 : : * Make a copy of the file on the new timeline.
5287 : : *
5288 : : * Writing WAL isn't allowed yet, so there are no locking
5289 : : * considerations. But we should be just as tense as XLogFileInit to
5290 : : * avoid emplacing a bogus file.
5291 : : */
1452 rhaas@postgresql.org 5292 : 40 : XLogFileCopy(newTLI, endLogSegNo, endTLI, endLogSegNo,
2960 andres@anarazel.de 5293 : 40 : XLogSegmentOffset(endOfLog, wal_segment_size));
5294 : : }
5295 : : else
5296 : : {
5297 : : /*
5298 : : * The switch happened at a segment boundary, so just create the next
5299 : : * segment on the new timeline.
5300 : : */
5301 : : int fd;
5302 : :
1452 rhaas@postgresql.org 5303 : 8 : fd = XLogFileInit(startLogSegNo, newTLI);
5304 : :
2305 peter@eisentraut.org 5305 [ - + ]: 8 : if (close(fd) != 0)
5306 : : {
2155 michael@paquier.xyz 5307 :UBC 0 : int save_errno = errno;
5308 : :
1452 rhaas@postgresql.org 5309 : 0 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
2155 michael@paquier.xyz 5310 : 0 : errno = save_errno;
3963 heikki.linnakangas@i 5311 [ # # ]: 0 : ereport(ERROR,
5312 : : (errcode_for_file_access(),
5313 : : errmsg("could not close file \"%s\": %m", xlogfname)));
5314 : : }
5315 : : }
5316 : :
5317 : : /*
5318 : : * Let's just make real sure there are not .ready or .done flags posted
5319 : : * for the new segment.
5320 : : */
1452 rhaas@postgresql.org 5321 :CBC 48 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
4022 fujii@postgresql.org 5322 : 48 : XLogArchiveCleanup(xlogfname);
7770 tgl@sss.pgh.pa.us 5323 : 48 : }
5324 : :
5325 : : /*
5326 : : * Perform cleanup actions at the conclusion of archive recovery.
5327 : : */
5328 : : static void
1452 rhaas@postgresql.org 5329 : 48 : CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog,
5330 : : TimeLineID newTLI)
5331 : : {
5332 : : /*
5333 : : * Execute the recovery_end_command, if any.
5334 : : */
1475 5335 [ + - + + ]: 48 : if (recoveryEndCommand && strcmp(recoveryEndCommand, "") != 0)
994 michael@paquier.xyz 5336 : 2 : ExecuteRecoveryCommand(recoveryEndCommand,
5337 : : "recovery_end_command",
5338 : : true,
5339 : : WAIT_EVENT_RECOVERY_END_COMMAND);
5340 : :
5341 : : /*
5342 : : * We switched to a new timeline. Clean up segments on the old timeline.
5343 : : *
5344 : : * If there are any higher-numbered segments on the old timeline, remove
5345 : : * them. They might contain valid WAL, but they might also be
5346 : : * pre-allocated files containing garbage. In any case, they are not part
5347 : : * of the new timeline's history so we don't need them.
5348 : : */
1452 rhaas@postgresql.org 5349 : 48 : RemoveNonParentXlogFiles(EndOfLog, newTLI);
5350 : :
5351 : : /*
5352 : : * If the switch happened in the middle of a segment, what to do with the
5353 : : * last, partial segment on the old timeline? If we don't archive it, and
5354 : : * the server that created the WAL never archives it either (e.g. because
5355 : : * it was hit by a meteor), it will never make it to the archive. That's
5356 : : * OK from our point of view, because the new segment that we created with
5357 : : * the new TLI contains all the WAL from the old timeline up to the switch
5358 : : * point. But if you later try to do PITR to the "missing" WAL on the old
5359 : : * timeline, recovery won't find it in the archive. It's physically
5360 : : * present in the new file with new TLI, but recovery won't look there
5361 : : * when it's recovering to the older timeline. On the other hand, if we
5362 : : * archive the partial segment, and the original server on that timeline
5363 : : * is still running and archives the completed version of the same segment
5364 : : * later, it will fail. (We used to do that in 9.4 and below, and it
5365 : : * caused such problems).
5366 : : *
5367 : : * As a compromise, we rename the last segment with the .partial suffix,
5368 : : * and archive it. Archive recovery will never try to read .partial
5369 : : * segments, so they will normally go unused. But in the odd PITR case,
5370 : : * the administrator can copy them manually to the pg_wal directory
5371 : : * (removing the suffix). They can be useful in debugging, too.
5372 : : *
5373 : : * If a .done or .ready file already exists for the old timeline, however,
5374 : : * we had already determined that the segment is complete, so we can let
5375 : : * it be archived normally. (In particular, if it was restored from the
5376 : : * archive to begin with, it's expected to have a .done file).
5377 : : */
1475 5378 [ + + + + ]: 88 : if (XLogSegmentOffset(EndOfLog, wal_segment_size) != 0 &&
5379 [ + + - + ]: 40 : XLogArchivingActive())
5380 : : {
5381 : : char origfname[MAXFNAMELEN];
5382 : : XLogSegNo endLogSegNo;
5383 : :
5384 : 9 : XLByteToPrevSeg(EndOfLog, endLogSegNo, wal_segment_size);
5385 : 9 : XLogFileName(origfname, EndOfLogTLI, endLogSegNo, wal_segment_size);
5386 : :
5387 [ + + ]: 9 : if (!XLogArchiveIsReadyOrDone(origfname))
5388 : : {
5389 : : char origpath[MAXPGPATH];
5390 : : char partialfname[MAXFNAMELEN];
5391 : : char partialpath[MAXPGPATH];
5392 : :
5393 : : /*
5394 : : * If we're summarizing WAL, we can't rename the partial file
5395 : : * until the summarizer finishes with it, else it will fail.
5396 : : */
458 5397 [ + + ]: 5 : if (summarize_wal)
5398 : 1 : WaitForWalSummarization(EndOfLog);
5399 : :
1475 5400 : 5 : XLogFilePath(origpath, EndOfLogTLI, endLogSegNo, wal_segment_size);
5401 : 5 : snprintf(partialfname, MAXFNAMELEN, "%s.partial", origfname);
5402 : 5 : snprintf(partialpath, MAXPGPATH, "%s.partial", origpath);
5403 : :
5404 : : /*
5405 : : * Make sure there's no .done or .ready file for the .partial
5406 : : * file.
5407 : : */
5408 : 5 : XLogArchiveCleanup(partialfname);
5409 : :
5410 : 5 : durable_rename(origpath, partialpath, ERROR);
5411 : 5 : XLogArchiveNotify(partialfname);
5412 : : }
5413 : : }
5414 : 48 : }
5415 : :
5416 : : /*
5417 : : * Check to see if required parameters are set high enough on this server
5418 : : * for various aspects of recovery operation.
5419 : : *
5420 : : * Note that all the parameters which this function tests need to be
5421 : : * listed in Administrator's Overview section in high-availability.sgml.
5422 : : * If you change them, don't forget to update the list.
5423 : : */
5424 : : static void
1349 heikki.linnakangas@i 5425 : 238 : CheckRequiredParameterValues(void)
5426 : : {
5427 : : /*
5428 : : * For archive recovery, the WAL must be generated with at least 'replica'
5429 : : * wal_level.
5430 : : */
5431 [ + + + + ]: 238 : if (ArchiveRecoveryRequested && ControlFile->wal_level == WAL_LEVEL_MINIMAL)
5432 : : {
5433 [ + - ]: 2 : ereport(FATAL,
5434 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5435 : : errmsg("WAL was generated with \"wal_level=minimal\", cannot continue recovering"),
5436 : : errdetail("This happens if you temporarily set \"wal_level=minimal\" on the server."),
5437 : : errhint("Use a backup taken after setting \"wal_level\" to higher than \"minimal\".")));
5438 : : }
5439 : :
5440 : : /*
5441 : : * For Hot Standby, the WAL must be generated with 'replica' mode, and we
5442 : : * must have at least as many backend slots as the primary.
5443 : : */
4254 5444 [ + + + + ]: 236 : if (ArchiveRecoveryRequested && EnableHotStandby)
5445 : : {
5446 : : /* We ignore autovacuum_worker_slots when we make this test. */
5661 5447 : 119 : RecoveryRequiresIntParameter("max_connections",
5448 : : MaxConnections,
5660 tgl@sss.pgh.pa.us 5449 : 119 : ControlFile->MaxConnections);
4498 rhaas@postgresql.org 5450 : 119 : RecoveryRequiresIntParameter("max_worker_processes",
5451 : : max_worker_processes,
5452 : 119 : ControlFile->max_worker_processes);
2449 michael@paquier.xyz 5453 : 119 : RecoveryRequiresIntParameter("max_wal_senders",
5454 : : max_wal_senders,
5455 : 119 : ControlFile->max_wal_senders);
4800 tgl@sss.pgh.pa.us 5456 : 119 : RecoveryRequiresIntParameter("max_prepared_transactions",
5457 : : max_prepared_xacts,
5660 5458 : 119 : ControlFile->max_prepared_xacts);
4800 5459 : 119 : RecoveryRequiresIntParameter("max_locks_per_transaction",
5460 : : max_locks_per_xact,
5660 5461 : 119 : ControlFile->max_locks_per_xact);
5462 : : }
5791 simon@2ndQuadrant.co 5463 : 236 : }
5464 : :
5465 : : /*
5466 : : * This must be called ONCE during postmaster or standalone-backend startup
5467 : : */
5468 : : void
8994 tgl@sss.pgh.pa.us 5469 : 907 : StartupXLOG(void)
5470 : : {
5471 : : XLogCtlInsert *Insert;
5472 : : CheckPoint checkPoint;
5473 : : bool wasShutdown;
5474 : : bool didCrash;
5475 : : bool haveTblspcMap;
5476 : : bool haveBackupLabel;
5477 : : XLogRecPtr EndOfLog;
5478 : : TimeLineID EndOfLogTLI;
5479 : : TimeLineID newTLI;
5480 : : bool performedWalRecovery;
5481 : : EndOfWalRecoveryInfo *endOfRecoveryInfo;
5482 : : XLogRecPtr abortedRecPtr;
5483 : : XLogRecPtr missingContrecPtr;
5484 : : TransactionId oldestActiveXID;
1916 fujii@postgresql.org 5485 : 907 : bool promoted = false;
5486 : : char timebuf[128];
5487 : :
5488 : : /*
5489 : : * We should have an aux process resource owner to use, and we should not
5490 : : * be in a transaction that's installed some other resowner.
5491 : : */
2658 tgl@sss.pgh.pa.us 5492 [ - + ]: 907 : Assert(AuxProcessResourceOwner != NULL);
5493 [ + - - + ]: 907 : Assert(CurrentResourceOwner == NULL ||
5494 : : CurrentResourceOwner == AuxProcessResourceOwner);
5495 : 907 : CurrentResourceOwner = AuxProcessResourceOwner;
5496 : :
5497 : : /*
5498 : : * Check that contents look valid.
5499 : : */
2180 peter@eisentraut.org 5500 [ - + ]: 907 : if (!XRecOffIsValid(ControlFile->checkPoint))
8134 tgl@sss.pgh.pa.us 5501 [ # # ]:UBC 0 : ereport(FATAL,
5502 : : (errcode(ERRCODE_DATA_CORRUPTED),
5503 : : errmsg("control file contains invalid checkpoint location")));
5504 : :
2180 peter@eisentraut.org 5505 [ + + - - :CBC 907 : switch (ControlFile->state)
+ + - ]
5506 : : {
5507 : 709 : case DB_SHUTDOWNED:
5508 : :
5509 : : /*
5510 : : * This is the expected case, so don't be chatty in standalone
5511 : : * mode
5512 : : */
5513 [ + + + + ]: 709 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
5514 : : (errmsg("database system was shut down at %s",
5515 : : str_time(ControlFile->time,
5516 : : timebuf, sizeof(timebuf)))));
5517 : 709 : break;
5518 : :
5519 : 28 : case DB_SHUTDOWNED_IN_RECOVERY:
5520 [ + - ]: 28 : ereport(LOG,
5521 : : (errmsg("database system was shut down in recovery at %s",
5522 : : str_time(ControlFile->time,
5523 : : timebuf, sizeof(timebuf)))));
5524 : 28 : break;
5525 : :
2180 peter@eisentraut.org 5526 :UBC 0 : case DB_SHUTDOWNING:
5527 [ # # ]: 0 : ereport(LOG,
5528 : : (errmsg("database system shutdown was interrupted; last known up at %s",
5529 : : str_time(ControlFile->time,
5530 : : timebuf, sizeof(timebuf)))));
5531 : 0 : break;
5532 : :
5533 : 0 : case DB_IN_CRASH_RECOVERY:
5534 [ # # ]: 0 : ereport(LOG,
5535 : : (errmsg("database system was interrupted while in recovery at %s",
5536 : : str_time(ControlFile->time,
5537 : : timebuf, sizeof(timebuf))),
5538 : : errhint("This probably means that some data is corrupted and"
5539 : : " you will have to use the last backup for recovery.")));
5540 : 0 : break;
5541 : :
2180 peter@eisentraut.org 5542 :CBC 6 : case DB_IN_ARCHIVE_RECOVERY:
5543 [ + - ]: 6 : ereport(LOG,
5544 : : (errmsg("database system was interrupted while in recovery at log time %s",
5545 : : str_time(ControlFile->checkPointCopy.time,
5546 : : timebuf, sizeof(timebuf))),
5547 : : errhint("If this has occurred more than once some data might be corrupted"
5548 : : " and you might need to choose an earlier recovery target.")));
5549 : 6 : break;
5550 : :
5551 : 164 : case DB_IN_PRODUCTION:
5552 [ + - ]: 164 : ereport(LOG,
5553 : : (errmsg("database system was interrupted; last known up at %s",
5554 : : str_time(ControlFile->time,
5555 : : timebuf, sizeof(timebuf)))));
5556 : 164 : break;
5557 : :
2180 peter@eisentraut.org 5558 :UBC 0 : default:
5559 [ # # ]: 0 : ereport(FATAL,
5560 : : (errcode(ERRCODE_DATA_CORRUPTED),
5561 : : errmsg("control file contains invalid database cluster state")));
5562 : : }
5563 : :
5564 : : /* This is just to allow attaching to startup process with a debugger */
5565 : : #ifdef XLOG_REPLAY_DELAY
5566 : : if (ControlFile->state != DB_SHUTDOWNED)
5567 : : pg_usleep(60000000L);
5568 : : #endif
5569 : :
5570 : : /*
5571 : : * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
5572 : : * In cases where someone has performed a copy for PITR, these directories
5573 : : * may have been excluded and need to be re-created.
5574 : : */
6196 tgl@sss.pgh.pa.us 5575 :CBC 907 : ValidateXLOGDirectoryStructure();
5576 : :
5577 : : /* Set up timeout handler needed to report startup progress. */
1463 rhaas@postgresql.org 5578 [ + + ]: 907 : if (!IsBootstrapProcessingMode())
5579 : 857 : RegisterTimeout(STARTUP_PROGRESS_TIMEOUT,
5580 : : startup_progress_timeout_handler);
5581 : :
5582 : : /*----------
5583 : : * If we previously crashed, perform a couple of actions:
5584 : : *
5585 : : * - The pg_wal directory may still include some temporary WAL segments
5586 : : * used when creating a new segment, so perform some clean up to not
5587 : : * bloat this path. This is done first as there is no point to sync
5588 : : * this temporary data.
5589 : : *
5590 : : * - There might be data which we had written, intending to fsync it, but
5591 : : * which we had not actually fsync'd yet. Therefore, a power failure in
5592 : : * the near future might cause earlier unflushed writes to be lost, even
5593 : : * though more recent data written to disk from here on would be
5594 : : * persisted. To avoid that, fsync the entire data directory.
5595 : : */
1349 heikki.linnakangas@i 5596 [ + + ]: 907 : if (ControlFile->state != DB_SHUTDOWNED &&
5597 [ + + ]: 198 : ControlFile->state != DB_SHUTDOWNED_IN_RECOVERY)
5598 : : {
5599 : 170 : RemoveTempXlogFiles();
5600 : 170 : SyncDataDirectory();
1300 andres@anarazel.de 5601 : 170 : didCrash = true;
5602 : : }
5603 : : else
5604 : 737 : didCrash = false;
5605 : :
5606 : : /*
5607 : : * Prepare for WAL recovery if needed.
5608 : : *
5609 : : * InitWalRecovery analyzes the control file and the backup label file, if
5610 : : * any. It updates the in-memory ControlFile buffer according to the
5611 : : * starting checkpoint, and sets InRecovery and ArchiveRecoveryRequested.
5612 : : * It also applies the tablespace map file, if any.
5613 : : */
1349 heikki.linnakangas@i 5614 : 907 : InitWalRecovery(ControlFile, &wasShutdown,
5615 : : &haveBackupLabel, &haveTblspcMap);
5616 : 907 : checkPoint = ControlFile->checkPointCopy;
5617 : :
5618 : : /* initialize shared memory variables from the checkpoint record */
689 5619 : 907 : TransamVariables->nextXid = checkPoint.nextXid;
5620 : 907 : TransamVariables->nextOid = checkPoint.nextOid;
5621 : 907 : TransamVariables->oidCount = 0;
7446 tgl@sss.pgh.pa.us 5622 : 907 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
3140 rhaas@postgresql.org 5623 : 907 : AdvanceOldestClogXid(checkPoint.oldestXid);
5731 tgl@sss.pgh.pa.us 5624 : 907 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
3149 5625 : 907 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
3591 mail@joeconway.com 5626 : 907 : SetCommitTsLimit(checkPoint.oldestCommitTsXid,
5627 : : checkPoint.newestCommitTsXid);
5628 : :
5629 : : /*
5630 : : * Clear out any old relcache cache files. This is *necessary* if we do
5631 : : * any WAL replay, since that would probably result in the cache files
5632 : : * being out of sync with database reality. In theory we could leave them
5633 : : * in place if the database had been cleanly shut down, but it seems
5634 : : * safest to just remove them always and let them be rebuilt during the
5635 : : * first backend startup. These files needs to be removed from all
5636 : : * directories including pg_tblspc, however the symlinks are created only
5637 : : * after reading tablespace_map file in case of archive recovery from
5638 : : * backup, so needs to clear old relcache files here after creating
5639 : : * symlinks.
5640 : : */
1349 heikki.linnakangas@i 5641 : 907 : RelationCacheInitFileRemove();
5642 : :
5643 : : /*
5644 : : * Initialize replication slots, before there's a chance to remove
5645 : : * required resources.
5646 : : */
4155 andres@anarazel.de 5647 : 907 : StartupReplicationSlots();
5648 : :
5649 : : /*
5650 : : * Startup logical state, needs to be setup now so we have proper data
5651 : : * during crash recovery.
5652 : : */
4256 rhaas@postgresql.org 5653 : 907 : StartupReorderBuffer();
5654 : :
5655 : : /*
5656 : : * Startup CLOG. This must be done after TransamVariables->nextXid has
5657 : : * been initialized and before we accept connections or begin WAL replay.
5658 : : */
1734 5659 : 907 : StartupCLOG();
5660 : :
5661 : : /*
5662 : : * Startup MultiXact. We need to do this early to be able to replay
5663 : : * truncations.
5664 : : */
4350 alvherre@alvh.no-ip. 5665 : 907 : StartupMultiXact();
5666 : :
5667 : : /*
5668 : : * Ditto for commit timestamps. Activate the facility if the setting is
5669 : : * enabled in the control file, as there should be no tracking of commit
5670 : : * timestamps done when the setting was disabled. This facility can be
5671 : : * started or stopped when replaying a XLOG_PARAMETER_CHANGE record.
5672 : : */
2588 michael@paquier.xyz 5673 [ + + ]: 907 : if (ControlFile->track_commit_timestamp)
3608 alvherre@alvh.no-ip. 5674 : 13 : StartupCommitTs();
5675 : :
5676 : : /*
5677 : : * Recover knowledge about replay progress of known replication partners.
5678 : : */
3834 andres@anarazel.de 5679 : 907 : StartupReplicationOrigin();
5680 : :
5681 : : /*
5682 : : * Initialize unlogged LSN. On a clean shutdown, it's restored from the
5683 : : * control file. On recovery, all unlogged relations are blown away, so
5684 : : * the unlogged LSN counter can be reset too.
5685 : : */
4641 heikki.linnakangas@i 5686 [ + + ]: 907 : if (ControlFile->state == DB_SHUTDOWNED)
606 nathan@postgresql.or 5687 : 702 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5688 : 702 : ControlFile->unloggedLSN);
5689 : : else
5690 : 205 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5691 : : FirstNormalUnloggedLSN);
5692 : :
5693 : : /*
5694 : : * Copy any missing timeline history files between 'now' and the recovery
5695 : : * target timeline from archive to pg_wal. While we don't need those files
5696 : : * ourselves - the history file of the recovery target timeline covers all
5697 : : * the previous timelines in the history too - a cascading standby server
5698 : : * might be interested in them. Or, if you archive the WAL from this
5699 : : * server to a different archive than the primary, it'd be good for all
5700 : : * the history files to get archived there after failover, so that you can
5701 : : * use one of the old timelines as a PITR target. Timeline history files
5702 : : * are small, so it's better to copy them unnecessarily than not copy them
5703 : : * and regret later.
5704 : : */
1349 heikki.linnakangas@i 5705 : 907 : restoreTimeLineHistoryFiles(checkPoint.ThisTimeLineID, recoveryTargetTLI);
5706 : :
5707 : : /*
5708 : : * Before running in recovery, scan pg_twophase and fill in its status to
5709 : : * be able to work on entries generated by redo. Doing a scan before
5710 : : * taking any recovery action has the merit to discard any 2PC files that
5711 : : * are newer than the first record to replay, saving from any conflicts at
5712 : : * replay. This avoids as well any subsequent scans when doing recovery
5713 : : * of the on-disk two-phase data.
5714 : : */
3128 simon@2ndQuadrant.co 5715 : 907 : restoreTwoPhaseData();
5716 : :
5717 : : /*
5718 : : * When starting with crash recovery, reset pgstat data - it might not be
5719 : : * valid. Otherwise restore pgstat data. It's safe to do this here,
5720 : : * because postmaster will not yet have started any other processes.
5721 : : *
5722 : : * NB: Restoring replication slot stats relies on slot state to have
5723 : : * already been restored from disk.
5724 : : *
5725 : : * TODO: With a bit of extra work we could just start with a pgstat file
5726 : : * associated with the checkpoint redo location we're starting from.
5727 : : */
1300 andres@anarazel.de 5728 [ + + ]: 907 : if (didCrash)
5729 : 170 : pgstat_discard_stats();
5730 : : else
224 michael@paquier.xyz 5731 : 737 : pgstat_restore_stats();
5732 : :
5024 simon@2ndQuadrant.co 5733 : 907 : lastFullPageWrites = checkPoint.fullPageWrites;
5734 : :
4494 heikki.linnakangas@i 5735 : 907 : RedoRecPtr = XLogCtl->RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
4008 5736 : 907 : doPageWrites = lastFullPageWrites;
5737 : :
5738 : : /* REDO */
7754 tgl@sss.pgh.pa.us 5739 [ + + ]: 907 : if (InRecovery)
5740 : : {
5741 : : /* Initialize state for RecoveryInProgress() */
1349 heikki.linnakangas@i 5742 [ - + ]: 205 : SpinLockAcquire(&XLogCtl->info_lck);
5743 [ + + ]: 205 : if (InArchiveRecovery)
5744 : 107 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
5745 : : else
5746 : 98 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
5747 : 205 : SpinLockRelease(&XLogCtl->info_lck);
5748 : :
5749 : : /*
5750 : : * Update pg_control to show that we are recovering and to show the
5751 : : * selected checkpoint as the place we are starting from. We also mark
5752 : : * pg_control with any minimum recovery stop point obtained from a
5753 : : * backup history file.
5754 : : *
5755 : : * No need to hold ControlFileLock yet, we aren't up far enough.
5756 : : */
5757 : 205 : UpdateControlFile();
5758 : :
5759 : : /*
5760 : : * If there was a backup label file, it's done its job and the info
5761 : : * has now been propagated into pg_control. We must get rid of the
5762 : : * label file so that if we crash during recovery, we'll pick up at
5763 : : * the latest recovery restartpoint instead of going all the way back
5764 : : * to the backup start point. It seems prudent though to just rename
5765 : : * the file out of the way rather than delete it completely.
5766 : : */
5767 [ + + ]: 205 : if (haveBackupLabel)
5768 : : {
5769 : 70 : unlink(BACKUP_LABEL_OLD);
5770 : 70 : durable_rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, FATAL);
5771 : : }
5772 : :
5773 : : /*
5774 : : * If there was a tablespace_map file, it's done its job and the
5775 : : * symlinks have been created. We must get rid of the map file so
5776 : : * that if we crash during recovery, we don't create symlinks again.
5777 : : * It seems prudent though to just rename the file out of the way
5778 : : * rather than delete it completely.
5779 : : */
5780 [ + + ]: 205 : if (haveTblspcMap)
5781 : : {
5782 : 2 : unlink(TABLESPACE_MAP_OLD);
5783 : 2 : durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, FATAL);
5784 : : }
5785 : :
5786 : : /*
5787 : : * Initialize our local copy of minRecoveryPoint. When doing crash
5788 : : * recovery we want to replay up to the end of WAL. Particularly, in
5789 : : * the case of a promoted standby minRecoveryPoint value in the
5790 : : * control file is only updated after the first checkpoint. However,
5791 : : * if the instance crashes before the first post-recovery checkpoint
5792 : : * is completed then recovery will use a stale location causing the
5793 : : * startup process to think that there are still invalid page
5794 : : * references when checking for data consistency.
5795 : : */
2671 michael@paquier.xyz 5796 [ + + ]: 205 : if (InArchiveRecovery)
5797 : : {
1349 heikki.linnakangas@i 5798 : 107 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
5799 : 107 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
5800 : : }
5801 : : else
5802 : : {
5803 : 98 : LocalMinRecoveryPoint = InvalidXLogRecPtr;
5804 : 98 : LocalMinRecoveryPointTLI = 0;
5805 : : }
5806 : :
5807 : : /* Check that the GUCs used to generate the WAL allow recovery */
5661 5808 : 205 : CheckRequiredParameterValues();
5809 : :
5810 : : /*
5811 : : * We're in recovery, so unlogged relations may be trashed and must be
5812 : : * reset. This should be done BEFORE allowing Hot Standby
5813 : : * connections, so that read-only backends don't try to read whatever
5814 : : * garbage is left over from before.
5815 : : */
5416 rhaas@postgresql.org 5816 : 205 : ResetUnloggedRelations(UNLOGGED_RELATION_CLEANUP);
5817 : :
5818 : : /*
5819 : : * Likewise, delete any saved transaction snapshot files that got left
5820 : : * behind by crashed backends.
5821 : : */
5119 tgl@sss.pgh.pa.us 5822 : 205 : DeleteAllExportedSnapshotFiles();
5823 : :
5824 : : /*
5825 : : * Initialize for Hot Standby, if enabled. We won't let backends in
5826 : : * yet, not until we've reached the min recovery point specified in
5827 : : * control file and we've established a recovery snapshot from a
5828 : : * running-xacts WAL record.
5829 : : */
4630 heikki.linnakangas@i 5830 [ + + + + ]: 205 : if (ArchiveRecoveryRequested && EnableHotStandby)
5831 : : {
5832 : : TransactionId *xids;
5833 : : int nxids;
5834 : :
5736 5835 [ + + ]: 101 : ereport(DEBUG1,
5836 : : (errmsg_internal("initializing for hot standby")));
5837 : :
5791 simon@2ndQuadrant.co 5838 : 101 : InitRecoveryTransactionEnvironment();
5839 : :
5840 [ + + ]: 101 : if (wasShutdown)
5841 : 26 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
5842 : : else
5843 : 75 : oldestActiveXID = checkPoint.oldestActiveXid;
5844 [ - + ]: 101 : Assert(TransactionIdIsValid(oldestActiveXID));
5845 : :
5846 : : /* Tell procarray about the range of xids it has to deal with */
689 heikki.linnakangas@i 5847 : 101 : ProcArrayInitRecovery(XidFromFullTransactionId(TransamVariables->nextXid));
5848 : :
5849 : : /*
5850 : : * Startup subtrans only. CLOG, MultiXact and commit timestamp
5851 : : * have already been started up and other SLRUs are not maintained
5852 : : * during recovery and need not be started yet.
5853 : : */
5791 simon@2ndQuadrant.co 5854 : 101 : StartupSUBTRANS(oldestActiveXID);
5855 : :
5856 : : /*
5857 : : * If we're beginning at a shutdown checkpoint, we know that
5858 : : * nothing was running on the primary at this point. So fake-up an
5859 : : * empty running-xacts record and use that here and now. Recover
5860 : : * additional standby state for prepared transactions.
5861 : : */
5676 heikki.linnakangas@i 5862 [ + + ]: 101 : if (wasShutdown)
5863 : : {
5864 : : RunningTransactionsData running;
5865 : : TransactionId latestCompletedXid;
5866 : :
5867 : : /* Update pg_subtrans entries for any prepared transactions */
487 5868 : 26 : StandbyRecoverPreparedTransactions();
5869 : :
5870 : : /*
5871 : : * Construct a RunningTransactions snapshot representing a
5872 : : * shut down server, with only prepared transactions still
5873 : : * alive. We're never overflowed at this point because all
5874 : : * subxids are listed with their parent prepared transactions.
5875 : : */
5676 5876 : 26 : running.xcnt = nxids;
4712 simon@2ndQuadrant.co 5877 : 26 : running.subxcnt = 0;
487 heikki.linnakangas@i 5878 : 26 : running.subxid_status = SUBXIDS_IN_SUBTRANS;
1903 andres@anarazel.de 5879 : 26 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5676 heikki.linnakangas@i 5880 : 26 : running.oldestRunningXid = oldestActiveXID;
1903 andres@anarazel.de 5881 : 26 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5646 simon@2ndQuadrant.co 5882 [ - + ]: 26 : TransactionIdRetreat(latestCompletedXid);
5645 5883 [ - + ]: 26 : Assert(TransactionIdIsNormal(latestCompletedXid));
5646 5884 : 26 : running.latestCompletedXid = latestCompletedXid;
5676 heikki.linnakangas@i 5885 : 26 : running.xids = xids;
5886 : :
5887 : 26 : ProcArrayApplyRecoveryInfo(&running);
5888 : : }
5889 : : }
5890 : :
5891 : : /*
5892 : : * We're all set for replaying the WAL now. Do it.
5893 : : */
1349 5894 : 205 : PerformWalRecovery();
5895 : 150 : performedWalRecovery = true;
5896 : : }
5897 : : else
1345 5898 : 702 : performedWalRecovery = false;
5899 : :
5900 : : /*
5901 : : * Finish WAL recovery.
5902 : : */
1349 5903 : 852 : endOfRecoveryInfo = FinishWalRecovery();
5904 : 852 : EndOfLog = endOfRecoveryInfo->endOfLog;
5905 : 852 : EndOfLogTLI = endOfRecoveryInfo->endOfLogTLI;
5906 : 852 : abortedRecPtr = endOfRecoveryInfo->abortedRecPtr;
5907 : 852 : missingContrecPtr = endOfRecoveryInfo->missingContrecPtr;
5908 : :
5909 : : /*
5910 : : * Reset ps status display, so as no information related to recovery shows
5911 : : * up.
5912 : : */
1131 michael@paquier.xyz 5913 : 852 : set_ps_display("");
5914 : :
5915 : : /*
5916 : : * When recovering from a backup (we are in recovery, and archive recovery
5917 : : * was requested), complain if we did not roll forward far enough to reach
5918 : : * the point where the database is consistent. For regular online
5919 : : * backup-from-primary, that means reaching the end-of-backup WAL record
5920 : : * (at which point we reset backupStartPoint to be Invalid), for
5921 : : * backup-from-replica (which can't inject records into the WAL stream),
5922 : : * that point is when we reach the minRecoveryPoint in pg_control (which
5923 : : * we purposefully copy last when backing up from a replica). For
5924 : : * pg_rewind (which creates a backup_label with a method of "pg_rewind")
5925 : : * or snapshot-style backups (which don't), backupEndRequired will be set
5926 : : * to false.
5927 : : *
5928 : : * Note: it is indeed okay to look at the local variable
5929 : : * LocalMinRecoveryPoint here, even though ControlFile->minRecoveryPoint
5930 : : * might be further ahead --- ControlFile->minRecoveryPoint cannot have
5931 : : * been advanced beyond the WAL we processed.
5932 : : */
5325 heikki.linnakangas@i 5933 [ + + ]: 852 : if (InRecovery &&
1349 5934 [ + - ]: 150 : (EndOfLog < LocalMinRecoveryPoint ||
5775 5935 [ - + ]: 150 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint)))
5936 : : {
5937 : : /*
5938 : : * Ran off end of WAL before reaching end-of-backup WAL record, or
5939 : : * minRecoveryPoint. That's a bad sign, indicating that you tried to
5940 : : * recover from an online backup but never called pg_backup_stop(), or
5941 : : * you didn't archive all the WAL needed.
5942 : : */
4630 heikki.linnakangas@i 5943 [ # # # # ]:UBC 0 : if (ArchiveRecoveryRequested || ControlFile->backupEndRequired)
5944 : : {
1300 sfrost@snowman.net 5945 [ # # # # ]: 0 : if (!XLogRecPtrIsInvalid(ControlFile->backupStartPoint) || ControlFile->backupEndRequired)
5192 heikki.linnakangas@i 5946 [ # # ]: 0 : ereport(FATAL,
5947 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5948 : : errmsg("WAL ends before end of online backup"),
5949 : : errhint("All WAL generated while online backup was taken must be available at recovery.")));
5950 : : else
5311 5951 [ # # ]: 0 : ereport(FATAL,
5952 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5953 : : errmsg("WAL ends before consistent recovery point")));
5954 : : }
5955 : : }
5956 : :
5957 : : /*
5958 : : * Reset unlogged relations to the contents of their INIT fork. This is
5959 : : * done AFTER recovery is complete so as to include any unlogged relations
5960 : : * created during recovery, but BEFORE recovery is marked as having
5961 : : * completed successfully. Otherwise we'd not retry if any of the post
5962 : : * end-of-recovery steps fail.
5963 : : */
1349 heikki.linnakangas@i 5964 [ + + ]:CBC 852 : if (InRecovery)
5965 : 150 : ResetUnloggedRelations(UNLOGGED_RELATION_INIT);
5966 : :
5967 : : /*
5968 : : * Pre-scan prepared transactions to find out the range of XIDs present.
5969 : : * This information is not quite needed yet, but it is positioned here so
5970 : : * as potential problems are detected before any on-disk change is done.
5971 : : */
2667 michael@paquier.xyz 5972 : 852 : oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
5973 : :
5974 : : /*
5975 : : * Allow ordinary WAL segment creation before possibly switching to a new
5976 : : * timeline, which creates a new segment, and after the last ReadRecord().
5977 : : */
1167 5978 : 852 : SetInstallXLogFileSegmentActive();
5979 : :
5980 : : /*
5981 : : * Consider whether we need to assign a new timeline ID.
5982 : : *
5983 : : * If we did archive recovery, we always assign a new ID. This handles a
5984 : : * couple of issues. If we stopped short of the end of WAL during
5985 : : * recovery, then we are clearly generating a new timeline and must assign
5986 : : * it a unique new ID. Even if we ran to the end, modifying the current
5987 : : * last segment is problematic because it may result in trying to
5988 : : * overwrite an already-archived copy of that segment, and we encourage
5989 : : * DBAs to make their archive_commands reject that. We can dodge the
5990 : : * problem by making the new active segment have a new timeline ID.
5991 : : *
5992 : : * In a normal crash recovery, we can just extend the timeline we were in.
5993 : : */
1349 heikki.linnakangas@i 5994 : 852 : newTLI = endOfRecoveryInfo->lastRecTLI;
4630 5995 [ + + ]: 852 : if (ArchiveRecoveryRequested)
5996 : : {
1447 rhaas@postgresql.org 5997 : 48 : newTLI = findNewestTimeLine(recoveryTargetTLI) + 1;
7768 tgl@sss.pgh.pa.us 5998 [ + - ]: 48 : ereport(LOG,
5999 : : (errmsg("selected new timeline ID: %u", newTLI)));
6000 : :
6001 : : /*
6002 : : * Make a writable copy of the last WAL segment. (Note that we also
6003 : : * have a copy of the last block of the old WAL in
6004 : : * endOfRecovery->lastPage; we will use that below.)
6005 : : */
1349 heikki.linnakangas@i 6006 : 48 : XLogInitNewTimeline(EndOfLogTLI, EndOfLog, newTLI);
6007 : :
6008 : : /*
6009 : : * Remove the signal files out of the way, so that we don't
6010 : : * accidentally re-enter archive recovery mode in a subsequent crash.
6011 : : */
6012 [ + + ]: 48 : if (endOfRecoveryInfo->standby_signal_file_found)
6013 : 45 : durable_unlink(STANDBY_SIGNAL_FILE, FATAL);
6014 : :
6015 [ + + ]: 48 : if (endOfRecoveryInfo->recovery_signal_file_found)
6016 : 3 : durable_unlink(RECOVERY_SIGNAL_FILE, FATAL);
6017 : :
6018 : : /*
6019 : : * Write the timeline history file, and have it archived. After this
6020 : : * point (or rather, as soon as the file is archived), the timeline
6021 : : * will appear as "taken" in the WAL archive and to any standby
6022 : : * servers. If we crash before actually switching to the new
6023 : : * timeline, standby servers will nevertheless think that we switched
6024 : : * to the new timeline, and will try to connect to the new timeline.
6025 : : * To minimize the window for that, try to do as little as possible
6026 : : * between here and writing the end-of-recovery record.
6027 : : */
1447 rhaas@postgresql.org 6028 : 48 : writeTimeLineHistory(newTLI, recoveryTargetTLI,
6029 : : EndOfLog, endOfRecoveryInfo->recoveryStopReason);
6030 : :
1349 heikki.linnakangas@i 6031 [ + - ]: 48 : ereport(LOG,
6032 : : (errmsg("archive recovery complete")));
6033 : : }
6034 : :
6035 : : /* Save the selected TimeLineID in shared memory, too */
458 rhaas@postgresql.org 6036 [ - + ]: 852 : SpinLockAcquire(&XLogCtl->info_lck);
1447 6037 : 852 : XLogCtl->InsertTimeLineID = newTLI;
1349 heikki.linnakangas@i 6038 : 852 : XLogCtl->PrevTimeLineID = endOfRecoveryInfo->lastRecTLI;
458 rhaas@postgresql.org 6039 : 852 : SpinLockRelease(&XLogCtl->info_lck);
6040 : :
6041 : : /*
6042 : : * Actually, if WAL ended in an incomplete record, skip the parts that
6043 : : * made it through and start writing after the portion that persisted.
6044 : : * (It's critical to first write an OVERWRITE_CONTRECORD message, which
6045 : : * we'll do as soon as we're open for writing new WAL.)
6046 : : */
1489 alvherre@alvh.no-ip. 6047 [ + + ]: 852 : if (!XLogRecPtrIsInvalid(missingContrecPtr))
6048 : : {
6049 : : /*
6050 : : * We should only have a missingContrecPtr if we're not switching to a
6051 : : * new timeline. When a timeline switch occurs, WAL is copied from the
6052 : : * old timeline to the new only up to the end of the last complete
6053 : : * record, so there can't be an incomplete WAL record that we need to
6054 : : * disregard.
6055 : : */
1155 rhaas@postgresql.org 6056 [ - + ]: 11 : Assert(newTLI == endOfRecoveryInfo->lastRecTLI);
1489 alvherre@alvh.no-ip. 6057 [ - + ]: 11 : Assert(!XLogRecPtrIsInvalid(abortedRecPtr));
6058 : 11 : EndOfLog = missingContrecPtr;
6059 : : }
6060 : :
6061 : : /*
6062 : : * Prepare to write WAL starting at EndOfLog location, and init xlog
6063 : : * buffer cache using the block containing the last record from the
6064 : : * previous incarnation.
6065 : : */
9130 vadim4o@yahoo.com 6066 : 852 : Insert = &XLogCtl->Insert;
1349 heikki.linnakangas@i 6067 : 852 : Insert->PrevBytePos = XLogRecPtrToBytePos(endOfRecoveryInfo->lastRec);
4485 6068 : 852 : Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
6069 : :
6070 : : /*
6071 : : * Tricky point here: lastPage contains the *last* block that the LastRec
6072 : : * record spans, not the one it starts in. The last block is indeed the
6073 : : * one we want to use.
6074 : : */
6075 [ + + ]: 852 : if (EndOfLog % XLOG_BLCKSZ != 0)
6076 : : {
6077 : : char *page;
6078 : : int len;
6079 : : int firstIdx;
6080 : :
6081 : 826 : firstIdx = XLogRecPtrToBufIdx(EndOfLog);
1349 6082 : 826 : len = EndOfLog - endOfRecoveryInfo->lastPageBeginPtr;
6083 [ - + ]: 826 : Assert(len < XLOG_BLCKSZ);
6084 : :
6085 : : /* Copy the valid part of the last block, and zero the rest */
4485 6086 : 826 : page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
1349 6087 : 826 : memcpy(page, endOfRecoveryInfo->lastPage, len);
4485 6088 : 826 : memset(page + len, 0, XLOG_BLCKSZ - len);
6089 : :
678 jdavis@postgresql.or 6090 : 826 : pg_atomic_write_u64(&XLogCtl->xlblocks[firstIdx], endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ);
66 akorotkov@postgresql 6091 : 826 : XLogCtl->InitializedUpTo = endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ;
6092 : : }
6093 : : else
6094 : : {
6095 : : /*
6096 : : * There is no partial block to copy. Just set InitializedUpTo, and
6097 : : * let the first attempt to insert a log record to initialize the next
6098 : : * buffer.
6099 : : */
6100 : 26 : XLogCtl->InitializedUpTo = EndOfLog;
6101 : : }
6102 : :
6103 : : /*
6104 : : * Update local and shared status. This is OK to do without any locks
6105 : : * because no other process can be reading or writing WAL yet.
6106 : : */
4485 heikki.linnakangas@i 6107 : 852 : LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
568 alvherre@alvh.no-ip. 6108 : 852 : pg_atomic_write_u64(&XLogCtl->logInsertResult, EndOfLog);
570 6109 : 852 : pg_atomic_write_u64(&XLogCtl->logWriteResult, EndOfLog);
6110 : 852 : pg_atomic_write_u64(&XLogCtl->logFlushResult, EndOfLog);
4485 heikki.linnakangas@i 6111 : 852 : XLogCtl->LogwrtRqst.Write = EndOfLog;
6112 : 852 : XLogCtl->LogwrtRqst.Flush = EndOfLog;
6113 : :
6114 : : /*
6115 : : * Preallocate additional log files, if wanted.
6116 : : */
1447 rhaas@postgresql.org 6117 : 852 : PreallocXlogFiles(EndOfLog, newTLI);
6118 : :
6119 : : /*
6120 : : * Okay, we're officially UP.
6121 : : */
9130 vadim4o@yahoo.com 6122 : 852 : InRecovery = false;
6123 : :
6124 : : /* start the archive_timeout timer and LSN running */
4485 heikki.linnakangas@i 6125 : 852 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
3231 andres@anarazel.de 6126 : 852 : XLogCtl->lastSegSwitchLSN = EndOfLog;
6127 : :
6128 : : /* also initialize latestCompletedXid, to nextXid - 1 */
5012 tgl@sss.pgh.pa.us 6129 : 852 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
689 heikki.linnakangas@i 6130 : 852 : TransamVariables->latestCompletedXid = TransamVariables->nextXid;
6131 : 852 : FullTransactionIdRetreat(&TransamVariables->latestCompletedXid);
5012 tgl@sss.pgh.pa.us 6132 : 852 : LWLockRelease(ProcArrayLock);
6133 : :
6134 : : /*
6135 : : * Start up subtrans, if not already done for hot standby. (commit
6136 : : * timestamps are started below, if necessary.)
6137 : : */
5791 simon@2ndQuadrant.co 6138 [ + + ]: 852 : if (standbyState == STANDBY_DISABLED)
6139 : 804 : StartupSUBTRANS(oldestActiveXID);
6140 : :
6141 : : /*
6142 : : * Perform end of recovery actions for any SLRUs that need it.
6143 : : */
5108 6144 : 852 : TrimCLOG();
4350 alvherre@alvh.no-ip. 6145 : 852 : TrimMultiXact();
6146 : :
6147 : : /*
6148 : : * Reload shared-memory state for prepared transactions. This needs to
6149 : : * happen before renaming the last partial segment of the old timeline as
6150 : : * it may be possible that we have to recover some transactions from it.
6151 : : */
7437 tgl@sss.pgh.pa.us 6152 : 852 : RecoverPreparedTransactions();
6153 : :
6154 : : /* Shut down xlogreader */
1349 heikki.linnakangas@i 6155 : 852 : ShutdownWalRecovery();
6156 : :
6157 : : /* Enable WAL writes for this backend only. */
1474 rhaas@postgresql.org 6158 : 852 : LocalSetXLogInsertAllowed();
6159 : :
6160 : : /* If necessary, write overwrite-contrecord before doing anything else */
6161 [ + + ]: 852 : if (!XLogRecPtrIsInvalid(abortedRecPtr))
6162 : : {
6163 [ - + ]: 11 : Assert(!XLogRecPtrIsInvalid(missingContrecPtr));
1349 heikki.linnakangas@i 6164 : 11 : CreateOverwriteContrecordRecord(abortedRecPtr, missingContrecPtr, newTLI);
6165 : : }
6166 : :
6167 : : /*
6168 : : * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
6169 : : * record before resource manager writes cleanup WAL records or checkpoint
6170 : : * record is written.
6171 : : */
1474 rhaas@postgresql.org 6172 : 852 : Insert->fullPageWrites = lastFullPageWrites;
6173 : 852 : UpdateFullPageWrites();
6174 : :
6175 : : /*
6176 : : * Emit checkpoint or end-of-recovery record in XLOG, if required.
6177 : : */
1349 heikki.linnakangas@i 6178 [ + + ]: 852 : if (performedWalRecovery)
1474 rhaas@postgresql.org 6179 : 150 : promoted = PerformRecoveryXLogAction();
6180 : :
6181 : : /*
6182 : : * If any of the critical GUCs have changed, log them before we allow
6183 : : * backends to write WAL.
6184 : : */
5661 heikki.linnakangas@i 6185 : 852 : XLogReportParameters();
6186 : :
6187 : : /* If this is archive recovery, perform post-recovery cleanup actions. */
1463 rhaas@postgresql.org 6188 [ + + ]: 852 : if (ArchiveRecoveryRequested)
1447 6189 : 48 : CleanupAfterArchiveRecovery(EndOfLogTLI, EndOfLog, newTLI);
6190 : :
6191 : : /*
6192 : : * Local WAL inserts enabled, so it's time to finish initialization of
6193 : : * commit timestamp.
6194 : : */
3981 alvherre@alvh.no-ip. 6195 : 852 : CompleteCommitTsInitialization();
6196 : :
6197 : : /* Clean up EndOfWalRecoveryInfo data to appease Valgrind leak checking */
86 tgl@sss.pgh.pa.us 6198 [ + + ]:GNC 852 : if (endOfRecoveryInfo->lastPage)
6199 : 837 : pfree(endOfRecoveryInfo->lastPage);
6200 : 852 : pfree(endOfRecoveryInfo->recoveryStopReason);
6201 : 852 : pfree(endOfRecoveryInfo);
6202 : :
6203 : : /*
6204 : : * All done with end-of-recovery actions.
6205 : : *
6206 : : * Now allow backends to write WAL and update the control file status in
6207 : : * consequence. SharedRecoveryState, that controls if backends can write
6208 : : * WAL, is updated while holding ControlFileLock to prevent other backends
6209 : : * to look at an inconsistent state of the control file in shared memory.
6210 : : * There is still a small window during which backends can write WAL and
6211 : : * the control file is still referring to a system not in DB_IN_PRODUCTION
6212 : : * state while looking at the on-disk control file.
6213 : : *
6214 : : * Also, we use info_lck to update SharedRecoveryState to ensure that
6215 : : * there are no race conditions concerning visibility of other recent
6216 : : * updates to shared memory.
6217 : : */
3380 peter_e@gmx.net 6218 :CBC 852 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6219 : 852 : ControlFile->state = DB_IN_PRODUCTION;
6220 : :
4053 andres@anarazel.de 6221 [ - + ]: 852 : SpinLockAcquire(&XLogCtl->info_lck);
2012 michael@paquier.xyz 6222 : 852 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_DONE;
4053 andres@anarazel.de 6223 : 852 : SpinLockRelease(&XLogCtl->info_lck);
6224 : :
3380 peter_e@gmx.net 6225 : 852 : UpdateControlFile();
6226 : 852 : LWLockRelease(ControlFileLock);
6227 : :
6228 : : /*
6229 : : * Shutdown the recovery environment. This must occur after
6230 : : * RecoverPreparedTransactions() (see notes in lock_twophase_recover())
6231 : : * and after switching SharedRecoveryState to RECOVERY_STATE_DONE so as
6232 : : * any session building a snapshot will not rely on KnownAssignedXids as
6233 : : * RecoveryInProgress() would return false at this stage. This is
6234 : : * particularly critical for prepared 2PC transactions, that would still
6235 : : * need to be included in snapshots once recovery has ended.
6236 : : */
1484 michael@paquier.xyz 6237 [ + + ]: 852 : if (standbyState != STANDBY_DISABLED)
6238 : 48 : ShutdownRecoveryTransactionEnvironment();
6239 : :
6240 : : /*
6241 : : * If there were cascading standby servers connected to us, nudge any wal
6242 : : * sender processes to notice that we've been promoted.
6243 : : */
933 andres@anarazel.de 6244 : 852 : WalSndWakeup(true, true);
6245 : :
6246 : : /*
6247 : : * If this was a promotion, request an (online) checkpoint now. This isn't
6248 : : * required for consistency, but the last restartpoint might be far back,
6249 : : * and in case of a crash, recovering from it might take a longer than is
6250 : : * appropriate now that we're not in standby mode anymore.
6251 : : */
1916 fujii@postgresql.org 6252 [ + + ]: 852 : if (promoted)
4542 simon@2ndQuadrant.co 6253 : 41 : RequestCheckpoint(CHECKPOINT_FORCE);
6095 heikki.linnakangas@i 6254 : 852 : }
6255 : :
6256 : : /*
6257 : : * Callback from PerformWalRecovery(), called when we switch from crash
6258 : : * recovery to archive recovery mode. Updates the control file accordingly.
6259 : : */
6260 : : void
1349 6261 : 2 : SwitchIntoArchiveRecovery(XLogRecPtr EndRecPtr, TimeLineID replayTLI)
6262 : : {
6263 : : /* initialize minRecoveryPoint to this record */
6264 : 2 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6265 : 2 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
6266 [ + - ]: 2 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6267 : : {
6268 : 2 : ControlFile->minRecoveryPoint = EndRecPtr;
6269 : 2 : ControlFile->minRecoveryPointTLI = replayTLI;
6270 : : }
6271 : : /* update local copy */
6272 : 2 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
6273 : 2 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
6274 : :
6275 : : /*
6276 : : * The startup process can update its local copy of minRecoveryPoint from
6277 : : * this point.
6278 : : */
6279 : 2 : updateMinRecoveryPoint = true;
6280 : :
6281 : 2 : UpdateControlFile();
6282 : :
6283 : : /*
6284 : : * We update SharedRecoveryState while holding the lock on ControlFileLock
6285 : : * so both states are consistent in shared memory.
6286 : : */
6287 [ - + ]: 2 : SpinLockAcquire(&XLogCtl->info_lck);
6288 : 2 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
6289 : 2 : SpinLockRelease(&XLogCtl->info_lck);
6290 : :
6291 : 2 : LWLockRelease(ControlFileLock);
6292 : 2 : }
6293 : :
6294 : : /*
6295 : : * Callback from PerformWalRecovery(), called when we reach the end of backup.
6296 : : * Updates the control file accordingly.
6297 : : */
6298 : : void
6299 : 70 : ReachedEndOfBackup(XLogRecPtr EndRecPtr, TimeLineID tli)
6300 : : {
6301 : : /*
6302 : : * We have reached the end of base backup, as indicated by pg_control. The
6303 : : * data on disk is now consistent (unless minRecoveryPoint is further
6304 : : * ahead, which can happen if we crashed during previous recovery). Reset
6305 : : * backupStartPoint and backupEndPoint, and update minRecoveryPoint to
6306 : : * make sure we don't allow starting up at an earlier point even if
6307 : : * recovery is stopped and restarted soon after this.
6308 : : */
6309 : 70 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6310 : :
6311 [ + + ]: 70 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6312 : : {
6313 : 66 : ControlFile->minRecoveryPoint = EndRecPtr;
6314 : 66 : ControlFile->minRecoveryPointTLI = tli;
6315 : : }
6316 : :
6317 : 70 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
6318 : 70 : ControlFile->backupEndPoint = InvalidXLogRecPtr;
6319 : 70 : ControlFile->backupEndRequired = false;
6320 : 70 : UpdateControlFile();
6321 : :
6322 : 70 : LWLockRelease(ControlFileLock);
5676 6323 : 70 : }
6324 : :
6325 : : /*
6326 : : * Perform whatever XLOG actions are necessary at end of REDO.
6327 : : *
6328 : : * The goal here is to make sure that we'll be able to recover properly if
6329 : : * we crash again. If we choose to write a checkpoint, we'll write a shutdown
6330 : : * checkpoint rather than an on-line one. This is not particularly critical,
6331 : : * but since we may be assigning a new TLI, using a shutdown checkpoint allows
6332 : : * us to have the rule that TLI only changes in shutdown checkpoints, which
6333 : : * allows some extra error checking in xlog_redo.
6334 : : */
6335 : : static bool
1475 rhaas@postgresql.org 6336 : 150 : PerformRecoveryXLogAction(void)
6337 : : {
6338 : 150 : bool promoted = false;
6339 : :
6340 : : /*
6341 : : * Perform a checkpoint to update all our recovery activity to disk.
6342 : : *
6343 : : * Note that we write a shutdown checkpoint rather than an on-line one.
6344 : : * This is not particularly critical, but since we may be assigning a new
6345 : : * TLI, using a shutdown checkpoint allows us to have the rule that TLI
6346 : : * only changes in shutdown checkpoints, which allows some extra error
6347 : : * checking in xlog_redo.
6348 : : *
6349 : : * In promotion, only create a lightweight end-of-recovery record instead
6350 : : * of a full checkpoint. A checkpoint is requested later, after we're
6351 : : * fully out of recovery mode and already accepting queries.
6352 : : */
6353 [ + + + - : 198 : if (ArchiveRecoveryRequested && IsUnderPostmaster &&
+ + ]
1349 heikki.linnakangas@i 6354 : 48 : PromoteIsTriggered())
6355 : : {
1475 rhaas@postgresql.org 6356 : 41 : promoted = true;
6357 : :
6358 : : /*
6359 : : * Insert a special WAL record to mark the end of recovery, since we
6360 : : * aren't doing a checkpoint. That means that the checkpointer process
6361 : : * may likely be in the middle of a time-smoothed restartpoint and
6362 : : * could continue to be for minutes after this. That sounds strange,
6363 : : * but the effect is roughly the same and it would be stranger to try
6364 : : * to come out of the restartpoint and then checkpoint. We request a
6365 : : * checkpoint later anyway, just for safety.
6366 : : */
6367 : 41 : CreateEndOfRecoveryRecord();
6368 : : }
6369 : : else
6370 : : {
6371 : 109 : RequestCheckpoint(CHECKPOINT_END_OF_RECOVERY |
6372 : : CHECKPOINT_FAST |
6373 : : CHECKPOINT_WAIT);
6374 : : }
6375 : :
6376 : 150 : return promoted;
6377 : : }
6378 : :
6379 : : /*
6380 : : * Is the system still in recovery?
6381 : : *
6382 : : * Unlike testing InRecovery, this works in any process that's connected to
6383 : : * shared memory.
6384 : : */
6385 : : bool
6095 heikki.linnakangas@i 6386 : 52107707 : RecoveryInProgress(void)
6387 : : {
6388 : : /*
6389 : : * We check shared state each time only until we leave recovery mode. We
6390 : : * can't re-enter recovery, so there's no need to keep checking after the
6391 : : * shared variable has once been seen false.
6392 : : */
6393 [ + + ]: 52107707 : if (!LocalRecoveryInProgress)
6394 : 49945886 : return false;
6395 : : else
6396 : : {
6397 : : /*
6398 : : * use volatile pointer to make sure we make a fresh read of the
6399 : : * shared variable.
6400 : : */
6401 : 2161821 : volatile XLogCtlData *xlogctl = XLogCtl;
6402 : :
2012 michael@paquier.xyz 6403 : 2161821 : LocalRecoveryInProgress = (xlogctl->SharedRecoveryState != RECOVERY_STATE_DONE);
6404 : :
6405 : : /*
6406 : : * Note: We don't need a memory barrier when we're still in recovery.
6407 : : * We might exit recovery immediately after return, so the caller
6408 : : * can't rely on 'true' meaning that we're still in recovery anyway.
6409 : : */
6410 : :
6095 heikki.linnakangas@i 6411 : 2161821 : return LocalRecoveryInProgress;
6412 : : }
6413 : : }
6414 : :
6415 : : /*
6416 : : * Returns current recovery state from shared memory.
6417 : : *
6418 : : * This returned state is kept consistent with the contents of the control
6419 : : * file. See details about the possible values of RecoveryState in xlog.h.
6420 : : */
6421 : : RecoveryState
2012 michael@paquier.xyz 6422 : 21448 : GetRecoveryState(void)
6423 : : {
6424 : : RecoveryState retval;
6425 : :
6426 [ - + ]: 21448 : SpinLockAcquire(&XLogCtl->info_lck);
6427 : 21448 : retval = XLogCtl->SharedRecoveryState;
6428 : 21448 : SpinLockRelease(&XLogCtl->info_lck);
6429 : :
6430 : 21448 : return retval;
6431 : : }
6432 : :
6433 : : /*
6434 : : * Is this process allowed to insert new WAL records?
6435 : : *
6436 : : * Ordinarily this is essentially equivalent to !RecoveryInProgress().
6437 : : * But we also have provisions for forcing the result "true" or "false"
6438 : : * within specific processes regardless of the global state.
6439 : : */
6440 : : bool
5967 tgl@sss.pgh.pa.us 6441 : 37891468 : XLogInsertAllowed(void)
6442 : : {
6443 : : /*
6444 : : * If value is "unconditionally true" or "unconditionally false", just
6445 : : * return it. This provides the normal fast path once recovery is known
6446 : : * done.
6447 : : */
6448 [ + + ]: 37891468 : if (LocalXLogInsertAllowed >= 0)
6449 : 37154670 : return (bool) LocalXLogInsertAllowed;
6450 : :
6451 : : /*
6452 : : * Else, must check to see if we're still in recovery.
6453 : : */
6454 [ + + ]: 736798 : if (RecoveryInProgress())
6455 : 728245 : return false;
6456 : :
6457 : : /*
6458 : : * On exit from recovery, reset to "unconditionally true", since there is
6459 : : * no need to keep checking.
6460 : : */
6461 : 8553 : LocalXLogInsertAllowed = 1;
6462 : 8553 : return true;
6463 : : }
6464 : :
6465 : : /*
6466 : : * Make XLogInsertAllowed() return true in the current process only.
6467 : : *
6468 : : * Note: it is allowed to switch LocalXLogInsertAllowed back to -1 later,
6469 : : * and even call LocalSetXLogInsertAllowed() again after that.
6470 : : *
6471 : : * Returns the previous value of LocalXLogInsertAllowed.
6472 : : */
6473 : : static int
6474 : 880 : LocalSetXLogInsertAllowed(void)
6475 : : {
1349 heikki.linnakangas@i 6476 : 880 : int oldXLogAllowed = LocalXLogInsertAllowed;
6477 : :
5967 tgl@sss.pgh.pa.us 6478 : 880 : LocalXLogInsertAllowed = 1;
6479 : :
1463 rhaas@postgresql.org 6480 : 880 : return oldXLogAllowed;
6481 : : }
6482 : :
6483 : : /*
6484 : : * Return the current Redo pointer from shared memory.
6485 : : *
6486 : : * As a side-effect, the local RedoRecPtr copy is updated.
6487 : : */
6488 : : XLogRecPtr
9069 vadim4o@yahoo.com 6489 : 208513 : GetRedoRecPtr(void)
6490 : : {
6491 : : XLogRecPtr ptr;
6492 : :
6493 : : /*
6494 : : * The possibly not up-to-date copy in XlogCtl is enough. Even if we
6495 : : * grabbed a WAL insertion lock to read the authoritative value in
6496 : : * Insert->RedoRecPtr, someone might update it just after we've released
6497 : : * the lock.
6498 : : */
4053 andres@anarazel.de 6499 [ + + ]: 208513 : SpinLockAcquire(&XLogCtl->info_lck);
6500 : 208513 : ptr = XLogCtl->RedoRecPtr;
6501 : 208513 : SpinLockRelease(&XLogCtl->info_lck);
6502 : :
4494 heikki.linnakangas@i 6503 [ + + ]: 208513 : if (RedoRecPtr < ptr)
6504 : 1580 : RedoRecPtr = ptr;
6505 : :
8627 tgl@sss.pgh.pa.us 6506 : 208513 : return RedoRecPtr;
6507 : : }
6508 : :
6509 : : /*
6510 : : * Return information needed to decide whether a modified block needs a
6511 : : * full-page image to be included in the WAL record.
6512 : : *
6513 : : * The returned values are cached copies from backend-private memory, and
6514 : : * possibly out-of-date or, indeed, uninitialized, in which case they will
6515 : : * be InvalidXLogRecPtr and false, respectively. XLogInsertRecord will
6516 : : * re-check them against up-to-date values, while holding the WAL insert lock.
6517 : : */
6518 : : void
4008 heikki.linnakangas@i 6519 : 14042395 : GetFullPageWriteInfo(XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
6520 : : {
6521 : 14042395 : *RedoRecPtr_p = RedoRecPtr;
6522 : 14042395 : *doPageWrites_p = doPageWrites;
6523 : 14042395 : }
6524 : :
6525 : : /*
6526 : : * GetInsertRecPtr -- Returns the current insert position.
6527 : : *
6528 : : * NOTE: The value *actually* returned is the position of the last full
6529 : : * xlog page. It lags behind the real insert position by at most 1 page.
6530 : : * For that, we don't need to scan through WAL insertion locks, and an
6531 : : * approximation is enough for the current usage of this function.
6532 : : */
6533 : : XLogRecPtr
6696 tgl@sss.pgh.pa.us 6534 : 6948 : GetInsertRecPtr(void)
6535 : : {
6536 : : XLogRecPtr recptr;
6537 : :
4053 andres@anarazel.de 6538 [ + + ]: 6948 : SpinLockAcquire(&XLogCtl->info_lck);
6539 : 6948 : recptr = XLogCtl->LogwrtRqst.Write;
6540 : 6948 : SpinLockRelease(&XLogCtl->info_lck);
6541 : :
6696 tgl@sss.pgh.pa.us 6542 : 6948 : return recptr;
6543 : : }
6544 : :
6545 : : /*
6546 : : * GetFlushRecPtr -- Returns the current flush position, ie, the last WAL
6547 : : * position known to be fsync'd to disk. This should only be used on a
6548 : : * system that is known not to be in recovery.
6549 : : */
6550 : : XLogRecPtr
1452 rhaas@postgresql.org 6551 : 175530 : GetFlushRecPtr(TimeLineID *insertTLI)
6552 : : {
1447 6553 [ - + ]: 175530 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6554 : :
572 alvherre@alvh.no-ip. 6555 : 175530 : RefreshXLogWriteResult(LogwrtResult);
6556 : :
6557 : : /*
6558 : : * If we're writing and flushing WAL, the time line can't be changing, so
6559 : : * no lock is required.
6560 : : */
1452 rhaas@postgresql.org 6561 [ + + ]: 175530 : if (insertTLI)
1447 6562 : 22754 : *insertTLI = XLogCtl->InsertTimeLineID;
6563 : :
3576 simon@2ndQuadrant.co 6564 : 175530 : return LogwrtResult.Flush;
6565 : : }
6566 : :
6567 : : /*
6568 : : * GetWALInsertionTimeLine -- Returns the current timeline of a system that
6569 : : * is not in recovery.
6570 : : */
6571 : : TimeLineID
1452 rhaas@postgresql.org 6572 : 112241 : GetWALInsertionTimeLine(void)
6573 : : {
6574 [ - + ]: 112241 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6575 : :
6576 : : /* Since the value can't be changing, no lock is required. */
1447 6577 : 112241 : return XLogCtl->InsertTimeLineID;
6578 : : }
6579 : :
6580 : : /*
6581 : : * GetWALInsertionTimeLineIfSet -- If the system is not in recovery, returns
6582 : : * the WAL insertion timeline; else, returns 0. Wherever possible, use
6583 : : * GetWALInsertionTimeLine() instead, since it's cheaper. Note that this
6584 : : * function decides recovery has ended as soon as the insert TLI is set, which
6585 : : * happens before we set XLogCtl->SharedRecoveryState to RECOVERY_STATE_DONE.
6586 : : */
6587 : : TimeLineID
458 rhaas@postgresql.org 6588 :UBC 0 : GetWALInsertionTimeLineIfSet(void)
6589 : : {
6590 : : TimeLineID insertTLI;
6591 : :
6592 [ # # ]: 0 : SpinLockAcquire(&XLogCtl->info_lck);
6593 : 0 : insertTLI = XLogCtl->InsertTimeLineID;
6594 : 0 : SpinLockRelease(&XLogCtl->info_lck);
6595 : :
6596 : 0 : return insertTLI;
6597 : : }
6598 : :
6599 : : /*
6600 : : * GetLastImportantRecPtr -- Returns the LSN of the last important record
6601 : : * inserted. All records not explicitly marked as unimportant are considered
6602 : : * important.
6603 : : *
6604 : : * The LSN is determined by computing the maximum of
6605 : : * WALInsertLocks[i].lastImportantAt.
6606 : : */
6607 : : XLogRecPtr
3231 andres@anarazel.de 6608 :CBC 1553 : GetLastImportantRecPtr(void)
6609 : : {
6610 : 1553 : XLogRecPtr res = InvalidXLogRecPtr;
6611 : : int i;
6612 : :
6613 [ + + ]: 13977 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
6614 : : {
6615 : : XLogRecPtr last_important;
6616 : :
6617 : : /*
6618 : : * Need to take a lock to prevent torn reads of the LSN, which are
6619 : : * possible on some of the supported platforms. WAL insert locks only
6620 : : * support exclusive mode, so we have to use that.
6621 : : */
6622 : 12424 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
6623 : 12424 : last_important = WALInsertLocks[i].l.lastImportantAt;
6624 : 12424 : LWLockRelease(&WALInsertLocks[i].l.lock);
6625 : :
6626 [ + + ]: 12424 : if (res < last_important)
6627 : 2668 : res = last_important;
6628 : : }
6629 : :
6630 : 1553 : return res;
6631 : : }
6632 : :
6633 : : /*
6634 : : * Get the time and LSN of the last xlog segment switch
6635 : : */
6636 : : pg_time_t
3231 andres@anarazel.de 6637 :UBC 0 : GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
6638 : : {
6639 : : pg_time_t result;
6640 : :
6641 : : /* Need WALWriteLock, but shared lock is sufficient */
7011 tgl@sss.pgh.pa.us 6642 : 0 : LWLockAcquire(WALWriteLock, LW_SHARED);
4485 heikki.linnakangas@i 6643 : 0 : result = XLogCtl->lastSegSwitchTime;
3231 andres@anarazel.de 6644 : 0 : *lastSwitchLSN = XLogCtl->lastSegSwitchLSN;
7011 tgl@sss.pgh.pa.us 6645 : 0 : LWLockRelease(WALWriteLock);
6646 : :
6647 : 0 : return result;
6648 : : }
6649 : :
6650 : : /*
6651 : : * This must be called ONCE during postmaster or standalone-backend shutdown
6652 : : */
6653 : : void
7990 peter_e@gmx.net 6654 :CBC 625 : ShutdownXLOG(int code, Datum arg)
6655 : : {
6656 : : /*
6657 : : * We should have an aux process resource owner to use, and we should not
6658 : : * be in a transaction that's installed some other resowner.
6659 : : */
2658 tgl@sss.pgh.pa.us 6660 [ - + ]: 625 : Assert(AuxProcessResourceOwner != NULL);
6661 [ + + - + ]: 625 : Assert(CurrentResourceOwner == NULL ||
6662 : : CurrentResourceOwner == AuxProcessResourceOwner);
6663 : 625 : CurrentResourceOwner = AuxProcessResourceOwner;
6664 : :
6665 : : /* Don't be chatty in standalone mode */
4519 6666 [ + + + + ]: 625 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
6667 : : (errmsg("shutting down")));
6668 : :
6669 : : /*
6670 : : * Signal walsenders to move to stopping state.
6671 : : */
3066 andres@anarazel.de 6672 : 625 : WalSndInitStopping();
6673 : :
6674 : : /*
6675 : : * Wait for WAL senders to be in stopping state. This prevents commands
6676 : : * from writing new WAL.
6677 : : */
6678 : 625 : WalSndWaitStopping();
6679 : :
6095 heikki.linnakangas@i 6680 [ + + ]: 625 : if (RecoveryInProgress())
108 nathan@postgresql.or 6681 :GNC 52 : CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_FAST);
6682 : : else
6683 : : {
6684 : : /*
6685 : : * If archiving is enabled, rotate the last XLOG file so that all the
6686 : : * remaining records are archived (postmaster wakes up the archiver
6687 : : * process one more time at the end of shutdown). The checkpoint
6688 : : * record will go to the next XLOG file and won't be archived (yet).
6689 : : */
1362 rhaas@postgresql.org 6690 [ + + - + :CBC 573 : if (XLogArchivingActive())
+ + ]
3231 andres@anarazel.de 6691 : 12 : RequestXLogSwitch(false);
6692 : :
108 nathan@postgresql.or 6693 :GNC 573 : CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_FAST);
6694 : : }
9518 vadim4o@yahoo.com 6695 :CBC 625 : }
6696 : :
6697 : : /*
6698 : : * Log start of a checkpoint.
6699 : : */
6700 : : static void
6095 heikki.linnakangas@i 6701 : 1409 : LogCheckpointStart(int flags, bool restartpoint)
6702 : : {
1788 peter@eisentraut.org 6703 [ + + ]: 1409 : if (restartpoint)
6704 [ + - - + : 197 : ereport(LOG,
+ + + + +
+ + + + +
- + + + ]
6705 : : /* translator: the placeholders show checkpoint options */
6706 : : (errmsg("restartpoint starting:%s%s%s%s%s%s%s%s",
6707 : : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6708 : : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6709 : : (flags & CHECKPOINT_FAST) ? " fast" : "",
6710 : : (flags & CHECKPOINT_FORCE) ? " force" : "",
6711 : : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6712 : : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6713 : : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6714 : : (flags & CHECKPOINT_FLUSH_UNLOGGED) ? " flush-unlogged" : "")));
6715 : : else
6716 [ + - + + : 1212 : ereport(LOG,
- + + + +
+ + + + +
+ + + + ]
6717 : : /* translator: the placeholders show checkpoint options */
6718 : : (errmsg("checkpoint starting:%s%s%s%s%s%s%s%s",
6719 : : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6720 : : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6721 : : (flags & CHECKPOINT_FAST) ? " fast" : "",
6722 : : (flags & CHECKPOINT_FORCE) ? " force" : "",
6723 : : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6724 : : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6725 : : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6726 : : (flags & CHECKPOINT_FLUSH_UNLOGGED) ? " flush-unlogged" : "")));
6694 tgl@sss.pgh.pa.us 6727 : 1409 : }
6728 : :
6729 : : /*
6730 : : * Log end of a checkpoint.
6731 : : */
6732 : : static void
6095 heikki.linnakangas@i 6733 : 1701 : LogCheckpointEnd(bool restartpoint)
6734 : : {
6735 : : long write_msecs,
6736 : : sync_msecs,
6737 : : total_msecs,
6738 : : longest_msecs,
6739 : : average_msecs;
6740 : : uint64 average_sync_time;
6741 : :
6694 tgl@sss.pgh.pa.us 6742 : 1701 : CheckpointStats.ckpt_end_t = GetCurrentTimestamp();
6743 : :
1812 6744 : 1701 : write_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_write_t,
6745 : : CheckpointStats.ckpt_sync_t);
6746 : :
6747 : 1701 : sync_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_sync_t,
6748 : : CheckpointStats.ckpt_sync_end_t);
6749 : :
6750 : : /* Accumulate checkpoint timing summary data, in milliseconds. */
728 michael@paquier.xyz 6751 : 1701 : PendingCheckpointerStats.write_time += write_msecs;
6752 : 1701 : PendingCheckpointerStats.sync_time += sync_msecs;
6753 : :
6754 : : /*
6755 : : * All of the published timing statistics are accounted for. Only
6756 : : * continue if a log message is to be written.
6757 : : */
4953 rhaas@postgresql.org 6758 [ + + ]: 1701 : if (!log_checkpoints)
6759 : 292 : return;
6760 : :
1812 tgl@sss.pgh.pa.us 6761 : 1409 : total_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_start_t,
6762 : : CheckpointStats.ckpt_end_t);
6763 : :
6764 : : /*
6765 : : * Timing values returned from CheckpointStats are in microseconds.
6766 : : * Convert to milliseconds for consistent printing.
6767 : : */
6768 : 1409 : longest_msecs = (long) ((CheckpointStats.ckpt_longest_sync + 999) / 1000);
6769 : :
5431 rhaas@postgresql.org 6770 : 1409 : average_sync_time = 0;
5314 bruce@momjian.us 6771 [ - + ]: 1409 : if (CheckpointStats.ckpt_sync_rels > 0)
5431 rhaas@postgresql.org 6772 :UBC 0 : average_sync_time = CheckpointStats.ckpt_agg_sync_time /
6773 : 0 : CheckpointStats.ckpt_sync_rels;
1812 tgl@sss.pgh.pa.us 6774 :CBC 1409 : average_msecs = (long) ((average_sync_time + 999) / 1000);
6775 : :
6776 : : /*
6777 : : * ControlFileLock is not required to see ControlFile->checkPoint and
6778 : : * ->checkPointCopy here as we are the only updator of those variables at
6779 : : * this moment.
6780 : : */
1788 peter@eisentraut.org 6781 [ + + ]: 1409 : if (restartpoint)
6782 [ + - ]: 197 : ereport(LOG,
6783 : : (errmsg("restartpoint complete: wrote %d buffers (%.1f%%), "
6784 : : "wrote %d SLRU buffers; %d WAL file(s) added, "
6785 : : "%d removed, %d recycled; write=%ld.%03d s, "
6786 : : "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6787 : : "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6788 : : "estimate=%d kB; lsn=%X/%08X, redo lsn=%X/%08X",
6789 : : CheckpointStats.ckpt_bufs_written,
6790 : : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6791 : : CheckpointStats.ckpt_slru_written,
6792 : : CheckpointStats.ckpt_segs_added,
6793 : : CheckpointStats.ckpt_segs_removed,
6794 : : CheckpointStats.ckpt_segs_recycled,
6795 : : write_msecs / 1000, (int) (write_msecs % 1000),
6796 : : sync_msecs / 1000, (int) (sync_msecs % 1000),
6797 : : total_msecs / 1000, (int) (total_msecs % 1000),
6798 : : CheckpointStats.ckpt_sync_rels,
6799 : : longest_msecs / 1000, (int) (longest_msecs % 1000),
6800 : : average_msecs / 1000, (int) (average_msecs % 1000),
6801 : : (int) (PrevCheckPointDistance / 1024.0),
6802 : : (int) (CheckPointDistanceEstimate / 1024.0),
6803 : : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6804 : : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6805 : : else
6806 [ + - ]: 1212 : ereport(LOG,
6807 : : (errmsg("checkpoint complete: wrote %d buffers (%.1f%%), "
6808 : : "wrote %d SLRU buffers; %d WAL file(s) added, "
6809 : : "%d removed, %d recycled; write=%ld.%03d s, "
6810 : : "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6811 : : "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6812 : : "estimate=%d kB; lsn=%X/%08X, redo lsn=%X/%08X",
6813 : : CheckpointStats.ckpt_bufs_written,
6814 : : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6815 : : CheckpointStats.ckpt_slru_written,
6816 : : CheckpointStats.ckpt_segs_added,
6817 : : CheckpointStats.ckpt_segs_removed,
6818 : : CheckpointStats.ckpt_segs_recycled,
6819 : : write_msecs / 1000, (int) (write_msecs % 1000),
6820 : : sync_msecs / 1000, (int) (sync_msecs % 1000),
6821 : : total_msecs / 1000, (int) (total_msecs % 1000),
6822 : : CheckpointStats.ckpt_sync_rels,
6823 : : longest_msecs / 1000, (int) (longest_msecs % 1000),
6824 : : average_msecs / 1000, (int) (average_msecs % 1000),
6825 : : (int) (PrevCheckPointDistance / 1024.0),
6826 : : (int) (CheckPointDistanceEstimate / 1024.0),
6827 : : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6828 : : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6829 : : }
6830 : :
6831 : : /*
6832 : : * Update the estimate of distance between checkpoints.
6833 : : *
6834 : : * The estimate is used to calculate the number of WAL segments to keep
6835 : : * preallocated, see XLOGfileslop().
6836 : : */
6837 : : static void
3899 heikki.linnakangas@i 6838 : 1701 : UpdateCheckPointDistanceEstimate(uint64 nbytes)
6839 : : {
6840 : : /*
6841 : : * To estimate the number of segments consumed between checkpoints, keep a
6842 : : * moving average of the amount of WAL generated in previous checkpoint
6843 : : * cycles. However, if the load is bursty, with quiet periods and busy
6844 : : * periods, we want to cater for the peak load. So instead of a plain
6845 : : * moving average, let the average decline slowly if the previous cycle
6846 : : * used less WAL than estimated, but bump it up immediately if it used
6847 : : * more.
6848 : : *
6849 : : * When checkpoints are triggered by max_wal_size, this should converge to
6850 : : * CheckpointSegments * wal_segment_size,
6851 : : *
6852 : : * Note: This doesn't pay any attention to what caused the checkpoint.
6853 : : * Checkpoints triggered manually with CHECKPOINT command, or by e.g.
6854 : : * starting a base backup, are counted the same as those created
6855 : : * automatically. The slow-decline will largely mask them out, if they are
6856 : : * not frequent. If they are frequent, it seems reasonable to count them
6857 : : * in as any others; if you issue a manual checkpoint every 5 minutes and
6858 : : * never let a timed checkpoint happen, it makes sense to base the
6859 : : * preallocation on that 5 minute interval rather than whatever
6860 : : * checkpoint_timeout is set to.
6861 : : */
6862 : 1701 : PrevCheckPointDistance = nbytes;
6863 [ + + ]: 1701 : if (CheckPointDistanceEstimate < nbytes)
6864 : 717 : CheckPointDistanceEstimate = nbytes;
6865 : : else
6866 : 984 : CheckPointDistanceEstimate =
6867 : 984 : (0.90 * CheckPointDistanceEstimate + 0.10 * (double) nbytes);
6694 tgl@sss.pgh.pa.us 6868 : 1701 : }
6869 : :
6870 : : /*
6871 : : * Update the ps display for a process running a checkpoint. Note that
6872 : : * this routine should not do any allocations so as it can be called
6873 : : * from a critical section.
6874 : : */
6875 : : static void
1778 michael@paquier.xyz 6876 : 3402 : update_checkpoint_display(int flags, bool restartpoint, bool reset)
6877 : : {
6878 : : /*
6879 : : * The status is reported only for end-of-recovery and shutdown
6880 : : * checkpoints or shutdown restartpoints. Updating the ps display is
6881 : : * useful in those situations as it may not be possible to rely on
6882 : : * pg_stat_activity to see the status of the checkpointer or the startup
6883 : : * process.
6884 : : */
6885 [ + + ]: 3402 : if ((flags & (CHECKPOINT_END_OF_RECOVERY | CHECKPOINT_IS_SHUTDOWN)) == 0)
6886 : 2158 : return;
6887 : :
6888 [ + + ]: 1244 : if (reset)
6889 : 622 : set_ps_display("");
6890 : : else
6891 : : {
6892 : : char activitymsg[128];
6893 : :
6894 [ + + ]: 1866 : snprintf(activitymsg, sizeof(activitymsg), "performing %s%s%s",
6895 [ + + ]: 622 : (flags & CHECKPOINT_END_OF_RECOVERY) ? "end-of-recovery " : "",
6896 [ + + ]: 622 : (flags & CHECKPOINT_IS_SHUTDOWN) ? "shutdown " : "",
6897 : : restartpoint ? "restartpoint" : "checkpoint");
6898 : 622 : set_ps_display(activitymsg);
6899 : : }
6900 : : }
6901 : :
6902 : :
6903 : : /*
6904 : : * Perform a checkpoint --- either during shutdown, or on-the-fly
6905 : : *
6906 : : * flags is a bitwise OR of the following:
6907 : : * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
6908 : : * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
6909 : : * CHECKPOINT_FAST: finish the checkpoint ASAP, ignoring
6910 : : * checkpoint_completion_target parameter.
6911 : : * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
6912 : : * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
6913 : : * CHECKPOINT_END_OF_RECOVERY).
6914 : : * CHECKPOINT_FLUSH_UNLOGGED: also flush buffers of unlogged tables.
6915 : : *
6916 : : * Note: flags contains other bits, of interest here only for logging purposes.
6917 : : * In particular note that this routine is synchronous and does not pay
6918 : : * attention to CHECKPOINT_WAIT.
6919 : : *
6920 : : * If !shutdown then we are writing an online checkpoint. An XLOG_CHECKPOINT_REDO
6921 : : * record is inserted into WAL at the logical location of the checkpoint, before
6922 : : * flushing anything to disk, and when the checkpoint is eventually completed,
6923 : : * and it is from this point that WAL replay will begin in the case of a recovery
6924 : : * from this checkpoint. Once everything is written to disk, an
6925 : : * XLOG_CHECKPOINT_ONLINE record is written to complete the checkpoint, and
6926 : : * points back to the earlier XLOG_CHECKPOINT_REDO record. This mechanism allows
6927 : : * other write-ahead log records to be written while the checkpoint is in
6928 : : * progress, but we must be very careful about order of operations. This function
6929 : : * may take many minutes to execute on a busy system.
6930 : : *
6931 : : * On the other hand, when shutdown is true, concurrent insertion into the
6932 : : * write-ahead log is impossible, so there is no need for two separate records.
6933 : : * In this case, we only insert an XLOG_CHECKPOINT_SHUTDOWN record, and it's
6934 : : * both the record marking the completion of the checkpoint and the location
6935 : : * from which WAL replay would begin if needed.
6936 : : *
6937 : : * Returns true if a new checkpoint was performed, or false if it was skipped
6938 : : * because the system was idle.
6939 : : */
6940 : : bool
6696 tgl@sss.pgh.pa.us 6941 : 1506 : CreateCheckPoint(int flags)
6942 : : {
6943 : : bool shutdown;
6944 : : CheckPoint checkPoint;
6945 : : XLogRecPtr recptr;
6946 : : XLogSegNo _logSegNo;
9329 bruce@momjian.us 6947 : 1506 : XLogCtlInsert *Insert = &XLogCtl->Insert;
6948 : : uint32 freespace;
6949 : : XLogRecPtr PriorRedoPtr;
6950 : : XLogRecPtr last_important_lsn;
6951 : : VirtualTransactionId *vxids;
6952 : : int nvxids;
1463 rhaas@postgresql.org 6953 : 1506 : int oldXLogAllowed = 0;
6954 : :
6955 : : /*
6956 : : * An end-of-recovery checkpoint is really a shutdown checkpoint, just
6957 : : * issued at a different time.
6958 : : */
5967 tgl@sss.pgh.pa.us 6959 [ + + ]: 1506 : if (flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY))
5968 heikki.linnakangas@i 6960 : 601 : shutdown = true;
6961 : : else
6962 : 905 : shutdown = false;
6963 : :
6964 : : /* sanity check */
5967 tgl@sss.pgh.pa.us 6965 [ + + - + ]: 1506 : if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
5967 tgl@sss.pgh.pa.us 6966 [ # # ]:UBC 0 : elog(ERROR, "can't create a checkpoint during recovery");
6967 : :
6968 : : /*
6969 : : * Prepare to accumulate statistics.
6970 : : *
6971 : : * Note: because it is possible for log_checkpoints to change while a
6972 : : * checkpoint proceeds, we always accumulate stats, even if
6973 : : * log_checkpoints is currently off.
6974 : : */
6694 tgl@sss.pgh.pa.us 6975 [ + - + - :CBC 16566 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ - + - +
+ ]
6976 : 1506 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
6977 : :
6978 : : /*
6979 : : * Let smgr prepare for checkpoint; this has to happen outside the
6980 : : * critical section and before we determine the REDO pointer. Note that
6981 : : * smgr must not do anything that'd have to be undone if we decide no
6982 : : * checkpoint is needed.
6983 : : */
1321 tmunro@postgresql.or 6984 : 1506 : SyncPreCheckpoint();
6985 : :
6986 : : /*
6987 : : * Use a critical section to force system panic if we have trouble.
6988 : : */
8794 tgl@sss.pgh.pa.us 6989 : 1506 : START_CRIT_SECTION();
6990 : :
9527 vadim4o@yahoo.com 6991 [ + + ]: 1506 : if (shutdown)
6992 : : {
6095 heikki.linnakangas@i 6993 : 601 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9527 vadim4o@yahoo.com 6994 : 601 : ControlFile->state = DB_SHUTDOWNING;
6995 : 601 : UpdateControlFile();
6095 heikki.linnakangas@i 6996 : 601 : LWLockRelease(ControlFileLock);
6997 : : }
6998 : :
6999 : : /* Begin filling in the checkpoint WAL record */
8194 tgl@sss.pgh.pa.us 7000 [ + - + - : 18072 : MemSet(&checkPoint, 0, sizeof(checkPoint));
+ - + - +
+ ]
6462 7001 : 1506 : checkPoint.time = (pg_time_t) time(NULL);
7002 : :
7003 : : /*
7004 : : * For Hot Standby, derive the oldestActiveXid before we fix the redo
7005 : : * pointer. This allows us to begin accumulating changes to assemble our
7006 : : * starting snapshot of locks and transactions.
7007 : : */
5108 simon@2ndQuadrant.co 7008 [ + + + + ]: 1506 : if (!shutdown && XLogStandbyInfoActive())
96 akapila@postgresql.o 7009 :GNC 857 : checkPoint.oldestActiveXid = GetOldestActiveTransactionId(false, true);
7010 : : else
5108 simon@2ndQuadrant.co 7011 :CBC 649 : checkPoint.oldestActiveXid = InvalidTransactionId;
7012 : :
7013 : : /*
7014 : : * Get location of last important record before acquiring insert locks (as
7015 : : * GetLastImportantRecPtr() also locks WAL locks).
7016 : : */
3231 andres@anarazel.de 7017 : 1506 : last_important_lsn = GetLastImportantRecPtr();
7018 : :
7019 : : /*
7020 : : * If this isn't a shutdown or forced checkpoint, and if there has been no
7021 : : * WAL activity requiring a checkpoint, skip it. The idea here is to
7022 : : * avoid inserting duplicate checkpoints when the system is idle.
7023 : : */
5968 heikki.linnakangas@i 7024 [ + + ]: 1506 : if ((flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY |
7025 : : CHECKPOINT_FORCE)) == 0)
7026 : : {
3231 andres@anarazel.de 7027 [ + + ]: 189 : if (last_important_lsn == ControlFile->checkPoint)
7028 : : {
8994 tgl@sss.pgh.pa.us 7029 [ - + ]: 2 : END_CRIT_SECTION();
3231 andres@anarazel.de 7030 [ - + ]: 2 : ereport(DEBUG1,
7031 : : (errmsg_internal("checkpoint skipped because system is idle")));
392 fujii@postgresql.org 7032 : 2 : return false;
7033 : : }
7034 : : }
7035 : :
7036 : : /*
7037 : : * An end-of-recovery checkpoint is created before anyone is allowed to
7038 : : * write WAL. To allow us to write the checkpoint record, temporarily
7039 : : * enable XLogInsertAllowed.
7040 : : */
5905 heikki.linnakangas@i 7041 [ + + ]: 1504 : if (flags & CHECKPOINT_END_OF_RECOVERY)
1463 rhaas@postgresql.org 7042 : 28 : oldXLogAllowed = LocalSetXLogInsertAllowed();
7043 : :
1447 7044 : 1504 : checkPoint.ThisTimeLineID = XLogCtl->InsertTimeLineID;
4641 heikki.linnakangas@i 7045 [ + + ]: 1504 : if (flags & CHECKPOINT_END_OF_RECOVERY)
7046 : 28 : checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
7047 : : else
1452 rhaas@postgresql.org 7048 : 1476 : checkPoint.PrevTimeLineID = checkPoint.ThisTimeLineID;
7049 : :
7050 : : /*
7051 : : * We must block concurrent insertions while examining insert state.
7052 : : */
739 7053 : 1504 : WALInsertLockAcquireExclusive();
7054 : :
7055 : 1504 : checkPoint.fullPageWrites = Insert->fullPageWrites;
466 7056 : 1504 : checkPoint.wal_level = wal_level;
7057 : :
739 7058 [ + + ]: 1504 : if (shutdown)
7059 : : {
7060 : 601 : XLogRecPtr curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
7061 : :
7062 : : /*
7063 : : * Compute new REDO record ptr = location of next XLOG record.
7064 : : *
7065 : : * Since this is a shutdown checkpoint, there can't be any concurrent
7066 : : * WAL insertion.
7067 : : */
7068 [ + - ]: 601 : freespace = INSERT_FREESPACE(curInsert);
7069 [ - + ]: 601 : if (freespace == 0)
7070 : : {
739 rhaas@postgresql.org 7071 [ # # ]:UBC 0 : if (XLogSegmentOffset(curInsert, wal_segment_size) == 0)
7072 : 0 : curInsert += SizeOfXLogLongPHD;
7073 : : else
7074 : 0 : curInsert += SizeOfXLogShortPHD;
7075 : : }
739 rhaas@postgresql.org 7076 :CBC 601 : checkPoint.redo = curInsert;
7077 : :
7078 : : /*
7079 : : * Here we update the shared RedoRecPtr for future XLogInsert calls;
7080 : : * this must be done while holding all the insertion locks.
7081 : : *
7082 : : * Note: if we fail to complete the checkpoint, RedoRecPtr will be
7083 : : * left pointing past where it really needs to point. This is okay;
7084 : : * the only consequence is that XLogInsert might back up whole buffers
7085 : : * that it didn't really need to. We can't postpone advancing
7086 : : * RedoRecPtr because XLogInserts that happen while we are dumping
7087 : : * buffers must assume that their buffer changes are not included in
7088 : : * the checkpoint.
7089 : : */
7090 : 601 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
7091 : : }
7092 : :
7093 : : /*
7094 : : * Now we can release the WAL insertion locks, allowing other xacts to
7095 : : * proceed while we are flushing disk buffers.
7096 : : */
4238 heikki.linnakangas@i 7097 : 1504 : WALInsertLockRelease();
7098 : :
7099 : : /*
7100 : : * If this is an online checkpoint, we have not yet determined the redo
7101 : : * point. We do so now by inserting the special XLOG_CHECKPOINT_REDO
7102 : : * record; the LSN at which it starts becomes the new redo pointer. We
7103 : : * don't do this for a shutdown checkpoint, because in that case no WAL
7104 : : * can be written between the redo point and the insertion of the
7105 : : * checkpoint record itself, so the checkpoint record itself serves to
7106 : : * mark the redo point.
7107 : : */
739 rhaas@postgresql.org 7108 [ + + ]: 1504 : if (!shutdown)
7109 : : {
7110 : : /* Include WAL level in record for WAL summarizer's benefit. */
7111 : 903 : XLogBeginInsert();
258 peter@eisentraut.org 7112 : 903 : XLogRegisterData(&wal_level, sizeof(wal_level));
739 rhaas@postgresql.org 7113 : 903 : (void) XLogInsert(RM_XLOG_ID, XLOG_CHECKPOINT_REDO);
7114 : :
7115 : : /*
7116 : : * XLogInsertRecord will have updated XLogCtl->Insert.RedoRecPtr in
7117 : : * shared memory and RedoRecPtr in backend-local memory, but we need
7118 : : * to copy that into the record that will be inserted when the
7119 : : * checkpoint is complete.
7120 : : */
7121 : 903 : checkPoint.redo = RedoRecPtr;
7122 : : }
7123 : :
7124 : : /* Update the info_lck-protected copy of RedoRecPtr as well */
4053 andres@anarazel.de 7125 [ - + ]: 1504 : SpinLockAcquire(&XLogCtl->info_lck);
7126 : 1504 : XLogCtl->RedoRecPtr = checkPoint.redo;
7127 : 1504 : SpinLockRelease(&XLogCtl->info_lck);
7128 : :
7129 : : /*
7130 : : * If enabled, log checkpoint start. We postpone this until now so as not
7131 : : * to log anything if we decided to skip the checkpoint.
7132 : : */
6694 tgl@sss.pgh.pa.us 7133 [ + + ]: 1504 : if (log_checkpoints)
6095 heikki.linnakangas@i 7134 : 1212 : LogCheckpointStart(flags, false);
7135 : :
7136 : : /* Update the process title */
1778 michael@paquier.xyz 7137 : 1504 : update_checkpoint_display(flags, false, false);
7138 : :
7139 : : TRACE_POSTGRESQL_CHECKPOINT_START(flags);
7140 : :
7141 : : /*
7142 : : * Get the other info we need for the checkpoint record.
7143 : : *
7144 : : * We don't need to save oldestClogXid in the checkpoint, it only matters
7145 : : * for the short period in which clog is being truncated, and if we crash
7146 : : * during that we'll redo the clog truncation and fix up oldestClogXid
7147 : : * there.
7148 : : */
4224 heikki.linnakangas@i 7149 : 1504 : LWLockAcquire(XidGenLock, LW_SHARED);
689 7150 : 1504 : checkPoint.nextXid = TransamVariables->nextXid;
7151 : 1504 : checkPoint.oldestXid = TransamVariables->oldestXid;
7152 : 1504 : checkPoint.oldestXidDB = TransamVariables->oldestXidDB;
4224 7153 : 1504 : LWLockRelease(XidGenLock);
7154 : :
3981 alvherre@alvh.no-ip. 7155 : 1504 : LWLockAcquire(CommitTsLock, LW_SHARED);
689 heikki.linnakangas@i 7156 : 1504 : checkPoint.oldestCommitTsXid = TransamVariables->oldestCommitTsXid;
7157 : 1504 : checkPoint.newestCommitTsXid = TransamVariables->newestCommitTsXid;
3981 alvherre@alvh.no-ip. 7158 : 1504 : LWLockRelease(CommitTsLock);
7159 : :
4224 heikki.linnakangas@i 7160 : 1504 : LWLockAcquire(OidGenLock, LW_SHARED);
689 7161 : 1504 : checkPoint.nextOid = TransamVariables->nextOid;
4224 7162 [ + + ]: 1504 : if (!shutdown)
689 7163 : 903 : checkPoint.nextOid += TransamVariables->oidCount;
4224 7164 : 1504 : LWLockRelease(OidGenLock);
7165 : :
7166 : 1504 : MultiXactGetCheckptMulti(shutdown,
7167 : : &checkPoint.nextMulti,
7168 : : &checkPoint.nextMultiOffset,
7169 : : &checkPoint.oldestMulti,
7170 : : &checkPoint.oldestMultiDB);
7171 : :
7172 : : /*
7173 : : * Having constructed the checkpoint record, ensure all shmem disk buffers
7174 : : * and commit-log buffers are flushed to disk.
7175 : : *
7176 : : * This I/O could fail for various reasons. If so, we will fail to
7177 : : * complete the checkpoint, but there is no reason to force a system
7178 : : * panic. Accordingly, exit critical section while doing it.
7179 : : */
7180 [ - + ]: 1504 : END_CRIT_SECTION();
7181 : :
7182 : : /*
7183 : : * In some cases there are groups of actions that must all occur on one
7184 : : * side or the other of a checkpoint record. Before flushing the
7185 : : * checkpoint record we must explicitly wait for any backend currently
7186 : : * performing those groups of actions.
7187 : : *
7188 : : * One example is end of transaction, so we must wait for any transactions
7189 : : * that are currently in commit critical sections. If an xact inserted
7190 : : * its commit record into XLOG just before the REDO point, then a crash
7191 : : * restart from the REDO point would not replay that record, which means
7192 : : * that our flushing had better include the xact's update of pg_xact. So
7193 : : * we wait till he's out of his commit critical section before proceeding.
7194 : : * See notes in RecordTransactionCommit().
7195 : : *
7196 : : * Because we've already released the insertion locks, this test is a bit
7197 : : * fuzzy: it is possible that we will wait for xacts we didn't really need
7198 : : * to wait for. But the delay should be short and it seems better to make
7199 : : * checkpoint take a bit longer than to hold off insertions longer than
7200 : : * necessary. (In fact, the whole reason we have this issue is that xact.c
7201 : : * does commit record XLOG insertion and clog update as two separate steps
7202 : : * protected by different locks, but again that seems best on grounds of
7203 : : * minimizing lock contention.)
7204 : : *
7205 : : * A transaction that has not yet set delayChkptFlags when we look cannot
7206 : : * be at risk, since it has not inserted its commit record yet; and one
7207 : : * that's already cleared it is not at risk either, since it's done fixing
7208 : : * clog and we will correctly flush the update below. So we cannot miss
7209 : : * any xacts we need to wait for.
7210 : : */
1313 rhaas@postgresql.org 7211 : 1504 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_START);
4711 simon@2ndQuadrant.co 7212 [ + + ]: 1504 : if (nvxids > 0)
7213 : : {
7214 : : do
7215 : : {
7216 : : /*
7217 : : * Keep absorbing fsync requests while we wait. There could even
7218 : : * be a deadlock if we don't, if the process that prevents the
7219 : : * checkpoint is trying to add a request to the queue.
7220 : : */
493 heikki.linnakangas@i 7221 : 21 : AbsorbSyncRequests();
7222 : :
745 tmunro@postgresql.or 7223 : 21 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_START);
6556 bruce@momjian.us 7224 : 21 : pg_usleep(10000L); /* wait for 10 msec */
745 tmunro@postgresql.or 7225 : 21 : pgstat_report_wait_end();
1313 rhaas@postgresql.org 7226 [ + + ]: 21 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7227 : : DELAY_CHKPT_START));
7228 : : }
4711 simon@2ndQuadrant.co 7229 : 1504 : pfree(vxids);
7230 : :
6696 tgl@sss.pgh.pa.us 7231 : 1504 : CheckPointGuts(checkPoint.redo, flags);
7232 : :
1313 rhaas@postgresql.org 7233 : 1504 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_COMPLETE);
7234 [ - + ]: 1504 : if (nvxids > 0)
7235 : : {
7236 : : do
7237 : : {
493 heikki.linnakangas@i 7238 :UBC 0 : AbsorbSyncRequests();
7239 : :
745 tmunro@postgresql.or 7240 : 0 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_COMPLETE);
1313 rhaas@postgresql.org 7241 : 0 : pg_usleep(10000L); /* wait for 10 msec */
745 tmunro@postgresql.or 7242 : 0 : pgstat_report_wait_end();
1313 rhaas@postgresql.org 7243 [ # # ]: 0 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7244 : : DELAY_CHKPT_COMPLETE));
7245 : : }
1313 rhaas@postgresql.org 7246 :CBC 1504 : pfree(vxids);
7247 : :
7248 : : /*
7249 : : * Take a snapshot of running transactions and write this to WAL. This
7250 : : * allows us to reconstruct the state of running transactions during
7251 : : * archive recovery, if required. Skip, if this info disabled.
7252 : : *
7253 : : * If we are shutting down, or Startup process is completing crash
7254 : : * recovery we don't need to write running xact data.
7255 : : */
5791 simon@2ndQuadrant.co 7256 [ + + + + ]: 1504 : if (!shutdown && XLogStandbyInfoActive())
4712 tgl@sss.pgh.pa.us 7257 : 855 : LogStandbySnapshot();
7258 : :
8206 7259 : 1504 : START_CRIT_SECTION();
7260 : :
7261 : : /*
7262 : : * Now insert the checkpoint record into XLOG.
7263 : : */
3994 heikki.linnakangas@i 7264 : 1504 : XLogBeginInsert();
258 peter@eisentraut.org 7265 : 1504 : XLogRegisterData(&checkPoint, sizeof(checkPoint));
8994 tgl@sss.pgh.pa.us 7266 [ + + ]: 1504 : recptr = XLogInsert(RM_XLOG_ID,
7267 : : shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
7268 : : XLOG_CHECKPOINT_ONLINE);
7269 : :
7270 : 1504 : XLogFlush(recptr);
7271 : :
7272 : : /*
7273 : : * We mustn't write any new WAL after a shutdown checkpoint, or it will be
7274 : : * overwritten at next startup. No-one should even try, this just allows
7275 : : * sanity-checking. In the case of an end-of-recovery checkpoint, we want
7276 : : * to just temporarily disable writing until the system has exited
7277 : : * recovery.
7278 : : */
5967 7279 [ + + ]: 1504 : if (shutdown)
7280 : : {
7281 [ + + ]: 601 : if (flags & CHECKPOINT_END_OF_RECOVERY)
1463 rhaas@postgresql.org 7282 : 28 : LocalXLogInsertAllowed = oldXLogAllowed;
7283 : : else
5722 bruce@momjian.us 7284 : 573 : LocalXLogInsertAllowed = 0; /* never again write WAL */
7285 : : }
7286 : :
7287 : : /*
7288 : : * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
7289 : : * = end of actual checkpoint record.
7290 : : */
4686 alvherre@alvh.no-ip. 7291 [ + + - + ]: 1504 : if (shutdown && checkPoint.redo != ProcLastRecPtr)
8134 tgl@sss.pgh.pa.us 7292 [ # # ]:UBC 0 : ereport(PANIC,
7293 : : (errmsg("concurrent write-ahead log activity while database system is shutting down")));
7294 : :
7295 : : /*
7296 : : * Remember the prior checkpoint's redo ptr for
7297 : : * UpdateCheckPointDistanceEstimate()
7298 : : */
3899 heikki.linnakangas@i 7299 :CBC 1504 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7300 : :
7301 : : /*
7302 : : * Update the control file.
7303 : : */
8794 tgl@sss.pgh.pa.us 7304 : 1504 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9527 vadim4o@yahoo.com 7305 [ + + ]: 1504 : if (shutdown)
7306 : 601 : ControlFile->state = DB_SHUTDOWNED;
8994 tgl@sss.pgh.pa.us 7307 : 1504 : ControlFile->checkPoint = ProcLastRecPtr;
7308 : 1504 : ControlFile->checkPointCopy = checkPoint;
7309 : : /* crash recovery should always recover to the end of WAL */
4687 alvherre@alvh.no-ip. 7310 : 1504 : ControlFile->minRecoveryPoint = InvalidXLogRecPtr;
4710 heikki.linnakangas@i 7311 : 1504 : ControlFile->minRecoveryPointTLI = 0;
7312 : :
7313 : : /*
7314 : : * Persist unloggedLSN value. It's reset on crash recovery, so this goes
7315 : : * unused on non-shutdown checkpoints, but seems useful to store it always
7316 : : * for debugging purposes.
7317 : : */
606 nathan@postgresql.or 7318 : 1504 : ControlFile->unloggedLSN = pg_atomic_read_membarrier_u64(&XLogCtl->unloggedLSN);
7319 : :
9527 vadim4o@yahoo.com 7320 : 1504 : UpdateControlFile();
8794 tgl@sss.pgh.pa.us 7321 : 1504 : LWLockRelease(ControlFileLock);
7322 : :
7323 : : /*
7324 : : * We are now done with critical updates; no need for system panic if we
7325 : : * have trouble while fooling with old log segments.
7326 : : */
8206 7327 [ - + ]: 1504 : END_CRIT_SECTION();
7328 : :
7329 : : /*
7330 : : * WAL summaries end when the next XLOG_CHECKPOINT_REDO or
7331 : : * XLOG_CHECKPOINT_SHUTDOWN record is reached. This is the first point
7332 : : * where (a) we're not inside of a critical section and (b) we can be
7333 : : * certain that the relevant record has been flushed to disk, which must
7334 : : * happen before it can be summarized.
7335 : : *
7336 : : * If this is a shutdown checkpoint, then this happens reasonably
7337 : : * promptly: we've only just inserted and flushed the
7338 : : * XLOG_CHECKPOINT_SHUTDOWN record. If this is not a shutdown checkpoint,
7339 : : * then this might not be very prompt at all: the XLOG_CHECKPOINT_REDO
7340 : : * record was written before we began flushing data to disk, and that
7341 : : * could be many minutes ago at this point. However, we don't XLogFlush()
7342 : : * after inserting that record, so we're not guaranteed that it's on disk
7343 : : * until after the above call that flushes the XLOG_CHECKPOINT_ONLINE
7344 : : * record.
7345 : : */
360 heikki.linnakangas@i 7346 : 1504 : WakeupWalSummarizer();
7347 : :
7348 : : /*
7349 : : * Let smgr do post-checkpoint cleanup (eg, deleting old files).
7350 : : */
2398 tmunro@postgresql.or 7351 : 1504 : SyncPostCheckpoint();
7352 : :
7353 : : /*
7354 : : * Update the average distance between checkpoints if the prior checkpoint
7355 : : * exists.
7356 : : */
3899 heikki.linnakangas@i 7357 [ + - ]: 1504 : if (PriorRedoPtr != InvalidXLogRecPtr)
7358 : 1504 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7359 : :
7360 : : INJECTION_POINT("checkpoint-before-old-wal-removal", NULL);
7361 : :
7362 : : /*
7363 : : * Delete old log files, those no longer needed for last checkpoint to
7364 : : * prevent the disk holding the xlog from growing full.
7365 : : */
2652 michael@paquier.xyz 7366 : 1504 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7367 : 1504 : KeepLogSeg(recptr, &_logSegNo);
250 akapila@postgresql.o 7368 [ + + ]: 1504 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED | RS_INVAL_IDLE_TIMEOUT,
7369 : : _logSegNo, InvalidOid,
7370 : : InvalidTransactionId))
7371 : : {
7372 : : /*
7373 : : * Some slots have been invalidated; recalculate the old-segment
7374 : : * horizon, starting again from RedoRecPtr.
7375 : : */
1564 alvherre@alvh.no-ip. 7376 : 3 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7377 : 3 : KeepLogSeg(recptr, &_logSegNo);
7378 : : }
2652 michael@paquier.xyz 7379 : 1504 : _logSegNo--;
1452 rhaas@postgresql.org 7380 : 1504 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, recptr,
7381 : : checkPoint.ThisTimeLineID);
7382 : :
7383 : : /*
7384 : : * Make more log segments if needed. (Do this after recycling old log
7385 : : * segments, since that may supply some of the needed files.)
7386 : : */
8994 tgl@sss.pgh.pa.us 7387 [ + + ]: 1504 : if (!shutdown)
1452 rhaas@postgresql.org 7388 : 903 : PreallocXlogFiles(recptr, checkPoint.ThisTimeLineID);
7389 : :
7390 : : /*
7391 : : * Truncate pg_subtrans if possible. We can throw away all data before
7392 : : * the oldest XMIN of any running transaction. No future transaction will
7393 : : * attempt to reference any pg_subtrans entry older than that (see Asserts
7394 : : * in subtrans.c). During recovery, though, we mustn't do this because
7395 : : * StartupSUBTRANS hasn't been called yet.
7396 : : */
5967 tgl@sss.pgh.pa.us 7397 [ + + ]: 1504 : if (!RecoveryInProgress())
1902 andres@anarazel.de 7398 : 1476 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7399 : :
7400 : : /* Real work is done; log and update stats. */
4953 rhaas@postgresql.org 7401 : 1504 : LogCheckpointEnd(false);
7402 : :
7403 : : /* Reset the process title */
1778 michael@paquier.xyz 7404 : 1504 : update_checkpoint_display(flags, false, true);
7405 : :
7406 : : TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
7407 : : NBuffers,
7408 : : CheckpointStats.ckpt_segs_added,
7409 : : CheckpointStats.ckpt_segs_removed,
7410 : : CheckpointStats.ckpt_segs_recycled);
7411 : :
392 fujii@postgresql.org 7412 : 1504 : return true;
7413 : : }
7414 : :
7415 : : /*
7416 : : * Mark the end of recovery in WAL though without running a full checkpoint.
7417 : : * We can expect that a restartpoint is likely to be in progress as we
7418 : : * do this, though we are unwilling to wait for it to complete.
7419 : : *
7420 : : * CreateRestartPoint() allows for the case where recovery may end before
7421 : : * the restartpoint completes so there is no concern of concurrent behaviour.
7422 : : */
7423 : : static void
4654 simon@2ndQuadrant.co 7424 : 41 : CreateEndOfRecoveryRecord(void)
7425 : : {
7426 : : xl_end_of_recovery xlrec;
7427 : : XLogRecPtr recptr;
7428 : :
7429 : : /* sanity check */
7430 [ - + ]: 41 : if (!RecoveryInProgress())
4654 simon@2ndQuadrant.co 7431 [ # # ]:UBC 0 : elog(ERROR, "can only be used to end recovery");
7432 : :
3965 heikki.linnakangas@i 7433 :CBC 41 : xlrec.end_time = GetCurrentTimestamp();
466 rhaas@postgresql.org 7434 : 41 : xlrec.wal_level = wal_level;
7435 : :
4238 heikki.linnakangas@i 7436 : 41 : WALInsertLockAcquireExclusive();
1447 rhaas@postgresql.org 7437 : 41 : xlrec.ThisTimeLineID = XLogCtl->InsertTimeLineID;
4641 heikki.linnakangas@i 7438 : 41 : xlrec.PrevTimeLineID = XLogCtl->PrevTimeLineID;
4238 7439 : 41 : WALInsertLockRelease();
7440 : :
4654 simon@2ndQuadrant.co 7441 : 41 : START_CRIT_SECTION();
7442 : :
3994 heikki.linnakangas@i 7443 : 41 : XLogBeginInsert();
258 peter@eisentraut.org 7444 : 41 : XLogRegisterData(&xlrec, sizeof(xl_end_of_recovery));
3994 heikki.linnakangas@i 7445 : 41 : recptr = XLogInsert(RM_XLOG_ID, XLOG_END_OF_RECOVERY);
7446 : :
4652 simon@2ndQuadrant.co 7447 : 41 : XLogFlush(recptr);
7448 : :
7449 : : /*
7450 : : * Update the control file so that crash recovery can follow the timeline
7451 : : * changes to this point.
7452 : : */
7453 : 41 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7454 : 41 : ControlFile->minRecoveryPoint = recptr;
1452 rhaas@postgresql.org 7455 : 41 : ControlFile->minRecoveryPointTLI = xlrec.ThisTimeLineID;
4652 simon@2ndQuadrant.co 7456 : 41 : UpdateControlFile();
7457 : 41 : LWLockRelease(ControlFileLock);
7458 : :
4654 7459 [ - + ]: 41 : END_CRIT_SECTION();
7460 : 41 : }
7461 : :
7462 : : /*
7463 : : * Write an OVERWRITE_CONTRECORD message.
7464 : : *
7465 : : * When on WAL replay we expect a continuation record at the start of a page
7466 : : * that is not there, recovery ends and WAL writing resumes at that point.
7467 : : * But it's wrong to resume writing new WAL back at the start of the record
7468 : : * that was broken, because downstream consumers of that WAL (physical
7469 : : * replicas) are not prepared to "rewind". So the first action after
7470 : : * finishing replay of all valid WAL must be to write a record of this type
7471 : : * at the point where the contrecord was missing; to support xlogreader
7472 : : * detecting the special case, XLP_FIRST_IS_OVERWRITE_CONTRECORD is also added
7473 : : * to the page header where the record occurs. xlogreader has an ad-hoc
7474 : : * mechanism to report metadata about the broken record, which is what we
7475 : : * use here.
7476 : : *
7477 : : * At replay time, XLP_FIRST_IS_OVERWRITE_CONTRECORD instructs xlogreader to
7478 : : * skip the record it was reading, and pass back the LSN of the skipped
7479 : : * record, so that its caller can verify (on "replay" of that record) that the
7480 : : * XLOG_OVERWRITE_CONTRECORD matches what was effectively overwritten.
7481 : : *
7482 : : * 'aborted_lsn' is the beginning position of the record that was incomplete.
7483 : : * It is included in the WAL record. 'pagePtr' and 'newTLI' point to the
7484 : : * beginning of the XLOG page where the record is to be inserted. They must
7485 : : * match the current WAL insert position, they're passed here just so that we
7486 : : * can verify that.
7487 : : */
7488 : : static XLogRecPtr
1349 heikki.linnakangas@i 7489 : 11 : CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn, XLogRecPtr pagePtr,
7490 : : TimeLineID newTLI)
7491 : : {
7492 : : xl_overwrite_contrecord xlrec;
7493 : : XLogRecPtr recptr;
7494 : : XLogPageHeader pagehdr;
7495 : : XLogRecPtr startPos;
7496 : :
7497 : : /* sanity checks */
1489 alvherre@alvh.no-ip. 7498 [ - + ]: 11 : if (!RecoveryInProgress())
1489 alvherre@alvh.no-ip. 7499 [ # # ]:UBC 0 : elog(ERROR, "can only be used at end of recovery");
1349 heikki.linnakangas@i 7500 [ - + ]:CBC 11 : if (pagePtr % XLOG_BLCKSZ != 0)
112 alvherre@kurilemu.de 7501 [ # # ]:UNC 0 : elog(ERROR, "invalid position for missing continuation record %X/%08X",
7502 : : LSN_FORMAT_ARGS(pagePtr));
7503 : :
7504 : : /* The current WAL insert position should be right after the page header */
1349 heikki.linnakangas@i 7505 :CBC 11 : startPos = pagePtr;
7506 [ + + ]: 11 : if (XLogSegmentOffset(startPos, wal_segment_size) == 0)
7507 : 1 : startPos += SizeOfXLogLongPHD;
7508 : : else
7509 : 10 : startPos += SizeOfXLogShortPHD;
7510 : 11 : recptr = GetXLogInsertRecPtr();
7511 [ - + ]: 11 : if (recptr != startPos)
112 alvherre@kurilemu.de 7512 [ # # ]:UNC 0 : elog(ERROR, "invalid WAL insert position %X/%08X for OVERWRITE_CONTRECORD",
7513 : : LSN_FORMAT_ARGS(recptr));
7514 : :
1489 alvherre@alvh.no-ip. 7515 :CBC 11 : START_CRIT_SECTION();
7516 : :
7517 : : /*
7518 : : * Initialize the XLOG page header (by GetXLogBuffer), and set the
7519 : : * XLP_FIRST_IS_OVERWRITE_CONTRECORD flag.
7520 : : *
7521 : : * No other backend is allowed to write WAL yet, so acquiring the WAL
7522 : : * insertion lock is just pro forma.
7523 : : */
1349 heikki.linnakangas@i 7524 : 11 : WALInsertLockAcquire();
7525 : 11 : pagehdr = (XLogPageHeader) GetXLogBuffer(pagePtr, newTLI);
7526 : 11 : pagehdr->xlp_info |= XLP_FIRST_IS_OVERWRITE_CONTRECORD;
7527 : 11 : WALInsertLockRelease();
7528 : :
7529 : : /*
7530 : : * Insert the XLOG_OVERWRITE_CONTRECORD record as the first record on the
7531 : : * page. We know it becomes the first record, because no other backend is
7532 : : * allowed to write WAL yet.
7533 : : */
1489 alvherre@alvh.no-ip. 7534 : 11 : XLogBeginInsert();
1349 heikki.linnakangas@i 7535 : 11 : xlrec.overwritten_lsn = aborted_lsn;
7536 : 11 : xlrec.overwrite_time = GetCurrentTimestamp();
258 peter@eisentraut.org 7537 : 11 : XLogRegisterData(&xlrec, sizeof(xl_overwrite_contrecord));
1489 alvherre@alvh.no-ip. 7538 : 11 : recptr = XLogInsert(RM_XLOG_ID, XLOG_OVERWRITE_CONTRECORD);
7539 : :
7540 : : /* check that the record was inserted to the right place */
1349 heikki.linnakangas@i 7541 [ - + ]: 11 : if (ProcLastRecPtr != startPos)
112 alvherre@kurilemu.de 7542 [ # # ]:UNC 0 : elog(ERROR, "OVERWRITE_CONTRECORD was inserted to unexpected position %X/%08X",
7543 : : LSN_FORMAT_ARGS(ProcLastRecPtr));
7544 : :
1489 alvherre@alvh.no-ip. 7545 :CBC 11 : XLogFlush(recptr);
7546 : :
7547 [ - + ]: 11 : END_CRIT_SECTION();
7548 : :
7549 : 11 : return recptr;
7550 : : }
7551 : :
7552 : : /*
7553 : : * Flush all data in shared memory to disk, and fsync
7554 : : *
7555 : : * This is the common code shared between regular checkpoints and
7556 : : * recovery restartpoints.
7557 : : */
7558 : : static void
6696 tgl@sss.pgh.pa.us 7559 : 1701 : CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
7560 : : {
5741 7561 : 1701 : CheckPointRelationMap();
774 akapila@postgresql.o 7562 : 1701 : CheckPointReplicationSlots(flags & CHECKPOINT_IS_SHUTDOWN);
4256 rhaas@postgresql.org 7563 : 1701 : CheckPointSnapBuild();
7564 : 1701 : CheckPointLogicalRewriteHeap();
3834 andres@anarazel.de 7565 : 1701 : CheckPointReplicationOrigin();
7566 : :
7567 : : /* Write out all dirty data in SLRUs and the main buffer pool */
7568 : : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_START(flags);
1858 tmunro@postgresql.or 7569 : 1701 : CheckpointStats.ckpt_write_t = GetCurrentTimestamp();
7570 : 1701 : CheckPointCLOG();
7571 : 1701 : CheckPointCommitTs();
7572 : 1701 : CheckPointSUBTRANS();
7573 : 1701 : CheckPointMultiXact();
7574 : 1701 : CheckPointPredicate();
7575 : 1701 : CheckPointBuffers(flags);
7576 : :
7577 : : /* Perform all queued up fsyncs */
7578 : : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_SYNC_START();
7579 : 1701 : CheckpointStats.ckpt_sync_t = GetCurrentTimestamp();
7580 : 1701 : ProcessSyncRequests();
7581 : 1701 : CheckpointStats.ckpt_sync_end_t = GetCurrentTimestamp();
7582 : : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_DONE();
7583 : :
7584 : : /* We deliberately delay 2PC checkpointing as long as possible */
7021 tgl@sss.pgh.pa.us 7585 : 1701 : CheckPointTwoPhase(checkPointRedo);
7586 : 1701 : }
7587 : :
7588 : : /*
7589 : : * Save a checkpoint for recovery restart if appropriate
7590 : : *
7591 : : * This function is called each time a checkpoint record is read from XLOG.
7592 : : * It must determine whether the checkpoint represents a safe restartpoint or
7593 : : * not. If so, the checkpoint record is stashed in shared memory so that
7594 : : * CreateRestartPoint can consult it. (Note that the latter function is
7595 : : * executed by the checkpointer, while this one will be executed by the
7596 : : * startup process.)
7597 : : */
7598 : : static void
1433 rhaas@postgresql.org 7599 : 698 : RecoveryRestartPoint(const CheckPoint *checkPoint, XLogReaderState *record)
7600 : : {
7601 : : /*
7602 : : * Also refrain from creating a restartpoint if we have seen any
7603 : : * references to non-existent pages. Restarting recovery from the
7604 : : * restartpoint would not see the references, so we would lose the
7605 : : * cross-check that the pages belonged to a relation that was dropped
7606 : : * later.
7607 : : */
5078 heikki.linnakangas@i 7608 [ - + ]: 698 : if (XLogHaveInvalidPages())
7609 : : {
686 michael@paquier.xyz 7610 [ # # ]:UBC 0 : elog(DEBUG2,
7611 : : "could not record restart point at %X/%08X because there are unresolved references to invalid pages",
7612 : : LSN_FORMAT_ARGS(checkPoint->redo));
5078 heikki.linnakangas@i 7613 : 0 : return;
7614 : : }
7615 : :
7616 : : /*
7617 : : * Copy the checkpoint record to shared memory, so that checkpointer can
7618 : : * work out the next time it wants to perform a restartpoint.
7619 : : */
4053 andres@anarazel.de 7620 [ - + ]:CBC 698 : SpinLockAcquire(&XLogCtl->info_lck);
1433 rhaas@postgresql.org 7621 : 698 : XLogCtl->lastCheckPointRecPtr = record->ReadRecPtr;
7622 : 698 : XLogCtl->lastCheckPointEndPtr = record->EndRecPtr;
4053 andres@anarazel.de 7623 : 698 : XLogCtl->lastCheckPoint = *checkPoint;
7624 : 698 : SpinLockRelease(&XLogCtl->info_lck);
7625 : : }
7626 : :
7627 : : /*
7628 : : * Establish a restartpoint if possible.
7629 : : *
7630 : : * This is similar to CreateCheckPoint, but is used during WAL recovery
7631 : : * to establish a point from which recovery can roll forward without
7632 : : * replaying the entire recovery log.
7633 : : *
7634 : : * Returns true if a new restartpoint was established. We can only establish
7635 : : * a restartpoint if we have replayed a safe checkpoint record since last
7636 : : * restartpoint.
7637 : : */
7638 : : bool
6095 heikki.linnakangas@i 7639 : 601 : CreateRestartPoint(int flags)
7640 : : {
7641 : : XLogRecPtr lastCheckPointRecPtr;
7642 : : XLogRecPtr lastCheckPointEndPtr;
7643 : : CheckPoint lastCheckPoint;
7644 : : XLogRecPtr PriorRedoPtr;
7645 : : XLogRecPtr receivePtr;
7646 : : XLogRecPtr replayPtr;
7647 : : TimeLineID replayTLI;
7648 : : XLogRecPtr endptr;
7649 : : XLogSegNo _logSegNo;
7650 : : TimestampTz xtime;
7651 : :
7652 : : /* Concurrent checkpoint/restartpoint cannot happen */
1267 michael@paquier.xyz 7653 [ + - - + ]: 601 : Assert(!IsUnderPostmaster || MyBackendType == B_CHECKPOINTER);
7654 : :
7655 : : /* Get a local copy of the last safe checkpoint record. */
4053 andres@anarazel.de 7656 [ - + ]: 601 : SpinLockAcquire(&XLogCtl->info_lck);
7657 : 601 : lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
3287 rhaas@postgresql.org 7658 : 601 : lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
4053 andres@anarazel.de 7659 : 601 : lastCheckPoint = XLogCtl->lastCheckPoint;
7660 : 601 : SpinLockRelease(&XLogCtl->info_lck);
7661 : :
7662 : : /*
7663 : : * Check that we're still in recovery mode. It's ok if we exit recovery
7664 : : * mode after this check, the restart point is valid anyway.
7665 : : */
6095 heikki.linnakangas@i 7666 [ - + ]: 601 : if (!RecoveryInProgress())
7667 : : {
6095 heikki.linnakangas@i 7668 [ # # ]:UBC 0 : ereport(DEBUG2,
7669 : : (errmsg_internal("skipping restartpoint, recovery has already ended")));
7670 : 0 : return false;
7671 : : }
7672 : :
7673 : : /*
7674 : : * If the last checkpoint record we've replayed is already our last
7675 : : * restartpoint, we can't perform a new restart point. We still update
7676 : : * minRecoveryPoint in that case, so that if this is a shutdown restart
7677 : : * point, we won't start up earlier than before. That's not strictly
7678 : : * necessary, but when hot standby is enabled, it would be rather weird if
7679 : : * the database opened up for read-only connections at a point-in-time
7680 : : * before the last shutdown. Such time travel is still possible in case of
7681 : : * immediate shutdown, though.
7682 : : *
7683 : : * We don't explicitly advance minRecoveryPoint when we do create a
7684 : : * restartpoint. It's assumed that flushing the buffers will do that as a
7685 : : * side-effect.
7686 : : */
6095 heikki.linnakangas@i 7687 [ + + ]:CBC 601 : if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
4686 alvherre@alvh.no-ip. 7688 [ + + ]: 269 : lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
7689 : : {
6095 heikki.linnakangas@i 7690 [ - + ]: 404 : ereport(DEBUG2,
7691 : : errmsg_internal("skipping restartpoint, already performed at %X/%08X",
7692 : : LSN_FORMAT_ARGS(lastCheckPoint.redo)));
7693 : :
7694 : 404 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5625 rhaas@postgresql.org 7695 [ + + ]: 404 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7696 : : {
7697 : 31 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7698 : 31 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7699 : 31 : UpdateControlFile();
7700 : 31 : LWLockRelease(ControlFileLock);
7701 : : }
6095 heikki.linnakangas@i 7702 : 404 : return false;
7703 : : }
7704 : :
7705 : : /*
7706 : : * Update the shared RedoRecPtr so that the startup process can calculate
7707 : : * the number of segments replayed since last restartpoint, and request a
7708 : : * restartpoint if it exceeds CheckPointSegments.
7709 : : *
7710 : : * Like in CreateCheckPoint(), hold off insertions to update it, although
7711 : : * during recovery this is just pro forma, because no WAL insertions are
7712 : : * happening.
7713 : : */
4238 7714 : 197 : WALInsertLockAcquireExclusive();
3899 7715 : 197 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
4238 7716 : 197 : WALInsertLockRelease();
7717 : :
7718 : : /* Also update the info_lck-protected copy */
4053 andres@anarazel.de 7719 [ - + ]: 197 : SpinLockAcquire(&XLogCtl->info_lck);
7720 : 197 : XLogCtl->RedoRecPtr = lastCheckPoint.redo;
7721 : 197 : SpinLockRelease(&XLogCtl->info_lck);
7722 : :
7723 : : /*
7724 : : * Prepare to accumulate statistics.
7725 : : *
7726 : : * Note: because it is possible for log_checkpoints to change while a
7727 : : * checkpoint proceeds, we always accumulate stats, even if
7728 : : * log_checkpoints is currently off.
7729 : : */
5381 rhaas@postgresql.org 7730 [ + - + - : 2167 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ - + - +
+ ]
7731 : 197 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
7732 : :
7733 [ + - ]: 197 : if (log_checkpoints)
6095 heikki.linnakangas@i 7734 : 197 : LogCheckpointStart(flags, true);
7735 : :
7736 : : /* Update the process title */
1778 michael@paquier.xyz 7737 : 197 : update_checkpoint_display(flags, true, false);
7738 : :
6095 heikki.linnakangas@i 7739 : 197 : CheckPointGuts(lastCheckPoint.redo, flags);
7740 : :
7741 : : /*
7742 : : * This location needs to be after CheckPointGuts() to ensure that some
7743 : : * work has already happened during this checkpoint.
7744 : : */
7745 : : INJECTION_POINT("create-restart-point", NULL);
7746 : :
7747 : : /*
7748 : : * Remember the prior checkpoint's redo ptr for
7749 : : * UpdateCheckPointDistanceEstimate()
7750 : : */
3899 7751 : 197 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7752 : :
7753 : : /*
7754 : : * Update pg_control, using current time. Check that it still shows an
7755 : : * older checkpoint, else do nothing; this is a quick hack to make sure
7756 : : * nothing really bad happens if somehow we get here after the
7757 : : * end-of-recovery checkpoint.
7758 : : */
6095 7759 : 197 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
1267 michael@paquier.xyz 7760 [ + - ]: 197 : if (ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
7761 : : {
7762 : : /*
7763 : : * Update the checkpoint information. We do this even if the cluster
7764 : : * does not show DB_IN_ARCHIVE_RECOVERY to match with the set of WAL
7765 : : * segments recycled below.
7766 : : */
5967 tgl@sss.pgh.pa.us 7767 : 197 : ControlFile->checkPoint = lastCheckPointRecPtr;
7768 : 197 : ControlFile->checkPointCopy = lastCheckPoint;
7769 : :
7770 : : /*
7771 : : * Ensure minRecoveryPoint is past the checkpoint record and update it
7772 : : * if the control file still shows DB_IN_ARCHIVE_RECOVERY. Normally,
7773 : : * this will have happened already while writing out dirty buffers,
7774 : : * but not necessarily - e.g. because no buffers were dirtied. We do
7775 : : * this because a backup performed in recovery uses minRecoveryPoint
7776 : : * to determine which WAL files must be included in the backup, and
7777 : : * the file (or files) containing the checkpoint record must be
7778 : : * included, at a minimum. Note that for an ordinary restart of
7779 : : * recovery there's no value in having the minimum recovery point any
7780 : : * earlier than this anyway, because redo will begin just after the
7781 : : * checkpoint record.
7782 : : */
1267 michael@paquier.xyz 7783 [ + - ]: 197 : if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY)
7784 : : {
7785 [ + + ]: 197 : if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
7786 : : {
7787 : 18 : ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
7788 : 18 : ControlFile->minRecoveryPointTLI = lastCheckPoint.ThisTimeLineID;
7789 : :
7790 : : /* update local copy */
7791 : 18 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
7792 : 18 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
7793 : : }
7794 [ + + ]: 197 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7795 : 21 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7796 : : }
5967 tgl@sss.pgh.pa.us 7797 : 197 : UpdateControlFile();
7798 : : }
6095 heikki.linnakangas@i 7799 : 197 : LWLockRelease(ControlFileLock);
7800 : :
7801 : : /*
7802 : : * Update the average distance between checkpoints/restartpoints if the
7803 : : * prior checkpoint exists.
7804 : : */
3899 7805 [ + - ]: 197 : if (PriorRedoPtr != InvalidXLogRecPtr)
7806 : 197 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7807 : :
7808 : : /*
7809 : : * Delete old log files, those no longer needed for last restartpoint to
7810 : : * prevent the disk holding the xlog from growing full.
7811 : : */
2652 michael@paquier.xyz 7812 : 197 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7813 : :
7814 : : /*
7815 : : * Retreat _logSegNo using the current end of xlog replayed or received,
7816 : : * whichever is later.
7817 : : */
2028 tmunro@postgresql.or 7818 : 197 : receivePtr = GetWalRcvFlushRecPtr(NULL, NULL);
2652 michael@paquier.xyz 7819 : 197 : replayPtr = GetXLogReplayRecPtr(&replayTLI);
7820 : 197 : endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
7821 : 197 : KeepLogSeg(endptr, &_logSegNo);
250 akapila@postgresql.o 7822 [ + + ]: 197 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED | RS_INVAL_IDLE_TIMEOUT,
7823 : : _logSegNo, InvalidOid,
7824 : : InvalidTransactionId))
7825 : : {
7826 : : /*
7827 : : * Some slots have been invalidated; recalculate the old-segment
7828 : : * horizon, starting again from RedoRecPtr.
7829 : : */
1564 alvherre@alvh.no-ip. 7830 : 1 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7831 : 1 : KeepLogSeg(endptr, &_logSegNo);
7832 : : }
2652 michael@paquier.xyz 7833 : 197 : _logSegNo--;
7834 : :
7835 : : /*
7836 : : * Try to recycle segments on a useful timeline. If we've been promoted
7837 : : * since the beginning of this restartpoint, use the new timeline chosen
7838 : : * at end of recovery. If we're still in recovery, use the timeline we're
7839 : : * currently replaying.
7840 : : *
7841 : : * There is no guarantee that the WAL segments will be useful on the
7842 : : * current timeline; if recovery proceeds to a new timeline right after
7843 : : * this, the pre-allocated WAL segments on this timeline will not be used,
7844 : : * and will go wasted until recycled on the next restartpoint. We'll live
7845 : : * with that.
7846 : : */
1452 rhaas@postgresql.org 7847 [ - + ]: 197 : if (!RecoveryInProgress())
1447 rhaas@postgresql.org 7848 :UBC 0 : replayTLI = XLogCtl->InsertTimeLineID;
7849 : :
1452 rhaas@postgresql.org 7850 :CBC 197 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, endptr, replayTLI);
7851 : :
7852 : : /*
7853 : : * Make more log segments if needed. (Do this after recycling old log
7854 : : * segments, since that may supply some of the needed files.)
7855 : : */
7856 : 197 : PreallocXlogFiles(endptr, replayTLI);
7857 : :
7858 : : /*
7859 : : * Truncate pg_subtrans if possible. We can throw away all data before
7860 : : * the oldest XMIN of any running transaction. No future transaction will
7861 : : * attempt to reference any pg_subtrans entry older than that (see Asserts
7862 : : * in subtrans.c). When hot standby is disabled, though, we mustn't do
7863 : : * this because StartupSUBTRANS hasn't been called yet.
7864 : : */
5537 simon@2ndQuadrant.co 7865 [ + - ]: 197 : if (EnableHotStandby)
1902 andres@anarazel.de 7866 : 197 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7867 : :
7868 : : /* Real work is done; log and update stats. */
4953 rhaas@postgresql.org 7869 : 197 : LogCheckpointEnd(true);
7870 : :
7871 : : /* Reset the process title */
1778 michael@paquier.xyz 7872 : 197 : update_checkpoint_display(flags, true, true);
7873 : :
5595 tgl@sss.pgh.pa.us 7874 : 197 : xtime = GetLatestXTime();
6095 heikki.linnakangas@i 7875 [ + - + - : 197 : ereport((log_checkpoints ? LOG : DEBUG2),
+ + ]
7876 : : errmsg("recovery restart point at %X/%08X",
7877 : : LSN_FORMAT_ARGS(lastCheckPoint.redo)),
7878 : : xtime ? errdetail("Last completed transaction was at log time %s.",
7879 : : timestamptz_to_str(xtime)) : 0);
7880 : :
7881 : : /*
7882 : : * Finally, execute archive_cleanup_command, if any.
7883 : : */
2528 peter_e@gmx.net 7884 [ + - - + ]: 197 : if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
994 michael@paquier.xyz 7885 :UBC 0 : ExecuteRecoveryCommand(archiveCleanupCommand,
7886 : : "archive_cleanup_command",
7887 : : false,
7888 : : WAIT_EVENT_ARCHIVE_CLEANUP_COMMAND);
7889 : :
6095 heikki.linnakangas@i 7890 :CBC 197 : return true;
7891 : : }
7892 : :
7893 : : /*
7894 : : * Report availability of WAL for the given target LSN
7895 : : * (typically a slot's restart_lsn)
7896 : : *
7897 : : * Returns one of the following enum values:
7898 : : *
7899 : : * * WALAVAIL_RESERVED means targetLSN is available and it is in the range of
7900 : : * max_wal_size.
7901 : : *
7902 : : * * WALAVAIL_EXTENDED means it is still available by preserving extra
7903 : : * segments beyond max_wal_size. If max_slot_wal_keep_size is smaller
7904 : : * than max_wal_size, this state is not returned.
7905 : : *
7906 : : * * WALAVAIL_UNRESERVED means it is being lost and the next checkpoint will
7907 : : * remove reserved segments. The walsender using this slot may return to the
7908 : : * above.
7909 : : *
7910 : : * * WALAVAIL_REMOVED means it has been removed. A replication stream on
7911 : : * a slot with this LSN cannot continue. (Any associated walsender
7912 : : * processes should have been terminated already.)
7913 : : *
7914 : : * * WALAVAIL_INVALID_LSN means the slot hasn't been set to reserve WAL.
7915 : : */
7916 : : WALAvailability
2029 alvherre@alvh.no-ip. 7917 : 414 : GetWALAvailability(XLogRecPtr targetLSN)
7918 : : {
7919 : : XLogRecPtr currpos; /* current write LSN */
7920 : : XLogSegNo currSeg; /* segid of currpos */
7921 : : XLogSegNo targetSeg; /* segid of targetLSN */
7922 : : XLogSegNo oldestSeg; /* actual oldest segid */
7923 : : XLogSegNo oldestSegMaxWalSize; /* oldest segid kept by max_wal_size */
7924 : : XLogSegNo oldestSlotSeg; /* oldest segid kept by slot */
7925 : : uint64 keepSegs;
7926 : :
7927 : : /*
7928 : : * slot does not reserve WAL. Either deactivated, or has never been active
7929 : : */
7930 [ + + ]: 414 : if (XLogRecPtrIsInvalid(targetLSN))
7931 : 31 : return WALAVAIL_INVALID_LSN;
7932 : :
7933 : : /*
7934 : : * Calculate the oldest segment currently reserved by all slots,
7935 : : * considering wal_keep_size and max_slot_wal_keep_size. Initialize
7936 : : * oldestSlotSeg to the current segment.
7937 : : */
1932 7938 : 383 : currpos = GetXLogWriteRecPtr();
7939 : 383 : XLByteToSeg(currpos, oldestSlotSeg, wal_segment_size);
2029 7940 : 383 : KeepLogSeg(currpos, &oldestSlotSeg);
7941 : :
7942 : : /*
7943 : : * Find the oldest extant segment file. We get 1 until checkpoint removes
7944 : : * the first WAL segment file since startup, which causes the status being
7945 : : * wrong under certain abnormal conditions but that doesn't actually harm.
7946 : : */
7947 : 383 : oldestSeg = XLogGetLastRemovedSegno() + 1;
7948 : :
7949 : : /* calculate oldest segment by max_wal_size */
7950 : 383 : XLByteToSeg(currpos, currSeg, wal_segment_size);
1951 7951 : 383 : keepSegs = ConvertToXSegs(max_wal_size_mb, wal_segment_size) + 1;
7952 : :
2029 7953 [ + + ]: 383 : if (currSeg > keepSegs)
7954 : 8 : oldestSegMaxWalSize = currSeg - keepSegs;
7955 : : else
7956 : 375 : oldestSegMaxWalSize = 1;
7957 : :
7958 : : /* the segment we care about */
1932 7959 : 383 : XLByteToSeg(targetLSN, targetSeg, wal_segment_size);
7960 : :
7961 : : /*
7962 : : * No point in returning reserved or extended status values if the
7963 : : * targetSeg is known to be lost.
7964 : : */
1951 7965 [ + + ]: 383 : if (targetSeg >= oldestSlotSeg)
7966 : : {
7967 : : /* show "reserved" when targetSeg is within max_wal_size */
7968 [ + + ]: 382 : if (targetSeg >= oldestSegMaxWalSize)
2029 7969 : 380 : return WALAVAIL_RESERVED;
7970 : :
7971 : : /* being retained by slots exceeding max_wal_size */
1951 7972 : 2 : return WALAVAIL_EXTENDED;
7973 : : }
7974 : :
7975 : : /* WAL segments are no longer retained but haven't been removed yet */
7976 [ + - ]: 1 : if (targetSeg >= oldestSeg)
7977 : 1 : return WALAVAIL_UNRESERVED;
7978 : :
7979 : : /* Definitely lost */
2029 alvherre@alvh.no-ip. 7980 :UBC 0 : return WALAVAIL_REMOVED;
7981 : : }
7982 : :
7983 : :
7984 : : /*
7985 : : * Retreat *logSegNo to the last segment that we need to retain because of
7986 : : * either wal_keep_size or replication slots.
7987 : : *
7988 : : * This is calculated by subtracting wal_keep_size from the given xlog
7989 : : * location, recptr and by making sure that that result is below the
7990 : : * requirement of replication slots. For the latter criterion we do consider
7991 : : * the effects of max_slot_wal_keep_size: reserve at most that much space back
7992 : : * from recptr.
7993 : : *
7994 : : * Note about replication slots: if this function calculates a value
7995 : : * that's further ahead than what slots need reserved, then affected
7996 : : * slots need to be invalidated and this function invoked again.
7997 : : * XXX it might be a good idea to rewrite this function so that
7998 : : * invalidation is optionally done here, instead.
7999 : : */
8000 : : static void
4873 heikki.linnakangas@i 8001 :CBC 2088 : KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
8002 : : {
8003 : : XLogSegNo currSegNo;
8004 : : XLogSegNo segno;
8005 : : XLogRecPtr keep;
8006 : :
2029 alvherre@alvh.no-ip. 8007 : 2088 : XLByteToSeg(recptr, currSegNo, wal_segment_size);
8008 : 2088 : segno = currSegNo;
8009 : :
8010 : : /* Calculate how many segments are kept by slots. */
8011 : 2088 : keep = XLogGetReplicationSlotMinimumLSN();
914 nathan@postgresql.or 8012 [ + + + + ]: 2088 : if (keep != InvalidXLogRecPtr && keep < recptr)
8013 : : {
2029 alvherre@alvh.no-ip. 8014 : 638 : XLByteToSeg(keep, segno, wal_segment_size);
8015 : :
8016 : : /*
8017 : : * Account for max_slot_wal_keep_size to avoid keeping more than
8018 : : * configured. However, don't do that during a binary upgrade: if
8019 : : * slots were to be invalidated because of this, it would not be
8020 : : * possible to preserve logical ones during the upgrade.
8021 : : */
108 akapila@postgresql.o 8022 [ + + + - ]: 638 : if (max_slot_wal_keep_size_mb >= 0 && !IsBinaryUpgrade)
8023 : : {
8024 : : uint64 slot_keep_segs;
8025 : :
2029 alvherre@alvh.no-ip. 8026 : 21 : slot_keep_segs =
8027 : 21 : ConvertToXSegs(max_slot_wal_keep_size_mb, wal_segment_size);
8028 : :
8029 [ + + ]: 21 : if (currSegNo - segno > slot_keep_segs)
8030 : 5 : segno = currSegNo - slot_keep_segs;
8031 : : }
8032 : : }
8033 : :
8034 : : /*
8035 : : * If WAL summarization is in use, don't remove WAL that has yet to be
8036 : : * summarized.
8037 : : */
489 rhaas@postgresql.org 8038 : 2088 : keep = GetOldestUnsummarizedLSN(NULL, NULL);
677 8039 [ + + ]: 2088 : if (keep != InvalidXLogRecPtr)
8040 : : {
8041 : : XLogSegNo unsummarized_segno;
8042 : :
8043 : 2 : XLByteToSeg(keep, unsummarized_segno, wal_segment_size);
8044 [ + - ]: 2 : if (unsummarized_segno < segno)
8045 : 2 : segno = unsummarized_segno;
8046 : : }
8047 : :
8048 : : /* but, keep at least wal_keep_size if that's set */
1925 fujii@postgresql.org 8049 [ + + ]: 2088 : if (wal_keep_size_mb > 0)
8050 : : {
8051 : : uint64 keep_segs;
8052 : :
8053 : 74 : keep_segs = ConvertToXSegs(wal_keep_size_mb, wal_segment_size);
8054 [ + - ]: 74 : if (currSegNo - segno < keep_segs)
8055 : : {
8056 : : /* avoid underflow, don't go below 1 */
8057 [ + + ]: 74 : if (currSegNo <= keep_segs)
8058 : 70 : segno = 1;
8059 : : else
8060 : 4 : segno = currSegNo - keep_segs;
8061 : : }
8062 : : }
8063 : :
8064 : : /* don't delete WAL segments newer than the calculated segment */
1932 alvherre@alvh.no-ip. 8065 [ + + ]: 2088 : if (segno < *logSegNo)
4873 heikki.linnakangas@i 8066 : 345 : *logSegNo = segno;
5214 simon@2ndQuadrant.co 8067 : 2088 : }
8068 : :
8069 : : /*
8070 : : * Write a NEXTOID log record
8071 : : */
8072 : : void
9124 vadim4o@yahoo.com 8073 : 591 : XLogPutNextOid(Oid nextOid)
8074 : : {
3994 heikki.linnakangas@i 8075 : 591 : XLogBeginInsert();
258 peter@eisentraut.org 8076 : 591 : XLogRegisterData(&nextOid, sizeof(Oid));
3994 heikki.linnakangas@i 8077 : 591 : (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
8078 : :
8079 : : /*
8080 : : * We need not flush the NEXTOID record immediately, because any of the
8081 : : * just-allocated OIDs could only reach disk as part of a tuple insert or
8082 : : * update that would have its own XLOG record that must follow the NEXTOID
8083 : : * record. Therefore, the standard buffer LSN interlock applied to those
8084 : : * records will ensure no such OID reaches disk before the NEXTOID record
8085 : : * does.
8086 : : *
8087 : : * Note, however, that the above statement only covers state "within" the
8088 : : * database. When we use a generated OID as a file or directory name, we
8089 : : * are in a sense violating the basic WAL rule, because that filesystem
8090 : : * change may reach disk before the NEXTOID WAL record does. The impact
8091 : : * of this is that if a database crash occurs immediately afterward, we
8092 : : * might after restart re-generate the same OID and find that it conflicts
8093 : : * with the leftover file or directory. But since for safety's sake we
8094 : : * always loop until finding a nonconflicting filename, this poses no real
8095 : : * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
8096 : : */
7487 tgl@sss.pgh.pa.us 8097 : 591 : }
8098 : :
8099 : : /*
8100 : : * Write an XLOG SWITCH record.
8101 : : *
8102 : : * Here we just blindly issue an XLogInsert request for the record.
8103 : : * All the magic happens inside XLogInsert.
8104 : : *
8105 : : * The return value is either the end+1 address of the switch record,
8106 : : * or the end+1 address of the prior segment if we did not need to
8107 : : * write a switch record because we are already at segment start.
8108 : : */
8109 : : XLogRecPtr
3231 andres@anarazel.de 8110 : 707 : RequestXLogSwitch(bool mark_unimportant)
8111 : : {
8112 : : XLogRecPtr RecPtr;
8113 : :
8114 : : /* XLOG SWITCH has no data */
3994 heikki.linnakangas@i 8115 : 707 : XLogBeginInsert();
8116 : :
3231 andres@anarazel.de 8117 [ - + ]: 707 : if (mark_unimportant)
3231 andres@anarazel.de 8118 :UBC 0 : XLogSetRecordFlags(XLOG_MARK_UNIMPORTANT);
3994 heikki.linnakangas@i 8119 :CBC 707 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH);
8120 : :
7022 tgl@sss.pgh.pa.us 8121 : 707 : return RecPtr;
8122 : : }
8123 : :
8124 : : /*
8125 : : * Write a RESTORE POINT record
8126 : : */
8127 : : XLogRecPtr
5375 simon@2ndQuadrant.co 8128 : 3 : XLogRestorePoint(const char *rpName)
8129 : : {
8130 : : XLogRecPtr RecPtr;
8131 : : xl_restore_point xlrec;
8132 : :
8133 : 3 : xlrec.rp_time = GetCurrentTimestamp();
4270 tgl@sss.pgh.pa.us 8134 : 3 : strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
8135 : :
3994 heikki.linnakangas@i 8136 : 3 : XLogBeginInsert();
258 peter@eisentraut.org 8137 : 3 : XLogRegisterData(&xlrec, sizeof(xl_restore_point));
8138 : :
3994 heikki.linnakangas@i 8139 : 3 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
8140 : :
5359 rhaas@postgresql.org 8141 [ + - ]: 3 : ereport(LOG,
8142 : : errmsg("restore point \"%s\" created at %X/%08X",
8143 : : rpName, LSN_FORMAT_ARGS(RecPtr)));
8144 : :
5375 simon@2ndQuadrant.co 8145 : 3 : return RecPtr;
8146 : : }
8147 : :
8148 : : /*
8149 : : * Check if any of the GUC parameters that are critical for hot standby
8150 : : * have changed, and update the value in pg_control file if necessary.
8151 : : */
8152 : : static void
5661 heikki.linnakangas@i 8153 : 852 : XLogReportParameters(void)
8154 : : {
8155 [ + + ]: 852 : if (wal_level != ControlFile->wal_level ||
4317 rhaas@postgresql.org 8156 [ + + ]: 631 : wal_log_hints != ControlFile->wal_log_hints ||
5661 heikki.linnakangas@i 8157 [ + + ]: 548 : MaxConnections != ControlFile->MaxConnections ||
4498 rhaas@postgresql.org 8158 [ + + ]: 547 : max_worker_processes != ControlFile->max_worker_processes ||
2449 michael@paquier.xyz 8159 [ + + ]: 546 : max_wal_senders != ControlFile->max_wal_senders ||
5661 heikki.linnakangas@i 8160 [ + + ]: 526 : max_prepared_xacts != ControlFile->max_prepared_xacts ||
3981 alvherre@alvh.no-ip. 8161 [ + - ]: 437 : max_locks_per_xact != ControlFile->max_locks_per_xact ||
8162 [ + + ]: 437 : track_commit_timestamp != ControlFile->track_commit_timestamp)
8163 : : {
8164 : : /*
8165 : : * The change in number of backend slots doesn't need to be WAL-logged
8166 : : * if archiving is not enabled, as you can't start archive recovery
8167 : : * with wal_level=minimal anyway. We don't really care about the
8168 : : * values in pg_control either if wal_level=minimal, but seems better
8169 : : * to keep them up-to-date to avoid confusion.
8170 : : */
5661 heikki.linnakangas@i 8171 [ + + + + ]: 427 : if (wal_level != ControlFile->wal_level || XLogIsNeeded())
8172 : : {
8173 : : xl_parameter_change xlrec;
8174 : : XLogRecPtr recptr;
8175 : :
8176 : 407 : xlrec.MaxConnections = MaxConnections;
4498 rhaas@postgresql.org 8177 : 407 : xlrec.max_worker_processes = max_worker_processes;
2449 michael@paquier.xyz 8178 : 407 : xlrec.max_wal_senders = max_wal_senders;
5661 heikki.linnakangas@i 8179 : 407 : xlrec.max_prepared_xacts = max_prepared_xacts;
8180 : 407 : xlrec.max_locks_per_xact = max_locks_per_xact;
8181 : 407 : xlrec.wal_level = wal_level;
4317 rhaas@postgresql.org 8182 : 407 : xlrec.wal_log_hints = wal_log_hints;
3981 alvherre@alvh.no-ip. 8183 : 407 : xlrec.track_commit_timestamp = track_commit_timestamp;
8184 : :
3994 heikki.linnakangas@i 8185 : 407 : XLogBeginInsert();
258 peter@eisentraut.org 8186 : 407 : XLogRegisterData(&xlrec, sizeof(xlrec));
8187 : :
3994 heikki.linnakangas@i 8188 : 407 : recptr = XLogInsert(RM_XLOG_ID, XLOG_PARAMETER_CHANGE);
4233 fujii@postgresql.org 8189 : 407 : XLogFlush(recptr);
8190 : : }
8191 : :
1967 tmunro@postgresql.or 8192 : 427 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8193 : :
5661 heikki.linnakangas@i 8194 : 427 : ControlFile->MaxConnections = MaxConnections;
4498 rhaas@postgresql.org 8195 : 427 : ControlFile->max_worker_processes = max_worker_processes;
2449 michael@paquier.xyz 8196 : 427 : ControlFile->max_wal_senders = max_wal_senders;
5661 heikki.linnakangas@i 8197 : 427 : ControlFile->max_prepared_xacts = max_prepared_xacts;
8198 : 427 : ControlFile->max_locks_per_xact = max_locks_per_xact;
8199 : 427 : ControlFile->wal_level = wal_level;
4317 rhaas@postgresql.org 8200 : 427 : ControlFile->wal_log_hints = wal_log_hints;
3981 alvherre@alvh.no-ip. 8201 : 427 : ControlFile->track_commit_timestamp = track_commit_timestamp;
5661 heikki.linnakangas@i 8202 : 427 : UpdateControlFile();
8203 : :
1967 tmunro@postgresql.or 8204 : 427 : LWLockRelease(ControlFileLock);
8205 : : }
5759 heikki.linnakangas@i 8206 : 852 : }
8207 : :
8208 : : /*
8209 : : * Update full_page_writes in shared memory, and write an
8210 : : * XLOG_FPW_CHANGE record if necessary.
8211 : : *
8212 : : * Note: this function assumes there is no other process running
8213 : : * concurrently that could update it.
8214 : : */
8215 : : void
5024 simon@2ndQuadrant.co 8216 : 1418 : UpdateFullPageWrites(void)
8217 : : {
8218 : 1418 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8219 : : bool recoveryInProgress;
8220 : :
8221 : : /*
8222 : : * Do nothing if full_page_writes has not been changed.
8223 : : *
8224 : : * It's safe to check the shared full_page_writes without the lock,
8225 : : * because we assume that there is no concurrently running process which
8226 : : * can update it.
8227 : : */
8228 [ + + ]: 1418 : if (fullPageWrites == Insert->fullPageWrites)
8229 : 973 : return;
8230 : :
8231 : : /*
8232 : : * Perform this outside critical section so that the WAL insert
8233 : : * initialization done by RecoveryInProgress() doesn't trigger an
8234 : : * assertion failure.
8235 : : */
2586 akapila@postgresql.o 8236 : 445 : recoveryInProgress = RecoveryInProgress();
8237 : :
4983 heikki.linnakangas@i 8238 : 445 : START_CRIT_SECTION();
8239 : :
8240 : : /*
8241 : : * It's always safe to take full page images, even when not strictly
8242 : : * required, but not the other round. So if we're setting full_page_writes
8243 : : * to true, first set it true and then write the WAL record. If we're
8244 : : * setting it to false, first write the WAL record and then set the global
8245 : : * flag.
8246 : : */
8247 [ + + ]: 445 : if (fullPageWrites)
8248 : : {
4238 8249 : 435 : WALInsertLockAcquireExclusive();
4983 8250 : 435 : Insert->fullPageWrites = true;
4238 8251 : 435 : WALInsertLockRelease();
8252 : : }
8253 : :
8254 : : /*
8255 : : * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
8256 : : * full_page_writes during archive recovery, if required.
8257 : : */
2586 akapila@postgresql.o 8258 [ + + - + ]: 445 : if (XLogStandbyInfoActive() && !recoveryInProgress)
8259 : : {
3994 heikki.linnakangas@i 8260 :UBC 0 : XLogBeginInsert();
258 peter@eisentraut.org 8261 : 0 : XLogRegisterData(&fullPageWrites, sizeof(bool));
8262 : :
3994 heikki.linnakangas@i 8263 : 0 : XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
8264 : : }
8265 : :
4983 heikki.linnakangas@i 8266 [ + + ]:CBC 445 : if (!fullPageWrites)
8267 : : {
4238 8268 : 10 : WALInsertLockAcquireExclusive();
4983 8269 : 10 : Insert->fullPageWrites = false;
4238 8270 : 10 : WALInsertLockRelease();
8271 : : }
4983 8272 [ - + ]: 445 : END_CRIT_SECTION();
8273 : : }
8274 : :
8275 : : /*
8276 : : * XLOG resource manager's routines
8277 : : *
8278 : : * Definitions of info values are in include/catalog/pg_control.h, though
8279 : : * not all record types are related to control file updates.
8280 : : *
8281 : : * NOTE: Some XLOG record types that are directly related to WAL recovery
8282 : : * are handled in xlogrecovery_redo().
8283 : : */
8284 : : void
3994 8285 : 40571 : xlog_redo(XLogReaderState *record)
8286 : : {
8287 : 40571 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
8288 : 40571 : XLogRecPtr lsn = record->EndRecPtr;
8289 : :
8290 : : /*
8291 : : * In XLOG rmgr, backup blocks are only used by XLOG_FPI and
8292 : : * XLOG_FPI_FOR_HINT records.
8293 : : */
3990 8294 [ + + + + : 40571 : Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
- + ]
8295 : : !XLogRecHasAnyBlockRefs(record));
8296 : :
8989 tgl@sss.pgh.pa.us 8297 [ + + ]: 40571 : if (info == XLOG_NEXTOID)
8298 : : {
8299 : : Oid nextOid;
8300 : :
8301 : : /*
8302 : : * We used to try to take the maximum of TransamVariables->nextOid and
8303 : : * the recorded nextOid, but that fails if the OID counter wraps
8304 : : * around. Since no OID allocation should be happening during replay
8305 : : * anyway, better to just believe the record exactly. We still take
8306 : : * OidGenLock while setting the variable, just in case.
8307 : : */
9124 vadim4o@yahoo.com 8308 : 90 : memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
5012 tgl@sss.pgh.pa.us 8309 : 90 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
689 heikki.linnakangas@i 8310 : 90 : TransamVariables->nextOid = nextOid;
8311 : 90 : TransamVariables->oidCount = 0;
5012 tgl@sss.pgh.pa.us 8312 : 90 : LWLockRelease(OidGenLock);
8313 : : }
8994 8314 [ + + ]: 40481 : else if (info == XLOG_CHECKPOINT_SHUTDOWN)
8315 : : {
8316 : : CheckPoint checkPoint;
8317 : : TimeLineID replayTLI;
8318 : :
8319 : 31 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8320 : : /* In a SHUTDOWN checkpoint, believe the counters exactly */
5012 8321 : 31 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
689 heikki.linnakangas@i 8322 : 31 : TransamVariables->nextXid = checkPoint.nextXid;
5012 tgl@sss.pgh.pa.us 8323 : 31 : LWLockRelease(XidGenLock);
8324 : 31 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
689 heikki.linnakangas@i 8325 : 31 : TransamVariables->nextOid = checkPoint.nextOid;
8326 : 31 : TransamVariables->oidCount = 0;
5012 tgl@sss.pgh.pa.us 8327 : 31 : LWLockRelease(OidGenLock);
7446 8328 : 31 : MultiXactSetNextMXact(checkPoint.nextMulti,
8329 : : checkPoint.nextMultiOffset);
8330 : :
3684 andres@anarazel.de 8331 : 31 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8332 : : checkPoint.oldestMultiDB);
8333 : :
8334 : : /*
8335 : : * No need to set oldestClogXid here as well; it'll be set when we
8336 : : * redo an xl_clog_truncate if it changed since initialization.
8337 : : */
5731 tgl@sss.pgh.pa.us 8338 : 31 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
8339 : :
8340 : : /*
8341 : : * If we see a shutdown checkpoint while waiting for an end-of-backup
8342 : : * record, the backup was canceled and the end-of-backup record will
8343 : : * never arrive.
8344 : : */
4630 heikki.linnakangas@i 8345 [ + - ]: 31 : if (ArchiveRecoveryRequested &&
5024 simon@2ndQuadrant.co 8346 [ - + ]: 31 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint) &&
5024 simon@2ndQuadrant.co 8347 [ # # ]:UBC 0 : XLogRecPtrIsInvalid(ControlFile->backupEndPoint))
5012 tgl@sss.pgh.pa.us 8348 [ # # ]: 0 : ereport(PANIC,
8349 : : (errmsg("online backup was canceled, recovery cannot continue")));
8350 : :
8351 : : /*
8352 : : * If we see a shutdown checkpoint, we know that nothing was running
8353 : : * on the primary at this point. So fake-up an empty running-xacts
8354 : : * record and use that here and now. Recover additional standby state
8355 : : * for prepared transactions.
8356 : : */
5791 simon@2ndQuadrant.co 8357 [ + + ]:CBC 31 : if (standbyState >= STANDBY_INITIALIZED)
8358 : : {
8359 : : TransactionId *xids;
8360 : : int nxids;
8361 : : TransactionId oldestActiveXID;
8362 : : TransactionId latestCompletedXid;
8363 : : RunningTransactionsData running;
8364 : :
5676 heikki.linnakangas@i 8365 : 29 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
8366 : :
8367 : : /* Update pg_subtrans entries for any prepared transactions */
487 8368 : 29 : StandbyRecoverPreparedTransactions();
8369 : :
8370 : : /*
8371 : : * Construct a RunningTransactions snapshot representing a shut
8372 : : * down server, with only prepared transactions still alive. We're
8373 : : * never overflowed at this point because all subxids are listed
8374 : : * with their parent prepared transactions.
8375 : : */
5676 8376 : 29 : running.xcnt = nxids;
4712 simon@2ndQuadrant.co 8377 : 29 : running.subxcnt = 0;
487 heikki.linnakangas@i 8378 : 29 : running.subxid_status = SUBXIDS_IN_SUBTRANS;
1903 andres@anarazel.de 8379 : 29 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5676 heikki.linnakangas@i 8380 : 29 : running.oldestRunningXid = oldestActiveXID;
1903 andres@anarazel.de 8381 : 29 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5646 simon@2ndQuadrant.co 8382 [ - + ]: 29 : TransactionIdRetreat(latestCompletedXid);
5645 8383 [ - + ]: 29 : Assert(TransactionIdIsNormal(latestCompletedXid));
5646 8384 : 29 : running.latestCompletedXid = latestCompletedXid;
5676 heikki.linnakangas@i 8385 : 29 : running.xids = xids;
8386 : :
8387 : 29 : ProcArrayApplyRecoveryInfo(&running);
8388 : : }
8389 : :
8390 : : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
1967 tmunro@postgresql.or 8391 : 31 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
1903 andres@anarazel.de 8392 : 31 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
1967 tmunro@postgresql.or 8393 : 31 : LWLockRelease(ControlFileLock);
8394 : :
8395 : : /*
8396 : : * We should've already switched to the new TLI before replaying this
8397 : : * record.
8398 : : */
1349 heikki.linnakangas@i 8399 : 31 : (void) GetCurrentReplayRecPtr(&replayTLI);
1452 rhaas@postgresql.org 8400 [ - + ]: 31 : if (checkPoint.ThisTimeLineID != replayTLI)
4707 heikki.linnakangas@i 8401 [ # # ]:UBC 0 : ereport(PANIC,
8402 : : (errmsg("unexpected timeline ID %u (should be %u) in shutdown checkpoint record",
8403 : : checkPoint.ThisTimeLineID, replayTLI)));
8404 : :
1433 rhaas@postgresql.org 8405 :CBC 31 : RecoveryRestartPoint(&checkPoint, record);
8406 : :
8407 : : /*
8408 : : * After replaying a checkpoint record, free all smgr objects.
8409 : : * Otherwise we would never do so for dropped relations, as the
8410 : : * startup does not process shared invalidation messages or call
8411 : : * AtEOXact_SMgr().
8412 : : */
47 michael@paquier.xyz 8413 : 31 : smgrdestroyall();
8414 : : }
8994 tgl@sss.pgh.pa.us 8415 [ + + ]: 40450 : else if (info == XLOG_CHECKPOINT_ONLINE)
8416 : : {
8417 : : CheckPoint checkPoint;
8418 : : TimeLineID replayTLI;
8419 : :
8420 : 667 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8421 : : /* In an ONLINE checkpoint, treat the XID counter as a minimum */
5012 8422 : 667 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
689 heikki.linnakangas@i 8423 [ - + ]: 667 : if (FullTransactionIdPrecedes(TransamVariables->nextXid,
8424 : : checkPoint.nextXid))
689 heikki.linnakangas@i 8425 :UBC 0 : TransamVariables->nextXid = checkPoint.nextXid;
5012 tgl@sss.pgh.pa.us 8426 :CBC 667 : LWLockRelease(XidGenLock);
8427 : :
8428 : : /*
8429 : : * We ignore the nextOid counter in an ONLINE checkpoint, preferring
8430 : : * to track OID assignment through XLOG_NEXTOID records. The nextOid
8431 : : * counter is from the start of the checkpoint and might well be stale
8432 : : * compared to later XLOG_NEXTOID records. We could try to take the
8433 : : * maximum of the nextOid counter and our latest value, but since
8434 : : * there's no particular guarantee about the speed with which the OID
8435 : : * counter wraps around, that's a risky thing to do. In any case,
8436 : : * users of the nextOid counter are required to avoid assignment of
8437 : : * duplicates, so that a somewhat out-of-date value should be safe.
8438 : : */
8439 : :
8440 : : /* Handle multixact */
7446 8441 : 667 : MultiXactAdvanceNextMXact(checkPoint.nextMulti,
8442 : : checkPoint.nextMultiOffset);
8443 : :
8444 : : /*
8445 : : * NB: This may perform multixact truncation when replaying WAL
8446 : : * generated by an older primary.
8447 : : */
3684 andres@anarazel.de 8448 : 667 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8449 : : checkPoint.oldestMultiDB);
689 heikki.linnakangas@i 8450 [ - + ]: 667 : if (TransactionIdPrecedes(TransamVariables->oldestXid,
8451 : : checkPoint.oldestXid))
5731 tgl@sss.pgh.pa.us 8452 :UBC 0 : SetTransactionIdLimit(checkPoint.oldestXid,
8453 : : checkPoint.oldestXidDB);
8454 : : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
1967 tmunro@postgresql.or 8455 :CBC 667 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
1903 andres@anarazel.de 8456 : 667 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
1967 tmunro@postgresql.or 8457 : 667 : LWLockRelease(ControlFileLock);
8458 : :
8459 : : /* TLI should not change in an on-line checkpoint */
1349 heikki.linnakangas@i 8460 : 667 : (void) GetCurrentReplayRecPtr(&replayTLI);
1452 rhaas@postgresql.org 8461 [ - + ]: 667 : if (checkPoint.ThisTimeLineID != replayTLI)
7929 tgl@sss.pgh.pa.us 8462 [ # # ]:UBC 0 : ereport(PANIC,
8463 : : (errmsg("unexpected timeline ID %u (should be %u) in online checkpoint record",
8464 : : checkPoint.ThisTimeLineID, replayTLI)));
8465 : :
1433 rhaas@postgresql.org 8466 :CBC 667 : RecoveryRestartPoint(&checkPoint, record);
8467 : :
8468 : : /*
8469 : : * After replaying a checkpoint record, free all smgr objects.
8470 : : * Otherwise we would never do so for dropped relations, as the
8471 : : * startup does not process shared invalidation messages or call
8472 : : * AtEOXact_SMgr().
8473 : : */
47 michael@paquier.xyz 8474 : 667 : smgrdestroyall();
8475 : : }
1489 alvherre@alvh.no-ip. 8476 [ + + ]: 39783 : else if (info == XLOG_OVERWRITE_CONTRECORD)
8477 : : {
8478 : : /* nothing to do here, handled in xlogrecovery_redo() */
8479 : : }
4654 simon@2ndQuadrant.co 8480 [ + + ]: 39782 : else if (info == XLOG_END_OF_RECOVERY)
8481 : : {
8482 : : xl_end_of_recovery xlrec;
8483 : : TimeLineID replayTLI;
8484 : :
8485 : 9 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
8486 : :
8487 : : /*
8488 : : * For Hot Standby, we could treat this like a Shutdown Checkpoint,
8489 : : * but this case is rarer and harder to test, so the benefit doesn't
8490 : : * outweigh the potential extra cost of maintenance.
8491 : : */
8492 : :
8493 : : /*
8494 : : * We should've already switched to the new TLI before replaying this
8495 : : * record.
8496 : : */
1349 heikki.linnakangas@i 8497 : 9 : (void) GetCurrentReplayRecPtr(&replayTLI);
1452 rhaas@postgresql.org 8498 [ - + ]: 9 : if (xlrec.ThisTimeLineID != replayTLI)
4654 simon@2ndQuadrant.co 8499 [ # # ]:UBC 0 : ereport(PANIC,
8500 : : (errmsg("unexpected timeline ID %u (should be %u) in end-of-recovery record",
8501 : : xlrec.ThisTimeLineID, replayTLI)));
8502 : : }
6735 tgl@sss.pgh.pa.us 8503 [ + - ]:CBC 39773 : else if (info == XLOG_NOOP)
8504 : : {
8505 : : /* nothing to do here */
8506 : : }
7022 8507 [ + + ]: 39773 : else if (info == XLOG_SWITCH)
8508 : : {
8509 : : /* nothing to do here */
8510 : : }
5375 simon@2ndQuadrant.co 8511 [ + + ]: 39337 : else if (info == XLOG_RESTORE_POINT)
8512 : : {
8513 : : /* nothing to do here, handled in xlogrecovery.c */
8514 : : }
3990 heikki.linnakangas@i 8515 [ + + + + ]: 39332 : else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
8516 : : {
8517 : : /*
8518 : : * XLOG_FPI records contain nothing else but one or more block
8519 : : * references. Every block reference must include a full-page image
8520 : : * even if full_page_writes was disabled when the record was generated
8521 : : * - otherwise there would be no point in this record.
8522 : : *
8523 : : * XLOG_FPI_FOR_HINT records are generated when a page needs to be
8524 : : * WAL-logged because of a hint bit update. They are only generated
8525 : : * when checksums and/or wal_log_hints are enabled. They may include
8526 : : * no full-page images if full_page_writes was disabled when they were
8527 : : * generated. In this case there is nothing to do here.
8528 : : *
8529 : : * No recovery conflicts are generated by these generic records - if a
8530 : : * resource manager needs to generate conflicts, it has to define a
8531 : : * separate WAL record type and redo routine.
8532 : : */
1319 tmunro@postgresql.or 8533 [ + + ]: 81878 : for (uint8 block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
8534 : : {
8535 : : Buffer buffer;
8536 : :
1559 fujii@postgresql.org 8537 [ + + ]: 43329 : if (!XLogRecHasBlockImage(record, block_id))
8538 : : {
8539 [ - + ]: 66 : if (info == XLOG_FPI)
1559 fujii@postgresql.org 8540 [ # # ]:UBC 0 : elog(ERROR, "XLOG_FPI record did not contain a full-page image");
1559 fujii@postgresql.org 8541 :CBC 66 : continue;
8542 : : }
8543 : :
2399 heikki.linnakangas@i 8544 [ - + ]: 43263 : if (XLogReadBufferForRedo(record, block_id, &buffer) != BLK_RESTORED)
2399 heikki.linnakangas@i 8545 [ # # ]:UBC 0 : elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
2399 heikki.linnakangas@i 8546 :CBC 43263 : UnlockReleaseBuffer(buffer);
8547 : : }
8548 : : }
5775 8549 [ + + ]: 783 : else if (info == XLOG_BACKUP_END)
8550 : : {
8551 : : /* nothing to do here, handled in xlogrecovery_redo() */
8552 : : }
5661 8553 [ + + ]: 701 : else if (info == XLOG_PARAMETER_CHANGE)
8554 : : {
8555 : : xl_parameter_change xlrec;
8556 : :
8557 : : /* Update our copy of the parameters in pg_control */
8558 : 33 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
8559 : :
8560 : : /*
8561 : : * Invalidate logical slots if we are in hot standby and the primary
8562 : : * does not have a WAL level sufficient for logical decoding. No need
8563 : : * to search for potentially conflicting logically slots if standby is
8564 : : * running with wal_level lower than logical, because in that case, we
8565 : : * would have either disallowed creation of logical slots or
8566 : : * invalidated existing ones.
8567 : : */
934 andres@anarazel.de 8568 [ + - + + ]: 33 : if (InRecovery && InHotStandby &&
8569 [ + + ]: 18 : xlrec.wal_level < WAL_LEVEL_LOGICAL &&
8570 [ + + ]: 8 : wal_level >= WAL_LEVEL_LOGICAL)
8571 : 4 : InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_LEVEL,
8572 : : 0, InvalidOid,
8573 : : InvalidTransactionId);
8574 : :
5656 heikki.linnakangas@i 8575 : 33 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
5661 8576 : 33 : ControlFile->MaxConnections = xlrec.MaxConnections;
4498 rhaas@postgresql.org 8577 : 33 : ControlFile->max_worker_processes = xlrec.max_worker_processes;
2449 michael@paquier.xyz 8578 : 33 : ControlFile->max_wal_senders = xlrec.max_wal_senders;
5661 heikki.linnakangas@i 8579 : 33 : ControlFile->max_prepared_xacts = xlrec.max_prepared_xacts;
8580 : 33 : ControlFile->max_locks_per_xact = xlrec.max_locks_per_xact;
8581 : 33 : ControlFile->wal_level = xlrec.wal_level;
3938 8582 : 33 : ControlFile->wal_log_hints = xlrec.wal_log_hints;
8583 : :
8584 : : /*
8585 : : * Update minRecoveryPoint to ensure that if recovery is aborted, we
8586 : : * recover back up to this point before allowing hot standby again.
8587 : : * This is important if the max_* settings are decreased, to ensure
8588 : : * you don't run queries against the WAL preceding the change. The
8589 : : * local copies cannot be updated as long as crash recovery is
8590 : : * happening and we expect all the WAL to be replayed.
8591 : : */
2671 michael@paquier.xyz 8592 [ + + ]: 33 : if (InArchiveRecovery)
8593 : : {
1349 heikki.linnakangas@i 8594 : 19 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
8595 : 19 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
8596 : : }
8597 [ + + + + ]: 33 : if (LocalMinRecoveryPoint != InvalidXLogRecPtr && LocalMinRecoveryPoint < lsn)
8598 : : {
8599 : : TimeLineID replayTLI;
8600 : :
8601 : 7 : (void) GetCurrentReplayRecPtr(&replayTLI);
5656 8602 : 7 : ControlFile->minRecoveryPoint = lsn;
1452 rhaas@postgresql.org 8603 : 7 : ControlFile->minRecoveryPointTLI = replayTLI;
8604 : : }
8605 : :
3679 alvherre@alvh.no-ip. 8606 : 33 : CommitTsParameterChange(xlrec.track_commit_timestamp,
8607 : 33 : ControlFile->track_commit_timestamp);
8608 : 33 : ControlFile->track_commit_timestamp = xlrec.track_commit_timestamp;
8609 : :
5661 heikki.linnakangas@i 8610 : 33 : UpdateControlFile();
5656 8611 : 33 : LWLockRelease(ControlFileLock);
8612 : :
8613 : : /* Check to see if any parameter change gives a problem on recovery */
5661 8614 : 33 : CheckRequiredParameterValues();
8615 : : }
5024 simon@2ndQuadrant.co 8616 [ - + ]: 668 : else if (info == XLOG_FPW_CHANGE)
8617 : : {
8618 : : bool fpw;
8619 : :
5024 simon@2ndQuadrant.co 8620 :UBC 0 : memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
8621 : :
8622 : : /*
8623 : : * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
8624 : : * do_pg_backup_start() and do_pg_backup_stop() can check whether
8625 : : * full_page_writes has been disabled during online backup.
8626 : : */
8627 [ # # ]: 0 : if (!fpw)
8628 : : {
4053 andres@anarazel.de 8629 [ # # ]: 0 : SpinLockAcquire(&XLogCtl->info_lck);
1433 rhaas@postgresql.org 8630 [ # # ]: 0 : if (XLogCtl->lastFpwDisableRecPtr < record->ReadRecPtr)
8631 : 0 : XLogCtl->lastFpwDisableRecPtr = record->ReadRecPtr;
4053 andres@anarazel.de 8632 : 0 : SpinLockRelease(&XLogCtl->info_lck);
8633 : : }
8634 : :
8635 : : /* Keep track of full_page_writes */
5024 simon@2ndQuadrant.co 8636 : 0 : lastFullPageWrites = fpw;
8637 : : }
8638 : : else if (info == XLOG_CHECKPOINT_REDO)
8639 : : {
8640 : : /* nothing to do here, just for informational purposes */
8641 : : }
9137 vadim4o@yahoo.com 8642 :CBC 40569 : }
8643 : :
8644 : : /*
8645 : : * Return the extra open flags used for opening a file, depending on the
8646 : : * value of the GUCs wal_sync_method, fsync and debug_io_direct.
8647 : : */
8648 : : static int
6375 magnus@hagander.net 8649 : 15394 : get_sync_bit(int method)
8650 : : {
5722 bruce@momjian.us 8651 : 15394 : int o_direct_flag = 0;
8652 : :
8653 : : /*
8654 : : * Use O_DIRECT if requested, except in walreceiver process. The WAL
8655 : : * written by walreceiver is normally read by the startup process soon
8656 : : * after it's written. Also, walreceiver performs unaligned writes, which
8657 : : * don't work with O_DIRECT, so it is required for correctness too.
8658 : : */
933 tmunro@postgresql.or 8659 [ + + + - ]: 15394 : if ((io_direct_flags & IO_DIRECT_WAL) && !AmWalReceiverProcess())
5729 heikki.linnakangas@i 8660 : 8 : o_direct_flag = PG_O_DIRECT;
8661 : :
8662 : : /* If fsync is disabled, never open in sync mode */
933 tmunro@postgresql.or 8663 [ + - ]: 15394 : if (!enableFsync)
8664 : 15394 : return o_direct_flag;
8665 : :
6375 magnus@hagander.net 8666 [ # # # # ]:UBC 0 : switch (method)
8667 : : {
8668 : : /*
8669 : : * enum values for all sync options are defined even if they are
8670 : : * not supported on the current platform. But if not, they are
8671 : : * not included in the enum option array, and therefore will never
8672 : : * be seen here.
8673 : : */
745 nathan@postgresql.or 8674 : 0 : case WAL_SYNC_METHOD_FSYNC:
8675 : : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8676 : : case WAL_SYNC_METHOD_FDATASYNC:
933 tmunro@postgresql.or 8677 : 0 : return o_direct_flag;
8678 : : #ifdef O_SYNC
745 nathan@postgresql.or 8679 : 0 : case WAL_SYNC_METHOD_OPEN:
1193 tmunro@postgresql.or 8680 : 0 : return O_SYNC | o_direct_flag;
8681 : : #endif
8682 : : #ifdef O_DSYNC
745 nathan@postgresql.or 8683 : 0 : case WAL_SYNC_METHOD_OPEN_DSYNC:
1193 tmunro@postgresql.or 8684 : 0 : return O_DSYNC | o_direct_flag;
8685 : : #endif
6377 magnus@hagander.net 8686 : 0 : default:
8687 : : /* can't happen (unless we are out of sync with option array) */
528 peter@eisentraut.org 8688 [ # # ]: 0 : elog(ERROR, "unrecognized \"wal_sync_method\": %d", method);
8689 : : return 0; /* silence warning */
8690 : : }
8691 : : }
8692 : :
8693 : : /*
8694 : : * GUC support
8695 : : */
8696 : : void
745 nathan@postgresql.or 8697 :CBC 1087 : assign_wal_sync_method(int new_wal_sync_method, void *extra)
8698 : : {
8699 [ - + ]: 1087 : if (wal_sync_method != new_wal_sync_method)
8700 : : {
8701 : : /*
8702 : : * To ensure that no blocks escape unsynced, force an fsync on the
8703 : : * currently open log segment (if any). Also, if the open flag is
8704 : : * changing, close the log file so it will be reopened (with new flag
8705 : : * bit) at next use.
8706 : : */
8991 tgl@sss.pgh.pa.us 8707 [ # # ]:UBC 0 : if (openLogFile >= 0)
8708 : : {
3145 rhaas@postgresql.org 8709 : 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN);
8991 tgl@sss.pgh.pa.us 8710 [ # # ]: 0 : if (pg_fsync(openLogFile) != 0)
8711 : : {
8712 : : char xlogfname[MAXFNAMELEN];
8713 : : int save_errno;
8714 : :
2155 michael@paquier.xyz 8715 : 0 : save_errno = errno;
1452 rhaas@postgresql.org 8716 : 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo,
8717 : : wal_segment_size);
2155 michael@paquier.xyz 8718 : 0 : errno = save_errno;
8134 tgl@sss.pgh.pa.us 8719 [ # # ]: 0 : ereport(PANIC,
8720 : : (errcode_for_file_access(),
8721 : : errmsg("could not fsync file \"%s\": %m", xlogfname)));
8722 : : }
8723 : :
3145 rhaas@postgresql.org 8724 : 0 : pgstat_report_wait_end();
745 nathan@postgresql.or 8725 [ # # ]: 0 : if (get_sync_bit(wal_sync_method) != get_sync_bit(new_wal_sync_method))
7074 bruce@momjian.us 8726 : 0 : XLogFileClose();
8727 : : }
8728 : : }
8991 tgl@sss.pgh.pa.us 8729 :CBC 1087 : }
8730 : :
8731 : :
8732 : : /*
8733 : : * Issue appropriate kind of fsync (if any) for an XLOG output file.
8734 : : *
8735 : : * 'fd' is a file descriptor for the XLOG file to be fsync'd.
8736 : : * 'segno' is for error reporting purposes.
8737 : : */
8738 : : void
1452 rhaas@postgresql.org 8739 : 168379 : issue_xlog_fsync(int fd, XLogSegNo segno, TimeLineID tli)
8740 : : {
2155 michael@paquier.xyz 8741 : 168379 : char *msg = NULL;
8742 : : instr_time start;
8743 : :
1452 rhaas@postgresql.org 8744 [ - + ]: 168379 : Assert(tli != 0);
8745 : :
8746 : : /*
8747 : : * Quick exit if fsync is disabled or write() has already synced the WAL
8748 : : * file.
8749 : : */
1693 fujii@postgresql.org 8750 [ + + ]: 168379 : if (!enableFsync ||
745 nathan@postgresql.or 8751 [ + - ]:GBC 1 : wal_sync_method == WAL_SYNC_METHOD_OPEN ||
8752 [ - + ]: 1 : wal_sync_method == WAL_SYNC_METHOD_OPEN_DSYNC)
1693 fujii@postgresql.org 8753 :CBC 168378 : return;
8754 : :
8755 : : /*
8756 : : * Measure I/O timing to sync the WAL file for pg_stat_io.
8757 : : */
243 michael@paquier.xyz 8758 :GBC 1 : start = pgstat_prepare_io_time(track_wal_io_timing);
8759 : :
2674 8760 : 1 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC);
745 nathan@postgresql.or 8761 [ - + - - ]: 1 : switch (wal_sync_method)
8762 : : {
745 nathan@postgresql.or 8763 :UBC 0 : case WAL_SYNC_METHOD_FSYNC:
5764 heikki.linnakangas@i 8764 [ # # ]: 0 : if (pg_fsync_no_writethrough(fd) != 0)
2155 michael@paquier.xyz 8765 : 0 : msg = _("could not fsync file \"%s\": %m");
8991 tgl@sss.pgh.pa.us 8766 : 0 : break;
8767 : : #ifdef HAVE_FSYNC_WRITETHROUGH
8768 : : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8769 : : if (pg_fsync_writethrough(fd) != 0)
8770 : : msg = _("could not fsync write-through file \"%s\": %m");
8771 : : break;
8772 : : #endif
745 nathan@postgresql.or 8773 :GBC 1 : case WAL_SYNC_METHOD_FDATASYNC:
5764 heikki.linnakangas@i 8774 [ - + ]: 1 : if (pg_fdatasync(fd) != 0)
2155 michael@paquier.xyz 8775 :UBC 0 : msg = _("could not fdatasync file \"%s\": %m");
8991 tgl@sss.pgh.pa.us 8776 :GBC 1 : break;
745 nathan@postgresql.or 8777 :UBC 0 : case WAL_SYNC_METHOD_OPEN:
8778 : : case WAL_SYNC_METHOD_OPEN_DSYNC:
8779 : : /* not reachable */
1693 fujii@postgresql.org 8780 : 0 : Assert(false);
8781 : : break;
8991 tgl@sss.pgh.pa.us 8782 : 0 : default:
572 dgustafsson@postgres 8783 [ # # ]: 0 : ereport(PANIC,
8784 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8785 : : errmsg_internal("unrecognized \"wal_sync_method\": %d", wal_sync_method));
8786 : : break;
8787 : : }
8788 : :
8789 : : /* PANIC if failed to fsync */
2155 michael@paquier.xyz 8790 [ - + ]:GBC 1 : if (msg)
8791 : : {
8792 : : char xlogfname[MAXFNAMELEN];
2155 michael@paquier.xyz 8793 :UBC 0 : int save_errno = errno;
8794 : :
1452 rhaas@postgresql.org 8795 : 0 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
2155 michael@paquier.xyz 8796 : 0 : errno = save_errno;
8797 [ # # ]: 0 : ereport(PANIC,
8798 : : (errcode_for_file_access(),
8799 : : errmsg(msg, xlogfname)));
8800 : : }
8801 : :
2155 michael@paquier.xyz 8802 :GBC 1 : pgstat_report_wait_end();
8803 : :
265 8804 : 1 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_FSYNC,
8805 : : start, 1, 0);
8806 : : }
8807 : :
8808 : : /*
8809 : : * do_pg_backup_start is the workhorse of the user-visible pg_backup_start()
8810 : : * function. It creates the necessary starting checkpoint and constructs the
8811 : : * backup state and tablespace map.
8812 : : *
8813 : : * Input parameters are "state" (the backup state), "fast" (if true, we do
8814 : : * the checkpoint in fast mode), and "tablespaces" (if non-NULL, indicates a
8815 : : * list of tablespaceinfo structs describing the cluster's tablespaces.).
8816 : : *
8817 : : * The tablespace map contents are appended to passed-in parameter
8818 : : * tablespace_map and the caller is responsible for including it in the backup
8819 : : * archive as 'tablespace_map'. The tablespace_map file is required mainly for
8820 : : * tar format in windows as native windows utilities are not able to create
8821 : : * symlinks while extracting files from tar. However for consistency and
8822 : : * platform-independence, we do it the same way everywhere.
8823 : : *
8824 : : * It fills in "state" with the information required for the backup, such
8825 : : * as the minimum WAL location that must be present to restore from this
8826 : : * backup (starttli) and the corresponding timeline ID (starttli).
8827 : : *
8828 : : * Every successfully started backup must be stopped by calling
8829 : : * do_pg_backup_stop() or do_pg_abort_backup(). There can be many
8830 : : * backups active at the same time.
8831 : : *
8832 : : * It is the responsibility of the caller of this function to verify the
8833 : : * permissions of the calling user!
8834 : : */
8835 : : void
1127 michael@paquier.xyz 8836 :CBC 167 : do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces,
8837 : : BackupState *state, StringInfo tblspcmapfile)
8838 : : {
8839 : : bool backup_started_in_recovery;
8840 : :
8841 [ - + ]: 167 : Assert(state != NULL);
5024 simon@2ndQuadrant.co 8842 : 167 : backup_started_in_recovery = RecoveryInProgress();
8843 : :
8844 : : /*
8845 : : * During recovery, we don't need to check WAL level. Because, if WAL
8846 : : * level is not sufficient, it's impossible to get here during recovery.
8847 : : */
8848 [ + + - + ]: 167 : if (!backup_started_in_recovery && !XLogIsNeeded())
6606 tgl@sss.pgh.pa.us 8849 [ # # ]:UBC 0 : ereport(ERROR,
8850 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8851 : : errmsg("WAL level not sufficient for making an online backup"),
8852 : : errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
8853 : :
5383 heikki.linnakangas@i 8854 [ + + ]:CBC 167 : if (strlen(backupidstr) > MAXPGPATH)
8855 [ + - ]: 1 : ereport(ERROR,
8856 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8857 : : errmsg("backup label too long (max %d bytes)",
8858 : : MAXPGPATH)));
8859 : :
482 dgustafsson@postgres 8860 : 166 : strlcpy(state->name, backupidstr, sizeof(state->name));
8861 : :
8862 : : /*
8863 : : * Mark backup active in shared memory. We must do full-page WAL writes
8864 : : * during an on-line backup even if not doing so at other times, because
8865 : : * it's quite possible for the backup dump to obtain a "torn" (partially
8866 : : * written) copy of a database page if it reads the page concurrently with
8867 : : * our write to the same page. This can be fixed as long as the first
8868 : : * write to the page in the WAL sequence is a full-page write. Hence, we
8869 : : * increment runningBackups then force a CHECKPOINT, to ensure there are
8870 : : * no dirty pages in shared memory that might get dumped while the backup
8871 : : * is in progress without having a corresponding WAL record. (Once the
8872 : : * backup is complete, we need not force full-page writes anymore, since
8873 : : * we expect that any pages not modified during the backup interval must
8874 : : * have been correctly captured by the backup.)
8875 : : *
8876 : : * Note that forcing full-page writes has no effect during an online
8877 : : * backup from the standby.
8878 : : *
8879 : : * We must hold all the insertion locks to change the value of
8880 : : * runningBackups, to ensure adequate interlocking against
8881 : : * XLogInsertRecord().
8882 : : */
4238 heikki.linnakangas@i 8883 : 166 : WALInsertLockAcquireExclusive();
1300 sfrost@snowman.net 8884 : 166 : XLogCtl->Insert.runningBackups++;
4238 heikki.linnakangas@i 8885 : 166 : WALInsertLockRelease();
8886 : :
8887 : : /*
8888 : : * Ensure we decrement runningBackups if we fail below. NB -- for this to
8889 : : * work correctly, it is critical that sessionBackupState is only updated
8890 : : * after this block is over.
8891 : : */
83 peter@eisentraut.org 8892 [ + - ]:GNC 166 : PG_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, BoolGetDatum(true));
8893 : : {
5314 bruce@momjian.us 8894 :CBC 166 : bool gotUniqueStartpoint = false;
8895 : : DIR *tblspcdir;
8896 : : struct dirent *de;
8897 : : tablespaceinfo *ti;
8898 : : int datadirpathlen;
8899 : :
8900 : : /*
8901 : : * Force an XLOG file switch before the checkpoint, to ensure that the
8902 : : * WAL segment the checkpoint is written to doesn't contain pages with
8903 : : * old timeline IDs. That would otherwise happen if you called
8904 : : * pg_backup_start() right after restoring from a PITR archive: the
8905 : : * first WAL segment containing the startup checkpoint has pages in
8906 : : * the beginning with the old timeline ID. That can cause trouble at
8907 : : * recovery: we won't have a history file covering the old timeline if
8908 : : * pg_wal directory was not included in the base backup and the WAL
8909 : : * archive was cleared too before starting the backup.
8910 : : *
8911 : : * This also ensures that we have emitted a WAL page header that has
8912 : : * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
8913 : : * Therefore, if a WAL archiver (such as pglesslog) is trying to
8914 : : * compress out removable backup blocks, it won't remove any that
8915 : : * occur after this point.
8916 : : *
8917 : : * During recovery, we skip forcing XLOG file switch, which means that
8918 : : * the backup taken during recovery is not available for the special
8919 : : * recovery case described above.
8920 : : */
5024 simon@2ndQuadrant.co 8921 [ + + ]: 166 : if (!backup_started_in_recovery)
3231 andres@anarazel.de 8922 : 160 : RequestXLogSwitch(false);
8923 : :
8924 : : do
8925 : : {
8926 : : bool checkpointfpw;
8927 : :
8928 : : /*
8929 : : * Force a CHECKPOINT. Aside from being necessary to prevent torn
8930 : : * page problems, this guarantees that two successive backup runs
8931 : : * will have different checkpoint positions and hence different
8932 : : * history file names, even if nothing happened in between.
8933 : : *
8934 : : * During recovery, establish a restartpoint if possible. We use
8935 : : * the last restartpoint as the backup starting checkpoint. This
8936 : : * means that two successive backup runs can have same checkpoint
8937 : : * positions.
8938 : : *
8939 : : * Since the fact that we are executing do_pg_backup_start()
8940 : : * during recovery means that checkpointer is running, we can use
8941 : : * RequestCheckpoint() to establish a restartpoint.
8942 : : *
8943 : : * We use CHECKPOINT_FAST only if requested by user (via passing
8944 : : * fast = true). Otherwise this can take awhile.
8945 : : */
5334 heikki.linnakangas@i 8946 [ + + ]: 166 : RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_WAIT |
8947 : : (fast ? CHECKPOINT_FAST : 0));
8948 : :
8949 : : /*
8950 : : * Now we need to fetch the checkpoint record location, and also
8951 : : * its REDO pointer. The oldest point in WAL that would be needed
8952 : : * to restore starting from the checkpoint is precisely the REDO
8953 : : * pointer.
8954 : : */
8955 : 166 : LWLockAcquire(ControlFileLock, LW_SHARED);
1127 michael@paquier.xyz 8956 : 166 : state->checkpointloc = ControlFile->checkPoint;
8957 : 166 : state->startpoint = ControlFile->checkPointCopy.redo;
8958 : 166 : state->starttli = ControlFile->checkPointCopy.ThisTimeLineID;
5024 simon@2ndQuadrant.co 8959 : 166 : checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
5334 heikki.linnakangas@i 8960 : 166 : LWLockRelease(ControlFileLock);
8961 : :
5024 simon@2ndQuadrant.co 8962 [ + + ]: 166 : if (backup_started_in_recovery)
8963 : : {
8964 : : XLogRecPtr recptr;
8965 : :
8966 : : /*
8967 : : * Check to see if all WAL replayed during online backup
8968 : : * (i.e., since last restartpoint used as backup starting
8969 : : * checkpoint) contain full-page writes.
8970 : : */
4053 andres@anarazel.de 8971 [ - + ]: 6 : SpinLockAcquire(&XLogCtl->info_lck);
8972 : 6 : recptr = XLogCtl->lastFpwDisableRecPtr;
8973 : 6 : SpinLockRelease(&XLogCtl->info_lck);
8974 : :
1127 michael@paquier.xyz 8975 [ + - - + ]: 6 : if (!checkpointfpw || state->startpoint <= recptr)
5024 simon@2ndQuadrant.co 8976 [ # # ]:UBC 0 : ereport(ERROR,
8977 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8978 : : errmsg("WAL generated with \"full_page_writes=off\" was replayed "
8979 : : "since last restartpoint"),
8980 : : errhint("This means that the backup being taken on the standby "
8981 : : "is corrupt and should not be used. "
8982 : : "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
8983 : : "and then try an online backup again.")));
8984 : :
8985 : : /*
8986 : : * During recovery, since we don't use the end-of-backup WAL
8987 : : * record and don't write the backup history file, the
8988 : : * starting WAL location doesn't need to be unique. This means
8989 : : * that two base backups started at the same time might use
8990 : : * the same checkpoint as starting locations.
8991 : : */
5024 simon@2ndQuadrant.co 8992 :CBC 6 : gotUniqueStartpoint = true;
8993 : : }
8994 : :
8995 : : /*
8996 : : * If two base backups are started at the same time (in WAL sender
8997 : : * processes), we need to make sure that they use different
8998 : : * checkpoints as starting locations, because we use the starting
8999 : : * WAL location as a unique identifier for the base backup in the
9000 : : * end-of-backup WAL record and when we write the backup history
9001 : : * file. Perhaps it would be better generate a separate unique ID
9002 : : * for each backup instead of forcing another checkpoint, but
9003 : : * taking a checkpoint right after another is not that expensive
9004 : : * either because only few buffers have been dirtied yet.
9005 : : */
4238 heikki.linnakangas@i 9006 : 166 : WALInsertLockAcquireExclusive();
1127 michael@paquier.xyz 9007 [ + - ]: 166 : if (XLogCtl->Insert.lastBackupStart < state->startpoint)
9008 : : {
9009 : 166 : XLogCtl->Insert.lastBackupStart = state->startpoint;
5334 heikki.linnakangas@i 9010 : 166 : gotUniqueStartpoint = true;
9011 : : }
4238 9012 : 166 : WALInsertLockRelease();
5314 bruce@momjian.us 9013 [ - + ]: 166 : } while (!gotUniqueStartpoint);
9014 : :
9015 : : /*
9016 : : * Construct tablespace_map file.
9017 : : */
3821 andrew@dunslane.net 9018 : 166 : datadirpathlen = strlen(DataDir);
9019 : :
9020 : : /* Collect information about all tablespaces */
419 michael@paquier.xyz 9021 : 166 : tblspcdir = AllocateDir(PG_TBLSPC_DIR);
9022 [ + + ]: 537 : while ((de = ReadDir(tblspcdir, PG_TBLSPC_DIR)) != NULL)
9023 : : {
9024 : : char fullpath[MAXPGPATH + sizeof(PG_TBLSPC_DIR)];
9025 : : char linkpath[MAXPGPATH];
3821 andrew@dunslane.net 9026 : 371 : char *relpath = NULL;
9027 : : char *s;
9028 : : PGFileType de_type;
9029 : : char *badp;
9030 : : Oid tsoid;
9031 : :
9032 : : /*
9033 : : * Try to parse the directory name as an unsigned integer.
9034 : : *
9035 : : * Tablespace directories should be positive integers that can be
9036 : : * represented in 32 bits, with no leading zeroes or trailing
9037 : : * garbage. If we come across a name that doesn't meet those
9038 : : * criteria, skip it.
9039 : : */
735 rhaas@postgresql.org 9040 [ + + - + ]: 371 : if (de->d_name[0] < '1' || de->d_name[1] > '9')
9041 : 332 : continue;
9042 : 39 : errno = 0;
9043 : 39 : tsoid = strtoul(de->d_name, &badp, 10);
9044 [ + - + - : 39 : if (*badp != '\0' || errno == EINVAL || errno == ERANGE)
- + ]
3821 andrew@dunslane.net 9045 :UBC 0 : continue;
9046 : :
419 michael@paquier.xyz 9047 :CBC 39 : snprintf(fullpath, sizeof(fullpath), "%s/%s", PG_TBLSPC_DIR, de->d_name);
9048 : :
923 rhaas@postgresql.org 9049 : 39 : de_type = get_dirent_type(fullpath, de, false, ERROR);
9050 : :
9051 [ + + ]: 39 : if (de_type == PGFILETYPE_LNK)
9052 : : {
9053 : : StringInfoData escapedpath;
9054 : : int rllen;
9055 : :
9056 : 25 : rllen = readlink(fullpath, linkpath, sizeof(linkpath));
9057 [ - + ]: 25 : if (rllen < 0)
9058 : : {
923 rhaas@postgresql.org 9059 [ # # ]:UBC 0 : ereport(WARNING,
9060 : : (errmsg("could not read symbolic link \"%s\": %m",
9061 : : fullpath)));
9062 : 0 : continue;
9063 : : }
923 rhaas@postgresql.org 9064 [ - + ]:CBC 25 : else if (rllen >= sizeof(linkpath))
9065 : : {
923 rhaas@postgresql.org 9066 [ # # ]:UBC 0 : ereport(WARNING,
9067 : : (errmsg("symbolic link \"%s\" target is too long",
9068 : : fullpath)));
9069 : 0 : continue;
9070 : : }
923 rhaas@postgresql.org 9071 :CBC 25 : linkpath[rllen] = '\0';
9072 : :
9073 : : /*
9074 : : * Relpath holds the relative path of the tablespace directory
9075 : : * when it's located within PGDATA, or NULL if it's located
9076 : : * elsewhere.
9077 : : */
9078 [ - + ]: 25 : if (rllen > datadirpathlen &&
923 rhaas@postgresql.org 9079 [ # # ]:UBC 0 : strncmp(linkpath, DataDir, datadirpathlen) == 0 &&
892 tgl@sss.pgh.pa.us 9080 [ # # ]: 0 : IS_DIR_SEP(linkpath[datadirpathlen]))
923 rhaas@postgresql.org 9081 : 0 : relpath = pstrdup(linkpath + datadirpathlen + 1);
9082 : :
9083 : : /*
9084 : : * Add a backslash-escaped version of the link path to the
9085 : : * tablespace map file.
9086 : : */
923 rhaas@postgresql.org 9087 :CBC 25 : initStringInfo(&escapedpath);
9088 [ + + ]: 594 : for (s = linkpath; *s; s++)
9089 : : {
9090 [ + - + - : 569 : if (*s == '\n' || *s == '\r' || *s == '\\')
- + ]
923 rhaas@postgresql.org 9091 :UBC 0 : appendStringInfoChar(&escapedpath, '\\');
923 rhaas@postgresql.org 9092 :CBC 569 : appendStringInfoChar(&escapedpath, *s);
9093 : : }
9094 : 25 : appendStringInfo(tblspcmapfile, "%s %s\n",
9095 : 25 : de->d_name, escapedpath.data);
9096 : 25 : pfree(escapedpath.data);
9097 : : }
9098 [ + - ]: 14 : else if (de_type == PGFILETYPE_DIR)
9099 : : {
9100 : : /*
9101 : : * It's possible to use allow_in_place_tablespaces to create
9102 : : * directories directly under pg_tblspc, for testing purposes
9103 : : * only.
9104 : : *
9105 : : * In this case, we store a relative path rather than an
9106 : : * absolute path into the tablespaceinfo.
9107 : : */
419 michael@paquier.xyz 9108 : 14 : snprintf(linkpath, sizeof(linkpath), "%s/%s",
9109 : 14 : PG_TBLSPC_DIR, de->d_name);
923 rhaas@postgresql.org 9110 : 14 : relpath = pstrdup(linkpath);
9111 : : }
9112 : : else
9113 : : {
9114 : : /* Skip any other file type that appears here. */
923 rhaas@postgresql.org 9115 :UBC 0 : continue;
9116 : : }
9117 : :
3821 andrew@dunslane.net 9118 :CBC 39 : ti = palloc(sizeof(tablespaceinfo));
735 rhaas@postgresql.org 9119 : 39 : ti->oid = tsoid;
1685 tgl@sss.pgh.pa.us 9120 : 39 : ti->path = pstrdup(linkpath);
923 rhaas@postgresql.org 9121 : 39 : ti->rpath = relpath;
1958 9122 : 39 : ti->size = -1;
9123 : :
3810 bruce@momjian.us 9124 [ + - ]: 39 : if (tablespaces)
9125 : 39 : *tablespaces = lappend(*tablespaces, ti);
9126 : : }
2884 tgl@sss.pgh.pa.us 9127 : 166 : FreeDir(tblspcdir);
9128 : :
1127 michael@paquier.xyz 9129 : 166 : state->starttime = (pg_time_t) time(NULL);
9130 : : }
83 peter@eisentraut.org 9131 [ - + ]:GNC 166 : PG_END_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, BoolGetDatum(true));
9132 : :
1127 michael@paquier.xyz 9133 :CBC 166 : state->started_in_recovery = backup_started_in_recovery;
9134 : :
9135 : : /*
9136 : : * Mark that the start phase has correctly finished for the backup.
9137 : : */
1300 sfrost@snowman.net 9138 : 166 : sessionBackupState = SESSION_BACKUP_RUNNING;
7755 tgl@sss.pgh.pa.us 9139 : 166 : }
9140 : :
9141 : : /*
9142 : : * Utility routine to fetch the session-level status of a backup running.
9143 : : */
9144 : : SessionBackupState
3139 teodor@sigaev.ru 9145 : 187 : get_backup_status(void)
9146 : : {
9147 : 187 : return sessionBackupState;
9148 : : }
9149 : :
9150 : : /*
9151 : : * do_pg_backup_stop
9152 : : *
9153 : : * Utility function called at the end of an online backup. It creates history
9154 : : * file (if required), resets sessionBackupState and so on. It can optionally
9155 : : * wait for WAL segments to be archived.
9156 : : *
9157 : : * "state" is filled with the information necessary to restore from this
9158 : : * backup with its stop LSN (stoppoint), its timeline ID (stoptli), etc.
9159 : : *
9160 : : * It is the responsibility of the caller of this function to verify the
9161 : : * permissions of the calling user!
9162 : : */
9163 : : void
1127 michael@paquier.xyz 9164 : 160 : do_pg_backup_stop(BackupState *state, bool waitforarchive)
9165 : : {
9166 : 160 : bool backup_stopped_in_recovery = false;
9167 : : char histfilepath[MAXPGPATH];
9168 : : char lastxlogfilename[MAXFNAMELEN];
9169 : : char histfilename[MAXFNAMELEN];
9170 : : XLogSegNo _logSegNo;
9171 : : FILE *fp;
9172 : : int seconds_before_warning;
6414 bruce@momjian.us 9173 : 160 : int waits = 0;
5671 simon@2ndQuadrant.co 9174 : 160 : bool reported_waiting = false;
9175 : :
1127 michael@paquier.xyz 9176 [ - + ]: 160 : Assert(state != NULL);
9177 : :
9178 : 160 : backup_stopped_in_recovery = RecoveryInProgress();
9179 : :
9180 : : /*
9181 : : * During recovery, we don't need to check WAL level. Because, if WAL
9182 : : * level is not sufficient, it's impossible to get here during recovery.
9183 : : */
9184 [ + + - + ]: 160 : if (!backup_stopped_in_recovery && !XLogIsNeeded())
6258 tgl@sss.pgh.pa.us 9185 [ # # ]:UBC 0 : ereport(ERROR,
9186 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9187 : : errmsg("WAL level not sufficient for making an online backup"),
9188 : : errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
9189 : :
9190 : : /*
9191 : : * OK to update backup counter and session-level lock.
9192 : : *
9193 : : * Note that CHECK_FOR_INTERRUPTS() must not occur while updating them,
9194 : : * otherwise they can be updated inconsistently, which might cause
9195 : : * do_pg_abort_backup() to fail.
9196 : : */
3205 fujii@postgresql.org 9197 :CBC 160 : WALInsertLockAcquireExclusive();
9198 : :
9199 : : /*
9200 : : * It is expected that each do_pg_backup_start() call is matched by
9201 : : * exactly one do_pg_backup_stop() call.
9202 : : */
1300 sfrost@snowman.net 9203 [ - + ]: 160 : Assert(XLogCtl->Insert.runningBackups > 0);
9204 : 160 : XLogCtl->Insert.runningBackups--;
9205 : :
9206 : : /*
9207 : : * Clean up session-level lock.
9208 : : *
9209 : : * You might think that WALInsertLockRelease() can be called before
9210 : : * cleaning up session-level lock because session-level lock doesn't need
9211 : : * to be protected with WAL insertion lock. But since
9212 : : * CHECK_FOR_INTERRUPTS() can occur in it, session-level lock must be
9213 : : * cleaned up before it.
9214 : : */
3139 teodor@sigaev.ru 9215 : 160 : sessionBackupState = SESSION_BACKUP_NONE;
9216 : :
2869 fujii@postgresql.org 9217 : 160 : WALInsertLockRelease();
9218 : :
9219 : : /*
9220 : : * If we are taking an online backup from the standby, we confirm that the
9221 : : * standby has not been promoted during the backup.
9222 : : */
1127 michael@paquier.xyz 9223 [ + + - + ]: 160 : if (state->started_in_recovery && !backup_stopped_in_recovery)
5024 simon@2ndQuadrant.co 9224 [ # # ]:UBC 0 : ereport(ERROR,
9225 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9226 : : errmsg("the standby was promoted during online backup"),
9227 : : errhint("This means that the backup being taken is corrupt "
9228 : : "and should not be used. "
9229 : : "Try taking another online backup.")));
9230 : :
9231 : : /*
9232 : : * During recovery, we don't write an end-of-backup record. We assume that
9233 : : * pg_control was backed up last and its minimum recovery point can be
9234 : : * available as the backup end location. Since we don't have an
9235 : : * end-of-backup record, we use the pg_control value to check whether
9236 : : * we've reached the end of backup when starting recovery from this
9237 : : * backup. We have no way of checking if pg_control wasn't backed up last
9238 : : * however.
9239 : : *
9240 : : * We don't force a switch to new WAL file but it is still possible to
9241 : : * wait for all the required files to be archived if waitforarchive is
9242 : : * true. This is okay if we use the backup to start a standby and fetch
9243 : : * the missing WAL using streaming replication. But in the case of an
9244 : : * archive recovery, a user should set waitforarchive to true and wait for
9245 : : * them to be archived to ensure that all the required files are
9246 : : * available.
9247 : : *
9248 : : * We return the current minimum recovery point as the backup end
9249 : : * location. Note that it can be greater than the exact backup end
9250 : : * location if the minimum recovery point is updated after the backup of
9251 : : * pg_control. This is harmless for current uses.
9252 : : *
9253 : : * XXX currently a backup history file is for informational and debug
9254 : : * purposes only. It's not essential for an online backup. Furthermore,
9255 : : * even if it's created, it will not be archived during recovery because
9256 : : * an archiver is not invoked. So it doesn't seem worthwhile to write a
9257 : : * backup history file during recovery.
9258 : : */
1127 michael@paquier.xyz 9259 [ + + ]:CBC 160 : if (backup_stopped_in_recovery)
9260 : : {
9261 : : XLogRecPtr recptr;
9262 : :
9263 : : /*
9264 : : * Check to see if all WAL replayed during online backup contain
9265 : : * full-page writes.
9266 : : */
4053 andres@anarazel.de 9267 [ - + ]: 6 : SpinLockAcquire(&XLogCtl->info_lck);
9268 : 6 : recptr = XLogCtl->lastFpwDisableRecPtr;
9269 : 6 : SpinLockRelease(&XLogCtl->info_lck);
9270 : :
1127 michael@paquier.xyz 9271 [ - + ]: 6 : if (state->startpoint <= recptr)
5024 simon@2ndQuadrant.co 9272 [ # # ]:UBC 0 : ereport(ERROR,
9273 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9274 : : errmsg("WAL generated with \"full_page_writes=off\" was replayed "
9275 : : "during online backup"),
9276 : : errhint("This means that the backup being taken on the standby "
9277 : : "is corrupt and should not be used. "
9278 : : "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
9279 : : "and then try an online backup again.")));
9280 : :
9281 : :
5024 simon@2ndQuadrant.co 9282 :CBC 6 : LWLockAcquire(ControlFileLock, LW_SHARED);
1127 michael@paquier.xyz 9283 : 6 : state->stoppoint = ControlFile->minRecoveryPoint;
9284 : 6 : state->stoptli = ControlFile->minRecoveryPointTLI;
5024 simon@2ndQuadrant.co 9285 : 6 : LWLockRelease(ControlFileLock);
9286 : : }
9287 : : else
9288 : : {
9289 : : char *history_file;
9290 : :
9291 : : /*
9292 : : * Write the backup-end xlog record
9293 : : */
3005 rhaas@postgresql.org 9294 : 154 : XLogBeginInsert();
258 peter@eisentraut.org 9295 : 154 : XLogRegisterData(&state->startpoint,
9296 : : sizeof(state->startpoint));
1127 michael@paquier.xyz 9297 : 154 : state->stoppoint = XLogInsert(RM_XLOG_ID, XLOG_BACKUP_END);
9298 : :
9299 : : /*
9300 : : * Given that we're not in recovery, InsertTimeLineID is set and can't
9301 : : * change, so we can read it without a lock.
9302 : : */
9303 : 154 : state->stoptli = XLogCtl->InsertTimeLineID;
9304 : :
9305 : : /*
9306 : : * Force a switch to a new xlog segment file, so that the backup is
9307 : : * valid as soon as archiver moves out the current segment file.
9308 : : */
3005 rhaas@postgresql.org 9309 : 154 : RequestXLogSwitch(false);
9310 : :
1127 michael@paquier.xyz 9311 : 154 : state->stoptime = (pg_time_t) time(NULL);
9312 : :
9313 : : /*
9314 : : * Write the backup history file
9315 : : */
9316 : 154 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9317 : 154 : BackupHistoryFilePath(histfilepath, state->stoptli, _logSegNo,
9318 : : state->startpoint, wal_segment_size);
3005 rhaas@postgresql.org 9319 : 154 : fp = AllocateFile(histfilepath, "w");
9320 [ - + ]: 154 : if (!fp)
3005 rhaas@postgresql.org 9321 [ # # ]:UBC 0 : ereport(ERROR,
9322 : : (errcode_for_file_access(),
9323 : : errmsg("could not create file \"%s\": %m",
9324 : : histfilepath)));
9325 : :
9326 : : /* Build and save the contents of the backup history file */
1127 michael@paquier.xyz 9327 :CBC 154 : history_file = build_backup_content(state, true);
1126 9328 : 154 : fprintf(fp, "%s", history_file);
1127 9329 : 154 : pfree(history_file);
9330 : :
3005 rhaas@postgresql.org 9331 [ + - + - : 154 : if (fflush(fp) || ferror(fp) || FreeFile(fp))
- + ]
3005 rhaas@postgresql.org 9332 [ # # ]:UBC 0 : ereport(ERROR,
9333 : : (errcode_for_file_access(),
9334 : : errmsg("could not write file \"%s\": %m",
9335 : : histfilepath)));
9336 : :
9337 : : /*
9338 : : * Clean out any no-longer-needed history files. As a side effect,
9339 : : * this will post a .ready file for the newly created history file,
9340 : : * notifying the archiver that history file may be archived
9341 : : * immediately.
9342 : : */
3005 rhaas@postgresql.org 9343 :CBC 154 : CleanupBackupHistory();
9344 : : }
9345 : :
9346 : : /*
9347 : : * If archiving is enabled, wait for all the required WAL files to be
9348 : : * archived before returning. If archiving isn't enabled, the required WAL
9349 : : * needs to be transported via streaming replication (hopefully with
9350 : : * wal_keep_size set high enough), or some more exotic mechanism like
9351 : : * polling and copying files from pg_wal with script. We have no knowledge
9352 : : * of those mechanisms, so it's up to the user to ensure that he gets all
9353 : : * the required WAL.
9354 : : *
9355 : : * We wait until both the last WAL file filled during backup and the
9356 : : * history file have been archived, and assume that the alphabetic sorting
9357 : : * property of the WAL files ensures any earlier WAL files are safely
9358 : : * archived as well.
9359 : : *
9360 : : * We wait forever, since archive_command is supposed to work and we
9361 : : * assume the admin wanted his backup to work completely. If you don't
9362 : : * wish to wait, then either waitforarchive should be passed in as false,
9363 : : * or you can set statement_timeout. Also, some notices are issued to
9364 : : * clue in anyone who might be doing this interactively.
9365 : : */
9366 : :
9367 [ + + ]: 160 : if (waitforarchive &&
1127 michael@paquier.xyz 9368 [ + + + + : 10 : ((!backup_stopped_in_recovery && XLogArchivingActive()) ||
- + + + +
+ ]
9369 [ + - - + : 1 : (backup_stopped_in_recovery && XLogArchivingAlways())))
- + ]
9370 : : {
9371 : 4 : XLByteToPrevSeg(state->stoppoint, _logSegNo, wal_segment_size);
9372 : 4 : XLogFileName(lastxlogfilename, state->stoptli, _logSegNo,
9373 : : wal_segment_size);
9374 : :
9375 : 4 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9376 : 4 : BackupHistoryFileName(histfilename, state->stoptli, _logSegNo,
9377 : : state->startpoint, wal_segment_size);
9378 : :
5592 bruce@momjian.us 9379 : 4 : seconds_before_warning = 60;
9380 : 4 : waits = 0;
9381 : :
9382 [ + + - + ]: 12 : while (XLogArchiveIsBusy(lastxlogfilename) ||
9383 : 4 : XLogArchiveIsBusy(histfilename))
9384 : : {
9385 [ - + ]: 4 : CHECK_FOR_INTERRUPTS();
9386 : :
9387 [ + - - + ]: 4 : if (!reported_waiting && waits > 5)
9388 : : {
5592 bruce@momjian.us 9389 [ # # ]:UBC 0 : ereport(NOTICE,
9390 : : (errmsg("base backup done, waiting for required WAL segments to be archived")));
9391 : 0 : reported_waiting = true;
9392 : : }
9393 : :
1574 michael@paquier.xyz 9394 :CBC 4 : (void) WaitLatch(MyLatch,
9395 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
9396 : : 1000L,
9397 : : WAIT_EVENT_BACKUP_WAIT_WAL_ARCHIVE);
9398 : 4 : ResetLatch(MyLatch);
9399 : :
5592 bruce@momjian.us 9400 [ - + ]: 4 : if (++waits >= seconds_before_warning)
9401 : : {
5592 bruce@momjian.us 9402 :UBC 0 : seconds_before_warning *= 2; /* This wraps in >10 years... */
9403 [ # # ]: 0 : ereport(WARNING,
9404 : : (errmsg("still waiting for all required WAL segments to be archived (%d seconds elapsed)",
9405 : : waits),
9406 : : errhint("Check that your \"archive_command\" is executing properly. "
9407 : : "You can safely cancel this backup, "
9408 : : "but the database backup will not be usable without all the WAL segments.")));
9409 : : }
9410 : : }
9411 : :
5592 bruce@momjian.us 9412 [ + + ]:CBC 4 : ereport(NOTICE,
9413 : : (errmsg("all required WAL segments have been archived")));
9414 : : }
5374 magnus@hagander.net 9415 [ + + ]: 156 : else if (waitforarchive)
5660 tgl@sss.pgh.pa.us 9416 [ + - ]: 6 : ereport(NOTICE,
9417 : : (errmsg("WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup")));
5405 magnus@hagander.net 9418 : 160 : }
9419 : :
9420 : :
9421 : : /*
9422 : : * do_pg_abort_backup: abort a running backup
9423 : : *
9424 : : * This does just the most basic steps of do_pg_backup_stop(), by taking the
9425 : : * system out of backup mode, thus making it a lot more safe to call from
9426 : : * an error handler.
9427 : : *
9428 : : * 'arg' indicates that it's being called during backup setup; so
9429 : : * sessionBackupState has not been modified yet, but runningBackups has
9430 : : * already been incremented. When it's false, then it's invoked as a
9431 : : * before_shmem_exit handler, and therefore we must not change state
9432 : : * unless sessionBackupState indicates that a backup is actually running.
9433 : : *
9434 : : * NB: This gets used as a PG_ENSURE_ERROR_CLEANUP callback and
9435 : : * before_shmem_exit handler, hence the odd-looking signature.
9436 : : */
9437 : : void
2139 rhaas@postgresql.org 9438 : 8 : do_pg_abort_backup(int code, Datum arg)
9439 : : {
1104 alvherre@alvh.no-ip. 9440 : 8 : bool during_backup_start = DatumGetBool(arg);
9441 : :
9442 : : /* If called during backup start, there shouldn't be one already running */
1099 9443 [ - + - - ]: 8 : Assert(!during_backup_start || sessionBackupState == SESSION_BACKUP_NONE);
9444 : :
1104 9445 [ + - + + ]: 8 : if (during_backup_start || sessionBackupState != SESSION_BACKUP_NONE)
9446 : : {
9447 : 6 : WALInsertLockAcquireExclusive();
9448 [ - + ]: 6 : Assert(XLogCtl->Insert.runningBackups > 0);
9449 : 6 : XLogCtl->Insert.runningBackups--;
9450 : :
9451 : 6 : sessionBackupState = SESSION_BACKUP_NONE;
9452 : 6 : WALInsertLockRelease();
9453 : :
9454 [ + - ]: 6 : if (!during_backup_start)
9455 [ + - ]: 6 : ereport(WARNING,
9456 : : errmsg("aborting backup due to backend exiting before pg_backup_stop was called"));
9457 : : }
2139 rhaas@postgresql.org 9458 : 8 : }
9459 : :
9460 : : /*
9461 : : * Register a handler that will warn about unterminated backups at end of
9462 : : * session, unless this has already been done.
9463 : : */
9464 : : void
9465 : 4 : register_persistent_abort_backup_handler(void)
9466 : : {
9467 : : static bool already_done = false;
9468 : :
9469 [ + + ]: 4 : if (already_done)
9470 : 1 : return;
83 peter@eisentraut.org 9471 :GNC 3 : before_shmem_exit(do_pg_abort_backup, BoolGetDatum(false));
2139 rhaas@postgresql.org 9472 :CBC 3 : already_done = true;
9473 : : }
9474 : :
9475 : : /*
9476 : : * Get latest WAL insert pointer
9477 : : */
9478 : : XLogRecPtr
5038 heikki.linnakangas@i 9479 : 1929 : GetXLogInsertRecPtr(void)
9480 : : {
4053 andres@anarazel.de 9481 : 1929 : XLogCtlInsert *Insert = &XLogCtl->Insert;
9482 : : uint64 current_bytepos;
9483 : :
4494 heikki.linnakangas@i 9484 [ - + ]: 1929 : SpinLockAcquire(&Insert->insertpos_lck);
9485 : 1929 : current_bytepos = Insert->CurrBytePos;
9486 : 1929 : SpinLockRelease(&Insert->insertpos_lck);
9487 : :
9488 : 1929 : return XLogBytePosToRecPtr(current_bytepos);
9489 : : }
9490 : :
9491 : : /*
9492 : : * Get latest WAL write pointer
9493 : : */
9494 : : XLogRecPtr
1349 9495 : 1435 : GetXLogWriteRecPtr(void)
9496 : : {
572 alvherre@alvh.no-ip. 9497 : 1435 : RefreshXLogWriteResult(LogwrtResult);
9498 : :
1349 heikki.linnakangas@i 9499 : 1435 : return LogwrtResult.Write;
9500 : : }
9501 : :
9502 : : /*
9503 : : * Returns the redo pointer of the last checkpoint or restartpoint. This is
9504 : : * the oldest point in WAL that we still need, if we have to restart recovery.
9505 : : */
9506 : : void
9507 : 376 : GetOldestRestartPoint(XLogRecPtr *oldrecptr, TimeLineID *oldtli)
9508 : : {
9509 : 376 : LWLockAcquire(ControlFileLock, LW_SHARED);
9510 : 376 : *oldrecptr = ControlFile->checkPointCopy.redo;
9511 : 376 : *oldtli = ControlFile->checkPointCopy.ThisTimeLineID;
9512 : 376 : LWLockRelease(ControlFileLock);
7157 tgl@sss.pgh.pa.us 9513 : 376 : }
9514 : :
9515 : : /* Thin wrapper around ShutdownWalRcv(). */
9516 : : void
1582 noah@leadboat.com 9517 : 1002 : XLogShutdownWalRcv(void)
9518 : : {
9519 : 1002 : ShutdownWalRcv();
9520 : :
9521 : 1002 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9522 : 1002 : XLogCtl->InstallXLogFileSegmentActive = false;
9523 : 1002 : LWLockRelease(ControlFileLock);
9524 : 1002 : }
9525 : :
9526 : : /* Enable WAL file recycling and preallocation. */
9527 : : void
1349 heikki.linnakangas@i 9528 : 1056 : SetInstallXLogFileSegmentActive(void)
9529 : : {
9530 : 1056 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9531 : 1056 : XLogCtl->InstallXLogFileSegmentActive = true;
9532 : 1056 : LWLockRelease(ControlFileLock);
3701 fujii@postgresql.org 9533 : 1056 : }
9534 : :
9535 : : bool
1349 heikki.linnakangas@i 9536 : 357 : IsInstallXLogFileSegmentActive(void)
9537 : : {
9538 : : bool result;
9539 : :
9540 : 357 : LWLockAcquire(ControlFileLock, LW_SHARED);
9541 : 357 : result = XLogCtl->InstallXLogFileSegmentActive;
9542 : 357 : LWLockRelease(ControlFileLock);
9543 : :
9544 : 357 : return result;
9545 : : }
9546 : :
9547 : : /*
9548 : : * Update the WalWriterSleeping flag.
9549 : : */
9550 : : void
4920 tgl@sss.pgh.pa.us 9551 : 473 : SetWalWriterSleeping(bool sleeping)
9552 : : {
4053 andres@anarazel.de 9553 [ - + ]: 473 : SpinLockAcquire(&XLogCtl->info_lck);
9554 : 473 : XLogCtl->WalWriterSleeping = sleeping;
9555 : 473 : SpinLockRelease(&XLogCtl->info_lck);
4920 tgl@sss.pgh.pa.us 9556 : 473 : }
|
