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
3180 andres@anarazel.de 748 :CBC 13854345 : XLogInsertRecord(XLogRecData *rdata,
749 : : XLogRecPtr fpw_lsn,
750 : : uint8 flags,
751 : : int num_fpi,
752 : : bool topxid_included)
753 : : {
8934 bruce@momjian.us 754 : 13854345 : XLogCtlInsert *Insert = &XLogCtl->Insert;
755 : : pg_crc32c rdata_crc;
756 : : bool inserted;
3957 heikki.linnakangas@i 757 : 13854345 : XLogRecord *rechdr = (XLogRecord *) rdata->data;
3228 tgl@sss.pgh.pa.us 758 : 13854345 : uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
688 rhaas@postgresql.org 759 : 13854345 : WalInsertClass class = WALINSERT_NORMAL;
760 : : XLogRecPtr StartPos;
761 : : XLogRecPtr EndPos;
2550 akapila@postgresql.o 762 : 13854345 : bool prevDoPageWrites = doPageWrites;
763 : : TimeLineID insertTLI;
764 : :
765 : : /* Does this record type require special handling? */
688 rhaas@postgresql.org 766 [ + + ]: 13854345 : if (unlikely(rechdr->xl_rmid == RM_XLOG_ID))
767 : : {
768 [ + + ]: 214671 : if (info == XLOG_SWITCH)
769 : 707 : class = WALINSERT_SPECIAL_SWITCH;
770 [ + + ]: 213964 : else if (info == XLOG_CHECKPOINT_REDO)
771 : 896 : class = WALINSERT_SPECIAL_CHECKPOINT;
772 : : }
773 : :
774 : : /* we assume that all of the record header is in the first chunk */
3943 heikki.linnakangas@i 775 [ - + ]: 13854345 : Assert(rdata->len >= SizeOfXLogRecord);
776 : :
777 : : /* cross-check on whether we should be here or not */
5916 tgl@sss.pgh.pa.us 778 [ - + ]: 13854345 : if (!XLogInsertAllowed())
5916 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 : : */
1396 rhaas@postgresql.org 785 :CBC 13854345 : 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 : : */
4987 heikki.linnakangas@i 819 : 13854345 : START_CRIT_SECTION();
820 : :
688 rhaas@postgresql.org 821 [ + + ]: 13854345 : if (likely(class == WALINSERT_NORMAL))
822 : : {
697 823 : 13852742 : 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 [ + + ]: 13852742 : if (RedoRecPtr != Insert->RedoRecPtr)
842 : : {
843 [ - + ]: 6147 : Assert(RedoRecPtr < Insert->RedoRecPtr);
844 : 6147 : RedoRecPtr = Insert->RedoRecPtr;
845 : : }
846 [ + + + + ]: 13852742 : doPageWrites = (Insert->fullPageWrites || Insert->runningBackups > 0);
847 : :
848 [ + + ]: 13852742 : if (doPageWrites &&
849 [ + + + + ]: 13610517 : (!prevDoPageWrites ||
850 [ + + ]: 12888867 : (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 : 6839 : WALInsertLockRelease();
857 [ - + ]: 6839 : END_CRIT_SECTION();
858 : 6839 : return InvalidXLogRecPtr;
859 : : }
860 : :
861 : : /*
862 : : * Reserve space for the record in the WAL. This also sets the xl_prev
863 : : * pointer.
864 : : */
3957 heikki.linnakangas@i 865 : 13845903 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
866 : : &rechdr->xl_prev);
867 : :
868 : : /* Normal records are always inserted. */
4443 869 : 13845903 : inserted = true;
870 : : }
688 rhaas@postgresql.org 871 [ + + ]: 1603 : 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 : : */
697 884 [ - + ]: 707 : Assert(fpw_lsn == InvalidXLogRecPtr);
885 : 707 : WALInsertLockAcquireExclusive();
886 : 707 : inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
887 : : }
888 : : else
889 : : {
688 890 [ - + ]: 896 : 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 [ - + ]: 896 : Assert(fpw_lsn == InvalidXLogRecPtr);
900 : 896 : WALInsertLockAcquireExclusive();
901 : 896 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
902 : : &rechdr->xl_prev);
903 : 896 : RedoRecPtr = Insert->RedoRecPtr = StartPos;
904 : 896 : inserted = true;
905 : : }
906 : :
4443 heikki.linnakangas@i 907 [ + + ]: 13847506 : if (inserted)
908 : : {
909 : : /*
910 : : * Now that xl_prev has been filled in, calculate CRC of the record
911 : : * header.
912 : : */
3943 913 : 13847439 : rdata_crc = rechdr->xl_crc;
914 : 13847439 : COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
3959 915 : 13847439 : FIN_CRC32C(rdata_crc);
4443 916 : 13847439 : 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 : : */
688 rhaas@postgresql.org 922 : 13847439 : 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 : : */
3180 andres@anarazel.de 931 [ + + ]: 13847439 : if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
932 : : {
3034 bruce@momjian.us 933 [ + + ]: 13761513 : int lockno = holdingAllLocks ? 0 : MyLockNo;
934 : :
3180 andres@anarazel.de 935 : 13761513 : 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 : : */
4187 heikki.linnakangas@i 950 : 13847506 : WALInsertLockRelease();
951 : :
4443 952 [ - + ]: 13847506 : END_CRIT_SECTION();
953 : :
1404 akapila@postgresql.o 954 : 13847506 : 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 [ + + ]: 13847506 : if (topxid_included)
961 : 219 : MarkSubxactTopXidLogged();
962 : :
963 : : /*
964 : : * Update shared LogwrtRqst.Write, if we crossed page boundary.
965 : : */
4443 heikki.linnakangas@i 966 [ + + ]: 13847506 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
967 : : {
4002 andres@anarazel.de 968 [ + + ]: 1648538 : SpinLockAcquire(&XLogCtl->info_lck);
969 : : /* advance global request to include new block(s) */
970 [ + + ]: 1648538 : if (XLogCtl->LogwrtRqst.Write < EndPos)
971 : 1585237 : XLogCtl->LogwrtRqst.Write = EndPos;
972 : 1648538 : SpinLockRelease(&XLogCtl->info_lck);
519 alvherre@alvh.no-ip. 973 : 1648538 : 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 : : */
688 rhaas@postgresql.org 981 [ + + ]: 13847506 : if (class == WALINSERT_SPECIAL_SWITCH)
982 : : {
983 : : TRACE_POSTGRESQL_WAL_SWITCH();
4443 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 : 640 : EndPos = StartPos + SizeOfXLogRecord;
994 [ - + ]: 640 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
995 : : {
2909 andres@anarazel.de 996 :UBC 0 : uint64 offset = XLogSegmentOffset(EndPos, wal_segment_size);
997 : :
998 [ # # ]: 0 : if (offset == EndPos % XLOG_BLCKSZ)
4443 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 : : */
4443 heikki.linnakangas@i 1075 :CBC 13847506 : ProcLastRecPtr = StartPos;
1076 : 13847506 : XactLastRecEnd = EndPos;
1077 : :
1078 : : /* Report WAL traffic to the instrumentation. */
1981 akapila@postgresql.o 1079 [ + + ]: 13847506 : if (inserted)
1080 : : {
1081 : 13847439 : pgWalUsage.wal_bytes += rechdr->xl_tot_len;
1082 : 13847439 : pgWalUsage.wal_records++;
1950 1083 : 13847439 : pgWalUsage.wal_fpi += num_fpi;
1084 : :
1085 : : /* Required for the flush of pending stats WAL data */
40 michael@paquier.xyz 1086 : 13847439 : pgstat_report_fixed = true;
1087 : : }
1088 : :
4443 heikki.linnakangas@i 1089 : 13847506 : 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 : 13846799 : ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos,
1112 : : XLogRecPtr *PrevPtr)
1113 : : {
4002 andres@anarazel.de 1114 : 13846799 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1115 : : uint64 startbytepos;
1116 : : uint64 endbytepos;
1117 : : uint64 prevbytepos;
1118 : :
4443 heikki.linnakangas@i 1119 : 13846799 : size = MAXALIGN(size);
1120 : :
1121 : : /* All (non xlog-switch) records should contain data. */
1122 [ - + ]: 13846799 : 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 [ + + ]: 13846799 : SpinLockAcquire(&Insert->insertpos_lck);
1135 : :
1136 : 13846799 : startbytepos = Insert->CurrBytePos;
1137 : 13846799 : endbytepos = startbytepos + size;
1138 : 13846799 : prevbytepos = Insert->PrevBytePos;
1139 : 13846799 : Insert->CurrBytePos = endbytepos;
1140 : 13846799 : Insert->PrevBytePos = startbytepos;
1141 : :
1142 : 13846799 : SpinLockRelease(&Insert->insertpos_lck);
1143 : :
1144 : 13846799 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1145 : 13846799 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1146 : 13846799 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1147 : :
1148 : : /*
1149 : : * Check that the conversions between "usable byte positions" and
1150 : : * XLogRecPtrs work consistently in both directions.
1151 : : */
1152 [ - + ]: 13846799 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1153 [ - + ]: 13846799 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1154 [ - + ]: 13846799 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1155 : 13846799 : }
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 : : {
4002 andres@anarazel.de 1169 : 707 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1170 : : uint64 startbytepos;
1171 : : uint64 endbytepos;
1172 : : uint64 prevbytepos;
3943 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 : : */
4443 1183 [ - + ]: 707 : SpinLockAcquire(&Insert->insertpos_lck);
1184 : :
1185 : 707 : startbytepos = Insert->CurrBytePos;
1186 : :
1187 : 707 : ptr = XLogBytePosToEndRecPtr(startbytepos);
2909 andres@anarazel.de 1188 [ + + ]: 707 : if (XLogSegmentOffset(ptr, wal_segment_size) == 0)
1189 : : {
4443 heikki.linnakangas@i 1190 : 67 : SpinLockRelease(&Insert->insertpos_lck);
1191 : 67 : *EndPos = *StartPos = ptr;
1192 : 67 : return false;
1193 : : }
1194 : :
1195 : 640 : endbytepos = startbytepos + size;
1196 : 640 : prevbytepos = Insert->PrevBytePos;
1197 : :
1198 : 640 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1199 : 640 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1200 : :
2909 andres@anarazel.de 1201 : 640 : segleft = wal_segment_size - XLogSegmentOffset(*EndPos, wal_segment_size);
1202 [ + - ]: 640 : if (segleft != wal_segment_size)
1203 : : {
1204 : : /* consume the rest of the segment */
4443 heikki.linnakangas@i 1205 : 640 : *EndPos += segleft;
1206 : 640 : endbytepos = XLogRecPtrToBytePos(*EndPos);
1207 : : }
1208 : 640 : Insert->CurrBytePos = endbytepos;
1209 : 640 : Insert->PrevBytePos = startbytepos;
1210 : :
1211 : 640 : SpinLockRelease(&Insert->insertpos_lck);
1212 : :
1213 : 640 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1214 : :
2909 andres@anarazel.de 1215 [ - + ]: 640 : Assert(XLogSegmentOffset(*EndPos, wal_segment_size) == 0);
4443 heikki.linnakangas@i 1216 [ - + ]: 640 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1217 [ - + ]: 640 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1218 [ - + ]: 640 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1219 : :
1220 : 640 : 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 : 13847439 : 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 : 13847439 : CurrPos = StartPos;
1401 rhaas@postgresql.org 1242 : 13847439 : currpos = GetXLogBuffer(CurrPos, tli);
4443 heikki.linnakangas@i 1243 [ + - ]: 13847439 : 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 [ - + ]: 13847439 : Assert(freespace >= sizeof(uint32));
1250 : :
1251 : : /* Copy record data */
1252 : 13847439 : written = 0;
1253 [ + + ]: 66320011 : while (rdata != NULL)
1254 : : {
368 peter@eisentraut.org 1255 : 52472572 : const char *rdata_data = rdata->data;
4443 heikki.linnakangas@i 1256 : 52472572 : int rdata_len = rdata->len;
1257 : :
1258 [ + + ]: 54240921 : while (rdata_len > freespace)
1259 : : {
1260 : : /*
1261 : : * Write what fits on this page, and continue on the next page.
1262 : : */
1263 [ + + - + ]: 1768349 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || freespace == 0);
1264 : 1768349 : memcpy(currpos, rdata_data, freespace);
1265 : 1768349 : rdata_data += freespace;
1266 : 1768349 : rdata_len -= freespace;
1267 : 1768349 : written += freespace;
1268 : 1768349 : 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 : : */
1401 rhaas@postgresql.org 1279 : 1768349 : currpos = GetXLogBuffer(CurrPos, tli);
4443 heikki.linnakangas@i 1280 : 1768349 : pagehdr = (XLogPageHeader) currpos;
1281 : 1768349 : pagehdr->xlp_rem_len = write_len - written;
1282 : 1768349 : pagehdr->xlp_info |= XLP_FIRST_IS_CONTRECORD;
1283 : :
1284 : : /* skip over the page header */
2909 andres@anarazel.de 1285 [ + + ]: 1768349 : if (XLogSegmentOffset(CurrPos, wal_segment_size) == 0)
1286 : : {
4443 heikki.linnakangas@i 1287 : 1187 : CurrPos += SizeOfXLogLongPHD;
1288 : 1187 : currpos += SizeOfXLogLongPHD;
1289 : : }
1290 : : else
1291 : : {
1292 : 1767162 : CurrPos += SizeOfXLogShortPHD;
1293 : 1767162 : currpos += SizeOfXLogShortPHD;
1294 : : }
1295 [ + - ]: 1768349 : freespace = INSERT_FREESPACE(CurrPos);
1296 : : }
1297 : :
1298 [ - + - - ]: 52472572 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || rdata_len == 0);
1299 : 52472572 : memcpy(currpos, rdata_data, rdata_len);
1300 : 52472572 : currpos += rdata_len;
1301 : 52472572 : CurrPos += rdata_len;
1302 : 52472572 : freespace -= rdata_len;
1303 : 52472572 : written += rdata_len;
1304 : :
1305 : 52472572 : rdata = rdata->next;
1306 : : }
1307 [ - + ]: 13847439 : 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 : : */
2909 andres@anarazel.de 1314 [ + + + - ]: 13847439 : if (isLogSwitch && XLogSegmentOffset(CurrPos, wal_segment_size) != 0)
1315 : : {
1316 : : /* An xlog-switch record doesn't contain any data besides the header */
4443 heikki.linnakangas@i 1317 [ - + ]: 640 : Assert(write_len == SizeOfXLogRecord);
1318 : :
1319 : : /* Assert that we did reserve the right amount of space */
2909 andres@anarazel.de 1320 [ - + ]: 640 : Assert(XLogSegmentOffset(EndPos, wal_segment_size) == 0);
1321 : :
1322 : : /* Use up all the remaining space on the current page */
4443 heikki.linnakangas@i 1323 : 640 : 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 [ + + ]: 529997 : 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 : : */
1401 rhaas@postgresql.org 1352 : 529357 : currpos = GetXLogBuffer(CurrPos, tli);
2717 tgl@sss.pgh.pa.us 1353 [ + - + - : 2117428 : MemSet(currpos, 0, SizeOfXLogShortPHD);
+ - + - +
+ ]
1354 : :
4443 heikki.linnakangas@i 1355 : 529357 : CurrPos += XLOG_BLCKSZ;
1356 : : }
1357 : : }
1358 : : else
1359 : : {
1360 : : /* Align the end position, so that the next record starts aligned */
3943 1361 : 13846799 : CurrPos = MAXALIGN64(CurrPos);
1362 : : }
1363 : :
4443 1364 [ - + ]: 13847439 : if (CurrPos != EndPos)
521 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"));
4443 heikki.linnakangas@i 1368 :CBC 13847439 : }
1369 : :
1370 : : /*
1371 : : * Acquire a WAL insertion lock, for inserting to WAL.
1372 : : */
1373 : : static void
4187 1374 : 13852753 : 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 [ + + ]: 13852753 : if (lockToTry == -1)
562 1392 : 7202 : lockToTry = MyProcNumber % NUM_XLOGINSERT_LOCKS;
4187 1393 : 13852753 : MyLockNo = lockToTry;
1394 : :
1395 : : /*
1396 : : * The insertingAt value is initially set to 0, as we don't know our
1397 : : * insert location yet.
1398 : : */
3690 andres@anarazel.de 1399 : 13852753 : immed = LWLockAcquire(&WALInsertLocks[MyLockNo].l.lock, LW_EXCLUSIVE);
4187 heikki.linnakangas@i 1400 [ + + ]: 13852753 : 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 : : */
3993 1410 : 18775 : lockToTry = (lockToTry + 1) % NUM_XLOGINSERT_LOCKS;
1411 : : }
4443 1412 : 13852753 : }
1413 : :
1414 : : /*
1415 : : * Acquire all WAL insertion locks, to prevent other backends from inserting
1416 : : * to WAL.
1417 : : */
1418 : : static void
4187 1419 : 4261 : 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 : : */
3993 1428 [ + + ]: 34088 : for (i = 0; i < NUM_XLOGINSERT_LOCKS - 1; i++)
1429 : : {
3690 andres@anarazel.de 1430 : 29827 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1431 : 29827 : LWLockUpdateVar(&WALInsertLocks[i].l.lock,
1432 : 29827 : &WALInsertLocks[i].l.insertingAt,
1433 : : PG_UINT64_MAX);
1434 : : }
1435 : : /* Variable value reset to 0 at release */
1436 : 4261 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1437 : :
4187 heikki.linnakangas@i 1438 : 4261 : holdingAllLocks = true;
4443 1439 : 4261 : }
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
4187 1448 : 13857014 : WALInsertLockRelease(void)
1449 : : {
1450 [ + + ]: 13857014 : if (holdingAllLocks)
1451 : : {
1452 : : int i;
1453 : :
3993 1454 [ + + ]: 38349 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
3690 andres@anarazel.de 1455 : 34088 : LWLockReleaseClearVar(&WALInsertLocks[i].l.lock,
1456 : 34088 : &WALInsertLocks[i].l.insertingAt,
1457 : : 0);
1458 : :
4187 heikki.linnakangas@i 1459 : 4261 : holdingAllLocks = false;
1460 : : }
1461 : : else
1462 : : {
3690 andres@anarazel.de 1463 : 13852753 : LWLockReleaseClearVar(&WALInsertLocks[MyLockNo].l.lock,
1464 : 13852753 : &WALInsertLocks[MyLockNo].l.insertingAt,
1465 : : 0);
1466 : : }
4443 heikki.linnakangas@i 1467 : 13857014 : }
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
4187 1474 : 2235189 : WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt)
1475 : : {
1476 [ + + ]: 2235189 : if (holdingAllLocks)
1477 : : {
1478 : : /*
1479 : : * We use the last lock to mark our actual position, see comments in
1480 : : * WALInsertLockAcquireExclusive.
1481 : : */
3993 1482 : 528124 : LWLockUpdateVar(&WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.lock,
2999 tgl@sss.pgh.pa.us 1483 : 528124 : &WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.insertingAt,
1484 : : insertingAt);
1485 : : }
1486 : : else
4187 heikki.linnakangas@i 1487 : 1707065 : LWLockUpdateVar(&WALInsertLocks[MyLockNo].l.lock,
1488 : 1707065 : &WALInsertLocks[MyLockNo].l.insertingAt,
1489 : : insertingAt);
4443 1490 : 2235189 : }
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 : 2111604 : WaitXLogInsertionsToFinish(XLogRecPtr upto)
1508 : : {
1509 : : uint64 bytepos;
1510 : : XLogRecPtr inserted;
1511 : : XLogRecPtr reservedUpto;
1512 : : XLogRecPtr finishedUpto;
4002 andres@anarazel.de 1513 : 2111604 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1514 : : int i;
1515 : :
4443 heikki.linnakangas@i 1516 [ - + ]: 2111604 : if (MyProc == NULL)
4443 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 : : */
517 alvherre@alvh.no-ip. 1523 :CBC 2111604 : inserted = pg_atomic_read_membarrier_u64(&XLogCtl->logInsertResult);
1524 [ + + ]: 2111604 : if (upto <= inserted)
1525 : 1665819 : return inserted;
1526 : :
1527 : : /* Read the current insert position */
4443 heikki.linnakangas@i 1528 [ + + ]: 445785 : SpinLockAcquire(&Insert->insertpos_lck);
1529 : 445785 : bytepos = Insert->CurrBytePos;
1530 : 445785 : SpinLockRelease(&Insert->insertpos_lck);
1531 : 445785 : 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 [ - + ]: 445785 : if (upto > reservedUpto)
1542 : : {
1737 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)));
4443 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 : : */
4443 heikki.linnakangas@i 1558 :CBC 445785 : finishedUpto = reservedUpto;
3993 1559 [ + + ]: 4012065 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1560 : : {
4141 bruce@momjian.us 1561 : 3566280 : 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 : : */
4187 heikki.linnakangas@i 1588 [ + + ]: 3684573 : if (LWLockWaitForVar(&WALInsertLocks[i].l.lock,
1589 : 3684573 : &WALInsertLocks[i].l.insertingAt,
1590 : : insertingat, &insertingat))
1591 : : {
1592 : : /* the lock was free, so no insertion in progress */
1593 : 2543781 : insertingat = InvalidXLogRecPtr;
1594 : 2543781 : break;
1595 : : }
1596 : :
1597 : : /*
1598 : : * This insertion is still in progress. Have to wait, unless the
1599 : : * inserter has proceeded past 'upto'.
1600 : : */
1601 [ + + ]: 1140792 : } while (insertingat < upto);
1602 : :
1603 [ + + + + ]: 3566280 : if (insertingat != InvalidXLogRecPtr && insertingat < finishedUpto)
1604 : 388762 : 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 : : */
517 alvherre@alvh.no-ip. 1612 : 445785 : finishedUpto = pg_atomic_monotonic_advance_u64(&XLogCtl->logInsertResult,
1613 : : finishedUpto);
1614 : :
4443 heikki.linnakangas@i 1615 : 445785 : 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 *
1401 rhaas@postgresql.org 1635 : 16145156 : 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 : : */
4443 heikki.linnakangas@i 1647 [ + + ]: 16145156 : if (ptr / XLOG_BLCKSZ == cachedPage)
1648 : : {
1649 [ - + ]: 13514383 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1650 [ - + ]: 13514383 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1651 : 13514383 : 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 : 2630773 : 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 : 2630773 : expectedEndPtr = ptr;
1675 : 2630773 : expectedEndPtr += XLOG_BLCKSZ - ptr % XLOG_BLCKSZ;
1676 : :
627 jdavis@postgresql.or 1677 : 2630773 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
4443 heikki.linnakangas@i 1678 [ + + ]: 2630773 : 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 : : */
3688 1696 [ + + ]: 2235189 : if (ptr % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
2909 andres@anarazel.de 1697 [ + - ]: 5302 : XLogSegmentOffset(ptr, wal_segment_size) > XLOG_BLCKSZ)
3688 heikki.linnakangas@i 1698 : 5302 : initializedUpto = ptr - SizeOfXLogShortPHD;
1699 [ + + ]: 2229887 : else if (ptr % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
2909 andres@anarazel.de 1700 [ + + ]: 717 : XLogSegmentOffset(ptr, wal_segment_size) < XLOG_BLCKSZ)
3688 heikki.linnakangas@i 1701 : 521 : initializedUpto = ptr - SizeOfXLogLongPHD;
1702 : : else
1703 : 2229366 : initializedUpto = ptr;
1704 : :
1705 : 2235189 : WALInsertLockUpdateInsertingAt(initializedUpto);
1706 : :
1401 rhaas@postgresql.org 1707 : 2235189 : AdvanceXLInsertBuffer(ptr, tli, false);
627 jdavis@postgresql.or 1708 : 2235189 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1709 : :
4443 heikki.linnakangas@i 1710 [ - + ]: 2235189 : if (expectedEndPtr != endptr)
61 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 : : */
4443 heikki.linnakangas@i 1720 :CBC 395584 : 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 : 2630773 : cachedPage = ptr / XLOG_BLCKSZ;
1728 : 2630773 : cachedPos = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1729 : :
1730 [ - + ]: 2630773 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1731 [ - + ]: 2630773 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1732 : :
1733 : 2630773 : 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
572 jdavis@postgresql.or 1751 : 99037 : WALReadFromBuffers(char *dstbuf, XLogRecPtr startptr, Size count,
1752 : : TimeLineID tli)
1753 : : {
1754 : 99037 : char *pdst = dstbuf;
1755 : 99037 : XLogRecPtr recptr = startptr;
1756 : : XLogRecPtr inserted;
568 1757 : 99037 : Size nbytes = count;
1758 : :
572 1759 [ + + + + ]: 99037 : if (RecoveryInProgress() || tli != GetWALInsertionTimeLine())
1760 : 845 : return 0;
1761 : :
1762 [ - + ]: 98192 : 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 : : */
517 alvherre@alvh.no-ip. 1768 : 98192 : inserted = pg_atomic_read_u64(&XLogCtl->logInsertResult);
1769 [ - + ]: 98192 : if (startptr + count > inserted)
517 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 : : */
572 jdavis@postgresql.or 1789 [ + + ]:CBC 121573 : while (nbytes > 0)
1790 : : {
1791 : 115011 : uint32 offset = recptr % XLOG_BLCKSZ;
1792 : 115011 : 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 : 115011 : 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 : 115011 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1810 [ + + ]: 115011 : if (expectedEndPtr != endptr)
1811 : 91630 : 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 : 23381 : page = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1819 : 23381 : psrc = page + offset;
1820 : 23381 : 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 : 23381 : pg_read_barrier();
1827 : :
1828 : : /* data copy */
1829 : 23381 : memcpy(pdst, psrc, npagebytes);
1830 : :
1831 : : /*
1832 : : * Ensure that the data copy and the second verification step are not
1833 : : * reordered.
1834 : : */
1835 : 23381 : 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 : 23381 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1842 [ - + ]: 23381 : if (expectedEndPtr != endptr)
572 jdavis@postgresql.or 1843 :UBC 0 : break;
1844 : :
572 jdavis@postgresql.or 1845 :CBC 23381 : pdst += npagebytes;
1846 : 23381 : recptr += npagebytes;
1847 : 23381 : nbytes -= npagebytes;
1848 : : }
1849 : :
1850 [ - + ]: 98192 : Assert(pdst - dstbuf <= count);
1851 : :
1852 : 98192 : 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
4443 heikki.linnakangas@i 1861 : 27697401 : XLogBytePosToRecPtr(uint64 bytepos)
1862 : : {
1863 : : uint64 fullsegs;
1864 : : uint64 fullpages;
1865 : : uint64 bytesleft;
1866 : : uint32 seg_offset;
1867 : : XLogRecPtr result;
1868 : :
1869 : 27697401 : fullsegs = bytepos / UsableBytesInSegment;
1870 : 27697401 : bytesleft = bytepos % UsableBytesInSegment;
1871 : :
1872 [ + + ]: 27697401 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1873 : : {
1874 : : /* fits on first page of segment */
1875 : 50569 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1876 : : }
1877 : : else
1878 : : {
1879 : : /* account for the first page on segment with long header */
1880 : 27646832 : seg_offset = XLOG_BLCKSZ;
1881 : 27646832 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1882 : :
1883 : 27646832 : fullpages = bytesleft / UsableBytesInPage;
1884 : 27646832 : bytesleft = bytesleft % UsableBytesInPage;
1885 : :
1886 : 27646832 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1887 : : }
1888 : :
2616 alvherre@alvh.no-ip. 1889 : 27697401 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1890 : :
4443 heikki.linnakangas@i 1891 : 27697401 : 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 : 14293931 : XLogBytePosToEndRecPtr(uint64 bytepos)
1902 : : {
1903 : : uint64 fullsegs;
1904 : : uint64 fullpages;
1905 : : uint64 bytesleft;
1906 : : uint32 seg_offset;
1907 : : XLogRecPtr result;
1908 : :
1909 : 14293931 : fullsegs = bytepos / UsableBytesInSegment;
1910 : 14293931 : bytesleft = bytepos % UsableBytesInSegment;
1911 : :
1912 [ + + ]: 14293931 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1913 : : {
1914 : : /* fits on first page of segment */
1915 [ + + ]: 80723 : if (bytesleft == 0)
1916 : 54202 : seg_offset = 0;
1917 : : else
1918 : 26521 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1919 : : }
1920 : : else
1921 : : {
1922 : : /* account for the first page on segment with long header */
1923 : 14213208 : seg_offset = XLOG_BLCKSZ;
1924 : 14213208 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1925 : :
1926 : 14213208 : fullpages = bytesleft / UsableBytesInPage;
1927 : 14213208 : bytesleft = bytesleft % UsableBytesInPage;
1928 : :
1929 [ + + ]: 14213208 : if (bytesleft == 0)
1930 : 13500 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft;
1931 : : else
1932 : 14199708 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1933 : : }
1934 : :
2616 alvherre@alvh.no-ip. 1935 : 14293931 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1936 : :
4443 heikki.linnakangas@i 1937 : 14293931 : return result;
1938 : : }
1939 : :
1940 : : /*
1941 : : * Convert an XLogRecPtr to a "usable byte position".
1942 : : */
1943 : : static uint64
1944 : 41544621 : XLogRecPtrToBytePos(XLogRecPtr ptr)
1945 : : {
1946 : : uint64 fullsegs;
1947 : : uint32 fullpages;
1948 : : uint32 offset;
1949 : : uint64 result;
1950 : :
2909 andres@anarazel.de 1951 : 41544621 : XLByteToSeg(ptr, fullsegs, wal_segment_size);
1952 : :
1953 : 41544621 : fullpages = (XLogSegmentOffset(ptr, wal_segment_size)) / XLOG_BLCKSZ;
4443 heikki.linnakangas@i 1954 : 41544621 : offset = ptr % XLOG_BLCKSZ;
1955 : :
1956 [ + + ]: 41544621 : if (fullpages == 0)
1957 : : {
1958 : 76561 : result = fullsegs * UsableBytesInSegment;
1959 [ + + ]: 76561 : if (offset > 0)
1960 : : {
1961 [ - + ]: 75242 : Assert(offset >= SizeOfXLogLongPHD);
1962 : 75242 : result += offset - SizeOfXLogLongPHD;
1963 : : }
1964 : : }
1965 : : else
1966 : : {
1967 : 41468060 : result = fullsegs * UsableBytesInSegment +
4141 bruce@momjian.us 1968 : 41468060 : (XLOG_BLCKSZ - SizeOfXLogLongPHD) + /* account for first page */
2999 tgl@sss.pgh.pa.us 1969 : 41468060 : (fullpages - 1) * UsableBytesInPage; /* full pages */
4443 heikki.linnakangas@i 1970 [ + + ]: 41468060 : if (offset > 0)
1971 : : {
1972 [ - + ]: 41454786 : Assert(offset >= SizeOfXLogShortPHD);
1973 : 41454786 : result += offset - SizeOfXLogShortPHD;
1974 : : }
1975 : : }
1976 : :
1977 : 41544621 : 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
1401 rhaas@postgresql.org 1988 : 2238351 : AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
1989 : : {
8943 tgl@sss.pgh.pa.us 1990 : 2238351 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1991 : : int nextidx;
1992 : : XLogRecPtr OldPageRqstPtr;
1993 : : XLogwrtRqst WriteRqst;
4443 heikki.linnakangas@i 1994 : 2238351 : XLogRecPtr NewPageEndPtr = InvalidXLogRecPtr;
1995 : : XLogRecPtr NewPageBeginPtr;
1996 : : XLogPageHeader NewPage;
1082 tgl@sss.pgh.pa.us 1997 : 2238351 : int npages pg_attribute_unused() = 0;
1998 : :
15 akorotkov@postgresql 1999 : 2238351 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2000 : :
2001 : : /*
2002 : : * Now that we have the lock, check if someone initialized the page
2003 : : * already.
2004 : : */
2005 [ + + + + ]: 6575582 : while (upto >= XLogCtl->InitializedUpTo || opportunistic)
2006 : : {
2007 : 4340393 : 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 : 4340393 : OldPageRqstPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[nextidx]);
2015 [ + + ]: 4340393 : 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 [ + + ]: 1996994 : if (opportunistic)
2022 : 3162 : break;
2023 : :
2024 : : /* Advance shared memory write request position */
4002 andres@anarazel.de 2025 [ + + ]: 1993832 : SpinLockAcquire(&XLogCtl->info_lck);
2026 [ + + ]: 1993832 : if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
2027 : 501253 : XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
2028 : 1993832 : 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 : : */
519 alvherre@alvh.no-ip. 2034 : 1993832 : RefreshXLogWriteResult(LogwrtResult);
4443 heikki.linnakangas@i 2035 [ + + ]: 1993832 : 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 : : */
15 akorotkov@postgresql 2043 : 1979610 : LWLockRelease(WALBufMappingLock);
2044 : :
4443 heikki.linnakangas@i 2045 : 1979610 : WaitXLogInsertionsToFinish(OldPageRqstPtr);
2046 : :
2047 : 1979610 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
2048 : :
521 alvherre@alvh.no-ip. 2049 : 1979610 : RefreshXLogWriteResult(LogwrtResult);
4443 heikki.linnakangas@i 2050 [ + + ]: 1979610 : if (LogwrtResult.Write >= OldPageRqstPtr)
2051 : : {
2052 : : /* OK, someone wrote it already */
2053 : 115476 : LWLockRelease(WALWriteLock);
2054 : : }
2055 : : else
2056 : : {
2057 : : /* Have to write it ourselves */
2058 : : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_START();
2059 : 1864134 : WriteRqst.Write = OldPageRqstPtr;
2060 : 1864134 : WriteRqst.Flush = 0;
1401 rhaas@postgresql.org 2061 : 1864134 : XLogWrite(WriteRqst, tli, false);
4443 heikki.linnakangas@i 2062 : 1864134 : LWLockRelease(WALWriteLock);
201 michael@paquier.xyz 2063 : 1864134 : 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 : : */
40 2070 : 1864134 : pgstat_report_fixed = true;
2071 : : }
2072 : : /* Re-acquire WALBufMappingLock and retry */
15 akorotkov@postgresql 2073 : 1979610 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2074 : 1979610 : 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 : 2357621 : NewPageBeginPtr = XLogCtl->InitializedUpTo;
4443 heikki.linnakangas@i 2083 : 2357621 : NewPageEndPtr = NewPageBeginPtr + XLOG_BLCKSZ;
2084 : :
15 akorotkov@postgresql 2085 [ - + ]: 2357621 : Assert(XLogRecPtrToBufIdx(NewPageBeginPtr) == nextidx);
2086 : :
4443 heikki.linnakangas@i 2087 : 2357621 : 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 : : */
627 jdavis@postgresql.or 2094 : 2357621 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], InvalidXLogRecPtr);
2095 : 2357621 : 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 : : */
206 peter@eisentraut.org 2101 [ + - + - : 2357621 : MemSet(NewPage, 0, XLOG_BLCKSZ);
+ - - + -
- ]
2102 : :
2103 : : /*
2104 : : * Fill the new page's header
2105 : : */
3759 bruce@momjian.us 2106 : 2357621 : NewPage->xlp_magic = XLOG_PAGE_MAGIC;
2107 : :
2108 : : /* NewPage->xlp_info = 0; */ /* done by memset */
1401 rhaas@postgresql.org 2109 : 2357621 : NewPage->xlp_tli = tli;
3759 bruce@momjian.us 2110 : 2357621 : 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 : : */
1053 alvherre@alvh.no-ip. 2127 [ + + ]: 2357621 : if (Insert->runningBackups == 0)
3759 bruce@momjian.us 2128 : 2233387 : NewPage->xlp_info |= XLP_BKP_REMOVABLE;
2129 : :
2130 : : /*
2131 : : * If first page of an XLOG segment file, make it a long header.
2132 : : */
2909 andres@anarazel.de 2133 [ + + ]: 2357621 : if ((XLogSegmentOffset(NewPage->xlp_pageaddr, wal_segment_size)) == 0)
2134 : : {
4443 heikki.linnakangas@i 2135 : 1752 : XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
2136 : :
2137 : 1752 : NewLongPage->xlp_sysid = ControlFile->system_identifier;
2909 andres@anarazel.de 2138 : 1752 : NewLongPage->xlp_seg_size = wal_segment_size;
4443 heikki.linnakangas@i 2139 : 1752 : NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
3759 bruce@momjian.us 2140 : 1752 : 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 : : */
4443 heikki.linnakangas@i 2148 : 2357621 : pg_write_barrier();
2149 : :
627 jdavis@postgresql.or 2150 : 2357621 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], NewPageEndPtr);
15 akorotkov@postgresql 2151 : 2357621 : XLogCtl->InitializedUpTo = NewPageEndPtr;
2152 : :
4443 heikki.linnakangas@i 2153 : 2357621 : npages++;
2154 : : }
15 akorotkov@postgresql 2155 : 2238351 : 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
9476 vadim4o@yahoo.com 2164 : 2238351 : }
2165 : :
2166 : : /*
2167 : : * Calculate CheckPointSegments based on max_wal_size_mb and
2168 : : * checkpoint_completion_target.
2169 : : */
2170 : : static void
3848 heikki.linnakangas@i 2171 : 7451 : 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 : : */
2909 andres@anarazel.de 2189 : 7451 : target = (double) ConvertToXSegs(max_wal_size_mb, wal_segment_size) /
2860 simon@2ndQuadrant.co 2190 : 7451 : (1.0 + CheckPointCompletionTarget);
2191 : :
2192 : : /* round down */
3848 heikki.linnakangas@i 2193 : 7451 : CheckPointSegments = (int) target;
2194 : :
2195 [ + + ]: 7451 : if (CheckPointSegments < 1)
2196 : 10 : CheckPointSegments = 1;
2197 : 7451 : }
2198 : :
2199 : : void
2200 : 5447 : assign_max_wal_size(int newval, void *extra)
2201 : : {
3077 simon@2ndQuadrant.co 2202 : 5447 : max_wal_size_mb = newval;
3848 heikki.linnakangas@i 2203 : 5447 : CalculateCheckpointSegments();
2204 : 5447 : }
2205 : :
2206 : : void
2207 : 1067 : assign_checkpoint_completion_target(double newval, void *extra)
2208 : : {
2209 : 1067 : CheckPointCompletionTarget = newval;
2210 : 1067 : CalculateCheckpointSegments();
2211 : 1067 : }
2212 : :
2213 : : bool
740 peter@eisentraut.org 2214 : 2055 : check_wal_segment_size(int *newval, void **extra, GucSource source)
2215 : : {
2216 [ + - + - : 2055 : if (!IsValidWalSegSize(*newval))
+ - - + ]
2217 : : {
740 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 : :
740 peter@eisentraut.org 2222 :CBC 2055 : 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
2089 michael@paquier.xyz 2230 : 1677 : 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 : 1677 : minSegNo = lastredoptr / wal_segment_size +
2909 andres@anarazel.de 2243 : 1677 : ConvertToXSegs(min_wal_size_mb, wal_segment_size) - 1;
2089 michael@paquier.xyz 2244 : 1677 : maxSegNo = lastredoptr / wal_segment_size +
2909 andres@anarazel.de 2245 : 1677 : 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 : : */
2860 simon@2ndQuadrant.co 2255 : 1677 : distance = (1.0 + CheckPointCompletionTarget) * CheckPointDistanceEstimate;
2256 : : /* add 10% for good measure. */
3848 heikki.linnakangas@i 2257 : 1677 : distance *= 1.10;
2258 : :
2089 michael@paquier.xyz 2259 : 1677 : recycleSegNo = (XLogSegNo) ceil(((double) lastredoptr + distance) /
2260 : : wal_segment_size);
2261 : :
3848 heikki.linnakangas@i 2262 [ + + ]: 1677 : if (recycleSegNo < minSegNo)
2263 : 1174 : recycleSegNo = minSegNo;
2264 [ + + ]: 1677 : if (recycleSegNo > maxSegNo)
2265 : 393 : recycleSegNo = maxSegNo;
2266 : :
2267 : 1677 : 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
4822 2280 : 4606 : XLogCheckpointNeeded(XLogSegNo new_segno)
2281 : : {
2282 : : XLogSegNo old_segno;
2283 : :
2909 andres@anarazel.de 2284 : 4606 : XLByteToSeg(RedoRecPtr, old_segno, wal_segment_size);
2285 : :
4822 heikki.linnakangas@i 2286 [ + + ]: 4606 : if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
6539 tgl@sss.pgh.pa.us 2287 : 2894 : return true;
2288 : 1712 : 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
1401 rhaas@postgresql.org 2304 : 1990750 : 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 */
7449 tgl@sss.pgh.pa.us 2315 [ - + ]: 1990750 : Assert(CritSectionCount > 0);
2316 : :
2317 : : /*
2318 : : * Update local LogwrtResult (caller probably did this already, but...)
2319 : : */
521 alvherre@alvh.no-ip. 2320 : 1990750 : 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 : : */
7320 tgl@sss.pgh.pa.us 2331 : 1990750 : npages = 0;
2332 : 1990750 : startidx = 0;
2333 : 1990750 : 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 : : */
4434 heikki.linnakangas@i 2340 : 1990750 : curridx = XLogRecPtrToBufIdx(LogwrtResult.Write);
2341 : :
4635 alvherre@alvh.no-ip. 2342 [ + + ]: 4294038 : 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 : : */
627 jdavis@postgresql.or 2349 : 2424893 : XLogRecPtr EndPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[curridx]);
2350 : :
4443 heikki.linnakangas@i 2351 [ - + ]: 2424893 : if (LogwrtResult.Write >= EndPtr)
61 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 */
4443 heikki.linnakangas@i 2357 :CBC 2424893 : LogwrtResult.Write = EndPtr;
4635 alvherre@alvh.no-ip. 2358 : 2424893 : ispartialpage = WriteRqst.Write < LogwrtResult.Write;
2359 : :
2909 andres@anarazel.de 2360 [ + + ]: 2424893 : 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 : : */
7320 tgl@sss.pgh.pa.us 2367 [ - + ]: 12708 : Assert(npages == 0);
8943 2368 [ + + ]: 12708 : if (openLogFile >= 0)
7023 bruce@momjian.us 2369 : 6017 : XLogFileClose();
2909 andres@anarazel.de 2370 : 12708 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2371 : : wal_segment_size);
1401 rhaas@postgresql.org 2372 : 12708 : openLogTLI = tli;
2373 : :
2374 : : /* create/use new log file */
2375 : 12708 : openLogFile = XLogFileInit(openLogSegNo, tli);
2021 tgl@sss.pgh.pa.us 2376 : 12708 : ReserveExternalFD();
2377 : : }
2378 : :
2379 : : /* Make sure we have the current logfile open */
8943 2380 [ - + ]: 2424893 : if (openLogFile < 0)
2381 : : {
2909 andres@anarazel.de 2382 :UBC 0 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2383 : : wal_segment_size);
1401 rhaas@postgresql.org 2384 : 0 : openLogTLI = tli;
2385 : 0 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2021 tgl@sss.pgh.pa.us 2386 : 0 : ReserveExternalFD();
2387 : : }
2388 : :
2389 : : /* Add current page to the set of pending pages-to-dump */
7320 tgl@sss.pgh.pa.us 2390 [ + + ]:CBC 2424893 : if (npages == 0)
2391 : : {
2392 : : /* first of group */
2393 : 2003075 : startidx = curridx;
2909 andres@anarazel.de 2394 : 2003075 : startoffset = XLogSegmentOffset(LogwrtResult.Write - XLOG_BLCKSZ,
2395 : : wal_segment_size);
2396 : : }
7320 tgl@sss.pgh.pa.us 2397 : 2424893 : 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 : : */
4635 alvherre@alvh.no-ip. 2405 : 2424893 : last_iteration = WriteRqst.Write <= LogwrtResult.Write;
2406 : :
7320 tgl@sss.pgh.pa.us 2407 [ + + ]: 4730598 : finishing_seg = !ispartialpage &&
2909 andres@anarazel.de 2408 [ + + ]: 2305705 : (startoffset + npages * XLOG_BLCKSZ) >= wal_segment_size;
2409 : :
6971 tgl@sss.pgh.pa.us 2410 [ + + ]: 2424893 : if (last_iteration ||
7320 2411 [ + + - + ]: 436394 : 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) */
7096 2421 : 2003075 : from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
2422 : 2003075 : nbytes = npages * (Size) XLOG_BLCKSZ;
4450 heikki.linnakangas@i 2423 : 2003075 : nleft = nbytes;
2424 : : do
2425 : : {
2426 : 2003075 : errno = 0;
2427 : :
2428 : : /*
2429 : : * Measure I/O timing to write WAL data, for pg_stat_io.
2430 : : */
192 michael@paquier.xyz 2431 : 2003075 : start = pgstat_prepare_io_time(track_wal_io_timing);
2432 : :
3094 rhaas@postgresql.org 2433 : 2003075 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
1073 tmunro@postgresql.or 2434 : 2003075 : written = pg_pwrite(openLogFile, from, nleft, startoffset);
3094 rhaas@postgresql.org 2435 : 2003075 : pgstat_report_wait_end();
2436 : :
214 michael@paquier.xyz 2437 : 2003075 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL,
2438 : : IOOP_WRITE, start, 1, written);
2439 : :
4450 heikki.linnakangas@i 2440 [ - + ]: 2003075 : if (written <= 0)
2441 : : {
2442 : : char xlogfname[MAXFNAMELEN];
2443 : : int save_errno;
2444 : :
4450 heikki.linnakangas@i 2445 [ # # ]:UBC 0 : if (errno == EINTR)
2446 : 0 : continue;
2447 : :
2104 michael@paquier.xyz 2448 : 0 : save_errno = errno;
1401 rhaas@postgresql.org 2449 : 0 : XLogFileName(xlogfname, tli, openLogSegNo,
2450 : : wal_segment_size);
2104 michael@paquier.xyz 2451 : 0 : errno = save_errno;
4450 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 : : }
4450 heikki.linnakangas@i 2457 :CBC 2003075 : nleft -= written;
2458 : 2003075 : from += written;
2495 tmunro@postgresql.or 2459 : 2003075 : startoffset += written;
4450 heikki.linnakangas@i 2460 [ - + ]: 2003075 : } while (nleft > 0);
2461 : :
7320 tgl@sss.pgh.pa.us 2462 : 2003075 : 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 : : */
4443 heikki.linnakangas@i 2477 [ + + ]: 2003075 : if (finishing_seg)
2478 : : {
1401 rhaas@postgresql.org 2479 : 1843 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2480 : :
2481 : : /* signal that we need to wakeup walsenders later */
4814 2482 : 1843 : WalSndWakeupRequest();
2483 : :
2999 tgl@sss.pgh.pa.us 2484 : 1843 : LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
2485 : :
7320 2486 [ + + - + : 1843 : if (XLogArchivingActive())
+ + ]
1401 rhaas@postgresql.org 2487 : 401 : XLogArchiveNotifySeg(openLogSegNo, tli);
2488 : :
4434 heikki.linnakangas@i 2489 : 1843 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
3180 andres@anarazel.de 2490 : 1843 : 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 : : */
4822 heikki.linnakangas@i 2499 [ + + + + ]: 1843 : if (IsUnderPostmaster && XLogCheckpointNeeded(openLogSegNo))
2500 : : {
6539 tgl@sss.pgh.pa.us 2501 : 234 : (void) GetRedoRecPtr();
4822 heikki.linnakangas@i 2502 [ + + ]: 234 : if (XLogCheckpointNeeded(openLogSegNo))
6643 tgl@sss.pgh.pa.us 2503 : 190 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
2504 : : }
2505 : : }
2506 : : }
2507 : :
8943 2508 [ + + ]: 2424893 : if (ispartialpage)
2509 : : {
2510 : : /* Only asked to write a partial page */
2511 : 119188 : LogwrtResult.Write = WriteRqst.Write;
2512 : 119188 : break;
2513 : : }
7320 2514 [ + + ]: 2305705 : curridx = NextBufIdx(curridx);
2515 : :
2516 : : /* If flexible, break out of loop as soon as we wrote something */
2517 [ + + + + ]: 2305705 : if (flexible && npages == 0)
2518 : 2417 : break;
2519 : : }
2520 : :
2521 [ - + ]: 1990750 : Assert(npages == 0);
2522 : :
2523 : : /*
2524 : : * If asked to flush, do so
2525 : : */
4635 alvherre@alvh.no-ip. 2526 [ + + ]: 1990750 : if (LogwrtResult.Flush < WriteRqst.Flush &&
2527 [ + + ]: 125959 : 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 : : */
694 nathan@postgresql.or 2534 [ + - ]: 125894 : if (wal_sync_method != WAL_SYNC_METHOD_OPEN &&
2535 [ + - ]: 125894 : wal_sync_method != WAL_SYNC_METHOD_OPEN_DSYNC)
2536 : : {
8940 tgl@sss.pgh.pa.us 2537 [ + + ]: 125894 : if (openLogFile >= 0 &&
2909 andres@anarazel.de 2538 [ + + ]: 125873 : !XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2539 : : wal_segment_size))
7023 bruce@momjian.us 2540 : 120 : XLogFileClose();
8940 tgl@sss.pgh.pa.us 2541 [ + + ]: 125894 : if (openLogFile < 0)
2542 : : {
2909 andres@anarazel.de 2543 : 141 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2544 : : wal_segment_size);
1401 rhaas@postgresql.org 2545 : 141 : openLogTLI = tli;
2546 : 141 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2021 tgl@sss.pgh.pa.us 2547 : 141 : ReserveExternalFD();
2548 : : }
2549 : :
1401 rhaas@postgresql.org 2550 : 125894 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2551 : : }
2552 : :
2553 : : /* signal that we need to wakeup walsenders later */
4814 2554 : 125894 : WalSndWakeupRequest();
2555 : :
8943 tgl@sss.pgh.pa.us 2556 : 125894 : 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 : : */
519 alvherre@alvh.no-ip. 2566 [ + + ]: 1990750 : SpinLockAcquire(&XLogCtl->info_lck);
2567 [ + + ]: 1990750 : if (XLogCtl->LogwrtRqst.Write < LogwrtResult.Write)
2568 : 111719 : XLogCtl->LogwrtRqst.Write = LogwrtResult.Write;
2569 [ + + ]: 1990750 : if (XLogCtl->LogwrtRqst.Flush < LogwrtResult.Flush)
2570 : 127342 : XLogCtl->LogwrtRqst.Flush = LogwrtResult.Flush;
2571 : 1990750 : 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 : 1990750 : pg_atomic_write_u64(&XLogCtl->logWriteResult, LogwrtResult.Write);
2579 : 1990750 : pg_write_barrier();
2580 : 1990750 : 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 : 1990750 : Flush = pg_atomic_read_u64(&XLogCtl->logFlushResult);
2589 : 1990750 : pg_read_barrier();
2590 : 1990750 : Write = pg_atomic_read_u64(&XLogCtl->logWriteResult);
517 2591 : 1990750 : pg_read_barrier();
2592 : 1990750 : Insert = pg_atomic_read_u64(&XLogCtl->logInsertResult);
2593 : :
2594 : : /* WAL written to disk is always ahead of WAL flushed */
519 2595 [ - + ]: 1990750 : Assert(Write >= Flush);
2596 : :
2597 : : /* WAL inserted to buffers is always ahead of WAL written */
517 2598 [ - + ]: 1990750 : Assert(Insert >= Write);
2599 : : }
2600 : : #endif
8943 tgl@sss.pgh.pa.us 2601 : 1990750 : }
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
5518 simon@2ndQuadrant.co 2609 : 29263 : XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN)
2610 : : {
5046 2611 : 29263 : XLogRecPtr WriteRqstPtr = asyncXactLSN;
2612 : : bool sleeping;
649 heikki.linnakangas@i 2613 : 29263 : bool wakeup = false;
2614 : : XLogRecPtr prevAsyncXactLSN;
2615 : :
4002 andres@anarazel.de 2616 [ + + ]: 29263 : SpinLockAcquire(&XLogCtl->info_lck);
2617 : 29263 : sleeping = XLogCtl->WalWriterSleeping;
649 heikki.linnakangas@i 2618 : 29263 : prevAsyncXactLSN = XLogCtl->asyncXactLSN;
4002 andres@anarazel.de 2619 [ + + ]: 29263 : if (XLogCtl->asyncXactLSN < asyncXactLSN)
2620 : 28799 : XLogCtl->asyncXactLSN = asyncXactLSN;
2621 : 29263 : 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 : : */
649 heikki.linnakangas@i 2627 [ + + ]: 29263 : if (asyncXactLSN <= prevAsyncXactLSN)
2628 : 464 : 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 [ + + ]: 28799 : if (sleeping)
2637 : 8 : wakeup = true;
2638 : : else
2639 : : {
2640 : : int flushblocks;
2641 : :
519 alvherre@alvh.no-ip. 2642 : 28791 : RefreshXLogWriteResult(LogwrtResult);
2643 : :
649 heikki.linnakangas@i 2644 : 28791 : flushblocks =
2645 : 28791 : WriteRqstPtr / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
2646 : :
2647 [ + - + + ]: 28791 : if (WalWriterFlushAfter == 0 || flushblocks >= WalWriterFlushAfter)
2648 : 3701 : wakeup = true;
2649 : : }
2650 : :
309 2651 [ + + ]: 28799 : if (wakeup)
2652 : : {
2653 : 3709 : volatile PROC_HDR *procglobal = ProcGlobal;
2654 : 3709 : ProcNumber walwriterProc = procglobal->walwriterProc;
2655 : :
2656 [ + + ]: 3709 : if (walwriterProc != INVALID_PROC_NUMBER)
2657 : 181 : 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
4236 rhaas@postgresql.org 2666 : 64526 : XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn)
2667 : : {
4002 andres@anarazel.de 2668 [ + + ]: 64526 : SpinLockAcquire(&XLogCtl->info_lck);
2669 : 64526 : XLogCtl->replicationSlotMinLSN = lsn;
2670 : 64526 : SpinLockRelease(&XLogCtl->info_lck);
4236 rhaas@postgresql.org 2671 : 64526 : }
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 : 2042 : XLogGetReplicationSlotMinimumLSN(void)
2680 : : {
2681 : : XLogRecPtr retval;
2682 : :
4002 andres@anarazel.de 2683 [ - + ]: 2042 : SpinLockAcquire(&XLogCtl->info_lck);
2684 : 2042 : retval = XLogCtl->replicationSlotMinLSN;
2685 : 2042 : SpinLockRelease(&XLogCtl->info_lck);
2686 : :
4236 rhaas@postgresql.org 2687 : 2042 : 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
6044 heikki.linnakangas@i 2700 : 104889 : UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
2701 : : {
2702 : : /* Quick check using our local copy of the variable */
1298 2703 [ + + + + : 104889 : if (!updateMinRecoveryPoint || (!force && lsn <= LocalMinRecoveryPoint))
+ + ]
6044 2704 : 98199 : 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 : : */
1298 2717 [ + + + + ]: 6690 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint) && InRecovery)
2718 : : {
2620 michael@paquier.xyz 2719 : 31 : updateMinRecoveryPoint = false;
2720 : 31 : return;
2721 : : }
2722 : :
6044 heikki.linnakangas@i 2723 : 6659 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2724 : :
2725 : : /* update local copy */
1298 2726 : 6659 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
2727 : 6659 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
2728 : :
2729 [ + + ]: 6659 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint))
2563 michael@paquier.xyz 2730 : 1 : updateMinRecoveryPoint = false;
1298 heikki.linnakangas@i 2731 [ + + + + ]: 6658 : 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 : 5307 : newMinRecoveryPoint = GetCurrentReplayRecPtr(&newMinRecoveryPointTLI);
4635 alvherre@alvh.no-ip. 2750 [ + + - + ]: 5307 : if (!force && newMinRecoveryPoint < lsn)
5916 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 */
4635 alvherre@alvh.no-ip. 2756 [ + + ]:CBC 5307 : if (ControlFile->minRecoveryPoint < newMinRecoveryPoint)
2757 : : {
6044 heikki.linnakangas@i 2758 : 4961 : ControlFile->minRecoveryPoint = newMinRecoveryPoint;
4659 2759 : 4961 : ControlFile->minRecoveryPointTLI = newMinRecoveryPointTLI;
6044 2760 : 4961 : UpdateControlFile();
1298 2761 : 4961 : LocalMinRecoveryPoint = newMinRecoveryPoint;
2762 : 4961 : LocalMinRecoveryPointTLI = newMinRecoveryPointTLI;
2763 : :
6044 2764 [ + + ]: 4961 : 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 : 6659 : 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
8943 tgl@sss.pgh.pa.us 2780 : 677629 : XLogFlush(XLogRecPtr record)
2781 : : {
2782 : : XLogRecPtr WriteRqstPtr;
2783 : : XLogwrtRqst WriteRqst;
1396 rhaas@postgresql.org 2784 : 677629 : 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 : : */
5916 tgl@sss.pgh.pa.us 2793 [ + + ]: 677629 : if (!XLogInsertAllowed())
2794 : : {
6044 heikki.linnakangas@i 2795 : 104452 : UpdateMinRecoveryPoint(record, false);
8943 tgl@sss.pgh.pa.us 2796 : 540924 : return;
2797 : : }
2798 : :
2799 : : /* Quick exit if already known flushed */
4635 alvherre@alvh.no-ip. 2800 [ + + ]: 573177 : if (record <= LogwrtResult.Flush)
8943 tgl@sss.pgh.pa.us 2801 : 436472 : 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 : 136705 : 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 : 136705 : 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 (;;)
8653 2829 : 2247 : {
2830 : : XLogRecPtr insertpos;
2831 : :
2832 : : /* done already? */
519 alvherre@alvh.no-ip. 2833 : 138952 : RefreshXLogWriteResult(LogwrtResult);
4635 2834 [ + + ]: 138952 : if (record <= LogwrtResult.Flush)
4968 heikki.linnakangas@i 2835 : 10120 : 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 : : */
519 alvherre@alvh.no-ip. 2841 [ + + ]: 128832 : SpinLockAcquire(&XLogCtl->info_lck);
2842 [ + + ]: 128832 : if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2843 : 8575 : WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2844 : 128832 : SpinLockRelease(&XLogCtl->info_lck);
4443 heikki.linnakangas@i 2845 : 128832 : 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 : : */
4959 2854 [ + + ]: 128832 : 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 : : */
4968 2861 : 2247 : continue;
2862 : : }
2863 : :
2864 : : /* Got the lock; recheck whether request is satisfied */
521 alvherre@alvh.no-ip. 2865 : 126585 : RefreshXLogWriteResult(LogwrtResult);
4635 2866 [ + + ]: 126585 : if (record <= LogwrtResult.Flush)
2867 : : {
4814 rhaas@postgresql.org 2868 : 3072 : LWLockRelease(WALWriteLock);
2869 : 3072 : 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 [ - + - - : 123513 : if (CommitDelay > 0 && enableFsync &&
- - ]
4814 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 : : */
4443 heikki.linnakangas@i 2896 : 0 : insertpos = WaitXLogInsertionsToFinish(insertpos);
2897 : : }
2898 : :
2899 : : /* try to write/flush later additions to XLOG as well */
4443 heikki.linnakangas@i 2900 :CBC 123513 : WriteRqst.Write = insertpos;
2901 : 123513 : WriteRqst.Flush = insertpos;
2902 : :
1401 rhaas@postgresql.org 2903 : 123513 : XLogWrite(WriteRqst, insertTLI, false);
2904 : :
8743 tgl@sss.pgh.pa.us 2905 : 123513 : LWLockRelease(WALWriteLock);
2906 : : /* done */
4968 heikki.linnakangas@i 2907 : 123513 : break;
2908 : : }
2909 : :
8943 tgl@sss.pgh.pa.us 2910 [ - + ]: 136705 : END_CRIT_SECTION();
2911 : :
2912 : : /* wake up walsenders now that we've released heavily contended locks */
882 andres@anarazel.de 2913 : 136705 : 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 : : */
4635 alvherre@alvh.no-ip. 2936 [ - + ]: 136705 : if (LogwrtResult.Flush < record)
5916 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 : : /*
2944 : : * Write & flush xlog, but without specifying exactly where to.
2945 : : *
2946 : : * We normally write only completed blocks; but if there is nothing to do on
2947 : : * that basis, we check for unwritten async commits in the current incomplete
2948 : : * block, and write through the latest one of those. Thus, if async commits
2949 : : * are not being used, we will write complete blocks only.
2950 : : *
2951 : : * If, based on the above, there's anything to write we do so immediately. But
2952 : : * to avoid calling fsync, fdatasync et. al. at a rate that'd impact
2953 : : * concurrent IO, we only flush WAL every wal_writer_delay ms, or if there's
2954 : : * more than wal_writer_flush_after unflushed blocks.
2955 : : *
2956 : : * We can guarantee that async commits reach disk after at most three
2957 : : * wal_writer_delay cycles. (When flushing complete blocks, we allow XLogWrite
2958 : : * to write "flexibly", meaning it can stop at the end of the buffer ring;
2959 : : * this makes a difference only with very high load or long wal_writer_delay,
2960 : : * but imposes one extra cycle for the worst case for async commits.)
2961 : : *
2962 : : * This routine is invoked periodically by the background walwriter process.
2963 : : *
2964 : : * Returns true if there was any work to do, even if we skipped flushing due
2965 : : * to wal_writer_delay/wal_writer_flush_after.
2966 : : */
2967 : : bool
6619 tgl@sss.pgh.pa.us 2968 :CBC 13060 : XLogBackgroundFlush(void)
2969 : : {
2970 : : XLogwrtRqst WriteRqst;
2971 : 13060 : bool flexible = true;
2972 : : static TimestampTz lastflush;
2973 : : TimestampTz now;
2974 : : int flushblocks;
2975 : : TimeLineID insertTLI;
2976 : :
2977 : : /* XLOG doesn't need flushing during recovery */
6044 heikki.linnakangas@i 2978 [ - + ]: 13060 : if (RecoveryInProgress())
4869 tgl@sss.pgh.pa.us 2979 :UBC 0 : return false;
2980 : :
2981 : : /*
2982 : : * Since we're not in recovery, InsertTimeLineID is set and can't change,
2983 : : * so we can read it without a lock.
2984 : : */
1396 rhaas@postgresql.org 2985 :CBC 13060 : insertTLI = XLogCtl->InsertTimeLineID;
2986 : :
2987 : : /* read updated LogwrtRqst */
4002 andres@anarazel.de 2988 [ - + ]: 13060 : SpinLockAcquire(&XLogCtl->info_lck);
3491 2989 : 13060 : WriteRqst = XLogCtl->LogwrtRqst;
4002 2990 : 13060 : SpinLockRelease(&XLogCtl->info_lck);
2991 : :
2992 : : /* back off to last completed page boundary */
3491 2993 : 13060 : WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
2994 : :
2995 : : /* if we have already flushed that far, consider async commit records */
519 alvherre@alvh.no-ip. 2996 : 13060 : RefreshXLogWriteResult(LogwrtResult);
3491 andres@anarazel.de 2997 [ + + ]: 13060 : if (WriteRqst.Write <= LogwrtResult.Flush)
2998 : : {
4002 2999 [ + + ]: 10522 : SpinLockAcquire(&XLogCtl->info_lck);
3491 3000 : 10522 : WriteRqst.Write = XLogCtl->asyncXactLSN;
4002 3001 : 10522 : SpinLockRelease(&XLogCtl->info_lck);
6619 tgl@sss.pgh.pa.us 3002 : 10522 : flexible = false; /* ensure it all gets written */
3003 : : }
3004 : :
3005 : : /*
3006 : : * If already known flushed, we're done. Just need to check if we are
3007 : : * holding an open file handle to a logfile that's no longer in use,
3008 : : * preventing the file from being deleted.
3009 : : */
3491 andres@anarazel.de 3010 [ + + ]: 13060 : if (WriteRqst.Write <= LogwrtResult.Flush)
3011 : : {
5541 bruce@momjian.us 3012 [ + + ]: 9898 : if (openLogFile >= 0)
3013 : : {
2909 andres@anarazel.de 3014 [ + + ]: 4650 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
3015 : : wal_segment_size))
3016 : : {
5568 magnus@hagander.net 3017 : 201 : XLogFileClose();
3018 : : }
3019 : : }
4869 tgl@sss.pgh.pa.us 3020 : 9898 : return false;
3021 : : }
3022 : :
3023 : : /*
3024 : : * Determine how far to flush WAL, based on the wal_writer_delay and
3025 : : * wal_writer_flush_after GUCs.
3026 : : *
3027 : : * Note that XLogSetAsyncXactLSN() performs similar calculation based on
3028 : : * wal_writer_flush_after, to decide when to wake us up. Make sure the
3029 : : * logic is the same in both places if you change this.
3030 : : */
3491 andres@anarazel.de 3031 : 3162 : now = GetCurrentTimestamp();
649 heikki.linnakangas@i 3032 : 3162 : flushblocks =
3491 andres@anarazel.de 3033 : 3162 : WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
3034 : :
3035 [ + - + + ]: 3162 : if (WalWriterFlushAfter == 0 || lastflush == 0)
3036 : : {
3037 : : /* first call, or block based limits disabled */
3038 : 260 : WriteRqst.Flush = WriteRqst.Write;
3039 : 260 : lastflush = now;
3040 : : }
3041 [ + + ]: 2902 : else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
3042 : : {
3043 : : /*
3044 : : * Flush the writes at least every WalWriterDelay ms. This is
3045 : : * important to bound the amount of time it takes for an asynchronous
3046 : : * commit to hit disk.
3047 : : */
3048 : 2755 : WriteRqst.Flush = WriteRqst.Write;
3049 : 2755 : lastflush = now;
3050 : : }
649 heikki.linnakangas@i 3051 [ + + ]: 147 : else if (flushblocks >= WalWriterFlushAfter)
3052 : : {
3053 : : /* exceeded wal_writer_flush_after blocks, flush */
3491 andres@anarazel.de 3054 : 135 : WriteRqst.Flush = WriteRqst.Write;
3055 : 135 : lastflush = now;
3056 : : }
3057 : : else
3058 : : {
3059 : : /* no flushing, this time round */
3060 : 12 : WriteRqst.Flush = 0;
3061 : : }
3062 : :
3063 : : #ifdef WAL_DEBUG
3064 : : if (XLOG_DEBUG)
3065 : : elog(LOG, "xlog bg flush request write %X/%08X; flush: %X/%08X, current is write %X/%08X; flush %X/%08X",
3066 : : LSN_FORMAT_ARGS(WriteRqst.Write),
3067 : : LSN_FORMAT_ARGS(WriteRqst.Flush),
3068 : : LSN_FORMAT_ARGS(LogwrtResult.Write),
3069 : : LSN_FORMAT_ARGS(LogwrtResult.Flush));
3070 : : #endif
3071 : :
6619 tgl@sss.pgh.pa.us 3072 : 3162 : START_CRIT_SECTION();
3073 : :
3074 : : /* now wait for any in-progress insertions to finish and get write lock */
3491 andres@anarazel.de 3075 : 3162 : WaitXLogInsertionsToFinish(WriteRqst.Write);
6619 tgl@sss.pgh.pa.us 3076 : 3162 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
521 alvherre@alvh.no-ip. 3077 : 3162 : RefreshXLogWriteResult(LogwrtResult);
3491 andres@anarazel.de 3078 [ + + ]: 3162 : if (WriteRqst.Write > LogwrtResult.Write ||
3079 [ + + ]: 128 : WriteRqst.Flush > LogwrtResult.Flush)
3080 : : {
1401 rhaas@postgresql.org 3081 : 3103 : XLogWrite(WriteRqst, insertTLI, flexible);
3082 : : }
6619 tgl@sss.pgh.pa.us 3083 : 3162 : LWLockRelease(WALWriteLock);
3084 : :
3085 [ - + ]: 3162 : END_CRIT_SECTION();
3086 : :
3087 : : /* wake up walsenders now that we've released heavily contended locks */
882 andres@anarazel.de 3088 : 3162 : WalSndWakeupProcessRequests(true, !RecoveryInProgress());
3089 : :
3090 : : /*
3091 : : * Great, done. To take some work off the critical path, try to initialize
3092 : : * as many of the no-longer-needed WAL buffers for future use as we can.
3093 : : */
1401 rhaas@postgresql.org 3094 : 3162 : AdvanceXLInsertBuffer(InvalidXLogRecPtr, insertTLI, true);
3095 : :
3096 : : /*
3097 : : * If we determined that we need to write data, but somebody else
3098 : : * wrote/flushed already, it should be considered as being active, to
3099 : : * avoid hibernating too early.
3100 : : */
3491 andres@anarazel.de 3101 : 3162 : return true;
3102 : : }
3103 : :
3104 : : /*
3105 : : * Test whether XLOG data has been flushed up to (at least) the given position.
3106 : : *
3107 : : * Returns true if a flush is still needed. (It may be that someone else
3108 : : * is already in process of flushing that far, however.)
3109 : : */
3110 : : bool
6674 tgl@sss.pgh.pa.us 3111 : 8599683 : XLogNeedsFlush(XLogRecPtr record)
3112 : : {
3113 : : /*
3114 : : * During recovery, we don't flush WAL but update minRecoveryPoint
3115 : : * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3116 : : * would need to be updated.
3117 : : */
6044 heikki.linnakangas@i 3118 [ + + ]: 8599683 : if (RecoveryInProgress())
3119 : : {
3120 : : /*
3121 : : * An invalid minRecoveryPoint means that we need to recover all the
3122 : : * WAL, i.e., we're doing crash recovery. We never modify the control
3123 : : * file's value in that case, so we can short-circuit future checks
3124 : : * here too. This triggers a quick exit path for the startup process,
3125 : : * which cannot update its local copy of minRecoveryPoint as long as
3126 : : * it has not replayed all WAL available when doing crash recovery.
3127 : : */
1298 3128 [ + + - + ]: 614243 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint) && InRecovery)
2620 michael@paquier.xyz 3129 :UBC 0 : updateMinRecoveryPoint = false;
3130 : :
3131 : : /* Quick exit if already known to be updated or cannot be updated */
1298 heikki.linnakangas@i 3132 [ + + - + ]:CBC 614243 : if (record <= LocalMinRecoveryPoint || !updateMinRecoveryPoint)
5740 simon@2ndQuadrant.co 3133 : 605537 : return false;
3134 : :
3135 : : /*
3136 : : * Update local copy of minRecoveryPoint. But if the lock is busy,
3137 : : * just return a conservative guess.
3138 : : */
3139 [ - + ]: 8706 : if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
5740 simon@2ndQuadrant.co 3140 :UBC 0 : return true;
1298 heikki.linnakangas@i 3141 :CBC 8706 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
3142 : 8706 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
5740 simon@2ndQuadrant.co 3143 : 8706 : LWLockRelease(ControlFileLock);
3144 : :
3145 : : /*
3146 : : * Check minRecoveryPoint for any other process than the startup
3147 : : * process doing crash recovery, which should not update the control
3148 : : * file value if crash recovery is still running.
3149 : : */
1298 heikki.linnakangas@i 3150 [ - + ]: 8706 : if (XLogRecPtrIsInvalid(LocalMinRecoveryPoint))
2563 michael@paquier.xyz 3151 :UBC 0 : updateMinRecoveryPoint = false;
3152 : :
3153 : : /* check again */
1298 heikki.linnakangas@i 3154 [ + + - + ]:CBC 8706 : if (record <= LocalMinRecoveryPoint || !updateMinRecoveryPoint)
2563 michael@paquier.xyz 3155 : 71 : return false;
3156 : : else
3157 : 8635 : return true;
3158 : : }
3159 : :
3160 : : /* Quick exit if already known flushed */
4635 alvherre@alvh.no-ip. 3161 [ + + ]: 7985440 : if (record <= LogwrtResult.Flush)
6674 tgl@sss.pgh.pa.us 3162 : 7834091 : return false;
3163 : :
3164 : : /* read LogwrtResult and update local state */
521 alvherre@alvh.no-ip. 3165 : 151349 : RefreshXLogWriteResult(LogwrtResult);
3166 : :
3167 : : /* check again */
4635 3168 [ + + ]: 151349 : if (record <= LogwrtResult.Flush)
6674 tgl@sss.pgh.pa.us 3169 : 3678 : return false;
3170 : :
3171 : 147671 : return true;
3172 : : }
3173 : :
3174 : : /*
3175 : : * Try to make a given XLOG file segment exist.
3176 : : *
3177 : : * logsegno: identify segment.
3178 : : *
3179 : : * *added: on return, true if this call raised the number of extant segments.
3180 : : *
3181 : : * path: on return, this char[MAXPGPATH] has the path to the logsegno file.
3182 : : *
3183 : : * Returns -1 or FD of opened file. A -1 here is not an error; a caller
3184 : : * wanting an open segment should attempt to open "path", which usually will
3185 : : * succeed. (This is weird, but it's efficient for the callers.)
3186 : : */
3187 : : static int
1401 rhaas@postgresql.org 3188 : 13808 : XLogFileInitInternal(XLogSegNo logsegno, TimeLineID logtli,
3189 : : bool *added, char *path)
3190 : : {
3191 : : char tmppath[MAXPGPATH];
3192 : : XLogSegNo installed_segno;
3193 : : XLogSegNo max_segno;
3194 : : int fd;
3195 : : int save_errno;
882 tmunro@postgresql.or 3196 : 13808 : int open_flags = O_RDWR | O_CREAT | O_EXCL | PG_BINARY;
3197 : : instr_time io_start;
3198 : :
1401 rhaas@postgresql.org 3199 [ - + ]: 13808 : Assert(logtli != 0);
3200 : :
3201 : 13808 : XLogFilePath(path, logtli, logsegno, wal_segment_size);
3202 : :
3203 : : /*
3204 : : * Try to use existent file (checkpoint maker may have created it already)
3205 : : */
1531 noah@leadboat.com 3206 : 13808 : *added = false;
918 tmunro@postgresql.or 3207 : 13808 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
694 nathan@postgresql.or 3208 : 13808 : get_sync_bit(wal_sync_method));
1531 noah@leadboat.com 3209 [ + + ]: 13808 : if (fd < 0)
3210 : : {
3211 [ - + ]: 1357 : if (errno != ENOENT)
1531 noah@leadboat.com 3212 [ # # ]:UBC 0 : ereport(ERROR,
3213 : : (errcode_for_file_access(),
3214 : : errmsg("could not open file \"%s\": %m", path)));
3215 : : }
3216 : : else
1531 noah@leadboat.com 3217 :CBC 12451 : return fd;
3218 : :
3219 : : /*
3220 : : * Initialize an empty (all zeroes) segment. NOTE: it is possible that
3221 : : * another process is doing the same thing. If so, we will end up
3222 : : * pre-creating an extra log segment. That seems OK, and better than
3223 : : * holding the lock throughout this lengthy process.
3224 : : */
6643 tgl@sss.pgh.pa.us 3225 [ + + ]: 1357 : elog(DEBUG2, "creating and filling new WAL file");
3226 : :
7369 3227 : 1357 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3228 : :
8939 3229 : 1357 : unlink(tmppath);
3230 : :
882 tmunro@postgresql.or 3231 [ - + ]: 1357 : if (io_direct_flags & IO_DIRECT_WAL_INIT)
882 tmunro@postgresql.or 3232 :UBC 0 : open_flags |= PG_O_DIRECT;
3233 : :
3234 : : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
882 tmunro@postgresql.or 3235 :CBC 1357 : fd = BasicOpenFile(tmppath, open_flags);
9476 vadim4o@yahoo.com 3236 [ - + ]: 1357 : if (fd < 0)
7449 tgl@sss.pgh.pa.us 3237 [ # # ]:UBC 0 : ereport(ERROR,
3238 : : (errcode_for_file_access(),
3239 : : errmsg("could not create file \"%s\": %m", tmppath)));
3240 : :
3241 : : /* Measure I/O timing when initializing segment */
192 michael@paquier.xyz 3242 :CBC 1357 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3243 : :
2349 tmunro@postgresql.or 3244 : 1357 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_WRITE);
3245 : 1357 : save_errno = 0;
3246 [ + - ]: 1357 : if (wal_init_zero)
3247 : : {
3248 : : ssize_t rc;
3249 : :
3250 : : /*
3251 : : * Zero-fill the file. With this setting, we do this the hard way to
3252 : : * ensure that all the file space has really been allocated. On
3253 : : * platforms that allow "holes" in files, just seeking to the end
3254 : : * doesn't allocate intermediate space. This way, we know that we
3255 : : * have all the space and (after the fsync below) that all the
3256 : : * indirect blocks are down on disk. Therefore, fdatasync(2) or
3257 : : * O_DSYNC will be sufficient to sync future writes to the log file.
3258 : : */
915 michael@paquier.xyz 3259 : 1357 : rc = pg_pwrite_zeros(fd, wal_segment_size, 0);
3260 : :
1033 3261 [ - + ]: 1357 : if (rc < 0)
1033 michael@paquier.xyz 3262 :UBC 0 : save_errno = errno;
3263 : : }
3264 : : else
3265 : : {
3266 : : /*
3267 : : * Otherwise, seeking to the end and writing a solitary byte is
3268 : : * enough.
3269 : : */
4385 jdavis@postgresql.or 3270 : 0 : errno = 0;
1033 michael@paquier.xyz 3271 [ # # ]: 0 : if (pg_pwrite(fd, "\0", 1, wal_segment_size - 1) != 1)
3272 : : {
3273 : : /* if write didn't set errno, assume no disk space */
2349 tmunro@postgresql.or 3274 [ # # ]: 0 : save_errno = errno ? errno : ENOSPC;
3275 : : }
3276 : : }
2349 tmunro@postgresql.or 3277 :CBC 1357 : pgstat_report_wait_end();
3278 : :
3279 : : /*
3280 : : * A full segment worth of data is written when using wal_init_zero. One
3281 : : * byte is written when not using it.
3282 : : */
214 michael@paquier.xyz 3283 : 1357 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_INIT, IOOP_WRITE,
3284 : : io_start, 1,
3285 [ + - ]: 1357 : wal_init_zero ? wal_segment_size : 1);
3286 : :
2349 tmunro@postgresql.or 3287 [ - + ]: 1357 : if (save_errno)
3288 : : {
3289 : : /*
3290 : : * If we fail to make the file, delete it to release disk space
3291 : : */
2349 tmunro@postgresql.or 3292 :UBC 0 : unlink(tmppath);
3293 : :
3294 : 0 : close(fd);
3295 : :
3296 : 0 : errno = save_errno;
3297 : :
3298 [ # # ]: 0 : ereport(ERROR,
3299 : : (errcode_for_file_access(),
3300 : : errmsg("could not write to file \"%s\": %m", tmppath)));
3301 : : }
3302 : :
3303 : : /* Measure I/O timing when flushing segment */
192 michael@paquier.xyz 3304 :CBC 1357 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3305 : :
3094 rhaas@postgresql.org 3306 : 1357 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_SYNC);
9038 tgl@sss.pgh.pa.us 3307 [ - + ]: 1357 : if (pg_fsync(fd) != 0)
3308 : : {
1107 drowley@postgresql.o 3309 :UBC 0 : save_errno = errno;
4666 heikki.linnakangas@i 3310 : 0 : close(fd);
2630 michael@paquier.xyz 3311 : 0 : errno = save_errno;
7449 tgl@sss.pgh.pa.us 3312 [ # # ]: 0 : ereport(ERROR,
3313 : : (errcode_for_file_access(),
3314 : : errmsg("could not fsync file \"%s\": %m", tmppath)));
3315 : : }
3094 rhaas@postgresql.org 3316 :CBC 1357 : pgstat_report_wait_end();
3317 : :
214 michael@paquier.xyz 3318 : 1357 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_INIT,
3319 : : IOOP_FSYNC, io_start, 1, 0);
3320 : :
2254 peter@eisentraut.org 3321 [ - + ]: 1357 : if (close(fd) != 0)
7449 tgl@sss.pgh.pa.us 3322 [ # # ]:UBC 0 : ereport(ERROR,
3323 : : (errcode_for_file_access(),
3324 : : errmsg("could not close file \"%s\": %m", tmppath)));
3325 : :
3326 : : /*
3327 : : * Now move the segment into place with its final name. Cope with
3328 : : * possibility that someone else has created the file while we were
3329 : : * filling ours: if so, use ours to pre-create a future log segment.
3330 : : */
4822 heikki.linnakangas@i 3331 :CBC 1357 : installed_segno = logsegno;
3332 : :
3333 : : /*
3334 : : * XXX: What should we use as max_segno? We used to use XLOGfileslop when
3335 : : * that was a constant, but that was always a bit dubious: normally, at a
3336 : : * checkpoint, XLOGfileslop was the offset from the checkpoint record, but
3337 : : * here, it was the offset from the insert location. We can't do the
3338 : : * normal XLOGfileslop calculation here because we don't have access to
3339 : : * the prior checkpoint's redo location. So somewhat arbitrarily, just use
3340 : : * CheckPointSegments.
3341 : : */
3848 3342 : 1357 : max_segno = logsegno + CheckPointSegments;
1401 rhaas@postgresql.org 3343 [ + - ]: 1357 : if (InstallXLogFileSegment(&installed_segno, tmppath, true, max_segno,
3344 : : logtli))
3345 : : {
1531 noah@leadboat.com 3346 : 1357 : *added = true;
3347 [ + + ]: 1357 : elog(DEBUG2, "done creating and filling new WAL file");
3348 : : }
3349 : : else
3350 : : {
3351 : : /*
3352 : : * No need for any more future segments, or InstallXLogFileSegment()
3353 : : * failed to rename the file into place. If the rename failed, a
3354 : : * caller opening the file may fail.
3355 : : */
8815 tgl@sss.pgh.pa.us 3356 :UBC 0 : unlink(tmppath);
1531 noah@leadboat.com 3357 [ # # ]: 0 : elog(DEBUG2, "abandoned new WAL file");
3358 : : }
3359 : :
1531 noah@leadboat.com 3360 :CBC 1357 : return -1;
3361 : : }
3362 : :
3363 : : /*
3364 : : * Create a new XLOG file segment, or open a pre-existing one.
3365 : : *
3366 : : * logsegno: identify segment to be created/opened.
3367 : : *
3368 : : * Returns FD of opened file.
3369 : : *
3370 : : * Note: errors here are ERROR not PANIC because we might or might not be
3371 : : * inside a critical section (eg, during checkpoint there is no reason to
3372 : : * take down the system on failure). They will promote to PANIC if we are
3373 : : * in a critical section.
3374 : : */
3375 : : int
1401 rhaas@postgresql.org 3376 : 13584 : XLogFileInit(XLogSegNo logsegno, TimeLineID logtli)
3377 : : {
3378 : : bool ignore_added;
3379 : : char path[MAXPGPATH];
3380 : : int fd;
3381 : :
3382 [ - + ]: 13584 : Assert(logtli != 0);
3383 : :
3384 : 13584 : fd = XLogFileInitInternal(logsegno, logtli, &ignore_added, path);
1531 noah@leadboat.com 3385 [ + + ]: 13584 : if (fd >= 0)
3386 : 12276 : return fd;
3387 : :
3388 : : /* Now open original target segment (might not be file I just made) */
918 tmunro@postgresql.or 3389 : 1308 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
694 nathan@postgresql.or 3390 : 1308 : get_sync_bit(wal_sync_method));
8815 tgl@sss.pgh.pa.us 3391 [ - + ]: 1308 : if (fd < 0)
7449 tgl@sss.pgh.pa.us 3392 [ # # ]:UBC 0 : ereport(ERROR,
3393 : : (errcode_for_file_access(),
3394 : : errmsg("could not open file \"%s\": %m", path)));
7178 neilc@samurai.com 3395 :CBC 1308 : return fd;
3396 : : }
3397 : :
3398 : : /*
3399 : : * Create a new XLOG file segment by copying a pre-existing one.
3400 : : *
3401 : : * destsegno: identify segment to be created.
3402 : : *
3403 : : * srcTLI, srcsegno: identify segment to be copied (could be from
3404 : : * a different timeline)
3405 : : *
3406 : : * upto: how much of the source file to copy (the rest is filled with
3407 : : * zeros)
3408 : : *
3409 : : * Currently this is only used during recovery, and so there are no locking
3410 : : * considerations. But we should be just as tense as XLogFileInit to avoid
3411 : : * emplacing a bogus file.
3412 : : */
3413 : : static void
1401 rhaas@postgresql.org 3414 : 37 : XLogFileCopy(TimeLineID destTLI, XLogSegNo destsegno,
3415 : : TimeLineID srcTLI, XLogSegNo srcsegno,
3416 : : int upto)
3417 : : {
3418 : : char path[MAXPGPATH];
3419 : : char tmppath[MAXPGPATH];
3420 : : PGAlignedXLogBlock buffer;
3421 : : int srcfd;
3422 : : int fd;
3423 : : int nbytes;
3424 : :
3425 : : /*
3426 : : * Open the source file
3427 : : */
2909 andres@anarazel.de 3428 : 37 : XLogFilePath(path, srcTLI, srcsegno, wal_segment_size);
2905 peter_e@gmx.net 3429 : 37 : srcfd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
7717 tgl@sss.pgh.pa.us 3430 [ - + ]: 37 : if (srcfd < 0)
7449 tgl@sss.pgh.pa.us 3431 [ # # ]:UBC 0 : ereport(ERROR,
3432 : : (errcode_for_file_access(),
3433 : : errmsg("could not open file \"%s\": %m", path)));
3434 : :
3435 : : /*
3436 : : * Copy into a temp file name.
3437 : : */
7369 tgl@sss.pgh.pa.us 3438 :CBC 37 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3439 : :
7717 3440 : 37 : unlink(tmppath);
3441 : :
3442 : : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
2905 peter_e@gmx.net 3443 : 37 : fd = OpenTransientFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
7717 tgl@sss.pgh.pa.us 3444 [ - + ]: 37 : if (fd < 0)
7449 tgl@sss.pgh.pa.us 3445 [ # # ]:UBC 0 : ereport(ERROR,
3446 : : (errcode_for_file_access(),
3447 : : errmsg("could not create file \"%s\": %m", tmppath)));
3448 : :
3449 : : /*
3450 : : * Do the data copying.
3451 : : */
2909 andres@anarazel.de 3452 [ + + ]:CBC 75813 : for (nbytes = 0; nbytes < wal_segment_size; nbytes += sizeof(buffer))
3453 : : {
3454 : : int nread;
3455 : :
3915 heikki.linnakangas@i 3456 : 75776 : nread = upto - nbytes;
3457 : :
3458 : : /*
3459 : : * The part that is not read from the source file is filled with
3460 : : * zeros.
3461 : : */
3462 [ + + ]: 75776 : if (nread < sizeof(buffer))
2562 tgl@sss.pgh.pa.us 3463 : 37 : memset(buffer.data, 0, sizeof(buffer));
3464 : :
3915 heikki.linnakangas@i 3465 [ + + ]: 75776 : if (nread > 0)
3466 : : {
3467 : : int r;
3468 : :
3469 [ + + ]: 2166 : if (nread > sizeof(buffer))
3470 : 2129 : nread = sizeof(buffer);
3094 rhaas@postgresql.org 3471 : 2166 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_READ);
2562 tgl@sss.pgh.pa.us 3472 : 2166 : r = read(srcfd, buffer.data, nread);
2607 michael@paquier.xyz 3473 [ - + ]: 2166 : if (r != nread)
3474 : : {
2607 michael@paquier.xyz 3475 [ # # ]:UBC 0 : if (r < 0)
3915 heikki.linnakangas@i 3476 [ # # ]: 0 : ereport(ERROR,
3477 : : (errcode_for_file_access(),
3478 : : errmsg("could not read file \"%s\": %m",
3479 : : path)));
3480 : : else
3481 [ # # ]: 0 : ereport(ERROR,
3482 : : (errcode(ERRCODE_DATA_CORRUPTED),
3483 : : errmsg("could not read file \"%s\": read %d of %zu",
3484 : : path, r, (Size) nread)));
3485 : : }
3094 rhaas@postgresql.org 3486 :CBC 2166 : pgstat_report_wait_end();
3487 : : }
7717 tgl@sss.pgh.pa.us 3488 : 75776 : errno = 0;
3094 rhaas@postgresql.org 3489 : 75776 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_WRITE);
2562 tgl@sss.pgh.pa.us 3490 [ - + ]: 75776 : if ((int) write(fd, buffer.data, sizeof(buffer)) != (int) sizeof(buffer))
3491 : : {
7717 tgl@sss.pgh.pa.us 3492 :UBC 0 : int save_errno = errno;
3493 : :
3494 : : /*
3495 : : * If we fail to make the file, delete it to release disk space
3496 : : */
3497 : 0 : unlink(tmppath);
3498 : : /* if write didn't set errno, assume problem is no disk space */
3499 [ # # ]: 0 : errno = save_errno ? save_errno : ENOSPC;
3500 : :
7449 3501 [ # # ]: 0 : ereport(ERROR,
3502 : : (errcode_for_file_access(),
3503 : : errmsg("could not write to file \"%s\": %m", tmppath)));
3504 : : }
3094 rhaas@postgresql.org 3505 :CBC 75776 : pgstat_report_wait_end();
3506 : : }
3507 : :
3508 : 37 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_SYNC);
7717 tgl@sss.pgh.pa.us 3509 [ - + ]: 37 : if (pg_fsync(fd) != 0)
2483 tmunro@postgresql.or 3510 [ # # ]:UBC 0 : ereport(data_sync_elevel(ERROR),
3511 : : (errcode_for_file_access(),
3512 : : errmsg("could not fsync file \"%s\": %m", tmppath)));
3094 rhaas@postgresql.org 3513 :CBC 37 : pgstat_report_wait_end();
3514 : :
2254 peter@eisentraut.org 3515 [ - + ]: 37 : if (CloseTransientFile(fd) != 0)
7449 tgl@sss.pgh.pa.us 3516 [ # # ]:UBC 0 : ereport(ERROR,
3517 : : (errcode_for_file_access(),
3518 : : errmsg("could not close file \"%s\": %m", tmppath)));
3519 : :
2254 peter@eisentraut.org 3520 [ - + ]:CBC 37 : if (CloseTransientFile(srcfd) != 0)
2373 michael@paquier.xyz 3521 [ # # ]:UBC 0 : ereport(ERROR,
3522 : : (errcode_for_file_access(),
3523 : : errmsg("could not close file \"%s\": %m", path)));
3524 : :
3525 : : /*
3526 : : * Now move the segment into place with its final name.
3527 : : */
1401 rhaas@postgresql.org 3528 [ - + ]:CBC 37 : if (!InstallXLogFileSegment(&destsegno, tmppath, false, 0, destTLI))
3720 fujii@postgresql.org 3529 [ # # ]:UBC 0 : elog(ERROR, "InstallXLogFileSegment should not have failed");
7717 tgl@sss.pgh.pa.us 3530 :CBC 37 : }
3531 : :
3532 : : /*
3533 : : * Install a new XLOG segment file as a current or future log segment.
3534 : : *
3535 : : * This is used both to install a newly-created segment (which has a temp
3536 : : * filename while it's being created) and to recycle an old segment.
3537 : : *
3538 : : * *segno: identify segment to install as (or first possible target).
3539 : : * When find_free is true, this is modified on return to indicate the
3540 : : * actual installation location or last segment searched.
3541 : : *
3542 : : * tmppath: initial name of file to install. It will be renamed into place.
3543 : : *
3544 : : * find_free: if true, install the new segment at the first empty segno
3545 : : * number at or after the passed numbers. If false, install the new segment
3546 : : * exactly where specified, deleting any existing segment file there.
3547 : : *
3548 : : * max_segno: maximum segment number to install the new file as. Fail if no
3549 : : * free slot is found between *segno and max_segno. (Ignored when find_free
3550 : : * is false.)
3551 : : *
3552 : : * tli: The timeline on which the new segment should be installed.
3553 : : *
3554 : : * Returns true if the file was installed successfully. false indicates that
3555 : : * max_segno limit was exceeded, the startup process has disabled this
3556 : : * function for now, or an error occurred while renaming the file into place.
3557 : : */
3558 : : static bool
4822 heikki.linnakangas@i 3559 : 2876 : InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
3560 : : bool find_free, XLogSegNo max_segno, TimeLineID tli)
3561 : : {
3562 : : char path[MAXPGPATH];
3563 : : struct stat stat_buf;
3564 : :
1401 rhaas@postgresql.org 3565 [ - + ]: 2876 : Assert(tli != 0);
3566 : :
3567 : 2876 : XLogFilePath(path, tli, *segno, wal_segment_size);
3568 : :
1531 noah@leadboat.com 3569 : 2876 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
3570 [ - + ]: 2876 : if (!XLogCtl->InstallXLogFileSegmentActive)
3571 : : {
1531 noah@leadboat.com 3572 :UBC 0 : LWLockRelease(ControlFileLock);
3573 : 0 : return false;
3574 : : }
3575 : :
8815 tgl@sss.pgh.pa.us 3576 [ + + ]:CBC 2876 : if (!find_free)
3577 : : {
3578 : : /* Force installation: get rid of any pre-existing segment file */
3085 teodor@sigaev.ru 3579 : 37 : durable_unlink(path, DEBUG1);
3580 : : }
3581 : : else
3582 : : {
3583 : : /* Find a free slot to put it in */
8260 tgl@sss.pgh.pa.us 3584 [ + + ]: 3999 : while (stat(path, &stat_buf) == 0)
3585 : : {
3848 heikki.linnakangas@i 3586 [ + + ]: 1269 : if ((*segno) >= max_segno)
3587 : : {
3588 : : /* Failed to find a free slot within specified range */
1531 noah@leadboat.com 3589 : 109 : LWLockRelease(ControlFileLock);
8815 tgl@sss.pgh.pa.us 3590 : 109 : return false;
3591 : : }
4822 heikki.linnakangas@i 3592 : 1160 : (*segno)++;
1401 rhaas@postgresql.org 3593 : 1160 : XLogFilePath(path, tli, *segno, wal_segment_size);
3594 : : }
3595 : : }
3596 : :
1159 michael@paquier.xyz 3597 [ + - - + ]: 2767 : Assert(access(path, F_OK) != 0 && errno == ENOENT);
3598 [ - + ]: 2767 : if (durable_rename(tmppath, path, LOG) != 0)
3599 : : {
1531 noah@leadboat.com 3600 :UBC 0 : LWLockRelease(ControlFileLock);
3601 : : /* durable_rename already emitted log message */
5837 heikki.linnakangas@i 3602 : 0 : return false;
3603 : : }
3604 : :
1531 noah@leadboat.com 3605 :CBC 2767 : LWLockRelease(ControlFileLock);
3606 : :
8815 tgl@sss.pgh.pa.us 3607 : 2767 : return true;
3608 : : }
3609 : :
3610 : : /*
3611 : : * Open a pre-existing logfile segment for writing.
3612 : : */
3613 : : int
1401 rhaas@postgresql.org 3614 : 141 : XLogFileOpen(XLogSegNo segno, TimeLineID tli)
3615 : : {
3616 : : char path[MAXPGPATH];
3617 : : int fd;
3618 : :
3619 : 141 : XLogFilePath(path, tli, segno, wal_segment_size);
3620 : :
918 tmunro@postgresql.or 3621 : 141 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
694 nathan@postgresql.or 3622 : 141 : get_sync_bit(wal_sync_method));
9476 vadim4o@yahoo.com 3623 [ - + ]: 141 : if (fd < 0)
8083 tgl@sss.pgh.pa.us 3624 [ # # ]:UBC 0 : ereport(PANIC,
3625 : : (errcode_for_file_access(),
3626 : : errmsg("could not open file \"%s\": %m", path)));
3627 : :
7717 tgl@sss.pgh.pa.us 3628 :CBC 141 : return fd;
3629 : : }
3630 : :
3631 : : /*
3632 : : * Close the current logfile segment for writing.
3633 : : */
3634 : : static void
7023 bruce@momjian.us 3635 : 6338 : XLogFileClose(void)
3636 : : {
3637 [ - + ]: 6338 : Assert(openLogFile >= 0);
3638 : :
3639 : : /*
3640 : : * WAL segment files will not be re-read in normal operation, so we advise
3641 : : * the OS to release any cached pages. But do not do so if WAL archiving
3642 : : * or streaming is active, because archiver and walsender process could
3643 : : * use the cache to read the WAL segment.
3644 : : */
3645 : : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
882 tmunro@postgresql.or 3646 [ + + + - ]: 6338 : if (!XLogIsNeeded() && (io_direct_flags & IO_DIRECT_WAL) == 0)
6082 tgl@sss.pgh.pa.us 3647 : 1569 : (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
3648 : : #endif
3649 : :
2254 peter@eisentraut.org 3650 [ - + ]: 6338 : if (close(openLogFile) != 0)
3651 : : {
3652 : : char xlogfname[MAXFNAMELEN];
2104 michael@paquier.xyz 3653 :UBC 0 : int save_errno = errno;
3654 : :
1401 rhaas@postgresql.org 3655 : 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo, wal_segment_size);
2104 michael@paquier.xyz 3656 : 0 : errno = save_errno;
7023 bruce@momjian.us 3657 [ # # ]: 0 : ereport(PANIC,
3658 : : (errcode_for_file_access(),
3659 : : errmsg("could not close file \"%s\": %m", xlogfname)));
3660 : : }
3661 : :
7023 bruce@momjian.us 3662 :CBC 6338 : openLogFile = -1;
2021 tgl@sss.pgh.pa.us 3663 : 6338 : ReleaseExternalFD();
7023 bruce@momjian.us 3664 : 6338 : }
3665 : :
3666 : : /*
3667 : : * Preallocate log files beyond the specified log endpoint.
3668 : : *
3669 : : * XXX this is currently extremely conservative, since it forces only one
3670 : : * future log segment to exist, and even that only if we are 75% done with
3671 : : * the current one. This is only appropriate for very low-WAL-volume systems.
3672 : : * High-volume systems will be OK once they've built up a sufficient set of
3673 : : * recycled log segments, but the startup transient is likely to include
3674 : : * a lot of segment creations by foreground processes, which is not so good.
3675 : : *
3676 : : * XLogFileInitInternal() can ereport(ERROR). All known causes indicate big
3677 : : * trouble; for example, a full filesystem is one cause. The checkpoint WAL
3678 : : * and/or ControlFile updates already completed. If a RequestCheckpoint()
3679 : : * initiated the present checkpoint and an ERROR ends this function, the
3680 : : * command that called RequestCheckpoint() fails. That's not ideal, but it's
3681 : : * not worth contorting more functions to use caller-specified elevel values.
3682 : : * (With or without RequestCheckpoint(), an ERROR forestalls some inessential
3683 : : * reporting and resource reclamation.)
3684 : : */
3685 : : static void
1401 rhaas@postgresql.org 3686 : 1926 : PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli)
3687 : : {
3688 : : XLogSegNo _logSegNo;
3689 : : int lf;
3690 : : bool added;
3691 : : char path[MAXPGPATH];
3692 : : uint64 offset;
3693 : :
1531 noah@leadboat.com 3694 [ + + ]: 1926 : if (!XLogCtl->InstallXLogFileSegmentActive)
3695 : 11 : return; /* unlocked check says no */
3696 : :
2909 andres@anarazel.de 3697 : 1915 : XLByteToPrevSeg(endptr, _logSegNo, wal_segment_size);
3698 : 1915 : offset = XLogSegmentOffset(endptr - 1, wal_segment_size);
3699 [ + + ]: 1915 : if (offset >= (uint32) (0.75 * wal_segment_size))
3700 : : {
4822 heikki.linnakangas@i 3701 : 224 : _logSegNo++;
1401 rhaas@postgresql.org 3702 : 224 : lf = XLogFileInitInternal(_logSegNo, tli, &added, path);
1531 noah@leadboat.com 3703 [ + + ]: 224 : if (lf >= 0)
3704 : 175 : close(lf);
3705 [ + + ]: 224 : if (added)
6643 tgl@sss.pgh.pa.us 3706 : 49 : CheckpointStats.ckpt_segs_added++;
3707 : : }
3708 : : }
3709 : :
3710 : : /*
3711 : : * Throws an error if the given log segment has already been removed or
3712 : : * recycled. The caller should only pass a segment that it knows to have
3713 : : * existed while the server has been running, as this function always
3714 : : * succeeds if no WAL segments have been removed since startup.
3715 : : * 'tli' is only used in the error message.
3716 : : *
3717 : : * Note: this function guarantees to keep errno unchanged on return.
3718 : : * This supports callers that use this to possibly deliver a better
3719 : : * error message about a missing file, while still being able to throw
3720 : : * a normal file-access error afterwards, if this does return.
3721 : : */
3722 : : void
4629 heikki.linnakangas@i 3723 : 122179 : CheckXLogRemoved(XLogSegNo segno, TimeLineID tli)
3724 : : {
2833 tgl@sss.pgh.pa.us 3725 : 122179 : int save_errno = errno;
3726 : : XLogSegNo lastRemovedSegNo;
3727 : :
4002 andres@anarazel.de 3728 [ + + ]: 122179 : SpinLockAcquire(&XLogCtl->info_lck);
3729 : 122179 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3730 : 122179 : SpinLockRelease(&XLogCtl->info_lck);
3731 : :
4629 heikki.linnakangas@i 3732 [ - + ]: 122179 : if (segno <= lastRemovedSegNo)
3733 : : {
3734 : : char filename[MAXFNAMELEN];
3735 : :
2909 andres@anarazel.de 3736 :UBC 0 : XLogFileName(filename, tli, segno, wal_segment_size);
2833 tgl@sss.pgh.pa.us 3737 : 0 : errno = save_errno;
4629 heikki.linnakangas@i 3738 [ # # ]: 0 : ereport(ERROR,
3739 : : (errcode_for_file_access(),
3740 : : errmsg("requested WAL segment %s has already been removed",
3741 : : filename)));
3742 : : }
2833 tgl@sss.pgh.pa.us 3743 :CBC 122179 : errno = save_errno;
5626 heikki.linnakangas@i 3744 : 122179 : }
3745 : :
3746 : : /*
3747 : : * Return the last WAL segment removed, or 0 if no segment has been removed
3748 : : * since startup.
3749 : : *
3750 : : * NB: the result can be out of date arbitrarily fast, the caller has to deal
3751 : : * with that.
3752 : : */
3753 : : XLogSegNo
4205 rhaas@postgresql.org 3754 : 944 : XLogGetLastRemovedSegno(void)
3755 : : {
3756 : : XLogSegNo lastRemovedSegNo;
3757 : :
4002 andres@anarazel.de 3758 [ - + ]: 944 : SpinLockAcquire(&XLogCtl->info_lck);
3759 : 944 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3760 : 944 : SpinLockRelease(&XLogCtl->info_lck);
3761 : :
4205 rhaas@postgresql.org 3762 : 944 : return lastRemovedSegNo;
3763 : : }
3764 : :
3765 : : /*
3766 : : * Return the oldest WAL segment on the given TLI that still exists in
3767 : : * XLOGDIR, or 0 if none.
3768 : : */
3769 : : XLogSegNo
626 3770 : 5 : XLogGetOldestSegno(TimeLineID tli)
3771 : : {
3772 : : DIR *xldir;
3773 : : struct dirent *xlde;
3774 : 5 : XLogSegNo oldest_segno = 0;
3775 : :
3776 : 5 : xldir = AllocateDir(XLOGDIR);
3777 [ + + ]: 33 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3778 : : {
3779 : : TimeLineID file_tli;
3780 : : XLogSegNo file_segno;
3781 : :
3782 : : /* Ignore files that are not XLOG segments. */
3783 [ + + ]: 28 : if (!IsXLogFileName(xlde->d_name))
3784 : 20 : continue;
3785 : :
3786 : : /* Parse filename to get TLI and segno. */
3787 : 8 : XLogFromFileName(xlde->d_name, &file_tli, &file_segno,
3788 : : wal_segment_size);
3789 : :
3790 : : /* Ignore anything that's not from the TLI of interest. */
3791 [ - + ]: 8 : if (tli != file_tli)
626 rhaas@postgresql.org 3792 :UBC 0 : continue;
3793 : :
3794 : : /* If it's the oldest so far, update oldest_segno. */
626 rhaas@postgresql.org 3795 [ + + + + ]:CBC 8 : if (oldest_segno == 0 || file_segno < oldest_segno)
3796 : 6 : oldest_segno = file_segno;
3797 : : }
3798 : :
3799 : 5 : FreeDir(xldir);
3800 : 5 : return oldest_segno;
3801 : : }
3802 : :
3803 : : /*
3804 : : * Update the last removed segno pointer in shared memory, to reflect that the
3805 : : * given XLOG file has been removed.
3806 : : */
3807 : : static void
5626 heikki.linnakangas@i 3808 : 2516 : UpdateLastRemovedPtr(char *filename)
3809 : : {
3810 : : uint32 tli;
3811 : : XLogSegNo segno;
3812 : :
2909 andres@anarazel.de 3813 : 2516 : XLogFromFileName(filename, &tli, &segno, wal_segment_size);
3814 : :
4002 3815 [ + + ]: 2516 : SpinLockAcquire(&XLogCtl->info_lck);
3816 [ + + ]: 2516 : if (segno > XLogCtl->lastRemovedSegNo)
3817 : 1112 : XLogCtl->lastRemovedSegNo = segno;
3818 : 2516 : SpinLockRelease(&XLogCtl->info_lck);
5626 heikki.linnakangas@i 3819 : 2516 : }
3820 : :
3821 : : /*
3822 : : * Remove all temporary log files in pg_wal
3823 : : *
3824 : : * This is called at the beginning of recovery after a previous crash,
3825 : : * at a point where no other processes write fresh WAL data.
3826 : : */
3827 : : static void
2612 michael@paquier.xyz 3828 : 169 : RemoveTempXlogFiles(void)
3829 : : {
3830 : : DIR *xldir;
3831 : : struct dirent *xlde;
3832 : :
3833 [ + + ]: 169 : elog(DEBUG2, "removing all temporary WAL segments");
3834 : :
3835 : 169 : xldir = AllocateDir(XLOGDIR);
3836 [ + + ]: 1094 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3837 : : {
3838 : : char path[MAXPGPATH];
3839 : :
3840 [ + - ]: 925 : if (strncmp(xlde->d_name, "xlogtemp.", 9) != 0)
3841 : 925 : continue;
3842 : :
2612 michael@paquier.xyz 3843 :UBC 0 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
3844 : 0 : unlink(path);
3845 [ # # ]: 0 : elog(DEBUG2, "removed temporary WAL segment \"%s\"", path);
3846 : : }
2612 michael@paquier.xyz 3847 :CBC 169 : FreeDir(xldir);
3848 : 169 : }
3849 : :
3850 : : /*
3851 : : * Recycle or remove all log files older or equal to passed segno.
3852 : : *
3853 : : * endptr is current (or recent) end of xlog, and lastredoptr is the
3854 : : * redo pointer of the last checkpoint. These are used to determine
3855 : : * whether we want to recycle rather than delete no-longer-wanted log files.
3856 : : *
3857 : : * insertTLI is the current timeline for XLOG insertion. Any recycled
3858 : : * segments should be reused for this timeline.
3859 : : */
3860 : : static void
1401 rhaas@postgresql.org 3861 : 1677 : RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr,
3862 : : TimeLineID insertTLI)
3863 : : {
3864 : : DIR *xldir;
3865 : : struct dirent *xlde;
3866 : : char lastoff[MAXFNAMELEN];
3867 : : XLogSegNo endlogSegNo;
3868 : : XLogSegNo recycleSegNo;
3869 : :
3870 : : /* Initialize info about where to try to recycle to */
1695 michael@paquier.xyz 3871 : 1677 : XLByteToSeg(endptr, endlogSegNo, wal_segment_size);
3872 : 1677 : recycleSegNo = XLOGfileslop(lastredoptr);
3873 : :
3874 : : /*
3875 : : * Construct a filename of the last segment to be kept. The timeline ID
3876 : : * doesn't matter, we ignore that in the comparison. (During recovery,
3877 : : * InsertTimeLineID isn't set, so we can't use that.)
3878 : : */
2909 andres@anarazel.de 3879 : 1677 : XLogFileName(lastoff, 0, segno, wal_segment_size);
3880 : :
5486 simon@2ndQuadrant.co 3881 [ + + ]: 1677 : elog(DEBUG2, "attempting to remove WAL segments older than log file %s",
3882 : : lastoff);
3883 : :
2833 tgl@sss.pgh.pa.us 3884 : 1677 : xldir = AllocateDir(XLOGDIR);
3885 : :
7369 3886 [ + + ]: 45131 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3887 : : {
3888 : : /* Ignore files that are not XLOG segments */
3774 heikki.linnakangas@i 3889 [ + + ]: 43454 : if (!IsXLogFileName(xlde->d_name) &&
3890 [ + + ]: 7151 : !IsPartialXLogFileName(xlde->d_name))
3799 3891 : 7147 : continue;
3892 : :
3893 : : /*
3894 : : * We ignore the timeline part of the XLOG segment identifiers in
3895 : : * deciding whether a segment is still needed. This ensures that we
3896 : : * won't prematurely remove a segment from a parent timeline. We could
3897 : : * probably be a little more proactive about removing segments of
3898 : : * non-parent timelines, but that would be a whole lot more
3899 : : * complicated.
3900 : : *
3901 : : * We use the alphanumeric sorting property of the filenames to decide
3902 : : * which ones are earlier than the lastoff segment.
3903 : : */
3904 [ + + ]: 36307 : if (strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
3905 : : {
4586 3906 [ + + ]: 30066 : if (XLogArchiveCheckDone(xlde->d_name))
3907 : : {
3908 : : /* Update the last removed location in shared memory first */
5626 3909 : 2516 : UpdateLastRemovedPtr(xlde->d_name);
3910 : :
1100 michael@paquier.xyz 3911 : 2516 : RemoveXlogFile(xlde, recycleSegNo, &endlogSegNo, insertTLI);
3912 : : }
3913 : : }
3914 : : }
3915 : :
3799 heikki.linnakangas@i 3916 : 1677 : FreeDir(xldir);
3917 : 1677 : }
3918 : :
3919 : : /*
3920 : : * Recycle or remove WAL files that are not part of the given timeline's
3921 : : * history.
3922 : : *
3923 : : * This is called during recovery, whenever we switch to follow a new
3924 : : * timeline, and at the end of recovery when we create a new timeline. We
3925 : : * wouldn't otherwise care about extra WAL files lying in pg_wal, but they
3926 : : * might be leftover pre-allocated or recycled WAL segments on the old timeline
3927 : : * that we haven't used yet, and contain garbage. If we just leave them in
3928 : : * pg_wal, they will eventually be archived, and we can't let that happen.
3929 : : * Files that belong to our timeline history are valid, because we have
3930 : : * successfully replayed them, but from others we can't be sure.
3931 : : *
3932 : : * 'switchpoint' is the current point in WAL where we switch to new timeline,
3933 : : * and 'newTLI' is the new timeline we switch to.
3934 : : */
3935 : : void
3936 : 57 : RemoveNonParentXlogFiles(XLogRecPtr switchpoint, TimeLineID newTLI)
3937 : : {
3938 : : DIR *xldir;
3939 : : struct dirent *xlde;
3940 : : char switchseg[MAXFNAMELEN];
3941 : : XLogSegNo endLogSegNo;
3942 : : XLogSegNo switchLogSegNo;
3943 : : XLogSegNo recycleSegNo;
3944 : :
3945 : : /*
3946 : : * Initialize info about where to begin the work. This will recycle,
3947 : : * somewhat arbitrarily, 10 future segments.
3948 : : */
1695 michael@paquier.xyz 3949 : 57 : XLByteToPrevSeg(switchpoint, switchLogSegNo, wal_segment_size);
3950 : 57 : XLByteToSeg(switchpoint, endLogSegNo, wal_segment_size);
3951 : 57 : recycleSegNo = endLogSegNo + 10;
3952 : :
3953 : : /*
3954 : : * Construct a filename of the last segment to be kept.
3955 : : */
3956 : 57 : XLogFileName(switchseg, newTLI, switchLogSegNo, wal_segment_size);
3957 : :
3799 heikki.linnakangas@i 3958 [ + + ]: 57 : elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
3959 : : switchseg);
3960 : :
2833 tgl@sss.pgh.pa.us 3961 : 57 : xldir = AllocateDir(XLOGDIR);
3962 : :
3799 heikki.linnakangas@i 3963 [ + + ]: 541 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3964 : : {
3965 : : /* Ignore files that are not XLOG segments */
3774 3966 [ + + ]: 484 : if (!IsXLogFileName(xlde->d_name))
3799 3967 : 303 : continue;
3968 : :
3969 : : /*
3970 : : * Remove files that are on a timeline older than the new one we're
3971 : : * switching to, but with a segment number >= the first segment on the
3972 : : * new timeline.
3973 : : */
3974 [ + + ]: 181 : if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
3975 [ + + ]: 117 : strcmp(xlde->d_name + 8, switchseg + 8) > 0)
3976 : : {
3977 : : /*
3978 : : * If the file has already been marked as .ready, however, don't
3979 : : * remove it yet. It should be OK to remove it - files that are
3980 : : * not part of our timeline history are not required for recovery
3981 : : * - but seems safer to let them be archived and removed later.
3982 : : */
3983 [ + - ]: 14 : if (!XLogArchiveIsReady(xlde->d_name))
1100 michael@paquier.xyz 3984 : 14 : RemoveXlogFile(xlde, recycleSegNo, &endLogSegNo, newTLI);
3985 : : }
3986 : : }
3987 : :
3799 heikki.linnakangas@i 3988 : 57 : FreeDir(xldir);
3989 : 57 : }
3990 : :
3991 : : /*
3992 : : * Recycle or remove a log file that's no longer needed.
3993 : : *
3994 : : * segment_de is the dirent structure of the segment to recycle or remove.
3995 : : * recycleSegNo is the segment number to recycle up to. endlogSegNo is
3996 : : * the segment number of the current (or recent) end of WAL.
3997 : : *
3998 : : * endlogSegNo gets incremented if the segment is recycled so as it is not
3999 : : * checked again with future callers of this function.
4000 : : *
4001 : : * insertTLI is the current timeline for XLOG insertion. Any recycled segments
4002 : : * should be used for this timeline.
4003 : : */
4004 : : static void
1100 michael@paquier.xyz 4005 : 2530 : RemoveXlogFile(const struct dirent *segment_de,
4006 : : XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo,
4007 : : TimeLineID insertTLI)
4008 : : {
4009 : : char path[MAXPGPATH];
4010 : : #ifdef WIN32
4011 : : char newpath[MAXPGPATH];
4012 : : #endif
4013 : 2530 : const char *segname = segment_de->d_name;
4014 : :
3799 heikki.linnakangas@i 4015 : 2530 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", segname);
4016 : :
4017 : : /*
4018 : : * Before deleting the file, see if it can be recycled as a future log
4019 : : * segment. Only recycle normal files, because we don't want to recycle
4020 : : * symbolic links pointing to a separate archive directory.
4021 : : */
2349 tmunro@postgresql.or 4022 [ + - ]: 2530 : if (wal_recycle &&
1695 michael@paquier.xyz 4023 [ + + ]: 2530 : *endlogSegNo <= recycleSegNo &&
1531 noah@leadboat.com 4024 [ + + + - ]: 3292 : XLogCtl->InstallXLogFileSegmentActive && /* callee rechecks this */
1100 michael@paquier.xyz 4025 [ + + ]: 2964 : get_dirent_type(path, segment_de, false, DEBUG2) == PGFILETYPE_REG &&
1695 4026 : 1482 : InstallXLogFileSegment(endlogSegNo, path,
4027 : : true, recycleSegNo, insertTLI))
4028 : : {
3799 heikki.linnakangas@i 4029 [ + + ]: 1373 : ereport(DEBUG2,
4030 : : (errmsg_internal("recycled write-ahead log file \"%s\"",
4031 : : segname)));
4032 : 1373 : CheckpointStats.ckpt_segs_recycled++;
4033 : : /* Needn't recheck that slot on future iterations */
1695 michael@paquier.xyz 4034 : 1373 : (*endlogSegNo)++;
4035 : : }
4036 : : else
4037 : : {
4038 : : /* No need for any more future segments, or recycling failed ... */
4039 : : int rc;
4040 : :
3799 heikki.linnakangas@i 4041 [ + + ]: 1157 : ereport(DEBUG2,
4042 : : (errmsg_internal("removing write-ahead log file \"%s\"",
4043 : : segname)));
4044 : :
4045 : : #ifdef WIN32
4046 : :
4047 : : /*
4048 : : * On Windows, if another process (e.g another backend) holds the file
4049 : : * open in FILE_SHARE_DELETE mode, unlink will succeed, but the file
4050 : : * will still show up in directory listing until the last handle is
4051 : : * closed. To avoid confusing the lingering deleted file for a live
4052 : : * WAL file that needs to be archived, rename it before deleting it.
4053 : : *
4054 : : * If another process holds the file open without FILE_SHARE_DELETE
4055 : : * flag, rename will fail. We'll try again at the next checkpoint.
4056 : : */
4057 : : snprintf(newpath, MAXPGPATH, "%s.deleted", path);
4058 : : if (rename(path, newpath) != 0)
4059 : : {
4060 : : ereport(LOG,
4061 : : (errcode_for_file_access(),
4062 : : errmsg("could not rename file \"%s\": %m",
4063 : : path)));
4064 : : return;
4065 : : }
4066 : : rc = durable_unlink(newpath, LOG);
4067 : : #else
3085 teodor@sigaev.ru 4068 : 1157 : rc = durable_unlink(path, LOG);
4069 : : #endif
3799 heikki.linnakangas@i 4070 [ - + ]: 1157 : if (rc != 0)
4071 : : {
4072 : : /* Message already logged by durable_unlink() */
3799 heikki.linnakangas@i 4073 :UBC 0 : return;
4074 : : }
3799 heikki.linnakangas@i 4075 :CBC 1157 : CheckpointStats.ckpt_segs_removed++;
4076 : : }
4077 : :
4078 : 2530 : XLogArchiveCleanup(segname);
4079 : : }
4080 : :
4081 : : /*
4082 : : * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
4083 : : * If the latter do not exist, recreate them.
4084 : : *
4085 : : * It is not the goal of this function to verify the contents of these
4086 : : * directories, but to help in cases where someone has performed a cluster
4087 : : * copy for PITR purposes but omitted pg_wal from the copy.
4088 : : *
4089 : : * We could also recreate pg_wal if it doesn't exist, but a deliberate
4090 : : * policy decision was made not to. It is fairly common for pg_wal to be
4091 : : * a symlink, and if that was the DBA's intent then automatically making a
4092 : : * plain directory would result in degraded performance with no notice.
4093 : : */
4094 : : static void
6145 tgl@sss.pgh.pa.us 4095 : 887 : ValidateXLOGDirectoryStructure(void)
4096 : : {
4097 : : char path[MAXPGPATH];
4098 : : struct stat stat_buf;
4099 : :
4100 : : /* Check for pg_wal; if it doesn't exist, error out */
4101 [ + - ]: 887 : if (stat(XLOGDIR, &stat_buf) != 0 ||
4102 [ - + ]: 887 : !S_ISDIR(stat_buf.st_mode))
5931 bruce@momjian.us 4103 [ # # ]:UBC 0 : ereport(FATAL,
4104 : : (errcode_for_file_access(),
4105 : : errmsg("required WAL directory \"%s\" does not exist",
4106 : : XLOGDIR)));
4107 : :
4108 : : /* Check for archive_status */
6145 tgl@sss.pgh.pa.us 4109 :CBC 887 : snprintf(path, MAXPGPATH, XLOGDIR "/archive_status");
4110 [ + + ]: 887 : if (stat(path, &stat_buf) == 0)
4111 : : {
4112 : : /* Check for weird cases where it exists but isn't a directory */
4113 [ - + ]: 886 : if (!S_ISDIR(stat_buf.st_mode))
5931 bruce@momjian.us 4114 [ # # ]:UBC 0 : ereport(FATAL,
4115 : : (errcode_for_file_access(),
4116 : : errmsg("required WAL directory \"%s\" does not exist",
4117 : : path)));
4118 : : }
4119 : : else
4120 : : {
6145 tgl@sss.pgh.pa.us 4121 [ + - ]:CBC 1 : ereport(LOG,
4122 : : (errmsg("creating missing WAL directory \"%s\"", path)));
2709 sfrost@snowman.net 4123 [ - + ]: 1 : if (MakePGDirectory(path) < 0)
5931 bruce@momjian.us 4124 [ # # ]:UBC 0 : ereport(FATAL,
4125 : : (errcode_for_file_access(),
4126 : : errmsg("could not create missing directory \"%s\": %m",
4127 : : path)));
4128 : : }
4129 : :
4130 : : /* Check for summaries */
626 rhaas@postgresql.org 4131 :CBC 887 : snprintf(path, MAXPGPATH, XLOGDIR "/summaries");
4132 [ + + ]: 887 : if (stat(path, &stat_buf) == 0)
4133 : : {
4134 : : /* Check for weird cases where it exists but isn't a directory */
4135 [ - + ]: 886 : if (!S_ISDIR(stat_buf.st_mode))
626 rhaas@postgresql.org 4136 [ # # ]:UBC 0 : ereport(FATAL,
4137 : : (errmsg("required WAL directory \"%s\" does not exist",
4138 : : path)));
4139 : : }
4140 : : else
4141 : : {
626 rhaas@postgresql.org 4142 [ + - ]:CBC 1 : ereport(LOG,
4143 : : (errmsg("creating missing WAL directory \"%s\"", path)));
4144 [ - + ]: 1 : if (MakePGDirectory(path) < 0)
626 rhaas@postgresql.org 4145 [ # # ]:UBC 0 : ereport(FATAL,
4146 : : (errmsg("could not create missing directory \"%s\": %m",
4147 : : path)));
4148 : : }
6145 tgl@sss.pgh.pa.us 4149 :CBC 887 : }
4150 : :
4151 : : /*
4152 : : * Remove previous backup history files. This also retries creation of
4153 : : * .ready files for any backup history files for which XLogArchiveNotify
4154 : : * failed earlier.
4155 : : */
4156 : : static void
7016 4157 : 155 : CleanupBackupHistory(void)
4158 : : {
4159 : : DIR *xldir;
4160 : : struct dirent *xlde;
4161 : : char path[MAXPGPATH + sizeof(XLOGDIR)];
4162 : :
7369 4163 : 155 : xldir = AllocateDir(XLOGDIR);
4164 : :
4165 [ + + ]: 1552 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4166 : : {
3774 heikki.linnakangas@i 4167 [ + + ]: 1242 : if (IsBackupHistoryFileName(xlde->d_name))
4168 : : {
6207 tgl@sss.pgh.pa.us 4169 [ + + ]: 163 : if (XLogArchiveCheckDone(xlde->d_name))
4170 : : {
3039 peter_e@gmx.net 4171 [ + + ]: 131 : elog(DEBUG2, "removing WAL backup history file \"%s\"",
4172 : : xlde->d_name);
3070 4173 : 131 : snprintf(path, sizeof(path), XLOGDIR "/%s", xlde->d_name);
7388 bruce@momjian.us 4174 : 131 : unlink(path);
4175 : 131 : XLogArchiveCleanup(xlde->d_name);
4176 : : }
4177 : : }
4178 : : }
4179 : :
4180 : 155 : FreeDir(xldir);
4181 : 155 : }
4182 : :
4183 : : /*
4184 : : * I/O routines for pg_control
4185 : : *
4186 : : * *ControlFile is a buffer in shared memory that holds an image of the
4187 : : * contents of pg_control. WriteControlFile() initializes pg_control
4188 : : * given a preloaded buffer, ReadControlFile() loads the buffer from
4189 : : * the pg_control file (during postmaster or standalone-backend startup),
4190 : : * and UpdateControlFile() rewrites pg_control after we modify xlog state.
4191 : : * InitControlFile() fills the buffer with initial values.
4192 : : *
4193 : : * For simplicity, WriteControlFile() initializes the fields of pg_control
4194 : : * that are related to checking backend/database compatibility, and
4195 : : * ReadControlFile() verifies they are correct. We could split out the
4196 : : * I/O and compatibility-check functions, but there seems no need currently.
4197 : : */
4198 : :
4199 : : static void
410 peter@eisentraut.org 4200 : 50 : InitControlFile(uint64 sysidentifier, uint32 data_checksum_version)
4201 : : {
4202 : : char mock_auth_nonce[MOCK_AUTH_NONCE_LEN];
4203 : :
4204 : : /*
4205 : : * Generate a random nonce. This is used for authentication requests that
4206 : : * will fail because the user does not exist. The nonce is used to create
4207 : : * a genuine-looking password challenge for the non-existent user, in lieu
4208 : : * of an actual stored password.
4209 : : */
1298 heikki.linnakangas@i 4210 [ - + ]: 50 : if (!pg_strong_random(mock_auth_nonce, MOCK_AUTH_NONCE_LEN))
1298 heikki.linnakangas@i 4211 [ # # ]:UBC 0 : ereport(PANIC,
4212 : : (errcode(ERRCODE_INTERNAL_ERROR),
4213 : : errmsg("could not generate secret authorization token")));
4214 : :
1298 heikki.linnakangas@i 4215 :CBC 50 : memset(ControlFile, 0, sizeof(ControlFileData));
4216 : : /* Initialize pg_control status fields */
4217 : 50 : ControlFile->system_identifier = sysidentifier;
4218 : 50 : memcpy(ControlFile->mock_authentication_nonce, mock_auth_nonce, MOCK_AUTH_NONCE_LEN);
4219 : 50 : ControlFile->state = DB_SHUTDOWNED;
4220 : 50 : ControlFile->unloggedLSN = FirstNormalUnloggedLSN;
4221 : :
4222 : : /* Set important parameter values for use when replaying WAL */
2028 peter@eisentraut.org 4223 : 50 : ControlFile->MaxConnections = MaxConnections;
4224 : 50 : ControlFile->max_worker_processes = max_worker_processes;
4225 : 50 : ControlFile->max_wal_senders = max_wal_senders;
4226 : 50 : ControlFile->max_prepared_xacts = max_prepared_xacts;
4227 : 50 : ControlFile->max_locks_per_xact = max_locks_per_xact;
4228 : 50 : ControlFile->wal_level = wal_level;
4229 : 50 : ControlFile->wal_log_hints = wal_log_hints;
4230 : 50 : ControlFile->track_commit_timestamp = track_commit_timestamp;
410 4231 : 50 : ControlFile->data_checksum_version = data_checksum_version;
2028 4232 : 50 : }
4233 : :
4234 : : static void
9051 tgl@sss.pgh.pa.us 4235 : 50 : WriteControlFile(void)
4236 : : {
4237 : : int fd;
4238 : : char buffer[PG_CONTROL_FILE_SIZE]; /* need not be aligned */
4239 : :
4240 : : /*
4241 : : * Initialize version and compatibility-check fields
4242 : : */
8943 4243 : 50 : ControlFile->pg_control_version = PG_CONTROL_VERSION;
4244 : 50 : ControlFile->catalog_version_no = CATALOG_VERSION_NO;
4245 : :
7278 4246 : 50 : ControlFile->maxAlign = MAXIMUM_ALIGNOF;
4247 : 50 : ControlFile->floatFormat = FLOATFORMAT_VALUE;
4248 : :
9051 4249 : 50 : ControlFile->blcksz = BLCKSZ;
4250 : 50 : ControlFile->relseg_size = RELSEG_SIZE;
7096 4251 : 50 : ControlFile->xlog_blcksz = XLOG_BLCKSZ;
2909 andres@anarazel.de 4252 : 50 : ControlFile->xlog_seg_size = wal_segment_size;
4253 : :
8539 lockhart@fourpalms.o 4254 : 50 : ControlFile->nameDataLen = NAMEDATALEN;
7466 tgl@sss.pgh.pa.us 4255 : 50 : ControlFile->indexMaxKeys = INDEX_MAX_KEYS;
4256 : :
6731 4257 : 50 : ControlFile->toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
4111 4258 : 50 : ControlFile->loblksize = LOBLKSIZE;
4259 : :
24 tgl@sss.pgh.pa.us 4260 :GNC 50 : ControlFile->float8ByVal = true; /* vestigial */
4261 : :
4262 : : /*
4263 : : * Initialize the default 'char' signedness.
4264 : : *
4265 : : * The signedness of the char type is implementation-defined. For instance
4266 : : * on x86 architecture CPUs, the char data type is typically treated as
4267 : : * signed by default, whereas on aarch architecture CPUs, it is typically
4268 : : * treated as unsigned by default. In v17 or earlier, we accidentally let
4269 : : * C implementation signedness affect persistent data. This led to
4270 : : * inconsistent results when comparing char data across different
4271 : : * platforms.
4272 : : *
4273 : : * This flag can be used as a hint to ensure consistent behavior for
4274 : : * pre-v18 data files that store data sorted by the 'char' type on disk,
4275 : : * especially in cross-platform replication scenarios.
4276 : : *
4277 : : * Newly created database clusters unconditionally set the default char
4278 : : * signedness to true. pg_upgrade changes this flag for clusters that were
4279 : : * initialized on signedness=false platforms. As a result,
4280 : : * signedness=false setting will become rare over time. If we had known
4281 : : * about this problem during the last development cycle that forced initdb
4282 : : * (v8.3), we would have made all clusters signed or all clusters
4283 : : * unsigned. Making pg_upgrade the only source of signedness=false will
4284 : : * cause the population of database clusters to converge toward that
4285 : : * retrospective ideal.
4286 : : */
197 msawada@postgresql.o 4287 :CBC 50 : ControlFile->default_char_signedness = true;
4288 : :
4289 : : /* Contents are protected with a CRC */
3959 heikki.linnakangas@i 4290 : 50 : INIT_CRC32C(ControlFile->crc);
4291 : 50 : COMP_CRC32C(ControlFile->crc,
4292 : : ControlFile,
4293 : : offsetof(ControlFileData, crc));
4294 : 50 : FIN_CRC32C(ControlFile->crc);
4295 : :
4296 : : /*
4297 : : * We write out PG_CONTROL_FILE_SIZE bytes into pg_control, zero-padding
4298 : : * the excess over sizeof(ControlFileData). This reduces the odds of
4299 : : * premature-EOF errors when reading pg_control. We'll still fail when we
4300 : : * check the contents of the file, but hopefully with a more specific
4301 : : * error than "couldn't read pg_control".
4302 : : */
2971 tgl@sss.pgh.pa.us 4303 : 50 : memset(buffer, 0, PG_CONTROL_FILE_SIZE);
9051 4304 : 50 : memcpy(buffer, ControlFile, sizeof(ControlFileData));
4305 : :
7369 4306 : 50 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4307 : : O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
9051 4308 [ - + ]: 50 : if (fd < 0)
8083 tgl@sss.pgh.pa.us 4309 [ # # ]:UBC 0 : ereport(PANIC,
4310 : : (errcode_for_file_access(),
4311 : : errmsg("could not create file \"%s\": %m",
4312 : : XLOG_CONTROL_FILE)));
4313 : :
8858 tgl@sss.pgh.pa.us 4314 :CBC 50 : errno = 0;
3094 rhaas@postgresql.org 4315 : 50 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE);
2971 tgl@sss.pgh.pa.us 4316 [ - + ]: 50 : if (write(fd, buffer, PG_CONTROL_FILE_SIZE) != PG_CONTROL_FILE_SIZE)
4317 : : {
4318 : : /* if write didn't set errno, assume problem is no disk space */
8858 tgl@sss.pgh.pa.us 4319 [ # # ]:UBC 0 : if (errno == 0)
4320 : 0 : errno = ENOSPC;
8083 4321 [ # # ]: 0 : ereport(PANIC,
4322 : : (errcode_for_file_access(),
4323 : : errmsg("could not write to file \"%s\": %m",
4324 : : XLOG_CONTROL_FILE)));
4325 : : }
3094 rhaas@postgresql.org 4326 :CBC 50 : pgstat_report_wait_end();
4327 : :
4328 : 50 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC);
9038 tgl@sss.pgh.pa.us 4329 [ - + ]: 50 : if (pg_fsync(fd) != 0)
8083 tgl@sss.pgh.pa.us 4330 [ # # ]:UBC 0 : ereport(PANIC,
4331 : : (errcode_for_file_access(),
4332 : : errmsg("could not fsync file \"%s\": %m",
4333 : : XLOG_CONTROL_FILE)));
3094 rhaas@postgresql.org 4334 :CBC 50 : pgstat_report_wait_end();
4335 : :
2254 peter@eisentraut.org 4336 [ - + ]: 50 : if (close(fd) != 0)
7894 tgl@sss.pgh.pa.us 4337 [ # # ]:UBC 0 : ereport(PANIC,
4338 : : (errcode_for_file_access(),
4339 : : errmsg("could not close file \"%s\": %m",
4340 : : XLOG_CONTROL_FILE)));
9051 tgl@sss.pgh.pa.us 4341 :CBC 50 : }
4342 : :
4343 : : static void
4344 : 937 : ReadControlFile(void)
4345 : : {
4346 : : pg_crc32c crc;
4347 : : int fd;
4348 : : char wal_segsz_str[20];
4349 : : int r;
4350 : :
4351 : : /*
4352 : : * Read data...
4353 : : */
7369 4354 : 937 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4355 : : O_RDWR | PG_BINARY);
9051 4356 [ - + ]: 937 : if (fd < 0)
8083 tgl@sss.pgh.pa.us 4357 [ # # ]:UBC 0 : ereport(PANIC,
4358 : : (errcode_for_file_access(),
4359 : : errmsg("could not open file \"%s\": %m",
4360 : : XLOG_CONTROL_FILE)));
4361 : :
3094 rhaas@postgresql.org 4362 :CBC 937 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_READ);
2668 magnus@hagander.net 4363 : 937 : r = read(fd, ControlFile, sizeof(ControlFileData));
4364 [ - + ]: 937 : if (r != sizeof(ControlFileData))
4365 : : {
2668 magnus@hagander.net 4366 [ # # ]:UBC 0 : if (r < 0)
4367 [ # # ]: 0 : ereport(PANIC,
4368 : : (errcode_for_file_access(),
4369 : : errmsg("could not read file \"%s\": %m",
4370 : : XLOG_CONTROL_FILE)));
4371 : : else
4372 [ # # ]: 0 : ereport(PANIC,
4373 : : (errcode(ERRCODE_DATA_CORRUPTED),
4374 : : errmsg("could not read file \"%s\": read %d of %zu",
4375 : : XLOG_CONTROL_FILE, r, sizeof(ControlFileData))));
4376 : : }
3094 rhaas@postgresql.org 4377 :CBC 937 : pgstat_report_wait_end();
4378 : :
9051 tgl@sss.pgh.pa.us 4379 : 937 : close(fd);
4380 : :
4381 : : /*
4382 : : * Check for expected pg_control format version. If this is wrong, the
4383 : : * CRC check will likely fail because we'll be checking the wrong number
4384 : : * of bytes. Complaining about wrong version will probably be more
4385 : : * enlightening than complaining about wrong CRC.
4386 : : */
4387 : :
6438 peter_e@gmx.net 4388 [ - + - - : 937 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION && ControlFile->pg_control_version % 65536 == 0 && ControlFile->pg_control_version / 65536 != 0)
- - ]
6438 peter_e@gmx.net 4389 [ # # ]:UBC 0 : ereport(FATAL,
4390 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4391 : : errmsg("database files are incompatible with server"),
4392 : : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
4393 : : " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
4394 : : ControlFile->pg_control_version, ControlFile->pg_control_version,
4395 : : PG_CONTROL_VERSION, PG_CONTROL_VERSION),
4396 : : errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
4397 : :
8943 tgl@sss.pgh.pa.us 4398 [ - + ]:CBC 937 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION)
8083 tgl@sss.pgh.pa.us 4399 [ # # ]:UBC 0 : ereport(FATAL,
4400 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4401 : : errmsg("database files are incompatible with server"),
4402 : : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
4403 : : " but the server was compiled with PG_CONTROL_VERSION %d.",
4404 : : ControlFile->pg_control_version, PG_CONTROL_VERSION),
4405 : : errhint("It looks like you need to initdb.")));
4406 : :
4407 : : /* Now check the CRC. */
3959 heikki.linnakangas@i 4408 :CBC 937 : INIT_CRC32C(crc);
4409 : 937 : COMP_CRC32C(crc,
4410 : : ControlFile,
4411 : : offsetof(ControlFileData, crc));
4412 : 937 : FIN_CRC32C(crc);
4413 : :
4414 [ - + ]: 937 : if (!EQ_CRC32C(crc, ControlFile->crc))
8083 tgl@sss.pgh.pa.us 4415 [ # # ]:UBC 0 : ereport(FATAL,
4416 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4417 : : errmsg("incorrect checksum in control file")));
4418 : :
4419 : : /*
4420 : : * Do compatibility checking immediately. If the database isn't
4421 : : * compatible with the backend executable, we want to abort before we can
4422 : : * possibly do any damage.
4423 : : */
8943 tgl@sss.pgh.pa.us 4424 [ - + ]:CBC 937 : if (ControlFile->catalog_version_no != CATALOG_VERSION_NO)
8083 tgl@sss.pgh.pa.us 4425 [ # # ]:UBC 0 : ereport(FATAL,
4426 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4427 : : errmsg("database files are incompatible with server"),
4428 : : /* translator: %s is a variable name and %d is its value */
4429 : : errdetail("The database cluster was initialized with %s %d,"
4430 : : " but the server was compiled with %s %d.",
4431 : : "CATALOG_VERSION_NO", ControlFile->catalog_version_no,
4432 : : "CATALOG_VERSION_NO", CATALOG_VERSION_NO),
4433 : : errhint("It looks like you need to initdb.")));
7278 tgl@sss.pgh.pa.us 4434 [ - + ]:CBC 937 : if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
7278 tgl@sss.pgh.pa.us 4435 [ # # ]:UBC 0 : ereport(FATAL,
4436 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4437 : : errmsg("database files are incompatible with server"),
4438 : : /* translator: %s is a variable name and %d is its value */
4439 : : errdetail("The database cluster was initialized with %s %d,"
4440 : : " but the server was compiled with %s %d.",
4441 : : "MAXALIGN", ControlFile->maxAlign,
4442 : : "MAXALIGN", MAXIMUM_ALIGNOF),
4443 : : errhint("It looks like you need to initdb.")));
7278 tgl@sss.pgh.pa.us 4444 [ - + ]:CBC 937 : if (ControlFile->floatFormat != FLOATFORMAT_VALUE)
7278 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 : : errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
4449 : : errhint("It looks like you need to initdb.")));
9051 tgl@sss.pgh.pa.us 4450 [ - + ]:CBC 937 : if (ControlFile->blcksz != BLCKSZ)
8083 tgl@sss.pgh.pa.us 4451 [ # # ]:UBC 0 : ereport(FATAL,
4452 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4453 : : errmsg("database files are incompatible with server"),
4454 : : /* translator: %s is a variable name and %d is its value */
4455 : : errdetail("The database cluster was initialized with %s %d,"
4456 : : " but the server was compiled with %s %d.",
4457 : : "BLCKSZ", ControlFile->blcksz,
4458 : : "BLCKSZ", BLCKSZ),
4459 : : errhint("It looks like you need to recompile or initdb.")));
9051 tgl@sss.pgh.pa.us 4460 [ - + ]:CBC 937 : if (ControlFile->relseg_size != RELSEG_SIZE)
8083 tgl@sss.pgh.pa.us 4461 [ # # ]:UBC 0 : ereport(FATAL,
4462 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4463 : : errmsg("database files are incompatible with server"),
4464 : : /* translator: %s is a variable name and %d is its value */
4465 : : errdetail("The database cluster was initialized with %s %d,"
4466 : : " but the server was compiled with %s %d.",
4467 : : "RELSEG_SIZE", ControlFile->relseg_size,
4468 : : "RELSEG_SIZE", RELSEG_SIZE),
4469 : : errhint("It looks like you need to recompile or initdb.")));
7096 tgl@sss.pgh.pa.us 4470 [ - + ]:CBC 937 : if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
7096 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 : : "XLOG_BLCKSZ", ControlFile->xlog_blcksz,
4478 : : "XLOG_BLCKSZ", XLOG_BLCKSZ),
4479 : : errhint("It looks like you need to recompile or initdb.")));
8539 lockhart@fourpalms.o 4480 [ - + ]:CBC 937 : if (ControlFile->nameDataLen != NAMEDATALEN)
8083 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 : : "NAMEDATALEN", ControlFile->nameDataLen,
4488 : : "NAMEDATALEN", NAMEDATALEN),
4489 : : errhint("It looks like you need to recompile or initdb.")));
7466 tgl@sss.pgh.pa.us 4490 [ - + ]:CBC 937 : if (ControlFile->indexMaxKeys != INDEX_MAX_KEYS)
8083 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 : : "INDEX_MAX_KEYS", ControlFile->indexMaxKeys,
4498 : : "INDEX_MAX_KEYS", INDEX_MAX_KEYS),
4499 : : errhint("It looks like you need to recompile or initdb.")));
6731 tgl@sss.pgh.pa.us 4500 [ - + ]:CBC 937 : if (ControlFile->toast_max_chunk_size != TOAST_MAX_CHUNK_SIZE)
6731 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 : : "TOAST_MAX_CHUNK_SIZE", ControlFile->toast_max_chunk_size,
4508 : : "TOAST_MAX_CHUNK_SIZE", (int) TOAST_MAX_CHUNK_SIZE),
4509 : : errhint("It looks like you need to recompile or initdb.")));
4111 tgl@sss.pgh.pa.us 4510 [ - + ]:CBC 937 : if (ControlFile->loblksize != LOBLKSIZE)
4111 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 : : "LOBLKSIZE", ControlFile->loblksize,
4518 : : "LOBLKSIZE", (int) LOBLKSIZE),
4519 : : errhint("It looks like you need to recompile or initdb.")));
4520 : :
24 tgl@sss.pgh.pa.us 4521 [ - + ]:GNC 937 : Assert(ControlFile->float8ByVal); /* vestigial, not worth an error msg */
4522 : :
2909 andres@anarazel.de 4523 :CBC 937 : wal_segment_size = ControlFile->xlog_seg_size;
4524 : :
4525 [ + - + - : 937 : if (!IsValidWalSegSize(wal_segment_size))
+ - - + ]
2909 andres@anarazel.de 4526 [ # # ]:UBC 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4527 : : errmsg_plural("invalid WAL segment size in control file (%d byte)",
4528 : : "invalid WAL segment size in control file (%d bytes)",
4529 : : wal_segment_size,
4530 : : wal_segment_size),
4531 : : errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.")));
4532 : :
2909 andres@anarazel.de 4533 :CBC 937 : snprintf(wal_segsz_str, sizeof(wal_segsz_str), "%d", wal_segment_size);
4534 : 937 : SetConfigOption("wal_segment_size", wal_segsz_str, PGC_INTERNAL,
4535 : : PGC_S_DYNAMIC_DEFAULT);
4536 : :
4537 : : /* check and update variables dependent on wal_segment_size */
4538 [ - + ]: 937 : if (ConvertToXSegs(min_wal_size_mb, wal_segment_size) < 2)
2909 andres@anarazel.de 4539 [ # # ]:UBC 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4540 : : /* translator: both %s are GUC names */
4541 : : errmsg("\"%s\" must be at least twice \"%s\"",
4542 : : "min_wal_size", "wal_segment_size")));
4543 : :
2909 andres@anarazel.de 4544 [ - + ]:CBC 937 : if (ConvertToXSegs(max_wal_size_mb, wal_segment_size) < 2)
2909 andres@anarazel.de 4545 [ # # ]:UBC 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4546 : : /* translator: both %s are GUC names */
4547 : : errmsg("\"%s\" must be at least twice \"%s\"",
4548 : : "max_wal_size", "wal_segment_size")));
4549 : :
2909 andres@anarazel.de 4550 :CBC 937 : UsableBytesInSegment =
4551 : 937 : (wal_segment_size / XLOG_BLCKSZ * UsableBytesInPage) -
4552 : : (SizeOfXLogLongPHD - SizeOfXLogShortPHD);
4553 : :
4554 : 937 : CalculateCheckpointSegments();
4555 : :
4556 : : /* Make the initdb settings visible as GUC variables, too */
2707 magnus@hagander.net 4557 [ + + ]: 937 : SetConfigOption("data_checksums", DataChecksumsEnabled() ? "yes" : "no",
4558 : : PGC_INTERNAL, PGC_S_DYNAMIC_DEFAULT);
9051 tgl@sss.pgh.pa.us 4559 : 937 : }
4560 : :
4561 : : /*
4562 : : * Utility wrapper to update the control file. Note that the control
4563 : : * file gets flushed.
4564 : : */
4565 : : static void
4566 : 8846 : UpdateControlFile(void)
4567 : : {
2350 peter@eisentraut.org 4568 : 8846 : update_controlfile(DataDir, ControlFile, true);
9467 vadim4o@yahoo.com 4569 : 8846 : }
4570 : :
4571 : : /*
4572 : : * Returns the unique system identifier from control file.
4573 : : */
4574 : : uint64
5713 heikki.linnakangas@i 4575 : 1345 : GetSystemIdentifier(void)
4576 : : {
4577 [ - + ]: 1345 : Assert(ControlFile != NULL);
4578 : 1345 : return ControlFile->system_identifier;
4579 : : }
4580 : :
4581 : : /*
4582 : : * Returns the random nonce from control file.
4583 : : */
4584 : : char *
3105 4585 : 1 : GetMockAuthenticationNonce(void)
4586 : : {
4587 [ - + ]: 1 : Assert(ControlFile != NULL);
4588 : 1 : return ControlFile->mock_authentication_nonce;
4589 : : }
4590 : :
4591 : : /*
4592 : : * Are checksums enabled for data pages?
4593 : : */
4594 : : bool
2707 magnus@hagander.net 4595 : 8916351 : DataChecksumsEnabled(void)
4596 : : {
4551 simon@2ndQuadrant.co 4597 [ - + ]: 8916351 : Assert(ControlFile != NULL);
4512 4598 : 8916351 : return (ControlFile->data_checksum_version > 0);
4599 : : }
4600 : :
4601 : : /*
4602 : : * Return true if the cluster was initialized on a platform where the
4603 : : * default signedness of char is "signed". This function exists for code
4604 : : * that deals with pre-v18 data files that store data sorted by the 'char'
4605 : : * type on disk (e.g., GIN and GiST indexes). See the comments in
4606 : : * WriteControlFile() for details.
4607 : : */
4608 : : bool
197 msawada@postgresql.o 4609 : 3 : GetDefaultCharSignedness(void)
4610 : : {
4611 : 3 : return ControlFile->default_char_signedness;
4612 : : }
4613 : :
4614 : : /*
4615 : : * Returns a fake LSN for unlogged relations.
4616 : : *
4617 : : * Each call generates an LSN that is greater than any previous value
4618 : : * returned. The current counter value is saved and restored across clean
4619 : : * shutdowns, but like unlogged relations, does not survive a crash. This can
4620 : : * be used in lieu of real LSN values returned by XLogInsert, if you need an
4621 : : * LSN-like increasing sequence of numbers without writing any WAL.
4622 : : */
4623 : : XLogRecPtr
4590 heikki.linnakangas@i 4624 : 33 : GetFakeLSNForUnloggedRel(void)
4625 : : {
555 nathan@postgresql.or 4626 : 33 : return pg_atomic_fetch_add_u64(&XLogCtl->unloggedLSN, 1);
4627 : : }
4628 : :
4629 : : /*
4630 : : * Auto-tune the number of XLOG buffers.
4631 : : *
4632 : : * The preferred setting for wal_buffers is about 3% of shared_buffers, with
4633 : : * a maximum of one XLOG segment (there is little reason to think that more
4634 : : * is helpful, at least so long as we force an fsync when switching log files)
4635 : : * and a minimum of 8 blocks (which was the default value prior to PostgreSQL
4636 : : * 9.1, when auto-tuning was added).
4637 : : *
4638 : : * This should not be called until NBuffers has received its final value.
4639 : : */
4640 : : static int
5266 tgl@sss.pgh.pa.us 4641 : 1028 : XLOGChooseNumBuffers(void)
4642 : : {
4643 : : int xbuffers;
4644 : :
4645 : 1028 : xbuffers = NBuffers / 32;
2909 andres@anarazel.de 4646 [ + + ]: 1028 : if (xbuffers > (wal_segment_size / XLOG_BLCKSZ))
4647 : 24 : xbuffers = (wal_segment_size / XLOG_BLCKSZ);
5266 tgl@sss.pgh.pa.us 4648 [ + + ]: 1028 : if (xbuffers < 8)
4649 : 409 : xbuffers = 8;
4650 : 1028 : return xbuffers;
4651 : : }
4652 : :
4653 : : /*
4654 : : * GUC check_hook for wal_buffers
4655 : : */
4656 : : bool
4657 : 2095 : check_wal_buffers(int *newval, void **extra, GucSource source)
4658 : : {
4659 : : /*
4660 : : * -1 indicates a request for auto-tune.
4661 : : */
4662 [ + + ]: 2095 : if (*newval == -1)
4663 : : {
4664 : : /*
4665 : : * If we haven't yet changed the boot_val default of -1, just let it
4666 : : * be. We'll fix it when XLOGShmemSize is called.
4667 : : */
4668 [ + - ]: 1067 : if (XLOGbuffers == -1)
4669 : 1067 : return true;
4670 : :
4671 : : /* Otherwise, substitute the auto-tune value */
5266 tgl@sss.pgh.pa.us 4672 :UBC 0 : *newval = XLOGChooseNumBuffers();
4673 : : }
4674 : :
4675 : : /*
4676 : : * We clamp manually-set values to at least 4 blocks. Prior to PostgreSQL
4677 : : * 9.1, a minimum of 4 was enforced by guc.c, but since that is no longer
4678 : : * the case, we just silently treat such values as a request for the
4679 : : * minimum. (We could throw an error instead, but that doesn't seem very
4680 : : * helpful.)
4681 : : */
5266 tgl@sss.pgh.pa.us 4682 [ - + ]:CBC 1028 : if (*newval < 4)
5266 tgl@sss.pgh.pa.us 4683 :UBC 0 : *newval = 4;
4684 : :
5266 tgl@sss.pgh.pa.us 4685 :CBC 1028 : return true;
4686 : : }
4687 : :
4688 : : /*
4689 : : * GUC check_hook for wal_consistency_checking
4690 : : */
4691 : : bool
1089 4692 : 1971 : check_wal_consistency_checking(char **newval, void **extra, GucSource source)
4693 : : {
4694 : : char *rawstring;
4695 : : List *elemlist;
4696 : : ListCell *l;
4697 : : bool newwalconsistency[RM_MAX_ID + 1];
4698 : :
4699 : : /* Initialize the array */
4700 [ + - + - : 65043 : MemSet(newwalconsistency, 0, (RM_MAX_ID + 1) * sizeof(bool));
+ - + - +
+ ]
4701 : :
4702 : : /* Need a modifiable copy of string */
4703 : 1971 : rawstring = pstrdup(*newval);
4704 : :
4705 : : /* Parse string into list of identifiers */
4706 [ - + ]: 1971 : if (!SplitIdentifierString(rawstring, ',', &elemlist))
4707 : : {
4708 : : /* syntax error in list */
1089 tgl@sss.pgh.pa.us 4709 :UBC 0 : GUC_check_errdetail("List syntax is invalid.");
4710 : 0 : pfree(rawstring);
4711 : 0 : list_free(elemlist);
4712 : 0 : return false;
4713 : : }
4714 : :
1089 tgl@sss.pgh.pa.us 4715 [ + + + + :CBC 2426 : foreach(l, elemlist)
+ + ]
4716 : : {
4717 : 455 : char *tok = (char *) lfirst(l);
4718 : : int rmid;
4719 : :
4720 : : /* Check for 'all'. */
4721 [ + + ]: 455 : if (pg_strcasecmp(tok, "all") == 0)
4722 : : {
4723 [ + + ]: 116421 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4724 [ + + + + ]: 115968 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL)
4725 : 4530 : newwalconsistency[rmid] = true;
4726 : : }
4727 : : else
4728 : : {
4729 : : /* Check if the token matches any known resource manager. */
4730 : 2 : bool found = false;
4731 : :
4732 [ + - ]: 36 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4733 : : {
4734 [ + - + + : 54 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL &&
+ + ]
4735 : 18 : pg_strcasecmp(tok, GetRmgr(rmid).rm_name) == 0)
4736 : : {
4737 : 2 : newwalconsistency[rmid] = true;
4738 : 2 : found = true;
4739 : 2 : break;
4740 : : }
4741 : : }
4742 [ - + ]: 2 : if (!found)
4743 : : {
4744 : : /*
4745 : : * During startup, it might be a not-yet-loaded custom
4746 : : * resource manager. Defer checking until
4747 : : * InitializeWalConsistencyChecking().
4748 : : */
1089 tgl@sss.pgh.pa.us 4749 [ # # ]:UBC 0 : if (!process_shared_preload_libraries_done)
4750 : : {
4751 : 0 : check_wal_consistency_checking_deferred = true;
4752 : : }
4753 : : else
4754 : : {
4755 : 0 : GUC_check_errdetail("Unrecognized key word: \"%s\".", tok);
4756 : 0 : pfree(rawstring);
4757 : 0 : list_free(elemlist);
4758 : 0 : return false;
4759 : : }
4760 : : }
4761 : : }
4762 : : }
4763 : :
1089 tgl@sss.pgh.pa.us 4764 :CBC 1971 : pfree(rawstring);
4765 : 1971 : list_free(elemlist);
4766 : :
4767 : : /* assign new value */
163 dgustafsson@postgres 4768 : 1971 : *extra = guc_malloc(LOG, (RM_MAX_ID + 1) * sizeof(bool));
4769 [ - + ]: 1971 : if (!*extra)
163 dgustafsson@postgres 4770 :UBC 0 : return false;
1089 tgl@sss.pgh.pa.us 4771 :CBC 1971 : memcpy(*extra, newwalconsistency, (RM_MAX_ID + 1) * sizeof(bool));
4772 : 1971 : return true;
4773 : : }
4774 : :
4775 : : /*
4776 : : * GUC assign_hook for wal_consistency_checking
4777 : : */
4778 : : void
4779 : 1970 : assign_wal_consistency_checking(const char *newval, void *extra)
4780 : : {
4781 : : /*
4782 : : * If some checks were deferred, it's possible that the checks will fail
4783 : : * later during InitializeWalConsistencyChecking(). But in that case, the
4784 : : * postmaster will exit anyway, so it's safe to proceed with the
4785 : : * assignment.
4786 : : *
4787 : : * Any built-in resource managers specified are assigned immediately,
4788 : : * which affects WAL created before shared_preload_libraries are
4789 : : * processed. Any custom resource managers specified won't be assigned
4790 : : * until after shared_preload_libraries are processed, but that's OK
4791 : : * because WAL for a custom resource manager can't be written before the
4792 : : * module is loaded anyway.
4793 : : */
4794 : 1970 : wal_consistency_checking = extra;
4795 : 1970 : }
4796 : :
4797 : : /*
4798 : : * InitializeWalConsistencyChecking: run after loading custom resource managers
4799 : : *
4800 : : * If any unknown resource managers were specified in the
4801 : : * wal_consistency_checking GUC, processing was deferred. Now that
4802 : : * shared_preload_libraries have been loaded, process wal_consistency_checking
4803 : : * again.
4804 : : */
4805 : : void
4806 : 878 : InitializeWalConsistencyChecking(void)
4807 : : {
4808 [ - + ]: 878 : Assert(process_shared_preload_libraries_done);
4809 : :
4810 [ - + ]: 878 : if (check_wal_consistency_checking_deferred)
4811 : : {
4812 : : struct config_generic *guc;
4813 : :
1089 tgl@sss.pgh.pa.us 4814 :UBC 0 : guc = find_option("wal_consistency_checking", false, false, ERROR);
4815 : :
4816 : 0 : check_wal_consistency_checking_deferred = false;
4817 : :
4818 : 0 : set_config_option_ext("wal_consistency_checking",
4819 : : wal_consistency_checking_string,
4820 : : guc->scontext, guc->source, guc->srole,
4821 : : GUC_ACTION_SET, true, ERROR, false);
4822 : :
4823 : : /* checking should not be deferred again */
4824 [ # # ]: 0 : Assert(!check_wal_consistency_checking_deferred);
4825 : : }
1089 tgl@sss.pgh.pa.us 4826 :CBC 878 : }
4827 : :
4828 : : /*
4829 : : * GUC show_hook for archive_command
4830 : : */
4831 : : const char *
4832 : 1694 : show_archive_command(void)
4833 : : {
4834 [ + + - + : 1694 : if (XLogArchivingActive())
+ + ]
4835 : 2 : return XLogArchiveCommand;
4836 : : else
4837 : 1692 : return "(disabled)";
4838 : : }
4839 : :
4840 : : /*
4841 : : * GUC show_hook for in_hot_standby
4842 : : */
4843 : : const char *
4844 : 13781 : show_in_hot_standby(void)
4845 : : {
4846 : : /*
4847 : : * We display the actual state based on shared memory, so that this GUC
4848 : : * reports up-to-date state if examined intra-query. The underlying
4849 : : * variable (in_hot_standby_guc) changes only when we transmit a new value
4850 : : * to the client.
4851 : : */
4852 [ + + ]: 13781 : return RecoveryInProgress() ? "on" : "off";
4853 : : }
4854 : :
4855 : : /*
4856 : : * Read the control file, set respective GUCs.
4857 : : *
4858 : : * This is to be called during startup, including a crash recovery cycle,
4859 : : * unless in bootstrap mode, where no control file yet exists. As there's no
4860 : : * usable shared memory yet (its sizing can depend on the contents of the
4861 : : * control file!), first store the contents in local memory. XLOGShmemInit()
4862 : : * will then copy it to shared memory later.
4863 : : *
4864 : : * reset just controls whether previous contents are to be expected (in the
4865 : : * reset case, there's a dangling pointer into old shared memory), or not.
4866 : : */
4867 : : void
2911 andres@anarazel.de 4868 : 887 : LocalProcessControlFile(bool reset)
4869 : : {
4870 [ + + - + ]: 887 : Assert(reset || ControlFile == NULL);
2915 4871 : 887 : ControlFile = palloc(sizeof(ControlFileData));
4872 : 887 : ReadControlFile();
4873 : 887 : }
4874 : :
4875 : : /*
4876 : : * Get the wal_level from the control file. For a standby, this value should be
4877 : : * considered as its active wal_level, because it may be different from what
4878 : : * was originally configured on standby.
4879 : : */
4880 : : WalLevel
882 4881 : 1 : GetActiveWalLevelOnStandby(void)
4882 : : {
4883 : 1 : return ControlFile->wal_level;
4884 : : }
4885 : :
4886 : : /*
4887 : : * Initialization of shared memory for XLOG
4888 : : */
4889 : : Size
9055 peter_e@gmx.net 4890 : 2938 : XLOGShmemSize(void)
4891 : : {
4892 : : Size size;
4893 : :
4894 : : /*
4895 : : * If the value of wal_buffers is -1, use the preferred auto-tune value.
4896 : : * This isn't an amazingly clean place to do this, but we must wait till
4897 : : * NBuffers has received its final value, and must do it before using the
4898 : : * value of XLOGbuffers to do anything important.
4899 : : *
4900 : : * We prefer to report this value's source as PGC_S_DYNAMIC_DEFAULT.
4901 : : * However, if the DBA explicitly set wal_buffers = -1 in the config file,
4902 : : * then PGC_S_DYNAMIC_DEFAULT will fail to override that and we must force
4903 : : * the matter with PGC_S_OVERRIDE.
4904 : : */
5266 tgl@sss.pgh.pa.us 4905 [ + + ]: 2938 : if (XLOGbuffers == -1)
4906 : : {
4907 : : char buf[32];
4908 : :
4909 : 1028 : snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
1186 4910 : 1028 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4911 : : PGC_S_DYNAMIC_DEFAULT);
4912 [ - + ]: 1028 : if (XLOGbuffers == -1) /* failed to apply it? */
1186 tgl@sss.pgh.pa.us 4913 :UBC 0 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4914 : : PGC_S_OVERRIDE);
4915 : : }
5341 tgl@sss.pgh.pa.us 4916 [ - + ]:CBC 2938 : Assert(XLOGbuffers > 0);
4917 : :
4918 : : /* XLogCtl */
7322 4919 : 2938 : size = sizeof(XLogCtlData);
4920 : :
4921 : : /* WAL insertion locks, plus alignment */
3993 heikki.linnakangas@i 4922 : 2938 : size = add_size(size, mul_size(sizeof(WALInsertLockPadded), NUM_XLOGINSERT_LOCKS + 1));
4923 : : /* xlblocks array */
627 jdavis@postgresql.or 4924 : 2938 : size = add_size(size, mul_size(sizeof(pg_atomic_uint64), XLOGbuffers));
4925 : : /* extra alignment padding for XLOG I/O buffers */
882 tmunro@postgresql.or 4926 : 2938 : size = add_size(size, Max(XLOG_BLCKSZ, PG_IO_ALIGN_SIZE));
4927 : : /* and the buffers themselves */
7096 tgl@sss.pgh.pa.us 4928 : 2938 : size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4929 : :
4930 : : /*
4931 : : * Note: we don't count ControlFileData, it comes out of the "slop factor"
4932 : : * added by CreateSharedMemoryAndSemaphores. This lets us use this
4933 : : * routine again below to compute the actual allocation size.
4934 : : */
4935 : :
7322 4936 : 2938 : return size;
4937 : : }
4938 : :
4939 : : void
9467 vadim4o@yahoo.com 4940 : 1029 : XLOGShmemInit(void)
4941 : : {
4942 : : bool foundCFile,
4943 : : foundXLog;
4944 : : char *allocptr;
4945 : : int i;
4946 : : ControlFileData *localControlFile;
4947 : :
4948 : : #ifdef WAL_DEBUG
4949 : :
4950 : : /*
4951 : : * Create a memory context for WAL debugging that's exempt from the normal
4952 : : * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
4953 : : * an allocation fails, but wal_debug is not for production use anyway.
4954 : : */
4955 : : if (walDebugCxt == NULL)
4956 : : {
4957 : : walDebugCxt = AllocSetContextCreate(TopMemoryContext,
4958 : : "WAL Debug",
4959 : : ALLOCSET_DEFAULT_SIZES);
4960 : : MemoryContextAllowInCriticalSection(walDebugCxt, true);
4961 : : }
4962 : : #endif
4963 : :
4964 : :
2911 andres@anarazel.de 4965 : 1029 : XLogCtl = (XLogCtlData *)
4966 : 1029 : ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
4967 : :
2915 4968 : 1029 : localControlFile = ControlFile;
9051 tgl@sss.pgh.pa.us 4969 : 1029 : ControlFile = (ControlFileData *)
7931 bruce@momjian.us 4970 : 1029 : ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
4971 : :
7320 tgl@sss.pgh.pa.us 4972 [ + - - + ]: 1029 : if (foundCFile || foundXLog)
4973 : : {
4974 : : /* both should be present or neither */
7320 tgl@sss.pgh.pa.us 4975 [ # # # # ]:UBC 0 : Assert(foundCFile && foundXLog);
4976 : :
4977 : : /* Initialize local copy of WALInsertLocks */
4062 rhaas@postgresql.org 4978 : 0 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks;
4979 : :
2911 andres@anarazel.de 4980 [ # # ]: 0 : if (localControlFile)
4981 : 0 : pfree(localControlFile);
7931 bruce@momjian.us 4982 : 0 : return;
4983 : : }
8943 tgl@sss.pgh.pa.us 4984 :CBC 1029 : memset(XLogCtl, 0, sizeof(XLogCtlData));
4985 : :
4986 : : /*
4987 : : * Already have read control file locally, unless in bootstrap mode. Move
4988 : : * contents into shared memory.
4989 : : */
2911 andres@anarazel.de 4990 [ + + ]: 1029 : if (localControlFile)
4991 : : {
4992 : 879 : memcpy(ControlFile, localControlFile, sizeof(ControlFileData));
4993 : 879 : pfree(localControlFile);
4994 : : }
4995 : :
4996 : : /*
4997 : : * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
4998 : : * multiple of the alignment for same, so no extra alignment padding is
4999 : : * needed here.
5000 : : */
4443 heikki.linnakangas@i 5001 : 1029 : allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
627 jdavis@postgresql.or 5002 : 1029 : XLogCtl->xlblocks = (pg_atomic_uint64 *) allocptr;
5003 : 1029 : allocptr += sizeof(pg_atomic_uint64) * XLOGbuffers;
5004 : :
5005 [ + + ]: 287467 : for (i = 0; i < XLOGbuffers; i++)
5006 : : {
5007 : 286438 : pg_atomic_init_u64(&XLogCtl->xlblocks[i], InvalidXLogRecPtr);
5008 : : }
5009 : :
5010 : : /* WAL insertion locks. Ensure they're aligned to the full padded size */
4187 heikki.linnakangas@i 5011 : 1029 : allocptr += sizeof(WALInsertLockPadded) -
2999 tgl@sss.pgh.pa.us 5012 : 1029 : ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
4187 heikki.linnakangas@i 5013 : 1029 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks =
5014 : : (WALInsertLockPadded *) allocptr;
3993 5015 : 1029 : allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
5016 : :
5017 [ + + ]: 9261 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
5018 : : {
3553 rhaas@postgresql.org 5019 : 8232 : LWLockInitialize(&WALInsertLocks[i].l.lock, LWTRANCHE_WAL_INSERT);
774 michael@paquier.xyz 5020 : 8232 : pg_atomic_init_u64(&WALInsertLocks[i].l.insertingAt, InvalidXLogRecPtr);
3180 andres@anarazel.de 5021 : 8232 : WALInsertLocks[i].l.lastImportantAt = InvalidXLogRecPtr;
5022 : : }
5023 : :
5024 : : /*
5025 : : * Align the start of the page buffers to a full xlog block size boundary.
5026 : : * This simplifies some calculations in XLOG insertion. It is also
5027 : : * required for O_DIRECT.
5028 : : */
4443 heikki.linnakangas@i 5029 : 1029 : allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
7322 tgl@sss.pgh.pa.us 5030 : 1029 : XLogCtl->pages = allocptr;
7096 5031 : 1029 : memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
5032 : :
5033 : : /*
5034 : : * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
5035 : : * in additional info.)
5036 : : */
8943 5037 : 1029 : XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
1961 michael@paquier.xyz 5038 : 1029 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
1531 noah@leadboat.com 5039 : 1029 : XLogCtl->InstallXLogFileSegmentActive = false;
4869 tgl@sss.pgh.pa.us 5040 : 1029 : XLogCtl->WalWriterSleeping = false;
5041 : :
4443 heikki.linnakangas@i 5042 : 1029 : SpinLockInit(&XLogCtl->Insert.insertpos_lck);
8743 tgl@sss.pgh.pa.us 5043 : 1029 : SpinLockInit(&XLogCtl->info_lck);
517 alvherre@alvh.no-ip. 5044 : 1029 : pg_atomic_init_u64(&XLogCtl->logInsertResult, InvalidXLogRecPtr);
519 5045 : 1029 : pg_atomic_init_u64(&XLogCtl->logWriteResult, InvalidXLogRecPtr);
5046 : 1029 : pg_atomic_init_u64(&XLogCtl->logFlushResult, InvalidXLogRecPtr);
555 nathan@postgresql.or 5047 : 1029 : pg_atomic_init_u64(&XLogCtl->unloggedLSN, InvalidXLogRecPtr);
5048 : : }
5049 : :
5050 : : /*
5051 : : * This func must be called ONCE on system install. It creates pg_control
5052 : : * and the initial XLOG segment.
5053 : : */
5054 : : void
410 peter@eisentraut.org 5055 : 50 : BootStrapXLOG(uint32 data_checksum_version)
5056 : : {
5057 : : CheckPoint checkPoint;
5058 : : char *buffer;
5059 : : XLogPageHeader page;
5060 : : XLogLongPageHeader longpage;
5061 : : XLogRecord *record;
5062 : : char *recptr;
5063 : : uint64 sysidentifier;
5064 : : struct timeval tv;
5065 : : pg_crc32c crc;
5066 : :
5067 : : /* allow ordinary WAL segment creation, like StartupXLOG() would */
1116 michael@paquier.xyz 5068 : 50 : SetInstallXLogFileSegmentActive();
5069 : :
5070 : : /*
5071 : : * Select a hopefully-unique system identifier code for this installation.
5072 : : * We use the result of gettimeofday(), including the fractional seconds
5073 : : * field, as being about as unique as we can easily get. (Think not to
5074 : : * use random(), since it hasn't been seeded and there's no portable way
5075 : : * to seed it other than the system clock value...) The upper half of the
5076 : : * uint64 value is just the tv_sec part, while the lower half contains the
5077 : : * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
5078 : : * PID for a little extra uniqueness. A person knowing this encoding can
5079 : : * determine the initialization time of the installation, which could
5080 : : * perhaps be useful sometimes.
5081 : : */
7878 tgl@sss.pgh.pa.us 5082 : 50 : gettimeofday(&tv, NULL);
5083 : 50 : sysidentifier = ((uint64) tv.tv_sec) << 32;
4151 5084 : 50 : sysidentifier |= ((uint64) tv.tv_usec) << 12;
5085 : 50 : sysidentifier |= getpid() & 0xFFF;
5086 : :
5087 : : /* page buffer must be aligned suitably for O_DIRECT */
4443 heikki.linnakangas@i 5088 : 50 : buffer = (char *) palloc(XLOG_BLCKSZ + XLOG_BLCKSZ);
5089 : 50 : page = (XLogPageHeader) TYPEALIGN(XLOG_BLCKSZ, buffer);
7096 tgl@sss.pgh.pa.us 5090 : 50 : memset(page, 0, XLOG_BLCKSZ);
5091 : :
5092 : : /*
5093 : : * Set up information for the initial checkpoint record
5094 : : *
5095 : : * The initial checkpoint record is written to the beginning of the WAL
5096 : : * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
5097 : : * used, so that we can use 0/0 to mean "before any valid WAL segment".
5098 : : */
2909 andres@anarazel.de 5099 : 50 : checkPoint.redo = wal_segment_size + SizeOfXLogLongPHD;
1401 rhaas@postgresql.org 5100 : 50 : checkPoint.ThisTimeLineID = BootstrapTimeLineID;
5101 : 50 : checkPoint.PrevTimeLineID = BootstrapTimeLineID;
4973 simon@2ndQuadrant.co 5102 : 50 : checkPoint.fullPageWrites = fullPageWrites;
411 rhaas@postgresql.org 5103 : 50 : checkPoint.wal_level = wal_level;
5104 : : checkPoint.nextXid =
2354 tmunro@postgresql.or 5105 : 50 : FullTransactionIdFromEpochAndXid(0, FirstNormalTransactionId);
1514 tgl@sss.pgh.pa.us 5106 : 50 : checkPoint.nextOid = FirstGenbkiObjectId;
7436 5107 : 50 : checkPoint.nextMulti = FirstMultiXactId;
7395 5108 : 50 : checkPoint.nextMultiOffset = 0;
5850 5109 : 50 : checkPoint.oldestXid = FirstNormalTransactionId;
1234 5110 : 50 : checkPoint.oldestXidDB = Template1DbOid;
4609 alvherre@alvh.no-ip. 5111 : 50 : checkPoint.oldestMulti = FirstMultiXactId;
1234 tgl@sss.pgh.pa.us 5112 : 50 : checkPoint.oldestMultiDB = Template1DbOid;
3540 mail@joeconway.com 5113 : 50 : checkPoint.oldestCommitTsXid = InvalidTransactionId;
5114 : 50 : checkPoint.newestCommitTsXid = InvalidTransactionId;
6411 tgl@sss.pgh.pa.us 5115 : 50 : checkPoint.time = (pg_time_t) time(NULL);
5740 simon@2ndQuadrant.co 5116 : 50 : checkPoint.oldestActiveXid = InvalidTransactionId;
5117 : :
638 heikki.linnakangas@i 5118 : 50 : TransamVariables->nextXid = checkPoint.nextXid;
5119 : 50 : TransamVariables->nextOid = checkPoint.nextOid;
5120 : 50 : TransamVariables->oidCount = 0;
7395 tgl@sss.pgh.pa.us 5121 : 50 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
3089 rhaas@postgresql.org 5122 : 50 : AdvanceOldestClogXid(checkPoint.oldestXid);
5680 tgl@sss.pgh.pa.us 5123 : 50 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
3098 5124 : 50 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
3930 alvherre@alvh.no-ip. 5125 : 50 : SetCommitTsLimit(InvalidTransactionId, InvalidTransactionId);
5126 : :
5127 : : /* Set up the XLOG page header */
9467 vadim4o@yahoo.com 5128 : 50 : page->xlp_magic = XLOG_PAGE_MAGIC;
7717 tgl@sss.pgh.pa.us 5129 : 50 : page->xlp_info = XLP_LONG_HEADER;
1401 rhaas@postgresql.org 5130 : 50 : page->xlp_tli = BootstrapTimeLineID;
2909 andres@anarazel.de 5131 : 50 : page->xlp_pageaddr = wal_segment_size;
7717 tgl@sss.pgh.pa.us 5132 : 50 : longpage = (XLogLongPageHeader) page;
5133 : 50 : longpage->xlp_sysid = sysidentifier;
2909 andres@anarazel.de 5134 : 50 : longpage->xlp_seg_size = wal_segment_size;
7094 tgl@sss.pgh.pa.us 5135 : 50 : longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5136 : :
5137 : : /* Insert the initial checkpoint record */
3943 heikki.linnakangas@i 5138 : 50 : recptr = ((char *) page + SizeOfXLogLongPHD);
5139 : 50 : record = (XLogRecord *) recptr;
4822 5140 : 50 : record->xl_prev = 0;
9467 vadim4o@yahoo.com 5141 : 50 : record->xl_xid = InvalidTransactionId;
3943 heikki.linnakangas@i 5142 : 50 : record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
8943 tgl@sss.pgh.pa.us 5143 : 50 : record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
9467 vadim4o@yahoo.com 5144 : 50 : record->xl_rmid = RM_XLOG_ID;
3943 heikki.linnakangas@i 5145 : 50 : recptr += SizeOfXLogRecord;
5146 : : /* fill the XLogRecordDataHeaderShort struct */
3084 tgl@sss.pgh.pa.us 5147 : 50 : *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
3943 heikki.linnakangas@i 5148 : 50 : *(recptr++) = sizeof(checkPoint);
5149 : 50 : memcpy(recptr, &checkPoint, sizeof(checkPoint));
5150 : 50 : recptr += sizeof(checkPoint);
5151 [ - + ]: 50 : Assert(recptr - (char *) record == record->xl_tot_len);
5152 : :
3959 5153 : 50 : INIT_CRC32C(crc);
3943 5154 : 50 : COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
3959 5155 : 50 : COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5156 : 50 : FIN_CRC32C(crc);
9018 vadim4o@yahoo.com 5157 : 50 : record->xl_crc = crc;
5158 : :
5159 : : /* Create first XLOG segment file */
1401 rhaas@postgresql.org 5160 : 50 : openLogTLI = BootstrapTimeLineID;
5161 : 50 : openLogFile = XLogFileInit(1, BootstrapTimeLineID);
5162 : :
5163 : : /*
5164 : : * We needn't bother with Reserve/ReleaseExternalFD here, since we'll
5165 : : * close the file again in a moment.
5166 : : */
5167 : :
5168 : : /* Write the first page with the initial record */
8858 tgl@sss.pgh.pa.us 5169 : 50 : errno = 0;
3094 rhaas@postgresql.org 5170 : 50 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
7096 tgl@sss.pgh.pa.us 5171 [ - + ]: 50 : if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5172 : : {
5173 : : /* if write didn't set errno, assume problem is no disk space */
8858 tgl@sss.pgh.pa.us 5174 [ # # ]:UBC 0 : if (errno == 0)
5175 : 0 : errno = ENOSPC;
8083 5176 [ # # ]: 0 : ereport(PANIC,
5177 : : (errcode_for_file_access(),
5178 : : errmsg("could not write bootstrap write-ahead log file: %m")));
5179 : : }
3094 rhaas@postgresql.org 5180 :CBC 50 : pgstat_report_wait_end();
5181 : :
5182 : 50 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
8943 tgl@sss.pgh.pa.us 5183 [ - + ]: 50 : if (pg_fsync(openLogFile) != 0)
8083 tgl@sss.pgh.pa.us 5184 [ # # ]:UBC 0 : ereport(PANIC,
5185 : : (errcode_for_file_access(),
5186 : : errmsg("could not fsync bootstrap write-ahead log file: %m")));
3094 rhaas@postgresql.org 5187 :CBC 50 : pgstat_report_wait_end();
5188 : :
2254 peter@eisentraut.org 5189 [ - + ]: 50 : if (close(openLogFile) != 0)
7894 tgl@sss.pgh.pa.us 5190 [ # # ]:UBC 0 : ereport(PANIC,
5191 : : (errcode_for_file_access(),
5192 : : errmsg("could not close bootstrap write-ahead log file: %m")));
5193 : :
8943 tgl@sss.pgh.pa.us 5194 :CBC 50 : openLogFile = -1;
5195 : :
5196 : : /* Now create pg_control */
410 peter@eisentraut.org 5197 : 50 : InitControlFile(sysidentifier, data_checksum_version);
8943 tgl@sss.pgh.pa.us 5198 : 50 : ControlFile->time = checkPoint.time;
9467 vadim4o@yahoo.com 5199 : 50 : ControlFile->checkPoint = checkPoint.redo;
8943 tgl@sss.pgh.pa.us 5200 : 50 : ControlFile->checkPointCopy = checkPoint;
5201 : :
5202 : : /* some additional ControlFile fields are set in WriteControlFile() */
9051 5203 : 50 : WriteControlFile();
5204 : :
5205 : : /* Bootstrap the commit log, too */
8778 5206 : 50 : BootStrapCLOG();
3930 alvherre@alvh.no-ip. 5207 : 50 : BootStrapCommitTs();
7737 tgl@sss.pgh.pa.us 5208 : 50 : BootStrapSUBTRANS();
7436 5209 : 50 : BootStrapMultiXact();
5210 : :
7322 5211 : 50 : pfree(buffer);
5212 : :
5213 : : /*
5214 : : * Force control file to be read - in contrast to normal processing we'd
5215 : : * otherwise never run the checks and GUC related initializations therein.
5216 : : */
2915 andres@anarazel.de 5217 : 50 : ReadControlFile();
9467 vadim4o@yahoo.com 5218 : 50 : }
5219 : :
5220 : : static char *
35 tgl@sss.pgh.pa.us 5221 :GNC 777 : str_time(pg_time_t tnow, char *buf, size_t bufsize)
5222 : : {
5223 : 777 : pg_strftime(buf, bufsize,
5224 : : "%Y-%m-%d %H:%M:%S %Z",
6608 tgl@sss.pgh.pa.us 5225 :CBC 777 : pg_localtime(&tnow, log_timezone));
5226 : :
9055 peter_e@gmx.net 5227 : 777 : return buf;
5228 : : }
5229 : :
5230 : : /*
5231 : : * Initialize the first WAL segment on new timeline.
5232 : : */
5233 : : static void
1298 heikki.linnakangas@i 5234 : 47 : XLogInitNewTimeline(TimeLineID endTLI, XLogRecPtr endOfLog, TimeLineID newTLI)
5235 : : {
5236 : : char xlogfname[MAXFNAMELEN];
5237 : : XLogSegNo endLogSegNo;
5238 : : XLogSegNo startLogSegNo;
5239 : :
5240 : : /* we always switch to a new timeline after archive recovery */
1401 rhaas@postgresql.org 5241 [ - + ]: 47 : Assert(endTLI != newTLI);
5242 : :
5243 : : /*
5244 : : * Update min recovery point one last time.
5245 : : */
5917 heikki.linnakangas@i 5246 : 47 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5247 : :
5248 : : /*
5249 : : * Calculate the last segment on the old timeline, and the first segment
5250 : : * on the new timeline. If the switch happens in the middle of a segment,
5251 : : * they are the same, but if the switch happens exactly at a segment
5252 : : * boundary, startLogSegNo will be endLogSegNo + 1.
5253 : : */
2909 andres@anarazel.de 5254 : 47 : XLByteToPrevSeg(endOfLog, endLogSegNo, wal_segment_size);
5255 : 47 : XLByteToSeg(endOfLog, startLogSegNo, wal_segment_size);
5256 : :
5257 : : /*
5258 : : * Initialize the starting WAL segment for the new timeline. If the switch
5259 : : * happens in the middle of a segment, copy data from the last WAL segment
5260 : : * of the old timeline up to the switch point, to the starting WAL segment
5261 : : * on the new timeline.
5262 : : */
3915 heikki.linnakangas@i 5263 [ + + ]: 47 : if (endLogSegNo == startLogSegNo)
5264 : : {
5265 : : /*
5266 : : * Make a copy of the file on the new timeline.
5267 : : *
5268 : : * Writing WAL isn't allowed yet, so there are no locking
5269 : : * considerations. But we should be just as tense as XLogFileInit to
5270 : : * avoid emplacing a bogus file.
5271 : : */
1401 rhaas@postgresql.org 5272 : 37 : XLogFileCopy(newTLI, endLogSegNo, endTLI, endLogSegNo,
2909 andres@anarazel.de 5273 : 37 : XLogSegmentOffset(endOfLog, wal_segment_size));
5274 : : }
5275 : : else
5276 : : {
5277 : : /*
5278 : : * The switch happened at a segment boundary, so just create the next
5279 : : * segment on the new timeline.
5280 : : */
5281 : : int fd;
5282 : :
1401 rhaas@postgresql.org 5283 : 10 : fd = XLogFileInit(startLogSegNo, newTLI);
5284 : :
2254 peter@eisentraut.org 5285 [ - + ]: 10 : if (close(fd) != 0)
5286 : : {
2104 michael@paquier.xyz 5287 :UBC 0 : int save_errno = errno;
5288 : :
1401 rhaas@postgresql.org 5289 : 0 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
2104 michael@paquier.xyz 5290 : 0 : errno = save_errno;
3912 heikki.linnakangas@i 5291 [ # # ]: 0 : ereport(ERROR,
5292 : : (errcode_for_file_access(),
5293 : : errmsg("could not close file \"%s\": %m", xlogfname)));
5294 : : }
5295 : : }
5296 : :
5297 : : /*
5298 : : * Let's just make real sure there are not .ready or .done flags posted
5299 : : * for the new segment.
5300 : : */
1401 rhaas@postgresql.org 5301 :CBC 47 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
3971 fujii@postgresql.org 5302 : 47 : XLogArchiveCleanup(xlogfname);
7719 tgl@sss.pgh.pa.us 5303 : 47 : }
5304 : :
5305 : : /*
5306 : : * Perform cleanup actions at the conclusion of archive recovery.
5307 : : */
5308 : : static void
1401 rhaas@postgresql.org 5309 : 47 : CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog,
5310 : : TimeLineID newTLI)
5311 : : {
5312 : : /*
5313 : : * Execute the recovery_end_command, if any.
5314 : : */
1424 5315 [ + - + + ]: 47 : if (recoveryEndCommand && strcmp(recoveryEndCommand, "") != 0)
943 michael@paquier.xyz 5316 : 2 : ExecuteRecoveryCommand(recoveryEndCommand,
5317 : : "recovery_end_command",
5318 : : true,
5319 : : WAIT_EVENT_RECOVERY_END_COMMAND);
5320 : :
5321 : : /*
5322 : : * We switched to a new timeline. Clean up segments on the old timeline.
5323 : : *
5324 : : * If there are any higher-numbered segments on the old timeline, remove
5325 : : * them. They might contain valid WAL, but they might also be
5326 : : * pre-allocated files containing garbage. In any case, they are not part
5327 : : * of the new timeline's history so we don't need them.
5328 : : */
1401 rhaas@postgresql.org 5329 : 47 : RemoveNonParentXlogFiles(EndOfLog, newTLI);
5330 : :
5331 : : /*
5332 : : * If the switch happened in the middle of a segment, what to do with the
5333 : : * last, partial segment on the old timeline? If we don't archive it, and
5334 : : * the server that created the WAL never archives it either (e.g. because
5335 : : * it was hit by a meteor), it will never make it to the archive. That's
5336 : : * OK from our point of view, because the new segment that we created with
5337 : : * the new TLI contains all the WAL from the old timeline up to the switch
5338 : : * point. But if you later try to do PITR to the "missing" WAL on the old
5339 : : * timeline, recovery won't find it in the archive. It's physically
5340 : : * present in the new file with new TLI, but recovery won't look there
5341 : : * when it's recovering to the older timeline. On the other hand, if we
5342 : : * archive the partial segment, and the original server on that timeline
5343 : : * is still running and archives the completed version of the same segment
5344 : : * later, it will fail. (We used to do that in 9.4 and below, and it
5345 : : * caused such problems).
5346 : : *
5347 : : * As a compromise, we rename the last segment with the .partial suffix,
5348 : : * and archive it. Archive recovery will never try to read .partial
5349 : : * segments, so they will normally go unused. But in the odd PITR case,
5350 : : * the administrator can copy them manually to the pg_wal directory
5351 : : * (removing the suffix). They can be useful in debugging, too.
5352 : : *
5353 : : * If a .done or .ready file already exists for the old timeline, however,
5354 : : * we had already determined that the segment is complete, so we can let
5355 : : * it be archived normally. (In particular, if it was restored from the
5356 : : * archive to begin with, it's expected to have a .done file).
5357 : : */
1424 5358 [ + + + + ]: 84 : if (XLogSegmentOffset(EndOfLog, wal_segment_size) != 0 &&
5359 [ + + - + ]: 37 : XLogArchivingActive())
5360 : : {
5361 : : char origfname[MAXFNAMELEN];
5362 : : XLogSegNo endLogSegNo;
5363 : :
5364 : 9 : XLByteToPrevSeg(EndOfLog, endLogSegNo, wal_segment_size);
5365 : 9 : XLogFileName(origfname, EndOfLogTLI, endLogSegNo, wal_segment_size);
5366 : :
5367 [ + + ]: 9 : if (!XLogArchiveIsReadyOrDone(origfname))
5368 : : {
5369 : : char origpath[MAXPGPATH];
5370 : : char partialfname[MAXFNAMELEN];
5371 : : char partialpath[MAXPGPATH];
5372 : :
5373 : : /*
5374 : : * If we're summarizing WAL, we can't rename the partial file
5375 : : * until the summarizer finishes with it, else it will fail.
5376 : : */
407 5377 [ + + ]: 5 : if (summarize_wal)
5378 : 1 : WaitForWalSummarization(EndOfLog);
5379 : :
1424 5380 : 5 : XLogFilePath(origpath, EndOfLogTLI, endLogSegNo, wal_segment_size);
5381 : 5 : snprintf(partialfname, MAXFNAMELEN, "%s.partial", origfname);
5382 : 5 : snprintf(partialpath, MAXPGPATH, "%s.partial", origpath);
5383 : :
5384 : : /*
5385 : : * Make sure there's no .done or .ready file for the .partial
5386 : : * file.
5387 : : */
5388 : 5 : XLogArchiveCleanup(partialfname);
5389 : :
5390 : 5 : durable_rename(origpath, partialpath, ERROR);
5391 : 5 : XLogArchiveNotify(partialfname);
5392 : : }
5393 : : }
5394 : 47 : }
5395 : :
5396 : : /*
5397 : : * Check to see if required parameters are set high enough on this server
5398 : : * for various aspects of recovery operation.
5399 : : *
5400 : : * Note that all the parameters which this function tests need to be
5401 : : * listed in Administrator's Overview section in high-availability.sgml.
5402 : : * If you change them, don't forget to update the list.
5403 : : */
5404 : : static void
1298 heikki.linnakangas@i 5405 : 234 : CheckRequiredParameterValues(void)
5406 : : {
5407 : : /*
5408 : : * For archive recovery, the WAL must be generated with at least 'replica'
5409 : : * wal_level.
5410 : : */
5411 [ + + + + ]: 234 : if (ArchiveRecoveryRequested && ControlFile->wal_level == WAL_LEVEL_MINIMAL)
5412 : : {
5413 [ + - ]: 2 : ereport(FATAL,
5414 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5415 : : errmsg("WAL was generated with \"wal_level=minimal\", cannot continue recovering"),
5416 : : errdetail("This happens if you temporarily set \"wal_level=minimal\" on the server."),
5417 : : errhint("Use a backup taken after setting \"wal_level\" to higher than \"minimal\".")));
5418 : : }
5419 : :
5420 : : /*
5421 : : * For Hot Standby, the WAL must be generated with 'replica' mode, and we
5422 : : * must have at least as many backend slots as the primary.
5423 : : */
4203 5424 [ + + + + ]: 232 : if (ArchiveRecoveryRequested && EnableHotStandby)
5425 : : {
5426 : : /* We ignore autovacuum_worker_slots when we make this test. */
5610 5427 : 115 : RecoveryRequiresIntParameter("max_connections",
5428 : : MaxConnections,
5609 tgl@sss.pgh.pa.us 5429 : 115 : ControlFile->MaxConnections);
4447 rhaas@postgresql.org 5430 : 115 : RecoveryRequiresIntParameter("max_worker_processes",
5431 : : max_worker_processes,
5432 : 115 : ControlFile->max_worker_processes);
2398 michael@paquier.xyz 5433 : 115 : RecoveryRequiresIntParameter("max_wal_senders",
5434 : : max_wal_senders,
5435 : 115 : ControlFile->max_wal_senders);
4749 tgl@sss.pgh.pa.us 5436 : 115 : RecoveryRequiresIntParameter("max_prepared_transactions",
5437 : : max_prepared_xacts,
5609 5438 : 115 : ControlFile->max_prepared_xacts);
4749 5439 : 115 : RecoveryRequiresIntParameter("max_locks_per_transaction",
5440 : : max_locks_per_xact,
5609 5441 : 115 : ControlFile->max_locks_per_xact);
5442 : : }
5740 simon@2ndQuadrant.co 5443 : 232 : }
5444 : :
5445 : : /*
5446 : : * This must be called ONCE during postmaster or standalone-backend startup
5447 : : */
5448 : : void
8943 tgl@sss.pgh.pa.us 5449 : 887 : StartupXLOG(void)
5450 : : {
5451 : : XLogCtlInsert *Insert;
5452 : : CheckPoint checkPoint;
5453 : : bool wasShutdown;
5454 : : bool didCrash;
5455 : : bool haveTblspcMap;
5456 : : bool haveBackupLabel;
5457 : : XLogRecPtr EndOfLog;
5458 : : TimeLineID EndOfLogTLI;
5459 : : TimeLineID newTLI;
5460 : : bool performedWalRecovery;
5461 : : EndOfWalRecoveryInfo *endOfRecoveryInfo;
5462 : : XLogRecPtr abortedRecPtr;
5463 : : XLogRecPtr missingContrecPtr;
5464 : : TransactionId oldestActiveXID;
1865 fujii@postgresql.org 5465 : 887 : bool promoted = false;
5466 : : char timebuf[128];
5467 : :
5468 : : /*
5469 : : * We should have an aux process resource owner to use, and we should not
5470 : : * be in a transaction that's installed some other resowner.
5471 : : */
2607 tgl@sss.pgh.pa.us 5472 [ - + ]: 887 : Assert(AuxProcessResourceOwner != NULL);
5473 [ + - - + ]: 887 : Assert(CurrentResourceOwner == NULL ||
5474 : : CurrentResourceOwner == AuxProcessResourceOwner);
5475 : 887 : CurrentResourceOwner = AuxProcessResourceOwner;
5476 : :
5477 : : /*
5478 : : * Check that contents look valid.
5479 : : */
2129 peter@eisentraut.org 5480 [ - + ]: 887 : if (!XRecOffIsValid(ControlFile->checkPoint))
8083 tgl@sss.pgh.pa.us 5481 [ # # ]:UBC 0 : ereport(FATAL,
5482 : : (errcode(ERRCODE_DATA_CORRUPTED),
5483 : : errmsg("control file contains invalid checkpoint location")));
5484 : :
2129 peter@eisentraut.org 5485 [ + + - - :CBC 887 : switch (ControlFile->state)
+ + - ]
5486 : : {
5487 : 690 : case DB_SHUTDOWNED:
5488 : :
5489 : : /*
5490 : : * This is the expected case, so don't be chatty in standalone
5491 : : * mode
5492 : : */
5493 [ + + + + ]: 690 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
5494 : : (errmsg("database system was shut down at %s",
5495 : : str_time(ControlFile->time,
5496 : : timebuf, sizeof(timebuf)))));
5497 : 690 : break;
5498 : :
5499 : 28 : case DB_SHUTDOWNED_IN_RECOVERY:
5500 [ + - ]: 28 : ereport(LOG,
5501 : : (errmsg("database system was shut down in recovery at %s",
5502 : : str_time(ControlFile->time,
5503 : : timebuf, sizeof(timebuf)))));
5504 : 28 : break;
5505 : :
2129 peter@eisentraut.org 5506 :UBC 0 : case DB_SHUTDOWNING:
5507 [ # # ]: 0 : ereport(LOG,
5508 : : (errmsg("database system shutdown was interrupted; last known up at %s",
5509 : : str_time(ControlFile->time,
5510 : : timebuf, sizeof(timebuf)))));
5511 : 0 : break;
5512 : :
5513 : 0 : case DB_IN_CRASH_RECOVERY:
5514 [ # # ]: 0 : ereport(LOG,
5515 : : (errmsg("database system was interrupted while in recovery at %s",
5516 : : str_time(ControlFile->time,
5517 : : timebuf, sizeof(timebuf))),
5518 : : errhint("This probably means that some data is corrupted and"
5519 : : " you will have to use the last backup for recovery.")));
5520 : 0 : break;
5521 : :
2129 peter@eisentraut.org 5522 :CBC 6 : case DB_IN_ARCHIVE_RECOVERY:
5523 [ + - ]: 6 : ereport(LOG,
5524 : : (errmsg("database system was interrupted while in recovery at log time %s",
5525 : : str_time(ControlFile->checkPointCopy.time,
5526 : : timebuf, sizeof(timebuf))),
5527 : : errhint("If this has occurred more than once some data might be corrupted"
5528 : : " and you might need to choose an earlier recovery target.")));
5529 : 6 : break;
5530 : :
5531 : 163 : case DB_IN_PRODUCTION:
5532 [ + - ]: 163 : ereport(LOG,
5533 : : (errmsg("database system was interrupted; last known up at %s",
5534 : : str_time(ControlFile->time,
5535 : : timebuf, sizeof(timebuf)))));
5536 : 163 : break;
5537 : :
2129 peter@eisentraut.org 5538 :UBC 0 : default:
5539 [ # # ]: 0 : ereport(FATAL,
5540 : : (errcode(ERRCODE_DATA_CORRUPTED),
5541 : : errmsg("control file contains invalid database cluster state")));
5542 : : }
5543 : :
5544 : : /* This is just to allow attaching to startup process with a debugger */
5545 : : #ifdef XLOG_REPLAY_DELAY
5546 : : if (ControlFile->state != DB_SHUTDOWNED)
5547 : : pg_usleep(60000000L);
5548 : : #endif
5549 : :
5550 : : /*
5551 : : * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
5552 : : * In cases where someone has performed a copy for PITR, these directories
5553 : : * may have been excluded and need to be re-created.
5554 : : */
6145 tgl@sss.pgh.pa.us 5555 :CBC 887 : ValidateXLOGDirectoryStructure();
5556 : :
5557 : : /* Set up timeout handler needed to report startup progress. */
1412 rhaas@postgresql.org 5558 [ + + ]: 887 : if (!IsBootstrapProcessingMode())
5559 : 837 : RegisterTimeout(STARTUP_PROGRESS_TIMEOUT,
5560 : : startup_progress_timeout_handler);
5561 : :
5562 : : /*----------
5563 : : * If we previously crashed, perform a couple of actions:
5564 : : *
5565 : : * - The pg_wal directory may still include some temporary WAL segments
5566 : : * used when creating a new segment, so perform some clean up to not
5567 : : * bloat this path. This is done first as there is no point to sync
5568 : : * this temporary data.
5569 : : *
5570 : : * - There might be data which we had written, intending to fsync it, but
5571 : : * which we had not actually fsync'd yet. Therefore, a power failure in
5572 : : * the near future might cause earlier unflushed writes to be lost, even
5573 : : * though more recent data written to disk from here on would be
5574 : : * persisted. To avoid that, fsync the entire data directory.
5575 : : */
1298 heikki.linnakangas@i 5576 [ + + ]: 887 : if (ControlFile->state != DB_SHUTDOWNED &&
5577 [ + + ]: 197 : ControlFile->state != DB_SHUTDOWNED_IN_RECOVERY)
5578 : : {
5579 : 169 : RemoveTempXlogFiles();
5580 : 169 : SyncDataDirectory();
1249 andres@anarazel.de 5581 : 169 : didCrash = true;
5582 : : }
5583 : : else
5584 : 718 : didCrash = false;
5585 : :
5586 : : /*
5587 : : * Prepare for WAL recovery if needed.
5588 : : *
5589 : : * InitWalRecovery analyzes the control file and the backup label file, if
5590 : : * any. It updates the in-memory ControlFile buffer according to the
5591 : : * starting checkpoint, and sets InRecovery and ArchiveRecoveryRequested.
5592 : : * It also applies the tablespace map file, if any.
5593 : : */
1298 heikki.linnakangas@i 5594 : 887 : InitWalRecovery(ControlFile, &wasShutdown,
5595 : : &haveBackupLabel, &haveTblspcMap);
5596 : 887 : checkPoint = ControlFile->checkPointCopy;
5597 : :
5598 : : /* initialize shared memory variables from the checkpoint record */
638 5599 : 887 : TransamVariables->nextXid = checkPoint.nextXid;
5600 : 887 : TransamVariables->nextOid = checkPoint.nextOid;
5601 : 887 : TransamVariables->oidCount = 0;
7395 tgl@sss.pgh.pa.us 5602 : 887 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
3089 rhaas@postgresql.org 5603 : 887 : AdvanceOldestClogXid(checkPoint.oldestXid);
5680 tgl@sss.pgh.pa.us 5604 : 887 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
3098 5605 : 887 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
3540 mail@joeconway.com 5606 : 887 : SetCommitTsLimit(checkPoint.oldestCommitTsXid,
5607 : : checkPoint.newestCommitTsXid);
5608 : :
5609 : : /*
5610 : : * Clear out any old relcache cache files. This is *necessary* if we do
5611 : : * any WAL replay, since that would probably result in the cache files
5612 : : * being out of sync with database reality. In theory we could leave them
5613 : : * in place if the database had been cleanly shut down, but it seems
5614 : : * safest to just remove them always and let them be rebuilt during the
5615 : : * first backend startup. These files needs to be removed from all
5616 : : * directories including pg_tblspc, however the symlinks are created only
5617 : : * after reading tablespace_map file in case of archive recovery from
5618 : : * backup, so needs to clear old relcache files here after creating
5619 : : * symlinks.
5620 : : */
1298 heikki.linnakangas@i 5621 : 887 : RelationCacheInitFileRemove();
5622 : :
5623 : : /*
5624 : : * Initialize replication slots, before there's a chance to remove
5625 : : * required resources.
5626 : : */
4104 andres@anarazel.de 5627 : 887 : StartupReplicationSlots();
5628 : :
5629 : : /*
5630 : : * Startup logical state, needs to be setup now so we have proper data
5631 : : * during crash recovery.
5632 : : */
4205 rhaas@postgresql.org 5633 : 887 : StartupReorderBuffer();
5634 : :
5635 : : /*
5636 : : * Startup CLOG. This must be done after TransamVariables->nextXid has
5637 : : * been initialized and before we accept connections or begin WAL replay.
5638 : : */
1683 5639 : 887 : StartupCLOG();
5640 : :
5641 : : /*
5642 : : * Startup MultiXact. We need to do this early to be able to replay
5643 : : * truncations.
5644 : : */
4299 alvherre@alvh.no-ip. 5645 : 887 : StartupMultiXact();
5646 : :
5647 : : /*
5648 : : * Ditto for commit timestamps. Activate the facility if the setting is
5649 : : * enabled in the control file, as there should be no tracking of commit
5650 : : * timestamps done when the setting was disabled. This facility can be
5651 : : * started or stopped when replaying a XLOG_PARAMETER_CHANGE record.
5652 : : */
2537 michael@paquier.xyz 5653 [ + + ]: 887 : if (ControlFile->track_commit_timestamp)
3557 alvherre@alvh.no-ip. 5654 : 13 : StartupCommitTs();
5655 : :
5656 : : /*
5657 : : * Recover knowledge about replay progress of known replication partners.
5658 : : */
3783 andres@anarazel.de 5659 : 887 : StartupReplicationOrigin();
5660 : :
5661 : : /*
5662 : : * Initialize unlogged LSN. On a clean shutdown, it's restored from the
5663 : : * control file. On recovery, all unlogged relations are blown away, so
5664 : : * the unlogged LSN counter can be reset too.
5665 : : */
4590 heikki.linnakangas@i 5666 [ + + ]: 887 : if (ControlFile->state == DB_SHUTDOWNED)
555 nathan@postgresql.or 5667 : 683 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5668 : 683 : ControlFile->unloggedLSN);
5669 : : else
5670 : 204 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5671 : : FirstNormalUnloggedLSN);
5672 : :
5673 : : /*
5674 : : * Copy any missing timeline history files between 'now' and the recovery
5675 : : * target timeline from archive to pg_wal. While we don't need those files
5676 : : * ourselves - the history file of the recovery target timeline covers all
5677 : : * the previous timelines in the history too - a cascading standby server
5678 : : * might be interested in them. Or, if you archive the WAL from this
5679 : : * server to a different archive than the primary, it'd be good for all
5680 : : * the history files to get archived there after failover, so that you can
5681 : : * use one of the old timelines as a PITR target. Timeline history files
5682 : : * are small, so it's better to copy them unnecessarily than not copy them
5683 : : * and regret later.
5684 : : */
1298 heikki.linnakangas@i 5685 : 887 : restoreTimeLineHistoryFiles(checkPoint.ThisTimeLineID, recoveryTargetTLI);
5686 : :
5687 : : /*
5688 : : * Before running in recovery, scan pg_twophase and fill in its status to
5689 : : * be able to work on entries generated by redo. Doing a scan before
5690 : : * taking any recovery action has the merit to discard any 2PC files that
5691 : : * are newer than the first record to replay, saving from any conflicts at
5692 : : * replay. This avoids as well any subsequent scans when doing recovery
5693 : : * of the on-disk two-phase data.
5694 : : */
3077 simon@2ndQuadrant.co 5695 : 887 : restoreTwoPhaseData();
5696 : :
5697 : : /*
5698 : : * When starting with crash recovery, reset pgstat data - it might not be
5699 : : * valid. Otherwise restore pgstat data. It's safe to do this here,
5700 : : * because postmaster will not yet have started any other processes.
5701 : : *
5702 : : * NB: Restoring replication slot stats relies on slot state to have
5703 : : * already been restored from disk.
5704 : : *
5705 : : * TODO: With a bit of extra work we could just start with a pgstat file
5706 : : * associated with the checkpoint redo location we're starting from.
5707 : : */
1249 andres@anarazel.de 5708 [ + + ]: 887 : if (didCrash)
5709 : 169 : pgstat_discard_stats();
5710 : : else
173 michael@paquier.xyz 5711 : 718 : pgstat_restore_stats();
5712 : :
4973 simon@2ndQuadrant.co 5713 : 887 : lastFullPageWrites = checkPoint.fullPageWrites;
5714 : :
4443 heikki.linnakangas@i 5715 : 887 : RedoRecPtr = XLogCtl->RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
3957 5716 : 887 : doPageWrites = lastFullPageWrites;
5717 : :
5718 : : /* REDO */
7703 tgl@sss.pgh.pa.us 5719 [ + + ]: 887 : if (InRecovery)
5720 : : {
5721 : : /* Initialize state for RecoveryInProgress() */
1298 heikki.linnakangas@i 5722 [ - + ]: 204 : SpinLockAcquire(&XLogCtl->info_lck);
5723 [ + + ]: 204 : if (InArchiveRecovery)
5724 : 106 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
5725 : : else
5726 : 98 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
5727 : 204 : SpinLockRelease(&XLogCtl->info_lck);
5728 : :
5729 : : /*
5730 : : * Update pg_control to show that we are recovering and to show the
5731 : : * selected checkpoint as the place we are starting from. We also mark
5732 : : * pg_control with any minimum recovery stop point obtained from a
5733 : : * backup history file.
5734 : : *
5735 : : * No need to hold ControlFileLock yet, we aren't up far enough.
5736 : : */
5737 : 204 : UpdateControlFile();
5738 : :
5739 : : /*
5740 : : * If there was a backup label file, it's done its job and the info
5741 : : * has now been propagated into pg_control. We must get rid of the
5742 : : * label file so that if we crash during recovery, we'll pick up at
5743 : : * the latest recovery restartpoint instead of going all the way back
5744 : : * to the backup start point. It seems prudent though to just rename
5745 : : * the file out of the way rather than delete it completely.
5746 : : */
5747 [ + + ]: 204 : if (haveBackupLabel)
5748 : : {
5749 : 69 : unlink(BACKUP_LABEL_OLD);
5750 : 69 : durable_rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, FATAL);
5751 : : }
5752 : :
5753 : : /*
5754 : : * If there was a tablespace_map file, it's done its job and the
5755 : : * symlinks have been created. We must get rid of the map file so
5756 : : * that if we crash during recovery, we don't create symlinks again.
5757 : : * It seems prudent though to just rename the file out of the way
5758 : : * rather than delete it completely.
5759 : : */
5760 [ + + ]: 204 : if (haveTblspcMap)
5761 : : {
5762 : 2 : unlink(TABLESPACE_MAP_OLD);
5763 : 2 : durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, FATAL);
5764 : : }
5765 : :
5766 : : /*
5767 : : * Initialize our local copy of minRecoveryPoint. When doing crash
5768 : : * recovery we want to replay up to the end of WAL. Particularly, in
5769 : : * the case of a promoted standby minRecoveryPoint value in the
5770 : : * control file is only updated after the first checkpoint. However,
5771 : : * if the instance crashes before the first post-recovery checkpoint
5772 : : * is completed then recovery will use a stale location causing the
5773 : : * startup process to think that there are still invalid page
5774 : : * references when checking for data consistency.
5775 : : */
2620 michael@paquier.xyz 5776 [ + + ]: 204 : if (InArchiveRecovery)
5777 : : {
1298 heikki.linnakangas@i 5778 : 106 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
5779 : 106 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
5780 : : }
5781 : : else
5782 : : {
5783 : 98 : LocalMinRecoveryPoint = InvalidXLogRecPtr;
5784 : 98 : LocalMinRecoveryPointTLI = 0;
5785 : : }
5786 : :
5787 : : /* Check that the GUCs used to generate the WAL allow recovery */
5610 5788 : 204 : CheckRequiredParameterValues();
5789 : :
5790 : : /*
5791 : : * We're in recovery, so unlogged relations may be trashed and must be
5792 : : * reset. This should be done BEFORE allowing Hot Standby
5793 : : * connections, so that read-only backends don't try to read whatever
5794 : : * garbage is left over from before.
5795 : : */
5365 rhaas@postgresql.org 5796 : 204 : ResetUnloggedRelations(UNLOGGED_RELATION_CLEANUP);
5797 : :
5798 : : /*
5799 : : * Likewise, delete any saved transaction snapshot files that got left
5800 : : * behind by crashed backends.
5801 : : */
5068 tgl@sss.pgh.pa.us 5802 : 204 : DeleteAllExportedSnapshotFiles();
5803 : :
5804 : : /*
5805 : : * Initialize for Hot Standby, if enabled. We won't let backends in
5806 : : * yet, not until we've reached the min recovery point specified in
5807 : : * control file and we've established a recovery snapshot from a
5808 : : * running-xacts WAL record.
5809 : : */
4579 heikki.linnakangas@i 5810 [ + + + + ]: 204 : if (ArchiveRecoveryRequested && EnableHotStandby)
5811 : : {
5812 : : TransactionId *xids;
5813 : : int nxids;
5814 : :
5685 5815 [ + + ]: 100 : ereport(DEBUG1,
5816 : : (errmsg_internal("initializing for hot standby")));
5817 : :
5740 simon@2ndQuadrant.co 5818 : 100 : InitRecoveryTransactionEnvironment();
5819 : :
5820 [ + + ]: 100 : if (wasShutdown)
5821 : 26 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
5822 : : else
5823 : 74 : oldestActiveXID = checkPoint.oldestActiveXid;
5824 [ - + ]: 100 : Assert(TransactionIdIsValid(oldestActiveXID));
5825 : :
5826 : : /* Tell procarray about the range of xids it has to deal with */
638 heikki.linnakangas@i 5827 : 100 : ProcArrayInitRecovery(XidFromFullTransactionId(TransamVariables->nextXid));
5828 : :
5829 : : /*
5830 : : * Startup subtrans only. CLOG, MultiXact and commit timestamp
5831 : : * have already been started up and other SLRUs are not maintained
5832 : : * during recovery and need not be started yet.
5833 : : */
5740 simon@2ndQuadrant.co 5834 : 100 : StartupSUBTRANS(oldestActiveXID);
5835 : :
5836 : : /*
5837 : : * If we're beginning at a shutdown checkpoint, we know that
5838 : : * nothing was running on the primary at this point. So fake-up an
5839 : : * empty running-xacts record and use that here and now. Recover
5840 : : * additional standby state for prepared transactions.
5841 : : */
5625 heikki.linnakangas@i 5842 [ + + ]: 100 : if (wasShutdown)
5843 : : {
5844 : : RunningTransactionsData running;
5845 : : TransactionId latestCompletedXid;
5846 : :
5847 : : /* Update pg_subtrans entries for any prepared transactions */
436 5848 : 26 : StandbyRecoverPreparedTransactions();
5849 : :
5850 : : /*
5851 : : * Construct a RunningTransactions snapshot representing a
5852 : : * shut down server, with only prepared transactions still
5853 : : * alive. We're never overflowed at this point because all
5854 : : * subxids are listed with their parent prepared transactions.
5855 : : */
5625 5856 : 26 : running.xcnt = nxids;
4661 simon@2ndQuadrant.co 5857 : 26 : running.subxcnt = 0;
436 heikki.linnakangas@i 5858 : 26 : running.subxid_status = SUBXIDS_IN_SUBTRANS;
1852 andres@anarazel.de 5859 : 26 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5625 heikki.linnakangas@i 5860 : 26 : running.oldestRunningXid = oldestActiveXID;
1852 andres@anarazel.de 5861 : 26 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5595 simon@2ndQuadrant.co 5862 [ - + ]: 26 : TransactionIdRetreat(latestCompletedXid);
5594 5863 [ - + ]: 26 : Assert(TransactionIdIsNormal(latestCompletedXid));
5595 5864 : 26 : running.latestCompletedXid = latestCompletedXid;
5625 heikki.linnakangas@i 5865 : 26 : running.xids = xids;
5866 : :
5867 : 26 : ProcArrayApplyRecoveryInfo(&running);
5868 : : }
5869 : : }
5870 : :
5871 : : /*
5872 : : * We're all set for replaying the WAL now. Do it.
5873 : : */
1298 5874 : 204 : PerformWalRecovery();
5875 : 149 : performedWalRecovery = true;
5876 : : }
5877 : : else
1294 5878 : 683 : performedWalRecovery = false;
5879 : :
5880 : : /*
5881 : : * Finish WAL recovery.
5882 : : */
1298 5883 : 832 : endOfRecoveryInfo = FinishWalRecovery();
5884 : 832 : EndOfLog = endOfRecoveryInfo->endOfLog;
5885 : 832 : EndOfLogTLI = endOfRecoveryInfo->endOfLogTLI;
5886 : 832 : abortedRecPtr = endOfRecoveryInfo->abortedRecPtr;
5887 : 832 : missingContrecPtr = endOfRecoveryInfo->missingContrecPtr;
5888 : :
5889 : : /*
5890 : : * Reset ps status display, so as no information related to recovery shows
5891 : : * up.
5892 : : */
1080 michael@paquier.xyz 5893 : 832 : set_ps_display("");
5894 : :
5895 : : /*
5896 : : * When recovering from a backup (we are in recovery, and archive recovery
5897 : : * was requested), complain if we did not roll forward far enough to reach
5898 : : * the point where the database is consistent. For regular online
5899 : : * backup-from-primary, that means reaching the end-of-backup WAL record
5900 : : * (at which point we reset backupStartPoint to be Invalid), for
5901 : : * backup-from-replica (which can't inject records into the WAL stream),
5902 : : * that point is when we reach the minRecoveryPoint in pg_control (which
5903 : : * we purposefully copy last when backing up from a replica). For
5904 : : * pg_rewind (which creates a backup_label with a method of "pg_rewind")
5905 : : * or snapshot-style backups (which don't), backupEndRequired will be set
5906 : : * to false.
5907 : : *
5908 : : * Note: it is indeed okay to look at the local variable
5909 : : * LocalMinRecoveryPoint here, even though ControlFile->minRecoveryPoint
5910 : : * might be further ahead --- ControlFile->minRecoveryPoint cannot have
5911 : : * been advanced beyond the WAL we processed.
5912 : : */
5274 heikki.linnakangas@i 5913 [ + + ]: 832 : if (InRecovery &&
1298 5914 [ + - ]: 149 : (EndOfLog < LocalMinRecoveryPoint ||
5724 5915 [ - + ]: 149 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint)))
5916 : : {
5917 : : /*
5918 : : * Ran off end of WAL before reaching end-of-backup WAL record, or
5919 : : * minRecoveryPoint. That's a bad sign, indicating that you tried to
5920 : : * recover from an online backup but never called pg_backup_stop(), or
5921 : : * you didn't archive all the WAL needed.
5922 : : */
4579 heikki.linnakangas@i 5923 [ # # # # ]:UBC 0 : if (ArchiveRecoveryRequested || ControlFile->backupEndRequired)
5924 : : {
1249 sfrost@snowman.net 5925 [ # # # # ]: 0 : if (!XLogRecPtrIsInvalid(ControlFile->backupStartPoint) || ControlFile->backupEndRequired)
5141 heikki.linnakangas@i 5926 [ # # ]: 0 : ereport(FATAL,
5927 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5928 : : errmsg("WAL ends before end of online backup"),
5929 : : errhint("All WAL generated while online backup was taken must be available at recovery.")));
5930 : : else
5260 5931 [ # # ]: 0 : ereport(FATAL,
5932 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5933 : : errmsg("WAL ends before consistent recovery point")));
5934 : : }
5935 : : }
5936 : :
5937 : : /*
5938 : : * Reset unlogged relations to the contents of their INIT fork. This is
5939 : : * done AFTER recovery is complete so as to include any unlogged relations
5940 : : * created during recovery, but BEFORE recovery is marked as having
5941 : : * completed successfully. Otherwise we'd not retry if any of the post
5942 : : * end-of-recovery steps fail.
5943 : : */
1298 heikki.linnakangas@i 5944 [ + + ]:CBC 832 : if (InRecovery)
5945 : 149 : ResetUnloggedRelations(UNLOGGED_RELATION_INIT);
5946 : :
5947 : : /*
5948 : : * Pre-scan prepared transactions to find out the range of XIDs present.
5949 : : * This information is not quite needed yet, but it is positioned here so
5950 : : * as potential problems are detected before any on-disk change is done.
5951 : : */
2616 michael@paquier.xyz 5952 : 832 : oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
5953 : :
5954 : : /*
5955 : : * Allow ordinary WAL segment creation before possibly switching to a new
5956 : : * timeline, which creates a new segment, and after the last ReadRecord().
5957 : : */
1116 5958 : 832 : SetInstallXLogFileSegmentActive();
5959 : :
5960 : : /*
5961 : : * Consider whether we need to assign a new timeline ID.
5962 : : *
5963 : : * If we did archive recovery, we always assign a new ID. This handles a
5964 : : * couple of issues. If we stopped short of the end of WAL during
5965 : : * recovery, then we are clearly generating a new timeline and must assign
5966 : : * it a unique new ID. Even if we ran to the end, modifying the current
5967 : : * last segment is problematic because it may result in trying to
5968 : : * overwrite an already-archived copy of that segment, and we encourage
5969 : : * DBAs to make their archive_commands reject that. We can dodge the
5970 : : * problem by making the new active segment have a new timeline ID.
5971 : : *
5972 : : * In a normal crash recovery, we can just extend the timeline we were in.
5973 : : */
1298 heikki.linnakangas@i 5974 : 832 : newTLI = endOfRecoveryInfo->lastRecTLI;
4579 5975 [ + + ]: 832 : if (ArchiveRecoveryRequested)
5976 : : {
1396 rhaas@postgresql.org 5977 : 47 : newTLI = findNewestTimeLine(recoveryTargetTLI) + 1;
7717 tgl@sss.pgh.pa.us 5978 [ + - ]: 47 : ereport(LOG,
5979 : : (errmsg("selected new timeline ID: %u", newTLI)));
5980 : :
5981 : : /*
5982 : : * Make a writable copy of the last WAL segment. (Note that we also
5983 : : * have a copy of the last block of the old WAL in
5984 : : * endOfRecovery->lastPage; we will use that below.)
5985 : : */
1298 heikki.linnakangas@i 5986 : 47 : XLogInitNewTimeline(EndOfLogTLI, EndOfLog, newTLI);
5987 : :
5988 : : /*
5989 : : * Remove the signal files out of the way, so that we don't
5990 : : * accidentally re-enter archive recovery mode in a subsequent crash.
5991 : : */
5992 [ + + ]: 47 : if (endOfRecoveryInfo->standby_signal_file_found)
5993 : 44 : durable_unlink(STANDBY_SIGNAL_FILE, FATAL);
5994 : :
5995 [ + + ]: 47 : if (endOfRecoveryInfo->recovery_signal_file_found)
5996 : 3 : durable_unlink(RECOVERY_SIGNAL_FILE, FATAL);
5997 : :
5998 : : /*
5999 : : * Write the timeline history file, and have it archived. After this
6000 : : * point (or rather, as soon as the file is archived), the timeline
6001 : : * will appear as "taken" in the WAL archive and to any standby
6002 : : * servers. If we crash before actually switching to the new
6003 : : * timeline, standby servers will nevertheless think that we switched
6004 : : * to the new timeline, and will try to connect to the new timeline.
6005 : : * To minimize the window for that, try to do as little as possible
6006 : : * between here and writing the end-of-recovery record.
6007 : : */
1396 rhaas@postgresql.org 6008 : 47 : writeTimeLineHistory(newTLI, recoveryTargetTLI,
6009 : : EndOfLog, endOfRecoveryInfo->recoveryStopReason);
6010 : :
1298 heikki.linnakangas@i 6011 [ + - ]: 47 : ereport(LOG,
6012 : : (errmsg("archive recovery complete")));
6013 : : }
6014 : :
6015 : : /* Save the selected TimeLineID in shared memory, too */
407 rhaas@postgresql.org 6016 [ - + ]: 832 : SpinLockAcquire(&XLogCtl->info_lck);
1396 6017 : 832 : XLogCtl->InsertTimeLineID = newTLI;
1298 heikki.linnakangas@i 6018 : 832 : XLogCtl->PrevTimeLineID = endOfRecoveryInfo->lastRecTLI;
407 rhaas@postgresql.org 6019 : 832 : SpinLockRelease(&XLogCtl->info_lck);
6020 : :
6021 : : /*
6022 : : * Actually, if WAL ended in an incomplete record, skip the parts that
6023 : : * made it through and start writing after the portion that persisted.
6024 : : * (It's critical to first write an OVERWRITE_CONTRECORD message, which
6025 : : * we'll do as soon as we're open for writing new WAL.)
6026 : : */
1438 alvherre@alvh.no-ip. 6027 [ + + ]: 832 : if (!XLogRecPtrIsInvalid(missingContrecPtr))
6028 : : {
6029 : : /*
6030 : : * We should only have a missingContrecPtr if we're not switching to a
6031 : : * new timeline. When a timeline switch occurs, WAL is copied from the
6032 : : * old timeline to the new only up to the end of the last complete
6033 : : * record, so there can't be an incomplete WAL record that we need to
6034 : : * disregard.
6035 : : */
1104 rhaas@postgresql.org 6036 [ - + ]: 11 : Assert(newTLI == endOfRecoveryInfo->lastRecTLI);
1438 alvherre@alvh.no-ip. 6037 [ - + ]: 11 : Assert(!XLogRecPtrIsInvalid(abortedRecPtr));
6038 : 11 : EndOfLog = missingContrecPtr;
6039 : : }
6040 : :
6041 : : /*
6042 : : * Prepare to write WAL starting at EndOfLog location, and init xlog
6043 : : * buffer cache using the block containing the last record from the
6044 : : * previous incarnation.
6045 : : */
9079 vadim4o@yahoo.com 6046 : 832 : Insert = &XLogCtl->Insert;
1298 heikki.linnakangas@i 6047 : 832 : Insert->PrevBytePos = XLogRecPtrToBytePos(endOfRecoveryInfo->lastRec);
4434 6048 : 832 : Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
6049 : :
6050 : : /*
6051 : : * Tricky point here: lastPage contains the *last* block that the LastRec
6052 : : * record spans, not the one it starts in. The last block is indeed the
6053 : : * one we want to use.
6054 : : */
6055 [ + + ]: 832 : if (EndOfLog % XLOG_BLCKSZ != 0)
6056 : : {
6057 : : char *page;
6058 : : int len;
6059 : : int firstIdx;
6060 : :
6061 : 804 : firstIdx = XLogRecPtrToBufIdx(EndOfLog);
1298 6062 : 804 : len = EndOfLog - endOfRecoveryInfo->lastPageBeginPtr;
6063 [ - + ]: 804 : Assert(len < XLOG_BLCKSZ);
6064 : :
6065 : : /* Copy the valid part of the last block, and zero the rest */
4434 6066 : 804 : page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
1298 6067 : 804 : memcpy(page, endOfRecoveryInfo->lastPage, len);
4434 6068 : 804 : memset(page + len, 0, XLOG_BLCKSZ - len);
6069 : :
627 jdavis@postgresql.or 6070 : 804 : pg_atomic_write_u64(&XLogCtl->xlblocks[firstIdx], endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ);
15 akorotkov@postgresql 6071 : 804 : XLogCtl->InitializedUpTo = endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ;
6072 : : }
6073 : : else
6074 : : {
6075 : : /*
6076 : : * There is no partial block to copy. Just set InitializedUpTo, and
6077 : : * let the first attempt to insert a log record to initialize the next
6078 : : * buffer.
6079 : : */
6080 : 28 : XLogCtl->InitializedUpTo = EndOfLog;
6081 : : }
6082 : :
6083 : : /*
6084 : : * Update local and shared status. This is OK to do without any locks
6085 : : * because no other process can be reading or writing WAL yet.
6086 : : */
4434 heikki.linnakangas@i 6087 : 832 : LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
517 alvherre@alvh.no-ip. 6088 : 832 : pg_atomic_write_u64(&XLogCtl->logInsertResult, EndOfLog);
519 6089 : 832 : pg_atomic_write_u64(&XLogCtl->logWriteResult, EndOfLog);
6090 : 832 : pg_atomic_write_u64(&XLogCtl->logFlushResult, EndOfLog);
4434 heikki.linnakangas@i 6091 : 832 : XLogCtl->LogwrtRqst.Write = EndOfLog;
6092 : 832 : XLogCtl->LogwrtRqst.Flush = EndOfLog;
6093 : :
6094 : : /*
6095 : : * Preallocate additional log files, if wanted.
6096 : : */
1396 rhaas@postgresql.org 6097 : 832 : PreallocXlogFiles(EndOfLog, newTLI);
6098 : :
6099 : : /*
6100 : : * Okay, we're officially UP.
6101 : : */
9079 vadim4o@yahoo.com 6102 : 832 : InRecovery = false;
6103 : :
6104 : : /* start the archive_timeout timer and LSN running */
4434 heikki.linnakangas@i 6105 : 832 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
3180 andres@anarazel.de 6106 : 832 : XLogCtl->lastSegSwitchLSN = EndOfLog;
6107 : :
6108 : : /* also initialize latestCompletedXid, to nextXid - 1 */
4961 tgl@sss.pgh.pa.us 6109 : 832 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
638 heikki.linnakangas@i 6110 : 832 : TransamVariables->latestCompletedXid = TransamVariables->nextXid;
6111 : 832 : FullTransactionIdRetreat(&TransamVariables->latestCompletedXid);
4961 tgl@sss.pgh.pa.us 6112 : 832 : LWLockRelease(ProcArrayLock);
6113 : :
6114 : : /*
6115 : : * Start up subtrans, if not already done for hot standby. (commit
6116 : : * timestamps are started below, if necessary.)
6117 : : */
5740 simon@2ndQuadrant.co 6118 [ + + ]: 832 : if (standbyState == STANDBY_DISABLED)
6119 : 785 : StartupSUBTRANS(oldestActiveXID);
6120 : :
6121 : : /*
6122 : : * Perform end of recovery actions for any SLRUs that need it.
6123 : : */
5057 6124 : 832 : TrimCLOG();
4299 alvherre@alvh.no-ip. 6125 : 832 : TrimMultiXact();
6126 : :
6127 : : /*
6128 : : * Reload shared-memory state for prepared transactions. This needs to
6129 : : * happen before renaming the last partial segment of the old timeline as
6130 : : * it may be possible that we have to recover some transactions from it.
6131 : : */
7386 tgl@sss.pgh.pa.us 6132 : 832 : RecoverPreparedTransactions();
6133 : :
6134 : : /* Shut down xlogreader */
1298 heikki.linnakangas@i 6135 : 832 : ShutdownWalRecovery();
6136 : :
6137 : : /* Enable WAL writes for this backend only. */
1423 rhaas@postgresql.org 6138 : 832 : LocalSetXLogInsertAllowed();
6139 : :
6140 : : /* If necessary, write overwrite-contrecord before doing anything else */
6141 [ + + ]: 832 : if (!XLogRecPtrIsInvalid(abortedRecPtr))
6142 : : {
6143 [ - + ]: 11 : Assert(!XLogRecPtrIsInvalid(missingContrecPtr));
1298 heikki.linnakangas@i 6144 : 11 : CreateOverwriteContrecordRecord(abortedRecPtr, missingContrecPtr, newTLI);
6145 : : }
6146 : :
6147 : : /*
6148 : : * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
6149 : : * record before resource manager writes cleanup WAL records or checkpoint
6150 : : * record is written.
6151 : : */
1423 rhaas@postgresql.org 6152 : 832 : Insert->fullPageWrites = lastFullPageWrites;
6153 : 832 : UpdateFullPageWrites();
6154 : :
6155 : : /*
6156 : : * Emit checkpoint or end-of-recovery record in XLOG, if required.
6157 : : */
1298 heikki.linnakangas@i 6158 [ + + ]: 832 : if (performedWalRecovery)
1423 rhaas@postgresql.org 6159 : 149 : promoted = PerformRecoveryXLogAction();
6160 : :
6161 : : /*
6162 : : * If any of the critical GUCs have changed, log them before we allow
6163 : : * backends to write WAL.
6164 : : */
5610 heikki.linnakangas@i 6165 : 832 : XLogReportParameters();
6166 : :
6167 : : /* If this is archive recovery, perform post-recovery cleanup actions. */
1412 rhaas@postgresql.org 6168 [ + + ]: 832 : if (ArchiveRecoveryRequested)
1396 6169 : 47 : CleanupAfterArchiveRecovery(EndOfLogTLI, EndOfLog, newTLI);
6170 : :
6171 : : /*
6172 : : * Local WAL inserts enabled, so it's time to finish initialization of
6173 : : * commit timestamp.
6174 : : */
3930 alvherre@alvh.no-ip. 6175 : 832 : CompleteCommitTsInitialization();
6176 : :
6177 : : /* Clean up EndOfWalRecoveryInfo data to appease Valgrind leak checking */
35 tgl@sss.pgh.pa.us 6178 [ + + ]:GNC 832 : if (endOfRecoveryInfo->lastPage)
6179 : 815 : pfree(endOfRecoveryInfo->lastPage);
6180 : 832 : pfree(endOfRecoveryInfo->recoveryStopReason);
6181 : 832 : pfree(endOfRecoveryInfo);
6182 : :
6183 : : /*
6184 : : * All done with end-of-recovery actions.
6185 : : *
6186 : : * Now allow backends to write WAL and update the control file status in
6187 : : * consequence. SharedRecoveryState, that controls if backends can write
6188 : : * WAL, is updated while holding ControlFileLock to prevent other backends
6189 : : * to look at an inconsistent state of the control file in shared memory.
6190 : : * There is still a small window during which backends can write WAL and
6191 : : * the control file is still referring to a system not in DB_IN_PRODUCTION
6192 : : * state while looking at the on-disk control file.
6193 : : *
6194 : : * Also, we use info_lck to update SharedRecoveryState to ensure that
6195 : : * there are no race conditions concerning visibility of other recent
6196 : : * updates to shared memory.
6197 : : */
3329 peter_e@gmx.net 6198 :CBC 832 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6199 : 832 : ControlFile->state = DB_IN_PRODUCTION;
6200 : :
4002 andres@anarazel.de 6201 [ - + ]: 832 : SpinLockAcquire(&XLogCtl->info_lck);
1961 michael@paquier.xyz 6202 : 832 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_DONE;
4002 andres@anarazel.de 6203 : 832 : SpinLockRelease(&XLogCtl->info_lck);
6204 : :
3329 peter_e@gmx.net 6205 : 832 : UpdateControlFile();
6206 : 832 : LWLockRelease(ControlFileLock);
6207 : :
6208 : : /*
6209 : : * Shutdown the recovery environment. This must occur after
6210 : : * RecoverPreparedTransactions() (see notes in lock_twophase_recover())
6211 : : * and after switching SharedRecoveryState to RECOVERY_STATE_DONE so as
6212 : : * any session building a snapshot will not rely on KnownAssignedXids as
6213 : : * RecoveryInProgress() would return false at this stage. This is
6214 : : * particularly critical for prepared 2PC transactions, that would still
6215 : : * need to be included in snapshots once recovery has ended.
6216 : : */
1433 michael@paquier.xyz 6217 [ + + ]: 832 : if (standbyState != STANDBY_DISABLED)
6218 : 47 : ShutdownRecoveryTransactionEnvironment();
6219 : :
6220 : : /*
6221 : : * If there were cascading standby servers connected to us, nudge any wal
6222 : : * sender processes to notice that we've been promoted.
6223 : : */
882 andres@anarazel.de 6224 : 832 : WalSndWakeup(true, true);
6225 : :
6226 : : /*
6227 : : * If this was a promotion, request an (online) checkpoint now. This isn't
6228 : : * required for consistency, but the last restartpoint might be far back,
6229 : : * and in case of a crash, recovering from it might take a longer than is
6230 : : * appropriate now that we're not in standby mode anymore.
6231 : : */
1865 fujii@postgresql.org 6232 [ + + ]: 832 : if (promoted)
4491 simon@2ndQuadrant.co 6233 : 40 : RequestCheckpoint(CHECKPOINT_FORCE);
6044 heikki.linnakangas@i 6234 : 832 : }
6235 : :
6236 : : /*
6237 : : * Callback from PerformWalRecovery(), called when we switch from crash
6238 : : * recovery to archive recovery mode. Updates the control file accordingly.
6239 : : */
6240 : : void
1298 6241 : 2 : SwitchIntoArchiveRecovery(XLogRecPtr EndRecPtr, TimeLineID replayTLI)
6242 : : {
6243 : : /* initialize minRecoveryPoint to this record */
6244 : 2 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6245 : 2 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
6246 [ + - ]: 2 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6247 : : {
6248 : 2 : ControlFile->minRecoveryPoint = EndRecPtr;
6249 : 2 : ControlFile->minRecoveryPointTLI = replayTLI;
6250 : : }
6251 : : /* update local copy */
6252 : 2 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
6253 : 2 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
6254 : :
6255 : : /*
6256 : : * The startup process can update its local copy of minRecoveryPoint from
6257 : : * this point.
6258 : : */
6259 : 2 : updateMinRecoveryPoint = true;
6260 : :
6261 : 2 : UpdateControlFile();
6262 : :
6263 : : /*
6264 : : * We update SharedRecoveryState while holding the lock on ControlFileLock
6265 : : * so both states are consistent in shared memory.
6266 : : */
6267 [ - + ]: 2 : SpinLockAcquire(&XLogCtl->info_lck);
6268 : 2 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
6269 : 2 : SpinLockRelease(&XLogCtl->info_lck);
6270 : :
6271 : 2 : LWLockRelease(ControlFileLock);
6272 : 2 : }
6273 : :
6274 : : /*
6275 : : * Callback from PerformWalRecovery(), called when we reach the end of backup.
6276 : : * Updates the control file accordingly.
6277 : : */
6278 : : void
6279 : 69 : ReachedEndOfBackup(XLogRecPtr EndRecPtr, TimeLineID tli)
6280 : : {
6281 : : /*
6282 : : * We have reached the end of base backup, as indicated by pg_control. The
6283 : : * data on disk is now consistent (unless minRecoveryPoint is further
6284 : : * ahead, which can happen if we crashed during previous recovery). Reset
6285 : : * backupStartPoint and backupEndPoint, and update minRecoveryPoint to
6286 : : * make sure we don't allow starting up at an earlier point even if
6287 : : * recovery is stopped and restarted soon after this.
6288 : : */
6289 : 69 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6290 : :
6291 [ + + ]: 69 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6292 : : {
6293 : 65 : ControlFile->minRecoveryPoint = EndRecPtr;
6294 : 65 : ControlFile->minRecoveryPointTLI = tli;
6295 : : }
6296 : :
6297 : 69 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
6298 : 69 : ControlFile->backupEndPoint = InvalidXLogRecPtr;
6299 : 69 : ControlFile->backupEndRequired = false;
6300 : 69 : UpdateControlFile();
6301 : :
6302 : 69 : LWLockRelease(ControlFileLock);
5625 6303 : 69 : }
6304 : :
6305 : : /*
6306 : : * Perform whatever XLOG actions are necessary at end of REDO.
6307 : : *
6308 : : * The goal here is to make sure that we'll be able to recover properly if
6309 : : * we crash again. If we choose to write a checkpoint, we'll write a shutdown
6310 : : * checkpoint rather than an on-line one. This is not particularly critical,
6311 : : * but since we may be assigning a new TLI, using a shutdown checkpoint allows
6312 : : * us to have the rule that TLI only changes in shutdown checkpoints, which
6313 : : * allows some extra error checking in xlog_redo.
6314 : : */
6315 : : static bool
1424 rhaas@postgresql.org 6316 : 149 : PerformRecoveryXLogAction(void)
6317 : : {
6318 : 149 : bool promoted = false;
6319 : :
6320 : : /*
6321 : : * Perform a checkpoint to update all our recovery activity to disk.
6322 : : *
6323 : : * Note that we write a shutdown checkpoint rather than an on-line one.
6324 : : * This is not particularly critical, but since we may be assigning a new
6325 : : * TLI, using a shutdown checkpoint allows us to have the rule that TLI
6326 : : * only changes in shutdown checkpoints, which allows some extra error
6327 : : * checking in xlog_redo.
6328 : : *
6329 : : * In promotion, only create a lightweight end-of-recovery record instead
6330 : : * of a full checkpoint. A checkpoint is requested later, after we're
6331 : : * fully out of recovery mode and already accepting queries.
6332 : : */
6333 [ + + + - : 196 : if (ArchiveRecoveryRequested && IsUnderPostmaster &&
+ + ]
1298 heikki.linnakangas@i 6334 : 47 : PromoteIsTriggered())
6335 : : {
1424 rhaas@postgresql.org 6336 : 40 : promoted = true;
6337 : :
6338 : : /*
6339 : : * Insert a special WAL record to mark the end of recovery, since we
6340 : : * aren't doing a checkpoint. That means that the checkpointer process
6341 : : * may likely be in the middle of a time-smoothed restartpoint and
6342 : : * could continue to be for minutes after this. That sounds strange,
6343 : : * but the effect is roughly the same and it would be stranger to try
6344 : : * to come out of the restartpoint and then checkpoint. We request a
6345 : : * checkpoint later anyway, just for safety.
6346 : : */
6347 : 40 : CreateEndOfRecoveryRecord();
6348 : : }
6349 : : else
6350 : : {
6351 : 109 : RequestCheckpoint(CHECKPOINT_END_OF_RECOVERY |
6352 : : CHECKPOINT_FAST |
6353 : : CHECKPOINT_WAIT);
6354 : : }
6355 : :
6356 : 149 : return promoted;
6357 : : }
6358 : :
6359 : : /*
6360 : : * Is the system still in recovery?
6361 : : *
6362 : : * Unlike testing InRecovery, this works in any process that's connected to
6363 : : * shared memory.
6364 : : */
6365 : : bool
6044 heikki.linnakangas@i 6366 : 59737666 : RecoveryInProgress(void)
6367 : : {
6368 : : /*
6369 : : * We check shared state each time only until we leave recovery mode. We
6370 : : * can't re-enter recovery, so there's no need to keep checking after the
6371 : : * shared variable has once been seen false.
6372 : : */
6373 [ + + ]: 59737666 : if (!LocalRecoveryInProgress)
6374 : 57572943 : return false;
6375 : : else
6376 : : {
6377 : : /*
6378 : : * use volatile pointer to make sure we make a fresh read of the
6379 : : * shared variable.
6380 : : */
6381 : 2164723 : volatile XLogCtlData *xlogctl = XLogCtl;
6382 : :
1961 michael@paquier.xyz 6383 : 2164723 : LocalRecoveryInProgress = (xlogctl->SharedRecoveryState != RECOVERY_STATE_DONE);
6384 : :
6385 : : /*
6386 : : * Note: We don't need a memory barrier when we're still in recovery.
6387 : : * We might exit recovery immediately after return, so the caller
6388 : : * can't rely on 'true' meaning that we're still in recovery anyway.
6389 : : */
6390 : :
6044 heikki.linnakangas@i 6391 : 2164723 : return LocalRecoveryInProgress;
6392 : : }
6393 : : }
6394 : :
6395 : : /*
6396 : : * Returns current recovery state from shared memory.
6397 : : *
6398 : : * This returned state is kept consistent with the contents of the control
6399 : : * file. See details about the possible values of RecoveryState in xlog.h.
6400 : : */
6401 : : RecoveryState
1961 michael@paquier.xyz 6402 : 28117 : GetRecoveryState(void)
6403 : : {
6404 : : RecoveryState retval;
6405 : :
6406 [ - + ]: 28117 : SpinLockAcquire(&XLogCtl->info_lck);
6407 : 28117 : retval = XLogCtl->SharedRecoveryState;
6408 : 28117 : SpinLockRelease(&XLogCtl->info_lck);
6409 : :
6410 : 28117 : return retval;
6411 : : }
6412 : :
6413 : : /*
6414 : : * Is this process allowed to insert new WAL records?
6415 : : *
6416 : : * Ordinarily this is essentially equivalent to !RecoveryInProgress().
6417 : : * But we also have provisions for forcing the result "true" or "false"
6418 : : * within specific processes regardless of the global state.
6419 : : */
6420 : : bool
5916 tgl@sss.pgh.pa.us 6421 : 29005830 : XLogInsertAllowed(void)
6422 : : {
6423 : : /*
6424 : : * If value is "unconditionally true" or "unconditionally false", just
6425 : : * return it. This provides the normal fast path once recovery is known
6426 : : * done.
6427 : : */
6428 [ + + ]: 29005830 : if (LocalXLogInsertAllowed >= 0)
6429 : 28893864 : return (bool) LocalXLogInsertAllowed;
6430 : :
6431 : : /*
6432 : : * Else, must check to see if we're still in recovery.
6433 : : */
6434 [ + + ]: 111966 : if (RecoveryInProgress())
6435 : 104452 : return false;
6436 : :
6437 : : /*
6438 : : * On exit from recovery, reset to "unconditionally true", since there is
6439 : : * no need to keep checking.
6440 : : */
6441 : 7514 : LocalXLogInsertAllowed = 1;
6442 : 7514 : return true;
6443 : : }
6444 : :
6445 : : /*
6446 : : * Make XLogInsertAllowed() return true in the current process only.
6447 : : *
6448 : : * Note: it is allowed to switch LocalXLogInsertAllowed back to -1 later,
6449 : : * and even call LocalSetXLogInsertAllowed() again after that.
6450 : : *
6451 : : * Returns the previous value of LocalXLogInsertAllowed.
6452 : : */
6453 : : static int
6454 : 860 : LocalSetXLogInsertAllowed(void)
6455 : : {
1298 heikki.linnakangas@i 6456 : 860 : int oldXLogAllowed = LocalXLogInsertAllowed;
6457 : :
5916 tgl@sss.pgh.pa.us 6458 : 860 : LocalXLogInsertAllowed = 1;
6459 : :
1412 rhaas@postgresql.org 6460 : 860 : return oldXLogAllowed;
6461 : : }
6462 : :
6463 : : /*
6464 : : * Return the current Redo pointer from shared memory.
6465 : : *
6466 : : * As a side-effect, the local RedoRecPtr copy is updated.
6467 : : */
6468 : : XLogRecPtr
9018 vadim4o@yahoo.com 6469 : 207582 : GetRedoRecPtr(void)
6470 : : {
6471 : : XLogRecPtr ptr;
6472 : :
6473 : : /*
6474 : : * The possibly not up-to-date copy in XlogCtl is enough. Even if we
6475 : : * grabbed a WAL insertion lock to read the authoritative value in
6476 : : * Insert->RedoRecPtr, someone might update it just after we've released
6477 : : * the lock.
6478 : : */
4002 andres@anarazel.de 6479 [ + + ]: 207582 : SpinLockAcquire(&XLogCtl->info_lck);
6480 : 207582 : ptr = XLogCtl->RedoRecPtr;
6481 : 207582 : SpinLockRelease(&XLogCtl->info_lck);
6482 : :
4443 heikki.linnakangas@i 6483 [ + + ]: 207582 : if (RedoRecPtr < ptr)
6484 : 1521 : RedoRecPtr = ptr;
6485 : :
8576 tgl@sss.pgh.pa.us 6486 : 207582 : return RedoRecPtr;
6487 : : }
6488 : :
6489 : : /*
6490 : : * Return information needed to decide whether a modified block needs a
6491 : : * full-page image to be included in the WAL record.
6492 : : *
6493 : : * The returned values are cached copies from backend-private memory, and
6494 : : * possibly out-of-date or, indeed, uninitialized, in which case they will
6495 : : * be InvalidXLogRecPtr and false, respectively. XLogInsertRecord will
6496 : : * re-check them against up-to-date values, while holding the WAL insert lock.
6497 : : */
6498 : : void
3957 heikki.linnakangas@i 6499 : 13996795 : GetFullPageWriteInfo(XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
6500 : : {
6501 : 13996795 : *RedoRecPtr_p = RedoRecPtr;
6502 : 13996795 : *doPageWrites_p = doPageWrites;
6503 : 13996795 : }
6504 : :
6505 : : /*
6506 : : * GetInsertRecPtr -- Returns the current insert position.
6507 : : *
6508 : : * NOTE: The value *actually* returned is the position of the last full
6509 : : * xlog page. It lags behind the real insert position by at most 1 page.
6510 : : * For that, we don't need to scan through WAL insertion locks, and an
6511 : : * approximation is enough for the current usage of this function.
6512 : : */
6513 : : XLogRecPtr
6645 tgl@sss.pgh.pa.us 6514 : 7007 : GetInsertRecPtr(void)
6515 : : {
6516 : : XLogRecPtr recptr;
6517 : :
4002 andres@anarazel.de 6518 [ + + ]: 7007 : SpinLockAcquire(&XLogCtl->info_lck);
6519 : 7007 : recptr = XLogCtl->LogwrtRqst.Write;
6520 : 7007 : SpinLockRelease(&XLogCtl->info_lck);
6521 : :
6645 tgl@sss.pgh.pa.us 6522 : 7007 : return recptr;
6523 : : }
6524 : :
6525 : : /*
6526 : : * GetFlushRecPtr -- Returns the current flush position, ie, the last WAL
6527 : : * position known to be fsync'd to disk. This should only be used on a
6528 : : * system that is known not to be in recovery.
6529 : : */
6530 : : XLogRecPtr
1401 rhaas@postgresql.org 6531 : 178816 : GetFlushRecPtr(TimeLineID *insertTLI)
6532 : : {
1396 6533 [ - + ]: 178816 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6534 : :
521 alvherre@alvh.no-ip. 6535 : 178816 : RefreshXLogWriteResult(LogwrtResult);
6536 : :
6537 : : /*
6538 : : * If we're writing and flushing WAL, the time line can't be changing, so
6539 : : * no lock is required.
6540 : : */
1401 rhaas@postgresql.org 6541 [ + + ]: 178816 : if (insertTLI)
1396 6542 : 22981 : *insertTLI = XLogCtl->InsertTimeLineID;
6543 : :
3525 simon@2ndQuadrant.co 6544 : 178816 : return LogwrtResult.Flush;
6545 : : }
6546 : :
6547 : : /*
6548 : : * GetWALInsertionTimeLine -- Returns the current timeline of a system that
6549 : : * is not in recovery.
6550 : : */
6551 : : TimeLineID
1401 rhaas@postgresql.org 6552 : 110439 : GetWALInsertionTimeLine(void)
6553 : : {
6554 [ - + ]: 110439 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6555 : :
6556 : : /* Since the value can't be changing, no lock is required. */
1396 6557 : 110439 : return XLogCtl->InsertTimeLineID;
6558 : : }
6559 : :
6560 : : /*
6561 : : * GetWALInsertionTimeLineIfSet -- If the system is not in recovery, returns
6562 : : * the WAL insertion timeline; else, returns 0. Wherever possible, use
6563 : : * GetWALInsertionTimeLine() instead, since it's cheaper. Note that this
6564 : : * function decides recovery has ended as soon as the insert TLI is set, which
6565 : : * happens before we set XLogCtl->SharedRecoveryState to RECOVERY_STATE_DONE.
6566 : : */
6567 : : TimeLineID
407 rhaas@postgresql.org 6568 :UBC 0 : GetWALInsertionTimeLineIfSet(void)
6569 : : {
6570 : : TimeLineID insertTLI;
6571 : :
6572 [ # # ]: 0 : SpinLockAcquire(&XLogCtl->info_lck);
6573 : 0 : insertTLI = XLogCtl->InsertTimeLineID;
6574 : 0 : SpinLockRelease(&XLogCtl->info_lck);
6575 : :
6576 : 0 : return insertTLI;
6577 : : }
6578 : :
6579 : : /*
6580 : : * GetLastImportantRecPtr -- Returns the LSN of the last important record
6581 : : * inserted. All records not explicitly marked as unimportant are considered
6582 : : * important.
6583 : : *
6584 : : * The LSN is determined by computing the maximum of
6585 : : * WALInsertLocks[i].lastImportantAt.
6586 : : */
6587 : : XLogRecPtr
3180 andres@anarazel.de 6588 :CBC 1524 : GetLastImportantRecPtr(void)
6589 : : {
6590 : 1524 : XLogRecPtr res = InvalidXLogRecPtr;
6591 : : int i;
6592 : :
6593 [ + + ]: 13716 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
6594 : : {
6595 : : XLogRecPtr last_important;
6596 : :
6597 : : /*
6598 : : * Need to take a lock to prevent torn reads of the LSN, which are
6599 : : * possible on some of the supported platforms. WAL insert locks only
6600 : : * support exclusive mode, so we have to use that.
6601 : : */
6602 : 12192 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
6603 : 12192 : last_important = WALInsertLocks[i].l.lastImportantAt;
6604 : 12192 : LWLockRelease(&WALInsertLocks[i].l.lock);
6605 : :
6606 [ + + ]: 12192 : if (res < last_important)
6607 : 2635 : res = last_important;
6608 : : }
6609 : :
6610 : 1524 : return res;
6611 : : }
6612 : :
6613 : : /*
6614 : : * Get the time and LSN of the last xlog segment switch
6615 : : */
6616 : : pg_time_t
3180 andres@anarazel.de 6617 :UBC 0 : GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
6618 : : {
6619 : : pg_time_t result;
6620 : :
6621 : : /* Need WALWriteLock, but shared lock is sufficient */
6960 tgl@sss.pgh.pa.us 6622 : 0 : LWLockAcquire(WALWriteLock, LW_SHARED);
4434 heikki.linnakangas@i 6623 : 0 : result = XLogCtl->lastSegSwitchTime;
3180 andres@anarazel.de 6624 : 0 : *lastSwitchLSN = XLogCtl->lastSegSwitchLSN;
6960 tgl@sss.pgh.pa.us 6625 : 0 : LWLockRelease(WALWriteLock);
6626 : :
6627 : 0 : return result;
6628 : : }
6629 : :
6630 : : /*
6631 : : * This must be called ONCE during postmaster or standalone-backend shutdown
6632 : : */
6633 : : void
7939 peter_e@gmx.net 6634 :CBC 607 : ShutdownXLOG(int code, Datum arg)
6635 : : {
6636 : : /*
6637 : : * We should have an aux process resource owner to use, and we should not
6638 : : * be in a transaction that's installed some other resowner.
6639 : : */
2607 tgl@sss.pgh.pa.us 6640 [ - + ]: 607 : Assert(AuxProcessResourceOwner != NULL);
6641 [ + + - + ]: 607 : Assert(CurrentResourceOwner == NULL ||
6642 : : CurrentResourceOwner == AuxProcessResourceOwner);
6643 : 607 : CurrentResourceOwner = AuxProcessResourceOwner;
6644 : :
6645 : : /* Don't be chatty in standalone mode */
4468 6646 [ + + + + ]: 607 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
6647 : : (errmsg("shutting down")));
6648 : :
6649 : : /*
6650 : : * Signal walsenders to move to stopping state.
6651 : : */
3015 andres@anarazel.de 6652 : 607 : WalSndInitStopping();
6653 : :
6654 : : /*
6655 : : * Wait for WAL senders to be in stopping state. This prevents commands
6656 : : * from writing new WAL.
6657 : : */
6658 : 607 : WalSndWaitStopping();
6659 : :
6044 heikki.linnakangas@i 6660 [ + + ]: 607 : if (RecoveryInProgress())
57 nathan@postgresql.or 6661 :GNC 52 : CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_FAST);
6662 : : else
6663 : : {
6664 : : /*
6665 : : * If archiving is enabled, rotate the last XLOG file so that all the
6666 : : * remaining records are archived (postmaster wakes up the archiver
6667 : : * process one more time at the end of shutdown). The checkpoint
6668 : : * record will go to the next XLOG file and won't be archived (yet).
6669 : : */
1311 rhaas@postgresql.org 6670 [ + + - + :CBC 555 : if (XLogArchivingActive())
+ + ]
3180 andres@anarazel.de 6671 : 12 : RequestXLogSwitch(false);
6672 : :
57 nathan@postgresql.or 6673 :GNC 555 : CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_FAST);
6674 : : }
9467 vadim4o@yahoo.com 6675 :CBC 607 : }
6676 : :
6677 : : /*
6678 : : * Log start of a checkpoint.
6679 : : */
6680 : : static void
6044 heikki.linnakangas@i 6681 : 1385 : LogCheckpointStart(int flags, bool restartpoint)
6682 : : {
1737 peter@eisentraut.org 6683 [ + + ]: 1385 : if (restartpoint)
6684 [ + - - + : 198 : ereport(LOG,
- + + + +
+ + + + +
- + + + ]
6685 : : /* translator: the placeholders show checkpoint options */
6686 : : (errmsg("restartpoint starting:%s%s%s%s%s%s%s%s",
6687 : : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6688 : : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6689 : : (flags & CHECKPOINT_FAST) ? " fast" : "",
6690 : : (flags & CHECKPOINT_FORCE) ? " force" : "",
6691 : : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6692 : : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6693 : : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6694 : : (flags & CHECKPOINT_FLUSH_UNLOGGED) ? " flush-unlogged" : "")));
6695 : : else
6696 [ + - + + : 1187 : ereport(LOG,
- + + + +
+ + + + +
+ + + + ]
6697 : : /* translator: the placeholders show checkpoint options */
6698 : : (errmsg("checkpoint starting:%s%s%s%s%s%s%s%s",
6699 : : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6700 : : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6701 : : (flags & CHECKPOINT_FAST) ? " fast" : "",
6702 : : (flags & CHECKPOINT_FORCE) ? " force" : "",
6703 : : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6704 : : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6705 : : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6706 : : (flags & CHECKPOINT_FLUSH_UNLOGGED) ? " flush-unlogged" : "")));
6643 tgl@sss.pgh.pa.us 6707 : 1385 : }
6708 : :
6709 : : /*
6710 : : * Log end of a checkpoint.
6711 : : */
6712 : : static void
6044 heikki.linnakangas@i 6713 : 1677 : LogCheckpointEnd(bool restartpoint)
6714 : : {
6715 : : long write_msecs,
6716 : : sync_msecs,
6717 : : total_msecs,
6718 : : longest_msecs,
6719 : : average_msecs;
6720 : : uint64 average_sync_time;
6721 : :
6643 tgl@sss.pgh.pa.us 6722 : 1677 : CheckpointStats.ckpt_end_t = GetCurrentTimestamp();
6723 : :
1761 6724 : 1677 : write_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_write_t,
6725 : : CheckpointStats.ckpt_sync_t);
6726 : :
6727 : 1677 : sync_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_sync_t,
6728 : : CheckpointStats.ckpt_sync_end_t);
6729 : :
6730 : : /* Accumulate checkpoint timing summary data, in milliseconds. */
677 michael@paquier.xyz 6731 : 1677 : PendingCheckpointerStats.write_time += write_msecs;
6732 : 1677 : PendingCheckpointerStats.sync_time += sync_msecs;
6733 : :
6734 : : /*
6735 : : * All of the published timing statistics are accounted for. Only
6736 : : * continue if a log message is to be written.
6737 : : */
4902 rhaas@postgresql.org 6738 [ + + ]: 1677 : if (!log_checkpoints)
6739 : 292 : return;
6740 : :
1761 tgl@sss.pgh.pa.us 6741 : 1385 : total_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_start_t,
6742 : : CheckpointStats.ckpt_end_t);
6743 : :
6744 : : /*
6745 : : * Timing values returned from CheckpointStats are in microseconds.
6746 : : * Convert to milliseconds for consistent printing.
6747 : : */
6748 : 1385 : longest_msecs = (long) ((CheckpointStats.ckpt_longest_sync + 999) / 1000);
6749 : :
5380 rhaas@postgresql.org 6750 : 1385 : average_sync_time = 0;
5263 bruce@momjian.us 6751 [ - + ]: 1385 : if (CheckpointStats.ckpt_sync_rels > 0)
5380 rhaas@postgresql.org 6752 :UBC 0 : average_sync_time = CheckpointStats.ckpt_agg_sync_time /
6753 : 0 : CheckpointStats.ckpt_sync_rels;
1761 tgl@sss.pgh.pa.us 6754 :CBC 1385 : average_msecs = (long) ((average_sync_time + 999) / 1000);
6755 : :
6756 : : /*
6757 : : * ControlFileLock is not required to see ControlFile->checkPoint and
6758 : : * ->checkPointCopy here as we are the only updator of those variables at
6759 : : * this moment.
6760 : : */
1737 peter@eisentraut.org 6761 [ + + ]: 1385 : if (restartpoint)
6762 [ + - ]: 198 : ereport(LOG,
6763 : : (errmsg("restartpoint complete: wrote %d buffers (%.1f%%), "
6764 : : "wrote %d SLRU buffers; %d WAL file(s) added, "
6765 : : "%d removed, %d recycled; write=%ld.%03d s, "
6766 : : "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6767 : : "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6768 : : "estimate=%d kB; lsn=%X/%08X, redo lsn=%X/%08X",
6769 : : CheckpointStats.ckpt_bufs_written,
6770 : : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6771 : : CheckpointStats.ckpt_slru_written,
6772 : : CheckpointStats.ckpt_segs_added,
6773 : : CheckpointStats.ckpt_segs_removed,
6774 : : CheckpointStats.ckpt_segs_recycled,
6775 : : write_msecs / 1000, (int) (write_msecs % 1000),
6776 : : sync_msecs / 1000, (int) (sync_msecs % 1000),
6777 : : total_msecs / 1000, (int) (total_msecs % 1000),
6778 : : CheckpointStats.ckpt_sync_rels,
6779 : : longest_msecs / 1000, (int) (longest_msecs % 1000),
6780 : : average_msecs / 1000, (int) (average_msecs % 1000),
6781 : : (int) (PrevCheckPointDistance / 1024.0),
6782 : : (int) (CheckPointDistanceEstimate / 1024.0),
6783 : : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6784 : : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6785 : : else
6786 [ + - ]: 1187 : ereport(LOG,
6787 : : (errmsg("checkpoint complete: wrote %d buffers (%.1f%%), "
6788 : : "wrote %d SLRU buffers; %d WAL file(s) added, "
6789 : : "%d removed, %d recycled; write=%ld.%03d s, "
6790 : : "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6791 : : "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6792 : : "estimate=%d kB; lsn=%X/%08X, redo lsn=%X/%08X",
6793 : : CheckpointStats.ckpt_bufs_written,
6794 : : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6795 : : CheckpointStats.ckpt_slru_written,
6796 : : CheckpointStats.ckpt_segs_added,
6797 : : CheckpointStats.ckpt_segs_removed,
6798 : : CheckpointStats.ckpt_segs_recycled,
6799 : : write_msecs / 1000, (int) (write_msecs % 1000),
6800 : : sync_msecs / 1000, (int) (sync_msecs % 1000),
6801 : : total_msecs / 1000, (int) (total_msecs % 1000),
6802 : : CheckpointStats.ckpt_sync_rels,
6803 : : longest_msecs / 1000, (int) (longest_msecs % 1000),
6804 : : average_msecs / 1000, (int) (average_msecs % 1000),
6805 : : (int) (PrevCheckPointDistance / 1024.0),
6806 : : (int) (CheckPointDistanceEstimate / 1024.0),
6807 : : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6808 : : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6809 : : }
6810 : :
6811 : : /*
6812 : : * Update the estimate of distance between checkpoints.
6813 : : *
6814 : : * The estimate is used to calculate the number of WAL segments to keep
6815 : : * preallocated, see XLOGfileslop().
6816 : : */
6817 : : static void
3848 heikki.linnakangas@i 6818 : 1677 : UpdateCheckPointDistanceEstimate(uint64 nbytes)
6819 : : {
6820 : : /*
6821 : : * To estimate the number of segments consumed between checkpoints, keep a
6822 : : * moving average of the amount of WAL generated in previous checkpoint
6823 : : * cycles. However, if the load is bursty, with quiet periods and busy
6824 : : * periods, we want to cater for the peak load. So instead of a plain
6825 : : * moving average, let the average decline slowly if the previous cycle
6826 : : * used less WAL than estimated, but bump it up immediately if it used
6827 : : * more.
6828 : : *
6829 : : * When checkpoints are triggered by max_wal_size, this should converge to
6830 : : * CheckpointSegments * wal_segment_size,
6831 : : *
6832 : : * Note: This doesn't pay any attention to what caused the checkpoint.
6833 : : * Checkpoints triggered manually with CHECKPOINT command, or by e.g.
6834 : : * starting a base backup, are counted the same as those created
6835 : : * automatically. The slow-decline will largely mask them out, if they are
6836 : : * not frequent. If they are frequent, it seems reasonable to count them
6837 : : * in as any others; if you issue a manual checkpoint every 5 minutes and
6838 : : * never let a timed checkpoint happen, it makes sense to base the
6839 : : * preallocation on that 5 minute interval rather than whatever
6840 : : * checkpoint_timeout is set to.
6841 : : */
6842 : 1677 : PrevCheckPointDistance = nbytes;
6843 [ + + ]: 1677 : if (CheckPointDistanceEstimate < nbytes)
6844 : 695 : CheckPointDistanceEstimate = nbytes;
6845 : : else
6846 : 982 : CheckPointDistanceEstimate =
6847 : 982 : (0.90 * CheckPointDistanceEstimate + 0.10 * (double) nbytes);
6643 tgl@sss.pgh.pa.us 6848 : 1677 : }
6849 : :
6850 : : /*
6851 : : * Update the ps display for a process running a checkpoint. Note that
6852 : : * this routine should not do any allocations so as it can be called
6853 : : * from a critical section.
6854 : : */
6855 : : static void
1727 michael@paquier.xyz 6856 : 3354 : update_checkpoint_display(int flags, bool restartpoint, bool reset)
6857 : : {
6858 : : /*
6859 : : * The status is reported only for end-of-recovery and shutdown
6860 : : * checkpoints or shutdown restartpoints. Updating the ps display is
6861 : : * useful in those situations as it may not be possible to rely on
6862 : : * pg_stat_activity to see the status of the checkpointer or the startup
6863 : : * process.
6864 : : */
6865 [ + + ]: 3354 : if ((flags & (CHECKPOINT_END_OF_RECOVERY | CHECKPOINT_IS_SHUTDOWN)) == 0)
6866 : 2144 : return;
6867 : :
6868 [ + + ]: 1210 : if (reset)
6869 : 605 : set_ps_display("");
6870 : : else
6871 : : {
6872 : : char activitymsg[128];
6873 : :
6874 [ + + ]: 1815 : snprintf(activitymsg, sizeof(activitymsg), "performing %s%s%s",
6875 [ + + ]: 605 : (flags & CHECKPOINT_END_OF_RECOVERY) ? "end-of-recovery " : "",
6876 [ + + ]: 605 : (flags & CHECKPOINT_IS_SHUTDOWN) ? "shutdown " : "",
6877 : : restartpoint ? "restartpoint" : "checkpoint");
6878 : 605 : set_ps_display(activitymsg);
6879 : : }
6880 : : }
6881 : :
6882 : :
6883 : : /*
6884 : : * Perform a checkpoint --- either during shutdown, or on-the-fly
6885 : : *
6886 : : * flags is a bitwise OR of the following:
6887 : : * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
6888 : : * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
6889 : : * CHECKPOINT_FAST: finish the checkpoint ASAP, ignoring
6890 : : * checkpoint_completion_target parameter.
6891 : : * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
6892 : : * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
6893 : : * CHECKPOINT_END_OF_RECOVERY).
6894 : : * CHECKPOINT_FLUSH_UNLOGGED: also flush buffers of unlogged tables.
6895 : : *
6896 : : * Note: flags contains other bits, of interest here only for logging purposes.
6897 : : * In particular note that this routine is synchronous and does not pay
6898 : : * attention to CHECKPOINT_WAIT.
6899 : : *
6900 : : * If !shutdown then we are writing an online checkpoint. An XLOG_CHECKPOINT_REDO
6901 : : * record is inserted into WAL at the logical location of the checkpoint, before
6902 : : * flushing anything to disk, and when the checkpoint is eventually completed,
6903 : : * and it is from this point that WAL replay will begin in the case of a recovery
6904 : : * from this checkpoint. Once everything is written to disk, an
6905 : : * XLOG_CHECKPOINT_ONLINE record is written to complete the checkpoint, and
6906 : : * points back to the earlier XLOG_CHECKPOINT_REDO record. This mechanism allows
6907 : : * other write-ahead log records to be written while the checkpoint is in
6908 : : * progress, but we must be very careful about order of operations. This function
6909 : : * may take many minutes to execute on a busy system.
6910 : : *
6911 : : * On the other hand, when shutdown is true, concurrent insertion into the
6912 : : * write-ahead log is impossible, so there is no need for two separate records.
6913 : : * In this case, we only insert an XLOG_CHECKPOINT_SHUTDOWN record, and it's
6914 : : * both the record marking the completion of the checkpoint and the location
6915 : : * from which WAL replay would begin if needed.
6916 : : *
6917 : : * Returns true if a new checkpoint was performed, or false if it was skipped
6918 : : * because the system was idle.
6919 : : */
6920 : : bool
6645 tgl@sss.pgh.pa.us 6921 : 1479 : CreateCheckPoint(int flags)
6922 : : {
6923 : : bool shutdown;
6924 : : CheckPoint checkPoint;
6925 : : XLogRecPtr recptr;
6926 : : XLogSegNo _logSegNo;
9278 bruce@momjian.us 6927 : 1479 : XLogCtlInsert *Insert = &XLogCtl->Insert;
6928 : : uint32 freespace;
6929 : : XLogRecPtr PriorRedoPtr;
6930 : : XLogRecPtr last_important_lsn;
6931 : : VirtualTransactionId *vxids;
6932 : : int nvxids;
1412 rhaas@postgresql.org 6933 : 1479 : int oldXLogAllowed = 0;
6934 : :
6935 : : /*
6936 : : * An end-of-recovery checkpoint is really a shutdown checkpoint, just
6937 : : * issued at a different time.
6938 : : */
5916 tgl@sss.pgh.pa.us 6939 [ + + ]: 1479 : if (flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY))
5917 heikki.linnakangas@i 6940 : 583 : shutdown = true;
6941 : : else
6942 : 896 : shutdown = false;
6943 : :
6944 : : /* sanity check */
5916 tgl@sss.pgh.pa.us 6945 [ + + - + ]: 1479 : if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
5916 tgl@sss.pgh.pa.us 6946 [ # # ]:UBC 0 : elog(ERROR, "can't create a checkpoint during recovery");
6947 : :
6948 : : /*
6949 : : * Prepare to accumulate statistics.
6950 : : *
6951 : : * Note: because it is possible for log_checkpoints to change while a
6952 : : * checkpoint proceeds, we always accumulate stats, even if
6953 : : * log_checkpoints is currently off.
6954 : : */
6643 tgl@sss.pgh.pa.us 6955 [ + - + - :CBC 16269 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ - + - +
+ ]
6956 : 1479 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
6957 : :
6958 : : /*
6959 : : * Let smgr prepare for checkpoint; this has to happen outside the
6960 : : * critical section and before we determine the REDO pointer. Note that
6961 : : * smgr must not do anything that'd have to be undone if we decide no
6962 : : * checkpoint is needed.
6963 : : */
1270 tmunro@postgresql.or 6964 : 1479 : SyncPreCheckpoint();
6965 : :
6966 : : /*
6967 : : * Use a critical section to force system panic if we have trouble.
6968 : : */
8743 tgl@sss.pgh.pa.us 6969 : 1479 : START_CRIT_SECTION();
6970 : :
9476 vadim4o@yahoo.com 6971 [ + + ]: 1479 : if (shutdown)
6972 : : {
6044 heikki.linnakangas@i 6973 : 583 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9476 vadim4o@yahoo.com 6974 : 583 : ControlFile->state = DB_SHUTDOWNING;
6975 : 583 : UpdateControlFile();
6044 heikki.linnakangas@i 6976 : 583 : LWLockRelease(ControlFileLock);
6977 : : }
6978 : :
6979 : : /* Begin filling in the checkpoint WAL record */
8143 tgl@sss.pgh.pa.us 6980 [ + - + - : 17748 : MemSet(&checkPoint, 0, sizeof(checkPoint));
+ - + - +
+ ]
6411 6981 : 1479 : checkPoint.time = (pg_time_t) time(NULL);
6982 : :
6983 : : /*
6984 : : * For Hot Standby, derive the oldestActiveXid before we fix the redo
6985 : : * pointer. This allows us to begin accumulating changes to assemble our
6986 : : * starting snapshot of locks and transactions.
6987 : : */
5057 simon@2ndQuadrant.co 6988 [ + + + + ]: 1479 : if (!shutdown && XLogStandbyInfoActive())
45 akapila@postgresql.o 6989 :GNC 850 : checkPoint.oldestActiveXid = GetOldestActiveTransactionId(false, true);
6990 : : else
5057 simon@2ndQuadrant.co 6991 :CBC 629 : checkPoint.oldestActiveXid = InvalidTransactionId;
6992 : :
6993 : : /*
6994 : : * Get location of last important record before acquiring insert locks (as
6995 : : * GetLastImportantRecPtr() also locks WAL locks).
6996 : : */
3180 andres@anarazel.de 6997 : 1479 : last_important_lsn = GetLastImportantRecPtr();
6998 : :
6999 : : /*
7000 : : * If this isn't a shutdown or forced checkpoint, and if there has been no
7001 : : * WAL activity requiring a checkpoint, skip it. The idea here is to
7002 : : * avoid inserting duplicate checkpoints when the system is idle.
7003 : : */
5917 heikki.linnakangas@i 7004 [ + + ]: 1479 : if ((flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY |
7005 : : CHECKPOINT_FORCE)) == 0)
7006 : : {
3180 andres@anarazel.de 7007 [ - + ]: 189 : if (last_important_lsn == ControlFile->checkPoint)
7008 : : {
8943 tgl@sss.pgh.pa.us 7009 [ # # ]:LBC (1) : END_CRIT_SECTION();
3180 andres@anarazel.de 7010 [ # # ]: (1) : ereport(DEBUG1,
7011 : : (errmsg_internal("checkpoint skipped because system is idle")));
341 fujii@postgresql.org 7012 : (1) : return false;
7013 : : }
7014 : : }
7015 : :
7016 : : /*
7017 : : * An end-of-recovery checkpoint is created before anyone is allowed to
7018 : : * write WAL. To allow us to write the checkpoint record, temporarily
7019 : : * enable XLogInsertAllowed.
7020 : : */
5854 heikki.linnakangas@i 7021 [ + + ]:CBC 1479 : if (flags & CHECKPOINT_END_OF_RECOVERY)
1412 rhaas@postgresql.org 7022 : 28 : oldXLogAllowed = LocalSetXLogInsertAllowed();
7023 : :
1396 7024 : 1479 : checkPoint.ThisTimeLineID = XLogCtl->InsertTimeLineID;
4590 heikki.linnakangas@i 7025 [ + + ]: 1479 : if (flags & CHECKPOINT_END_OF_RECOVERY)
7026 : 28 : checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
7027 : : else
1401 rhaas@postgresql.org 7028 : 1451 : checkPoint.PrevTimeLineID = checkPoint.ThisTimeLineID;
7029 : :
7030 : : /*
7031 : : * We must block concurrent insertions while examining insert state.
7032 : : */
688 7033 : 1479 : WALInsertLockAcquireExclusive();
7034 : :
7035 : 1479 : checkPoint.fullPageWrites = Insert->fullPageWrites;
415 7036 : 1479 : checkPoint.wal_level = wal_level;
7037 : :
688 7038 [ + + ]: 1479 : if (shutdown)
7039 : : {
7040 : 583 : XLogRecPtr curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
7041 : :
7042 : : /*
7043 : : * Compute new REDO record ptr = location of next XLOG record.
7044 : : *
7045 : : * Since this is a shutdown checkpoint, there can't be any concurrent
7046 : : * WAL insertion.
7047 : : */
7048 [ + - ]: 583 : freespace = INSERT_FREESPACE(curInsert);
7049 [ - + ]: 583 : if (freespace == 0)
7050 : : {
688 rhaas@postgresql.org 7051 [ # # ]:UBC 0 : if (XLogSegmentOffset(curInsert, wal_segment_size) == 0)
7052 : 0 : curInsert += SizeOfXLogLongPHD;
7053 : : else
7054 : 0 : curInsert += SizeOfXLogShortPHD;
7055 : : }
688 rhaas@postgresql.org 7056 :CBC 583 : checkPoint.redo = curInsert;
7057 : :
7058 : : /*
7059 : : * Here we update the shared RedoRecPtr for future XLogInsert calls;
7060 : : * this must be done while holding all the insertion locks.
7061 : : *
7062 : : * Note: if we fail to complete the checkpoint, RedoRecPtr will be
7063 : : * left pointing past where it really needs to point. This is okay;
7064 : : * the only consequence is that XLogInsert might back up whole buffers
7065 : : * that it didn't really need to. We can't postpone advancing
7066 : : * RedoRecPtr because XLogInserts that happen while we are dumping
7067 : : * buffers must assume that their buffer changes are not included in
7068 : : * the checkpoint.
7069 : : */
7070 : 583 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
7071 : : }
7072 : :
7073 : : /*
7074 : : * Now we can release the WAL insertion locks, allowing other xacts to
7075 : : * proceed while we are flushing disk buffers.
7076 : : */
4187 heikki.linnakangas@i 7077 : 1479 : WALInsertLockRelease();
7078 : :
7079 : : /*
7080 : : * If this is an online checkpoint, we have not yet determined the redo
7081 : : * point. We do so now by inserting the special XLOG_CHECKPOINT_REDO
7082 : : * record; the LSN at which it starts becomes the new redo pointer. We
7083 : : * don't do this for a shutdown checkpoint, because in that case no WAL
7084 : : * can be written between the redo point and the insertion of the
7085 : : * checkpoint record itself, so the checkpoint record itself serves to
7086 : : * mark the redo point.
7087 : : */
688 rhaas@postgresql.org 7088 [ + + ]: 1479 : if (!shutdown)
7089 : : {
7090 : : /* Include WAL level in record for WAL summarizer's benefit. */
7091 : 896 : XLogBeginInsert();
207 peter@eisentraut.org 7092 : 896 : XLogRegisterData(&wal_level, sizeof(wal_level));
688 rhaas@postgresql.org 7093 : 896 : (void) XLogInsert(RM_XLOG_ID, XLOG_CHECKPOINT_REDO);
7094 : :
7095 : : /*
7096 : : * XLogInsertRecord will have updated XLogCtl->Insert.RedoRecPtr in
7097 : : * shared memory and RedoRecPtr in backend-local memory, but we need
7098 : : * to copy that into the record that will be inserted when the
7099 : : * checkpoint is complete.
7100 : : */
7101 : 896 : checkPoint.redo = RedoRecPtr;
7102 : : }
7103 : :
7104 : : /* Update the info_lck-protected copy of RedoRecPtr as well */
4002 andres@anarazel.de 7105 [ - + ]: 1479 : SpinLockAcquire(&XLogCtl->info_lck);
7106 : 1479 : XLogCtl->RedoRecPtr = checkPoint.redo;
7107 : 1479 : SpinLockRelease(&XLogCtl->info_lck);
7108 : :
7109 : : /*
7110 : : * If enabled, log checkpoint start. We postpone this until now so as not
7111 : : * to log anything if we decided to skip the checkpoint.
7112 : : */
6643 tgl@sss.pgh.pa.us 7113 [ + + ]: 1479 : if (log_checkpoints)
6044 heikki.linnakangas@i 7114 : 1187 : LogCheckpointStart(flags, false);
7115 : :
7116 : : /* Update the process title */
1727 michael@paquier.xyz 7117 : 1479 : update_checkpoint_display(flags, false, false);
7118 : :
7119 : : TRACE_POSTGRESQL_CHECKPOINT_START(flags);
7120 : :
7121 : : /*
7122 : : * Get the other info we need for the checkpoint record.
7123 : : *
7124 : : * We don't need to save oldestClogXid in the checkpoint, it only matters
7125 : : * for the short period in which clog is being truncated, and if we crash
7126 : : * during that we'll redo the clog truncation and fix up oldestClogXid
7127 : : * there.
7128 : : */
4173 heikki.linnakangas@i 7129 : 1479 : LWLockAcquire(XidGenLock, LW_SHARED);
638 7130 : 1479 : checkPoint.nextXid = TransamVariables->nextXid;
7131 : 1479 : checkPoint.oldestXid = TransamVariables->oldestXid;
7132 : 1479 : checkPoint.oldestXidDB = TransamVariables->oldestXidDB;
4173 7133 : 1479 : LWLockRelease(XidGenLock);
7134 : :
3930 alvherre@alvh.no-ip. 7135 : 1479 : LWLockAcquire(CommitTsLock, LW_SHARED);
638 heikki.linnakangas@i 7136 : 1479 : checkPoint.oldestCommitTsXid = TransamVariables->oldestCommitTsXid;
7137 : 1479 : checkPoint.newestCommitTsXid = TransamVariables->newestCommitTsXid;
3930 alvherre@alvh.no-ip. 7138 : 1479 : LWLockRelease(CommitTsLock);
7139 : :
4173 heikki.linnakangas@i 7140 : 1479 : LWLockAcquire(OidGenLock, LW_SHARED);
638 7141 : 1479 : checkPoint.nextOid = TransamVariables->nextOid;
4173 7142 [ + + ]: 1479 : if (!shutdown)
638 7143 : 896 : checkPoint.nextOid += TransamVariables->oidCount;
4173 7144 : 1479 : LWLockRelease(OidGenLock);
7145 : :
7146 : 1479 : MultiXactGetCheckptMulti(shutdown,
7147 : : &checkPoint.nextMulti,
7148 : : &checkPoint.nextMultiOffset,
7149 : : &checkPoint.oldestMulti,
7150 : : &checkPoint.oldestMultiDB);
7151 : :
7152 : : /*
7153 : : * Having constructed the checkpoint record, ensure all shmem disk buffers
7154 : : * and commit-log buffers are flushed to disk.
7155 : : *
7156 : : * This I/O could fail for various reasons. If so, we will fail to
7157 : : * complete the checkpoint, but there is no reason to force a system
7158 : : * panic. Accordingly, exit critical section while doing it.
7159 : : */
7160 [ - + ]: 1479 : END_CRIT_SECTION();
7161 : :
7162 : : /*
7163 : : * In some cases there are groups of actions that must all occur on one
7164 : : * side or the other of a checkpoint record. Before flushing the
7165 : : * checkpoint record we must explicitly wait for any backend currently
7166 : : * performing those groups of actions.
7167 : : *
7168 : : * One example is end of transaction, so we must wait for any transactions
7169 : : * that are currently in commit critical sections. If an xact inserted
7170 : : * its commit record into XLOG just before the REDO point, then a crash
7171 : : * restart from the REDO point would not replay that record, which means
7172 : : * that our flushing had better include the xact's update of pg_xact. So
7173 : : * we wait till he's out of his commit critical section before proceeding.
7174 : : * See notes in RecordTransactionCommit().
7175 : : *
7176 : : * Because we've already released the insertion locks, this test is a bit
7177 : : * fuzzy: it is possible that we will wait for xacts we didn't really need
7178 : : * to wait for. But the delay should be short and it seems better to make
7179 : : * checkpoint take a bit longer than to hold off insertions longer than
7180 : : * necessary. (In fact, the whole reason we have this issue is that xact.c
7181 : : * does commit record XLOG insertion and clog update as two separate steps
7182 : : * protected by different locks, but again that seems best on grounds of
7183 : : * minimizing lock contention.)
7184 : : *
7185 : : * A transaction that has not yet set delayChkptFlags when we look cannot
7186 : : * be at risk, since it has not inserted its commit record yet; and one
7187 : : * that's already cleared it is not at risk either, since it's done fixing
7188 : : * clog and we will correctly flush the update below. So we cannot miss
7189 : : * any xacts we need to wait for.
7190 : : */
1262 rhaas@postgresql.org 7191 : 1479 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_START);
4660 simon@2ndQuadrant.co 7192 [ + + ]: 1479 : if (nvxids > 0)
7193 : : {
7194 : : do
7195 : : {
7196 : : /*
7197 : : * Keep absorbing fsync requests while we wait. There could even
7198 : : * be a deadlock if we don't, if the process that prevents the
7199 : : * checkpoint is trying to add a request to the queue.
7200 : : */
442 heikki.linnakangas@i 7201 : 28 : AbsorbSyncRequests();
7202 : :
694 tmunro@postgresql.or 7203 : 28 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_START);
6505 bruce@momjian.us 7204 : 28 : pg_usleep(10000L); /* wait for 10 msec */
694 tmunro@postgresql.or 7205 : 28 : pgstat_report_wait_end();
1262 rhaas@postgresql.org 7206 [ + + ]: 28 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7207 : : DELAY_CHKPT_START));
7208 : : }
4660 simon@2ndQuadrant.co 7209 : 1479 : pfree(vxids);
7210 : :
6645 tgl@sss.pgh.pa.us 7211 : 1479 : CheckPointGuts(checkPoint.redo, flags);
7212 : :
1262 rhaas@postgresql.org 7213 : 1479 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_COMPLETE);
7214 [ - + ]: 1479 : if (nvxids > 0)
7215 : : {
7216 : : do
7217 : : {
442 heikki.linnakangas@i 7218 :UBC 0 : AbsorbSyncRequests();
7219 : :
694 tmunro@postgresql.or 7220 : 0 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_COMPLETE);
1262 rhaas@postgresql.org 7221 : 0 : pg_usleep(10000L); /* wait for 10 msec */
694 tmunro@postgresql.or 7222 : 0 : pgstat_report_wait_end();
1262 rhaas@postgresql.org 7223 [ # # ]: 0 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7224 : : DELAY_CHKPT_COMPLETE));
7225 : : }
1262 rhaas@postgresql.org 7226 :CBC 1479 : pfree(vxids);
7227 : :
7228 : : /*
7229 : : * Take a snapshot of running transactions and write this to WAL. This
7230 : : * allows us to reconstruct the state of running transactions during
7231 : : * archive recovery, if required. Skip, if this info disabled.
7232 : : *
7233 : : * If we are shutting down, or Startup process is completing crash
7234 : : * recovery we don't need to write running xact data.
7235 : : */
5740 simon@2ndQuadrant.co 7236 [ + + + + ]: 1479 : if (!shutdown && XLogStandbyInfoActive())
4661 tgl@sss.pgh.pa.us 7237 : 850 : LogStandbySnapshot();
7238 : :
8155 7239 : 1479 : START_CRIT_SECTION();
7240 : :
7241 : : /*
7242 : : * Now insert the checkpoint record into XLOG.
7243 : : */
3943 heikki.linnakangas@i 7244 : 1479 : XLogBeginInsert();
207 peter@eisentraut.org 7245 : 1479 : XLogRegisterData(&checkPoint, sizeof(checkPoint));
8943 tgl@sss.pgh.pa.us 7246 [ + + ]: 1479 : recptr = XLogInsert(RM_XLOG_ID,
7247 : : shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
7248 : : XLOG_CHECKPOINT_ONLINE);
7249 : :
7250 : 1479 : XLogFlush(recptr);
7251 : :
7252 : : /*
7253 : : * We mustn't write any new WAL after a shutdown checkpoint, or it will be
7254 : : * overwritten at next startup. No-one should even try, this just allows
7255 : : * sanity-checking. In the case of an end-of-recovery checkpoint, we want
7256 : : * to just temporarily disable writing until the system has exited
7257 : : * recovery.
7258 : : */
5916 7259 [ + + ]: 1479 : if (shutdown)
7260 : : {
7261 [ + + ]: 583 : if (flags & CHECKPOINT_END_OF_RECOVERY)
1412 rhaas@postgresql.org 7262 : 28 : LocalXLogInsertAllowed = oldXLogAllowed;
7263 : : else
5671 bruce@momjian.us 7264 : 555 : LocalXLogInsertAllowed = 0; /* never again write WAL */
7265 : : }
7266 : :
7267 : : /*
7268 : : * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
7269 : : * = end of actual checkpoint record.
7270 : : */
4635 alvherre@alvh.no-ip. 7271 [ + + - + ]: 1479 : if (shutdown && checkPoint.redo != ProcLastRecPtr)
8083 tgl@sss.pgh.pa.us 7272 [ # # ]:UBC 0 : ereport(PANIC,
7273 : : (errmsg("concurrent write-ahead log activity while database system is shutting down")));
7274 : :
7275 : : /*
7276 : : * Remember the prior checkpoint's redo ptr for
7277 : : * UpdateCheckPointDistanceEstimate()
7278 : : */
3848 heikki.linnakangas@i 7279 :CBC 1479 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7280 : :
7281 : : /*
7282 : : * Update the control file.
7283 : : */
8743 tgl@sss.pgh.pa.us 7284 : 1479 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9476 vadim4o@yahoo.com 7285 [ + + ]: 1479 : if (shutdown)
7286 : 583 : ControlFile->state = DB_SHUTDOWNED;
8943 tgl@sss.pgh.pa.us 7287 : 1479 : ControlFile->checkPoint = ProcLastRecPtr;
7288 : 1479 : ControlFile->checkPointCopy = checkPoint;
7289 : : /* crash recovery should always recover to the end of WAL */
4636 alvherre@alvh.no-ip. 7290 : 1479 : ControlFile->minRecoveryPoint = InvalidXLogRecPtr;
4659 heikki.linnakangas@i 7291 : 1479 : ControlFile->minRecoveryPointTLI = 0;
7292 : :
7293 : : /*
7294 : : * Persist unloggedLSN value. It's reset on crash recovery, so this goes
7295 : : * unused on non-shutdown checkpoints, but seems useful to store it always
7296 : : * for debugging purposes.
7297 : : */
555 nathan@postgresql.or 7298 : 1479 : ControlFile->unloggedLSN = pg_atomic_read_membarrier_u64(&XLogCtl->unloggedLSN);
7299 : :
9476 vadim4o@yahoo.com 7300 : 1479 : UpdateControlFile();
8743 tgl@sss.pgh.pa.us 7301 : 1479 : LWLockRelease(ControlFileLock);
7302 : :
7303 : : /*
7304 : : * We are now done with critical updates; no need for system panic if we
7305 : : * have trouble while fooling with old log segments.
7306 : : */
8155 7307 [ - + ]: 1479 : END_CRIT_SECTION();
7308 : :
7309 : : /*
7310 : : * WAL summaries end when the next XLOG_CHECKPOINT_REDO or
7311 : : * XLOG_CHECKPOINT_SHUTDOWN record is reached. This is the first point
7312 : : * where (a) we're not inside of a critical section and (b) we can be
7313 : : * certain that the relevant record has been flushed to disk, which must
7314 : : * happen before it can be summarized.
7315 : : *
7316 : : * If this is a shutdown checkpoint, then this happens reasonably
7317 : : * promptly: we've only just inserted and flushed the
7318 : : * XLOG_CHECKPOINT_SHUTDOWN record. If this is not a shutdown checkpoint,
7319 : : * then this might not be very prompt at all: the XLOG_CHECKPOINT_REDO
7320 : : * record was written before we began flushing data to disk, and that
7321 : : * could be many minutes ago at this point. However, we don't XLogFlush()
7322 : : * after inserting that record, so we're not guaranteed that it's on disk
7323 : : * until after the above call that flushes the XLOG_CHECKPOINT_ONLINE
7324 : : * record.
7325 : : */
309 heikki.linnakangas@i 7326 : 1479 : WakeupWalSummarizer();
7327 : :
7328 : : /*
7329 : : * Let smgr do post-checkpoint cleanup (eg, deleting old files).
7330 : : */
2347 tmunro@postgresql.or 7331 : 1479 : SyncPostCheckpoint();
7332 : :
7333 : : /*
7334 : : * Update the average distance between checkpoints if the prior checkpoint
7335 : : * exists.
7336 : : */
3848 heikki.linnakangas@i 7337 [ + - ]: 1479 : if (PriorRedoPtr != InvalidXLogRecPtr)
7338 : 1479 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7339 : :
7340 : : #ifdef USE_INJECTION_POINTS
7341 : : INJECTION_POINT("checkpoint-before-old-wal-removal", NULL);
7342 : : #endif
7343 : :
7344 : : /*
7345 : : * Delete old log files, those no longer needed for last checkpoint to
7346 : : * prevent the disk holding the xlog from growing full.
7347 : : */
2601 michael@paquier.xyz 7348 : 1479 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7349 : 1479 : KeepLogSeg(recptr, &_logSegNo);
199 akapila@postgresql.o 7350 [ + + ]: 1479 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED | RS_INVAL_IDLE_TIMEOUT,
7351 : : _logSegNo, InvalidOid,
7352 : : InvalidTransactionId))
7353 : : {
7354 : : /*
7355 : : * Some slots have been invalidated; recalculate the old-segment
7356 : : * horizon, starting again from RedoRecPtr.
7357 : : */
1513 alvherre@alvh.no-ip. 7358 : 3 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7359 : 3 : KeepLogSeg(recptr, &_logSegNo);
7360 : : }
2601 michael@paquier.xyz 7361 : 1479 : _logSegNo--;
1401 rhaas@postgresql.org 7362 : 1479 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, recptr,
7363 : : checkPoint.ThisTimeLineID);
7364 : :
7365 : : /*
7366 : : * Make more log segments if needed. (Do this after recycling old log
7367 : : * segments, since that may supply some of the needed files.)
7368 : : */
8943 tgl@sss.pgh.pa.us 7369 [ + + ]: 1479 : if (!shutdown)
1401 rhaas@postgresql.org 7370 : 896 : PreallocXlogFiles(recptr, checkPoint.ThisTimeLineID);
7371 : :
7372 : : /*
7373 : : * Truncate pg_subtrans if possible. We can throw away all data before
7374 : : * the oldest XMIN of any running transaction. No future transaction will
7375 : : * attempt to reference any pg_subtrans entry older than that (see Asserts
7376 : : * in subtrans.c). During recovery, though, we mustn't do this because
7377 : : * StartupSUBTRANS hasn't been called yet.
7378 : : */
5916 tgl@sss.pgh.pa.us 7379 [ + + ]: 1479 : if (!RecoveryInProgress())
1851 andres@anarazel.de 7380 : 1451 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7381 : :
7382 : : /* Real work is done; log and update stats. */
4902 rhaas@postgresql.org 7383 : 1479 : LogCheckpointEnd(false);
7384 : :
7385 : : /* Reset the process title */
1727 michael@paquier.xyz 7386 : 1479 : update_checkpoint_display(flags, false, true);
7387 : :
7388 : : TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
7389 : : NBuffers,
7390 : : CheckpointStats.ckpt_segs_added,
7391 : : CheckpointStats.ckpt_segs_removed,
7392 : : CheckpointStats.ckpt_segs_recycled);
7393 : :
341 fujii@postgresql.org 7394 : 1479 : return true;
7395 : : }
7396 : :
7397 : : /*
7398 : : * Mark the end of recovery in WAL though without running a full checkpoint.
7399 : : * We can expect that a restartpoint is likely to be in progress as we
7400 : : * do this, though we are unwilling to wait for it to complete.
7401 : : *
7402 : : * CreateRestartPoint() allows for the case where recovery may end before
7403 : : * the restartpoint completes so there is no concern of concurrent behaviour.
7404 : : */
7405 : : static void
4603 simon@2ndQuadrant.co 7406 : 40 : CreateEndOfRecoveryRecord(void)
7407 : : {
7408 : : xl_end_of_recovery xlrec;
7409 : : XLogRecPtr recptr;
7410 : :
7411 : : /* sanity check */
7412 [ - + ]: 40 : if (!RecoveryInProgress())
4603 simon@2ndQuadrant.co 7413 [ # # ]:UBC 0 : elog(ERROR, "can only be used to end recovery");
7414 : :
3914 heikki.linnakangas@i 7415 :CBC 40 : xlrec.end_time = GetCurrentTimestamp();
415 rhaas@postgresql.org 7416 : 40 : xlrec.wal_level = wal_level;
7417 : :
4187 heikki.linnakangas@i 7418 : 40 : WALInsertLockAcquireExclusive();
1396 rhaas@postgresql.org 7419 : 40 : xlrec.ThisTimeLineID = XLogCtl->InsertTimeLineID;
4590 heikki.linnakangas@i 7420 : 40 : xlrec.PrevTimeLineID = XLogCtl->PrevTimeLineID;
4187 7421 : 40 : WALInsertLockRelease();
7422 : :
4603 simon@2ndQuadrant.co 7423 : 40 : START_CRIT_SECTION();
7424 : :
3943 heikki.linnakangas@i 7425 : 40 : XLogBeginInsert();
207 peter@eisentraut.org 7426 : 40 : XLogRegisterData(&xlrec, sizeof(xl_end_of_recovery));
3943 heikki.linnakangas@i 7427 : 40 : recptr = XLogInsert(RM_XLOG_ID, XLOG_END_OF_RECOVERY);
7428 : :
4601 simon@2ndQuadrant.co 7429 : 40 : XLogFlush(recptr);
7430 : :
7431 : : /*
7432 : : * Update the control file so that crash recovery can follow the timeline
7433 : : * changes to this point.
7434 : : */
7435 : 40 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7436 : 40 : ControlFile->minRecoveryPoint = recptr;
1401 rhaas@postgresql.org 7437 : 40 : ControlFile->minRecoveryPointTLI = xlrec.ThisTimeLineID;
4601 simon@2ndQuadrant.co 7438 : 40 : UpdateControlFile();
7439 : 40 : LWLockRelease(ControlFileLock);
7440 : :
4603 7441 [ - + ]: 40 : END_CRIT_SECTION();
7442 : 40 : }
7443 : :
7444 : : /*
7445 : : * Write an OVERWRITE_CONTRECORD message.
7446 : : *
7447 : : * When on WAL replay we expect a continuation record at the start of a page
7448 : : * that is not there, recovery ends and WAL writing resumes at that point.
7449 : : * But it's wrong to resume writing new WAL back at the start of the record
7450 : : * that was broken, because downstream consumers of that WAL (physical
7451 : : * replicas) are not prepared to "rewind". So the first action after
7452 : : * finishing replay of all valid WAL must be to write a record of this type
7453 : : * at the point where the contrecord was missing; to support xlogreader
7454 : : * detecting the special case, XLP_FIRST_IS_OVERWRITE_CONTRECORD is also added
7455 : : * to the page header where the record occurs. xlogreader has an ad-hoc
7456 : : * mechanism to report metadata about the broken record, which is what we
7457 : : * use here.
7458 : : *
7459 : : * At replay time, XLP_FIRST_IS_OVERWRITE_CONTRECORD instructs xlogreader to
7460 : : * skip the record it was reading, and pass back the LSN of the skipped
7461 : : * record, so that its caller can verify (on "replay" of that record) that the
7462 : : * XLOG_OVERWRITE_CONTRECORD matches what was effectively overwritten.
7463 : : *
7464 : : * 'aborted_lsn' is the beginning position of the record that was incomplete.
7465 : : * It is included in the WAL record. 'pagePtr' and 'newTLI' point to the
7466 : : * beginning of the XLOG page where the record is to be inserted. They must
7467 : : * match the current WAL insert position, they're passed here just so that we
7468 : : * can verify that.
7469 : : */
7470 : : static XLogRecPtr
1298 heikki.linnakangas@i 7471 : 11 : CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn, XLogRecPtr pagePtr,
7472 : : TimeLineID newTLI)
7473 : : {
7474 : : xl_overwrite_contrecord xlrec;
7475 : : XLogRecPtr recptr;
7476 : : XLogPageHeader pagehdr;
7477 : : XLogRecPtr startPos;
7478 : :
7479 : : /* sanity checks */
1438 alvherre@alvh.no-ip. 7480 [ - + ]: 11 : if (!RecoveryInProgress())
1438 alvherre@alvh.no-ip. 7481 [ # # ]:UBC 0 : elog(ERROR, "can only be used at end of recovery");
1298 heikki.linnakangas@i 7482 [ - + ]:CBC 11 : if (pagePtr % XLOG_BLCKSZ != 0)
61 alvherre@kurilemu.de 7483 [ # # ]:UNC 0 : elog(ERROR, "invalid position for missing continuation record %X/%08X",
7484 : : LSN_FORMAT_ARGS(pagePtr));
7485 : :
7486 : : /* The current WAL insert position should be right after the page header */
1298 heikki.linnakangas@i 7487 :CBC 11 : startPos = pagePtr;
7488 [ + + ]: 11 : if (XLogSegmentOffset(startPos, wal_segment_size) == 0)
7489 : 1 : startPos += SizeOfXLogLongPHD;
7490 : : else
7491 : 10 : startPos += SizeOfXLogShortPHD;
7492 : 11 : recptr = GetXLogInsertRecPtr();
7493 [ - + ]: 11 : if (recptr != startPos)
61 alvherre@kurilemu.de 7494 [ # # ]:UNC 0 : elog(ERROR, "invalid WAL insert position %X/%08X for OVERWRITE_CONTRECORD",
7495 : : LSN_FORMAT_ARGS(recptr));
7496 : :
1438 alvherre@alvh.no-ip. 7497 :CBC 11 : START_CRIT_SECTION();
7498 : :
7499 : : /*
7500 : : * Initialize the XLOG page header (by GetXLogBuffer), and set the
7501 : : * XLP_FIRST_IS_OVERWRITE_CONTRECORD flag.
7502 : : *
7503 : : * No other backend is allowed to write WAL yet, so acquiring the WAL
7504 : : * insertion lock is just pro forma.
7505 : : */
1298 heikki.linnakangas@i 7506 : 11 : WALInsertLockAcquire();
7507 : 11 : pagehdr = (XLogPageHeader) GetXLogBuffer(pagePtr, newTLI);
7508 : 11 : pagehdr->xlp_info |= XLP_FIRST_IS_OVERWRITE_CONTRECORD;
7509 : 11 : WALInsertLockRelease();
7510 : :
7511 : : /*
7512 : : * Insert the XLOG_OVERWRITE_CONTRECORD record as the first record on the
7513 : : * page. We know it becomes the first record, because no other backend is
7514 : : * allowed to write WAL yet.
7515 : : */
1438 alvherre@alvh.no-ip. 7516 : 11 : XLogBeginInsert();
1298 heikki.linnakangas@i 7517 : 11 : xlrec.overwritten_lsn = aborted_lsn;
7518 : 11 : xlrec.overwrite_time = GetCurrentTimestamp();
207 peter@eisentraut.org 7519 : 11 : XLogRegisterData(&xlrec, sizeof(xl_overwrite_contrecord));
1438 alvherre@alvh.no-ip. 7520 : 11 : recptr = XLogInsert(RM_XLOG_ID, XLOG_OVERWRITE_CONTRECORD);
7521 : :
7522 : : /* check that the record was inserted to the right place */
1298 heikki.linnakangas@i 7523 [ - + ]: 11 : if (ProcLastRecPtr != startPos)
61 alvherre@kurilemu.de 7524 [ # # ]:UNC 0 : elog(ERROR, "OVERWRITE_CONTRECORD was inserted to unexpected position %X/%08X",
7525 : : LSN_FORMAT_ARGS(ProcLastRecPtr));
7526 : :
1438 alvherre@alvh.no-ip. 7527 :CBC 11 : XLogFlush(recptr);
7528 : :
7529 [ - + ]: 11 : END_CRIT_SECTION();
7530 : :
7531 : 11 : return recptr;
7532 : : }
7533 : :
7534 : : /*
7535 : : * Flush all data in shared memory to disk, and fsync
7536 : : *
7537 : : * This is the common code shared between regular checkpoints and
7538 : : * recovery restartpoints.
7539 : : */
7540 : : static void
6645 tgl@sss.pgh.pa.us 7541 : 1677 : CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
7542 : : {
5690 7543 : 1677 : CheckPointRelationMap();
723 akapila@postgresql.o 7544 : 1677 : CheckPointReplicationSlots(flags & CHECKPOINT_IS_SHUTDOWN);
4205 rhaas@postgresql.org 7545 : 1677 : CheckPointSnapBuild();
7546 : 1677 : CheckPointLogicalRewriteHeap();
3783 andres@anarazel.de 7547 : 1677 : CheckPointReplicationOrigin();
7548 : :
7549 : : /* Write out all dirty data in SLRUs and the main buffer pool */
7550 : : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_START(flags);
1807 tmunro@postgresql.or 7551 : 1677 : CheckpointStats.ckpt_write_t = GetCurrentTimestamp();
7552 : 1677 : CheckPointCLOG();
7553 : 1677 : CheckPointCommitTs();
7554 : 1677 : CheckPointSUBTRANS();
7555 : 1677 : CheckPointMultiXact();
7556 : 1677 : CheckPointPredicate();
7557 : 1677 : CheckPointBuffers(flags);
7558 : :
7559 : : /* Perform all queued up fsyncs */
7560 : : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_SYNC_START();
7561 : 1677 : CheckpointStats.ckpt_sync_t = GetCurrentTimestamp();
7562 : 1677 : ProcessSyncRequests();
7563 : 1677 : CheckpointStats.ckpt_sync_end_t = GetCurrentTimestamp();
7564 : : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_DONE();
7565 : :
7566 : : /* We deliberately delay 2PC checkpointing as long as possible */
6970 tgl@sss.pgh.pa.us 7567 : 1677 : CheckPointTwoPhase(checkPointRedo);
7568 : 1677 : }
7569 : :
7570 : : /*
7571 : : * Save a checkpoint for recovery restart if appropriate
7572 : : *
7573 : : * This function is called each time a checkpoint record is read from XLOG.
7574 : : * It must determine whether the checkpoint represents a safe restartpoint or
7575 : : * not. If so, the checkpoint record is stashed in shared memory so that
7576 : : * CreateRestartPoint can consult it. (Note that the latter function is
7577 : : * executed by the checkpointer, while this one will be executed by the
7578 : : * startup process.)
7579 : : */
7580 : : static void
1382 rhaas@postgresql.org 7581 : 694 : RecoveryRestartPoint(const CheckPoint *checkPoint, XLogReaderState *record)
7582 : : {
7583 : : /*
7584 : : * Also refrain from creating a restartpoint if we have seen any
7585 : : * references to non-existent pages. Restarting recovery from the
7586 : : * restartpoint would not see the references, so we would lose the
7587 : : * cross-check that the pages belonged to a relation that was dropped
7588 : : * later.
7589 : : */
5027 heikki.linnakangas@i 7590 [ - + ]: 694 : if (XLogHaveInvalidPages())
7591 : : {
635 michael@paquier.xyz 7592 [ # # ]:UBC 0 : elog(DEBUG2,
7593 : : "could not record restart point at %X/%08X because there are unresolved references to invalid pages",
7594 : : LSN_FORMAT_ARGS(checkPoint->redo));
5027 heikki.linnakangas@i 7595 : 0 : return;
7596 : : }
7597 : :
7598 : : /*
7599 : : * Copy the checkpoint record to shared memory, so that checkpointer can
7600 : : * work out the next time it wants to perform a restartpoint.
7601 : : */
4002 andres@anarazel.de 7602 [ - + ]:CBC 694 : SpinLockAcquire(&XLogCtl->info_lck);
1382 rhaas@postgresql.org 7603 : 694 : XLogCtl->lastCheckPointRecPtr = record->ReadRecPtr;
7604 : 694 : XLogCtl->lastCheckPointEndPtr = record->EndRecPtr;
4002 andres@anarazel.de 7605 : 694 : XLogCtl->lastCheckPoint = *checkPoint;
7606 : 694 : SpinLockRelease(&XLogCtl->info_lck);
7607 : : }
7608 : :
7609 : : /*
7610 : : * Establish a restartpoint if possible.
7611 : : *
7612 : : * This is similar to CreateCheckPoint, but is used during WAL recovery
7613 : : * to establish a point from which recovery can roll forward without
7614 : : * replaying the entire recovery log.
7615 : : *
7616 : : * Returns true if a new restartpoint was established. We can only establish
7617 : : * a restartpoint if we have replayed a safe checkpoint record since last
7618 : : * restartpoint.
7619 : : */
7620 : : bool
6044 heikki.linnakangas@i 7621 : 588 : CreateRestartPoint(int flags)
7622 : : {
7623 : : XLogRecPtr lastCheckPointRecPtr;
7624 : : XLogRecPtr lastCheckPointEndPtr;
7625 : : CheckPoint lastCheckPoint;
7626 : : XLogRecPtr PriorRedoPtr;
7627 : : XLogRecPtr receivePtr;
7628 : : XLogRecPtr replayPtr;
7629 : : TimeLineID replayTLI;
7630 : : XLogRecPtr endptr;
7631 : : XLogSegNo _logSegNo;
7632 : : TimestampTz xtime;
7633 : :
7634 : : /* Concurrent checkpoint/restartpoint cannot happen */
1216 michael@paquier.xyz 7635 [ + - - + ]: 588 : Assert(!IsUnderPostmaster || MyBackendType == B_CHECKPOINTER);
7636 : :
7637 : : /* Get a local copy of the last safe checkpoint record. */
4002 andres@anarazel.de 7638 [ - + ]: 588 : SpinLockAcquire(&XLogCtl->info_lck);
7639 : 588 : lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
3236 rhaas@postgresql.org 7640 : 588 : lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
4002 andres@anarazel.de 7641 : 588 : lastCheckPoint = XLogCtl->lastCheckPoint;
7642 : 588 : SpinLockRelease(&XLogCtl->info_lck);
7643 : :
7644 : : /*
7645 : : * Check that we're still in recovery mode. It's ok if we exit recovery
7646 : : * mode after this check, the restart point is valid anyway.
7647 : : */
6044 heikki.linnakangas@i 7648 [ - + ]: 588 : if (!RecoveryInProgress())
7649 : : {
6044 heikki.linnakangas@i 7650 [ # # ]:UBC 0 : ereport(DEBUG2,
7651 : : (errmsg_internal("skipping restartpoint, recovery has already ended")));
7652 : 0 : return false;
7653 : : }
7654 : :
7655 : : /*
7656 : : * If the last checkpoint record we've replayed is already our last
7657 : : * restartpoint, we can't perform a new restart point. We still update
7658 : : * minRecoveryPoint in that case, so that if this is a shutdown restart
7659 : : * point, we won't start up earlier than before. That's not strictly
7660 : : * necessary, but when hot standby is enabled, it would be rather weird if
7661 : : * the database opened up for read-only connections at a point-in-time
7662 : : * before the last shutdown. Such time travel is still possible in case of
7663 : : * immediate shutdown, though.
7664 : : *
7665 : : * We don't explicitly advance minRecoveryPoint when we do create a
7666 : : * restartpoint. It's assumed that flushing the buffers will do that as a
7667 : : * side-effect.
7668 : : */
6044 heikki.linnakangas@i 7669 [ + + ]:CBC 588 : if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
4635 alvherre@alvh.no-ip. 7670 [ + + ]: 256 : lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
7671 : : {
6044 heikki.linnakangas@i 7672 [ - + ]: 390 : ereport(DEBUG2,
7673 : : errmsg_internal("skipping restartpoint, already performed at %X/%08X",
7674 : : LSN_FORMAT_ARGS(lastCheckPoint.redo)));
7675 : :
7676 : 390 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5574 rhaas@postgresql.org 7677 [ + + ]: 390 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7678 : : {
7679 : 30 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7680 : 30 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7681 : 30 : UpdateControlFile();
7682 : 30 : LWLockRelease(ControlFileLock);
7683 : : }
6044 heikki.linnakangas@i 7684 : 390 : return false;
7685 : : }
7686 : :
7687 : : /*
7688 : : * Update the shared RedoRecPtr so that the startup process can calculate
7689 : : * the number of segments replayed since last restartpoint, and request a
7690 : : * restartpoint if it exceeds CheckPointSegments.
7691 : : *
7692 : : * Like in CreateCheckPoint(), hold off insertions to update it, although
7693 : : * during recovery this is just pro forma, because no WAL insertions are
7694 : : * happening.
7695 : : */
4187 7696 : 198 : WALInsertLockAcquireExclusive();
3848 7697 : 198 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
4187 7698 : 198 : WALInsertLockRelease();
7699 : :
7700 : : /* Also update the info_lck-protected copy */
4002 andres@anarazel.de 7701 [ - + ]: 198 : SpinLockAcquire(&XLogCtl->info_lck);
7702 : 198 : XLogCtl->RedoRecPtr = lastCheckPoint.redo;
7703 : 198 : SpinLockRelease(&XLogCtl->info_lck);
7704 : :
7705 : : /*
7706 : : * Prepare to accumulate statistics.
7707 : : *
7708 : : * Note: because it is possible for log_checkpoints to change while a
7709 : : * checkpoint proceeds, we always accumulate stats, even if
7710 : : * log_checkpoints is currently off.
7711 : : */
5330 rhaas@postgresql.org 7712 [ + - + - : 2178 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
+ - + - +
+ ]
7713 : 198 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
7714 : :
7715 [ + - ]: 198 : if (log_checkpoints)
6044 heikki.linnakangas@i 7716 : 198 : LogCheckpointStart(flags, true);
7717 : :
7718 : : /* Update the process title */
1727 michael@paquier.xyz 7719 : 198 : update_checkpoint_display(flags, true, false);
7720 : :
6044 heikki.linnakangas@i 7721 : 198 : CheckPointGuts(lastCheckPoint.redo, flags);
7722 : :
7723 : : /*
7724 : : * This location needs to be after CheckPointGuts() to ensure that some
7725 : : * work has already happened during this checkpoint.
7726 : : */
7727 : : INJECTION_POINT("create-restart-point", NULL);
7728 : :
7729 : : /*
7730 : : * Remember the prior checkpoint's redo ptr for
7731 : : * UpdateCheckPointDistanceEstimate()
7732 : : */
3848 7733 : 198 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7734 : :
7735 : : /*
7736 : : * Update pg_control, using current time. Check that it still shows an
7737 : : * older checkpoint, else do nothing; this is a quick hack to make sure
7738 : : * nothing really bad happens if somehow we get here after the
7739 : : * end-of-recovery checkpoint.
7740 : : */
6044 7741 : 198 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
1216 michael@paquier.xyz 7742 [ + - ]: 198 : if (ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
7743 : : {
7744 : : /*
7745 : : * Update the checkpoint information. We do this even if the cluster
7746 : : * does not show DB_IN_ARCHIVE_RECOVERY to match with the set of WAL
7747 : : * segments recycled below.
7748 : : */
5916 tgl@sss.pgh.pa.us 7749 : 198 : ControlFile->checkPoint = lastCheckPointRecPtr;
7750 : 198 : ControlFile->checkPointCopy = lastCheckPoint;
7751 : :
7752 : : /*
7753 : : * Ensure minRecoveryPoint is past the checkpoint record and update it
7754 : : * if the control file still shows DB_IN_ARCHIVE_RECOVERY. Normally,
7755 : : * this will have happened already while writing out dirty buffers,
7756 : : * but not necessarily - e.g. because no buffers were dirtied. We do
7757 : : * this because a backup performed in recovery uses minRecoveryPoint
7758 : : * to determine which WAL files must be included in the backup, and
7759 : : * the file (or files) containing the checkpoint record must be
7760 : : * included, at a minimum. Note that for an ordinary restart of
7761 : : * recovery there's no value in having the minimum recovery point any
7762 : : * earlier than this anyway, because redo will begin just after the
7763 : : * checkpoint record.
7764 : : */
1216 michael@paquier.xyz 7765 [ + - ]: 198 : if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY)
7766 : : {
7767 [ + + ]: 198 : if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
7768 : : {
7769 : 19 : ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
7770 : 19 : ControlFile->minRecoveryPointTLI = lastCheckPoint.ThisTimeLineID;
7771 : :
7772 : : /* update local copy */
7773 : 19 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
7774 : 19 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
7775 : : }
7776 [ + + ]: 198 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7777 : 22 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7778 : : }
5916 tgl@sss.pgh.pa.us 7779 : 198 : UpdateControlFile();
7780 : : }
6044 heikki.linnakangas@i 7781 : 198 : LWLockRelease(ControlFileLock);
7782 : :
7783 : : /*
7784 : : * Update the average distance between checkpoints/restartpoints if the
7785 : : * prior checkpoint exists.
7786 : : */
3848 7787 [ + - ]: 198 : if (PriorRedoPtr != InvalidXLogRecPtr)
7788 : 198 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7789 : :
7790 : : /*
7791 : : * Delete old log files, those no longer needed for last restartpoint to
7792 : : * prevent the disk holding the xlog from growing full.
7793 : : */
2601 michael@paquier.xyz 7794 : 198 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7795 : :
7796 : : /*
7797 : : * Retreat _logSegNo using the current end of xlog replayed or received,
7798 : : * whichever is later.
7799 : : */
1977 tmunro@postgresql.or 7800 : 198 : receivePtr = GetWalRcvFlushRecPtr(NULL, NULL);
2601 michael@paquier.xyz 7801 : 198 : replayPtr = GetXLogReplayRecPtr(&replayTLI);
7802 : 198 : endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
7803 : 198 : KeepLogSeg(endptr, &_logSegNo);
199 akapila@postgresql.o 7804 [ + + ]: 198 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED | RS_INVAL_IDLE_TIMEOUT,
7805 : : _logSegNo, InvalidOid,
7806 : : InvalidTransactionId))
7807 : : {
7808 : : /*
7809 : : * Some slots have been invalidated; recalculate the old-segment
7810 : : * horizon, starting again from RedoRecPtr.
7811 : : */
1513 alvherre@alvh.no-ip. 7812 : 1 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7813 : 1 : KeepLogSeg(endptr, &_logSegNo);
7814 : : }
2601 michael@paquier.xyz 7815 : 198 : _logSegNo--;
7816 : :
7817 : : /*
7818 : : * Try to recycle segments on a useful timeline. If we've been promoted
7819 : : * since the beginning of this restartpoint, use the new timeline chosen
7820 : : * at end of recovery. If we're still in recovery, use the timeline we're
7821 : : * currently replaying.
7822 : : *
7823 : : * There is no guarantee that the WAL segments will be useful on the
7824 : : * current timeline; if recovery proceeds to a new timeline right after
7825 : : * this, the pre-allocated WAL segments on this timeline will not be used,
7826 : : * and will go wasted until recycled on the next restartpoint. We'll live
7827 : : * with that.
7828 : : */
1401 rhaas@postgresql.org 7829 [ - + ]: 198 : if (!RecoveryInProgress())
1396 rhaas@postgresql.org 7830 :UBC 0 : replayTLI = XLogCtl->InsertTimeLineID;
7831 : :
1401 rhaas@postgresql.org 7832 :CBC 198 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, endptr, replayTLI);
7833 : :
7834 : : /*
7835 : : * Make more log segments if needed. (Do this after recycling old log
7836 : : * segments, since that may supply some of the needed files.)
7837 : : */
7838 : 198 : PreallocXlogFiles(endptr, replayTLI);
7839 : :
7840 : : /*
7841 : : * Truncate pg_subtrans if possible. We can throw away all data before
7842 : : * the oldest XMIN of any running transaction. No future transaction will
7843 : : * attempt to reference any pg_subtrans entry older than that (see Asserts
7844 : : * in subtrans.c). When hot standby is disabled, though, we mustn't do
7845 : : * this because StartupSUBTRANS hasn't been called yet.
7846 : : */
5486 simon@2ndQuadrant.co 7847 [ + - ]: 198 : if (EnableHotStandby)
1851 andres@anarazel.de 7848 : 198 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7849 : :
7850 : : /* Real work is done; log and update stats. */
4902 rhaas@postgresql.org 7851 : 198 : LogCheckpointEnd(true);
7852 : :
7853 : : /* Reset the process title */
1727 michael@paquier.xyz 7854 : 198 : update_checkpoint_display(flags, true, true);
7855 : :
5544 tgl@sss.pgh.pa.us 7856 : 198 : xtime = GetLatestXTime();
6044 heikki.linnakangas@i 7857 [ + - + - : 198 : ereport((log_checkpoints ? LOG : DEBUG2),
+ + ]
7858 : : errmsg("recovery restart point at %X/%08X",
7859 : : LSN_FORMAT_ARGS(lastCheckPoint.redo)),
7860 : : xtime ? errdetail("Last completed transaction was at log time %s.",
7861 : : timestamptz_to_str(xtime)) : 0);
7862 : :
7863 : : /*
7864 : : * Finally, execute archive_cleanup_command, if any.
7865 : : */
2477 peter_e@gmx.net 7866 [ + - - + ]: 198 : if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
943 michael@paquier.xyz 7867 :UBC 0 : ExecuteRecoveryCommand(archiveCleanupCommand,
7868 : : "archive_cleanup_command",
7869 : : false,
7870 : : WAIT_EVENT_ARCHIVE_CLEANUP_COMMAND);
7871 : :
6044 heikki.linnakangas@i 7872 :CBC 198 : return true;
7873 : : }
7874 : :
7875 : : /*
7876 : : * Report availability of WAL for the given target LSN
7877 : : * (typically a slot's restart_lsn)
7878 : : *
7879 : : * Returns one of the following enum values:
7880 : : *
7881 : : * * WALAVAIL_RESERVED means targetLSN is available and it is in the range of
7882 : : * max_wal_size.
7883 : : *
7884 : : * * WALAVAIL_EXTENDED means it is still available by preserving extra
7885 : : * segments beyond max_wal_size. If max_slot_wal_keep_size is smaller
7886 : : * than max_wal_size, this state is not returned.
7887 : : *
7888 : : * * WALAVAIL_UNRESERVED means it is being lost and the next checkpoint will
7889 : : * remove reserved segments. The walsender using this slot may return to the
7890 : : * above.
7891 : : *
7892 : : * * WALAVAIL_REMOVED means it has been removed. A replication stream on
7893 : : * a slot with this LSN cannot continue. (Any associated walsender
7894 : : * processes should have been terminated already.)
7895 : : *
7896 : : * * WALAVAIL_INVALID_LSN means the slot hasn't been set to reserve WAL.
7897 : : */
7898 : : WALAvailability
1978 alvherre@alvh.no-ip. 7899 : 381 : GetWALAvailability(XLogRecPtr targetLSN)
7900 : : {
7901 : : XLogRecPtr currpos; /* current write LSN */
7902 : : XLogSegNo currSeg; /* segid of currpos */
7903 : : XLogSegNo targetSeg; /* segid of targetLSN */
7904 : : XLogSegNo oldestSeg; /* actual oldest segid */
7905 : : XLogSegNo oldestSegMaxWalSize; /* oldest segid kept by max_wal_size */
7906 : : XLogSegNo oldestSlotSeg; /* oldest segid kept by slot */
7907 : : uint64 keepSegs;
7908 : :
7909 : : /*
7910 : : * slot does not reserve WAL. Either deactivated, or has never been active
7911 : : */
7912 [ + + ]: 381 : if (XLogRecPtrIsInvalid(targetLSN))
7913 : 20 : return WALAVAIL_INVALID_LSN;
7914 : :
7915 : : /*
7916 : : * Calculate the oldest segment currently reserved by all slots,
7917 : : * considering wal_keep_size and max_slot_wal_keep_size. Initialize
7918 : : * oldestSlotSeg to the current segment.
7919 : : */
1881 7920 : 361 : currpos = GetXLogWriteRecPtr();
7921 : 361 : XLByteToSeg(currpos, oldestSlotSeg, wal_segment_size);
1978 7922 : 361 : KeepLogSeg(currpos, &oldestSlotSeg);
7923 : :
7924 : : /*
7925 : : * Find the oldest extant segment file. We get 1 until checkpoint removes
7926 : : * the first WAL segment file since startup, which causes the status being
7927 : : * wrong under certain abnormal conditions but that doesn't actually harm.
7928 : : */
7929 : 361 : oldestSeg = XLogGetLastRemovedSegno() + 1;
7930 : :
7931 : : /* calculate oldest segment by max_wal_size */
7932 : 361 : XLByteToSeg(currpos, currSeg, wal_segment_size);
1900 7933 : 361 : keepSegs = ConvertToXSegs(max_wal_size_mb, wal_segment_size) + 1;
7934 : :
1978 7935 [ + + ]: 361 : if (currSeg > keepSegs)
7936 : 8 : oldestSegMaxWalSize = currSeg - keepSegs;
7937 : : else
7938 : 353 : oldestSegMaxWalSize = 1;
7939 : :
7940 : : /* the segment we care about */
1881 7941 : 361 : XLByteToSeg(targetLSN, targetSeg, wal_segment_size);
7942 : :
7943 : : /*
7944 : : * No point in returning reserved or extended status values if the
7945 : : * targetSeg is known to be lost.
7946 : : */
1900 7947 [ + + ]: 361 : if (targetSeg >= oldestSlotSeg)
7948 : : {
7949 : : /* show "reserved" when targetSeg is within max_wal_size */
7950 [ + + ]: 360 : if (targetSeg >= oldestSegMaxWalSize)
1978 7951 : 358 : return WALAVAIL_RESERVED;
7952 : :
7953 : : /* being retained by slots exceeding max_wal_size */
1900 7954 : 2 : return WALAVAIL_EXTENDED;
7955 : : }
7956 : :
7957 : : /* WAL segments are no longer retained but haven't been removed yet */
7958 [ + - ]: 1 : if (targetSeg >= oldestSeg)
7959 : 1 : return WALAVAIL_UNRESERVED;
7960 : :
7961 : : /* Definitely lost */
1978 alvherre@alvh.no-ip. 7962 :UBC 0 : return WALAVAIL_REMOVED;
7963 : : }
7964 : :
7965 : :
7966 : : /*
7967 : : * Retreat *logSegNo to the last segment that we need to retain because of
7968 : : * either wal_keep_size or replication slots.
7969 : : *
7970 : : * This is calculated by subtracting wal_keep_size from the given xlog
7971 : : * location, recptr and by making sure that that result is below the
7972 : : * requirement of replication slots. For the latter criterion we do consider
7973 : : * the effects of max_slot_wal_keep_size: reserve at most that much space back
7974 : : * from recptr.
7975 : : *
7976 : : * Note about replication slots: if this function calculates a value
7977 : : * that's further ahead than what slots need reserved, then affected
7978 : : * slots need to be invalidated and this function invoked again.
7979 : : * XXX it might be a good idea to rewrite this function so that
7980 : : * invalidation is optionally done here, instead.
7981 : : */
7982 : : static void
4822 heikki.linnakangas@i 7983 :CBC 2042 : KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
7984 : : {
7985 : : XLogSegNo currSegNo;
7986 : : XLogSegNo segno;
7987 : : XLogRecPtr keep;
7988 : :
1978 alvherre@alvh.no-ip. 7989 : 2042 : XLByteToSeg(recptr, currSegNo, wal_segment_size);
7990 : 2042 : segno = currSegNo;
7991 : :
7992 : : /* Calculate how many segments are kept by slots. */
7993 : 2042 : keep = XLogGetReplicationSlotMinimumLSN();
863 nathan@postgresql.or 7994 [ + + + + ]: 2042 : if (keep != InvalidXLogRecPtr && keep < recptr)
7995 : : {
1978 alvherre@alvh.no-ip. 7996 : 621 : XLByteToSeg(keep, segno, wal_segment_size);
7997 : :
7998 : : /*
7999 : : * Account for max_slot_wal_keep_size to avoid keeping more than
8000 : : * configured. However, don't do that during a binary upgrade: if
8001 : : * slots were to be invalidated because of this, it would not be
8002 : : * possible to preserve logical ones during the upgrade.
8003 : : */
57 akapila@postgresql.o 8004 [ + + + - ]: 621 : if (max_slot_wal_keep_size_mb >= 0 && !IsBinaryUpgrade)
8005 : : {
8006 : : uint64 slot_keep_segs;
8007 : :
1978 alvherre@alvh.no-ip. 8008 : 21 : slot_keep_segs =
8009 : 21 : ConvertToXSegs(max_slot_wal_keep_size_mb, wal_segment_size);
8010 : :
8011 [ + + ]: 21 : if (currSegNo - segno > slot_keep_segs)
8012 : 5 : segno = currSegNo - slot_keep_segs;
8013 : : }
8014 : : }
8015 : :
8016 : : /*
8017 : : * If WAL summarization is in use, don't remove WAL that has yet to be
8018 : : * summarized.
8019 : : */
438 rhaas@postgresql.org 8020 : 2042 : keep = GetOldestUnsummarizedLSN(NULL, NULL);
626 8021 [ + + ]: 2042 : if (keep != InvalidXLogRecPtr)
8022 : : {
8023 : : XLogSegNo unsummarized_segno;
8024 : :
8025 : 2 : XLByteToSeg(keep, unsummarized_segno, wal_segment_size);
8026 [ + - ]: 2 : if (unsummarized_segno < segno)
8027 : 2 : segno = unsummarized_segno;
8028 : : }
8029 : :
8030 : : /* but, keep at least wal_keep_size if that's set */
1874 fujii@postgresql.org 8031 [ + + ]: 2042 : if (wal_keep_size_mb > 0)
8032 : : {
8033 : : uint64 keep_segs;
8034 : :
8035 : 69 : keep_segs = ConvertToXSegs(wal_keep_size_mb, wal_segment_size);
8036 [ + - ]: 69 : if (currSegNo - segno < keep_segs)
8037 : : {
8038 : : /* avoid underflow, don't go below 1 */
8039 [ + + ]: 69 : if (currSegNo <= keep_segs)
8040 : 65 : segno = 1;
8041 : : else
8042 : 4 : segno = currSegNo - keep_segs;
8043 : : }
8044 : : }
8045 : :
8046 : : /* don't delete WAL segments newer than the calculated segment */
1881 alvherre@alvh.no-ip. 8047 [ + + ]: 2042 : if (segno < *logSegNo)
4822 heikki.linnakangas@i 8048 : 354 : *logSegNo = segno;
5163 simon@2ndQuadrant.co 8049 : 2042 : }
8050 : :
8051 : : /*
8052 : : * Write a NEXTOID log record
8053 : : */
8054 : : void
9073 vadim4o@yahoo.com 8055 : 582 : XLogPutNextOid(Oid nextOid)
8056 : : {
3943 heikki.linnakangas@i 8057 : 582 : XLogBeginInsert();
207 peter@eisentraut.org 8058 : 582 : XLogRegisterData(&nextOid, sizeof(Oid));
3943 heikki.linnakangas@i 8059 : 582 : (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
8060 : :
8061 : : /*
8062 : : * We need not flush the NEXTOID record immediately, because any of the
8063 : : * just-allocated OIDs could only reach disk as part of a tuple insert or
8064 : : * update that would have its own XLOG record that must follow the NEXTOID
8065 : : * record. Therefore, the standard buffer LSN interlock applied to those
8066 : : * records will ensure no such OID reaches disk before the NEXTOID record
8067 : : * does.
8068 : : *
8069 : : * Note, however, that the above statement only covers state "within" the
8070 : : * database. When we use a generated OID as a file or directory name, we
8071 : : * are in a sense violating the basic WAL rule, because that filesystem
8072 : : * change may reach disk before the NEXTOID WAL record does. The impact
8073 : : * of this is that if a database crash occurs immediately afterward, we
8074 : : * might after restart re-generate the same OID and find that it conflicts
8075 : : * with the leftover file or directory. But since for safety's sake we
8076 : : * always loop until finding a nonconflicting filename, this poses no real
8077 : : * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
8078 : : */
7436 tgl@sss.pgh.pa.us 8079 : 582 : }
8080 : :
8081 : : /*
8082 : : * Write an XLOG SWITCH record.
8083 : : *
8084 : : * Here we just blindly issue an XLogInsert request for the record.
8085 : : * All the magic happens inside XLogInsert.
8086 : : *
8087 : : * The return value is either the end+1 address of the switch record,
8088 : : * or the end+1 address of the prior segment if we did not need to
8089 : : * write a switch record because we are already at segment start.
8090 : : */
8091 : : XLogRecPtr
3180 andres@anarazel.de 8092 : 707 : RequestXLogSwitch(bool mark_unimportant)
8093 : : {
8094 : : XLogRecPtr RecPtr;
8095 : :
8096 : : /* XLOG SWITCH has no data */
3943 heikki.linnakangas@i 8097 : 707 : XLogBeginInsert();
8098 : :
3180 andres@anarazel.de 8099 [ - + ]: 707 : if (mark_unimportant)
3180 andres@anarazel.de 8100 :UBC 0 : XLogSetRecordFlags(XLOG_MARK_UNIMPORTANT);
3943 heikki.linnakangas@i 8101 :CBC 707 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH);
8102 : :
6971 tgl@sss.pgh.pa.us 8103 : 707 : return RecPtr;
8104 : : }
8105 : :
8106 : : /*
8107 : : * Write a RESTORE POINT record
8108 : : */
8109 : : XLogRecPtr
5324 simon@2ndQuadrant.co 8110 : 3 : XLogRestorePoint(const char *rpName)
8111 : : {
8112 : : XLogRecPtr RecPtr;
8113 : : xl_restore_point xlrec;
8114 : :
8115 : 3 : xlrec.rp_time = GetCurrentTimestamp();
4219 tgl@sss.pgh.pa.us 8116 : 3 : strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
8117 : :
3943 heikki.linnakangas@i 8118 : 3 : XLogBeginInsert();
207 peter@eisentraut.org 8119 : 3 : XLogRegisterData(&xlrec, sizeof(xl_restore_point));
8120 : :
3943 heikki.linnakangas@i 8121 : 3 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
8122 : :
5308 rhaas@postgresql.org 8123 [ + - ]: 3 : ereport(LOG,
8124 : : errmsg("restore point \"%s\" created at %X/%08X",
8125 : : rpName, LSN_FORMAT_ARGS(RecPtr)));
8126 : :
5324 simon@2ndQuadrant.co 8127 : 3 : return RecPtr;
8128 : : }
8129 : :
8130 : : /*
8131 : : * Check if any of the GUC parameters that are critical for hot standby
8132 : : * have changed, and update the value in pg_control file if necessary.
8133 : : */
8134 : : static void
5610 heikki.linnakangas@i 8135 : 832 : XLogReportParameters(void)
8136 : : {
8137 [ + + ]: 832 : if (wal_level != ControlFile->wal_level ||
4266 rhaas@postgresql.org 8138 [ + + ]: 616 : wal_log_hints != ControlFile->wal_log_hints ||
5610 heikki.linnakangas@i 8139 [ + + ]: 534 : MaxConnections != ControlFile->MaxConnections ||
4447 rhaas@postgresql.org 8140 [ + + ]: 533 : max_worker_processes != ControlFile->max_worker_processes ||
2398 michael@paquier.xyz 8141 [ + + ]: 532 : max_wal_senders != ControlFile->max_wal_senders ||
5610 heikki.linnakangas@i 8142 [ + + ]: 512 : max_prepared_xacts != ControlFile->max_prepared_xacts ||
3930 alvherre@alvh.no-ip. 8143 [ + - ]: 426 : max_locks_per_xact != ControlFile->max_locks_per_xact ||
8144 [ + + ]: 426 : track_commit_timestamp != ControlFile->track_commit_timestamp)
8145 : : {
8146 : : /*
8147 : : * The change in number of backend slots doesn't need to be WAL-logged
8148 : : * if archiving is not enabled, as you can't start archive recovery
8149 : : * with wal_level=minimal anyway. We don't really care about the
8150 : : * values in pg_control either if wal_level=minimal, but seems better
8151 : : * to keep them up-to-date to avoid confusion.
8152 : : */
5610 heikki.linnakangas@i 8153 [ + + + + ]: 418 : if (wal_level != ControlFile->wal_level || XLogIsNeeded())
8154 : : {
8155 : : xl_parameter_change xlrec;
8156 : : XLogRecPtr recptr;
8157 : :
8158 : 398 : xlrec.MaxConnections = MaxConnections;
4447 rhaas@postgresql.org 8159 : 398 : xlrec.max_worker_processes = max_worker_processes;
2398 michael@paquier.xyz 8160 : 398 : xlrec.max_wal_senders = max_wal_senders;
5610 heikki.linnakangas@i 8161 : 398 : xlrec.max_prepared_xacts = max_prepared_xacts;
8162 : 398 : xlrec.max_locks_per_xact = max_locks_per_xact;
8163 : 398 : xlrec.wal_level = wal_level;
4266 rhaas@postgresql.org 8164 : 398 : xlrec.wal_log_hints = wal_log_hints;
3930 alvherre@alvh.no-ip. 8165 : 398 : xlrec.track_commit_timestamp = track_commit_timestamp;
8166 : :
3943 heikki.linnakangas@i 8167 : 398 : XLogBeginInsert();
207 peter@eisentraut.org 8168 : 398 : XLogRegisterData(&xlrec, sizeof(xlrec));
8169 : :
3943 heikki.linnakangas@i 8170 : 398 : recptr = XLogInsert(RM_XLOG_ID, XLOG_PARAMETER_CHANGE);
4182 fujii@postgresql.org 8171 : 398 : XLogFlush(recptr);
8172 : : }
8173 : :
1916 tmunro@postgresql.or 8174 : 418 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8175 : :
5610 heikki.linnakangas@i 8176 : 418 : ControlFile->MaxConnections = MaxConnections;
4447 rhaas@postgresql.org 8177 : 418 : ControlFile->max_worker_processes = max_worker_processes;
2398 michael@paquier.xyz 8178 : 418 : ControlFile->max_wal_senders = max_wal_senders;
5610 heikki.linnakangas@i 8179 : 418 : ControlFile->max_prepared_xacts = max_prepared_xacts;
8180 : 418 : ControlFile->max_locks_per_xact = max_locks_per_xact;
8181 : 418 : ControlFile->wal_level = wal_level;
4266 rhaas@postgresql.org 8182 : 418 : ControlFile->wal_log_hints = wal_log_hints;
3930 alvherre@alvh.no-ip. 8183 : 418 : ControlFile->track_commit_timestamp = track_commit_timestamp;
5610 heikki.linnakangas@i 8184 : 418 : UpdateControlFile();
8185 : :
1916 tmunro@postgresql.or 8186 : 418 : LWLockRelease(ControlFileLock);
8187 : : }
5708 heikki.linnakangas@i 8188 : 832 : }
8189 : :
8190 : : /*
8191 : : * Update full_page_writes in shared memory, and write an
8192 : : * XLOG_FPW_CHANGE record if necessary.
8193 : : *
8194 : : * Note: this function assumes there is no other process running
8195 : : * concurrently that could update it.
8196 : : */
8197 : : void
4973 simon@2ndQuadrant.co 8198 : 1380 : UpdateFullPageWrites(void)
8199 : : {
8200 : 1380 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8201 : : bool recoveryInProgress;
8202 : :
8203 : : /*
8204 : : * Do nothing if full_page_writes has not been changed.
8205 : : *
8206 : : * It's safe to check the shared full_page_writes without the lock,
8207 : : * because we assume that there is no concurrently running process which
8208 : : * can update it.
8209 : : */
8210 [ + + ]: 1380 : if (fullPageWrites == Insert->fullPageWrites)
8211 : 940 : return;
8212 : :
8213 : : /*
8214 : : * Perform this outside critical section so that the WAL insert
8215 : : * initialization done by RecoveryInProgress() doesn't trigger an
8216 : : * assertion failure.
8217 : : */
2535 akapila@postgresql.o 8218 : 440 : recoveryInProgress = RecoveryInProgress();
8219 : :
4932 heikki.linnakangas@i 8220 : 440 : START_CRIT_SECTION();
8221 : :
8222 : : /*
8223 : : * It's always safe to take full page images, even when not strictly
8224 : : * required, but not the other round. So if we're setting full_page_writes
8225 : : * to true, first set it true and then write the WAL record. If we're
8226 : : * setting it to false, first write the WAL record and then set the global
8227 : : * flag.
8228 : : */
8229 [ + + ]: 440 : if (fullPageWrites)
8230 : : {
4187 8231 : 430 : WALInsertLockAcquireExclusive();
4932 8232 : 430 : Insert->fullPageWrites = true;
4187 8233 : 430 : WALInsertLockRelease();
8234 : : }
8235 : :
8236 : : /*
8237 : : * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
8238 : : * full_page_writes during archive recovery, if required.
8239 : : */
2535 akapila@postgresql.o 8240 [ + + - + ]: 440 : if (XLogStandbyInfoActive() && !recoveryInProgress)
8241 : : {
3943 heikki.linnakangas@i 8242 :UBC 0 : XLogBeginInsert();
207 peter@eisentraut.org 8243 : 0 : XLogRegisterData(&fullPageWrites, sizeof(bool));
8244 : :
3943 heikki.linnakangas@i 8245 : 0 : XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
8246 : : }
8247 : :
4932 heikki.linnakangas@i 8248 [ + + ]:CBC 440 : if (!fullPageWrites)
8249 : : {
4187 8250 : 10 : WALInsertLockAcquireExclusive();
4932 8251 : 10 : Insert->fullPageWrites = false;
4187 8252 : 10 : WALInsertLockRelease();
8253 : : }
4932 8254 [ - + ]: 440 : END_CRIT_SECTION();
8255 : : }
8256 : :
8257 : : /*
8258 : : * XLOG resource manager's routines
8259 : : *
8260 : : * Definitions of info values are in include/catalog/pg_control.h, though
8261 : : * not all record types are related to control file updates.
8262 : : *
8263 : : * NOTE: Some XLOG record types that are directly related to WAL recovery
8264 : : * are handled in xlogrecovery_redo().
8265 : : */
8266 : : void
3943 8267 : 41435 : xlog_redo(XLogReaderState *record)
8268 : : {
8269 : 41435 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
8270 : 41435 : XLogRecPtr lsn = record->EndRecPtr;
8271 : :
8272 : : /*
8273 : : * In XLOG rmgr, backup blocks are only used by XLOG_FPI and
8274 : : * XLOG_FPI_FOR_HINT records.
8275 : : */
3939 8276 [ + + + + : 41435 : Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
- + ]
8277 : : !XLogRecHasAnyBlockRefs(record));
8278 : :
8938 tgl@sss.pgh.pa.us 8279 [ + + ]: 41435 : if (info == XLOG_NEXTOID)
8280 : : {
8281 : : Oid nextOid;
8282 : :
8283 : : /*
8284 : : * We used to try to take the maximum of TransamVariables->nextOid and
8285 : : * the recorded nextOid, but that fails if the OID counter wraps
8286 : : * around. Since no OID allocation should be happening during replay
8287 : : * anyway, better to just believe the record exactly. We still take
8288 : : * OidGenLock while setting the variable, just in case.
8289 : : */
9073 vadim4o@yahoo.com 8290 : 90 : memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
4961 tgl@sss.pgh.pa.us 8291 : 90 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
638 heikki.linnakangas@i 8292 : 90 : TransamVariables->nextOid = nextOid;
8293 : 90 : TransamVariables->oidCount = 0;
4961 tgl@sss.pgh.pa.us 8294 : 90 : LWLockRelease(OidGenLock);
8295 : : }
8943 8296 [ + + ]: 41345 : else if (info == XLOG_CHECKPOINT_SHUTDOWN)
8297 : : {
8298 : : CheckPoint checkPoint;
8299 : : TimeLineID replayTLI;
8300 : :
8301 : 30 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8302 : : /* In a SHUTDOWN checkpoint, believe the counters exactly */
4961 8303 : 30 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
638 heikki.linnakangas@i 8304 : 30 : TransamVariables->nextXid = checkPoint.nextXid;
4961 tgl@sss.pgh.pa.us 8305 : 30 : LWLockRelease(XidGenLock);
8306 : 30 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
638 heikki.linnakangas@i 8307 : 30 : TransamVariables->nextOid = checkPoint.nextOid;
8308 : 30 : TransamVariables->oidCount = 0;
4961 tgl@sss.pgh.pa.us 8309 : 30 : LWLockRelease(OidGenLock);
7395 8310 : 30 : MultiXactSetNextMXact(checkPoint.nextMulti,
8311 : : checkPoint.nextMultiOffset);
8312 : :
3633 andres@anarazel.de 8313 : 30 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8314 : : checkPoint.oldestMultiDB);
8315 : :
8316 : : /*
8317 : : * No need to set oldestClogXid here as well; it'll be set when we
8318 : : * redo an xl_clog_truncate if it changed since initialization.
8319 : : */
5680 tgl@sss.pgh.pa.us 8320 : 30 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
8321 : :
8322 : : /*
8323 : : * If we see a shutdown checkpoint while waiting for an end-of-backup
8324 : : * record, the backup was canceled and the end-of-backup record will
8325 : : * never arrive.
8326 : : */
4579 heikki.linnakangas@i 8327 [ + - ]: 30 : if (ArchiveRecoveryRequested &&
4973 simon@2ndQuadrant.co 8328 [ - + ]: 30 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint) &&
4973 simon@2ndQuadrant.co 8329 [ # # ]:UBC 0 : XLogRecPtrIsInvalid(ControlFile->backupEndPoint))
4961 tgl@sss.pgh.pa.us 8330 [ # # ]: 0 : ereport(PANIC,
8331 : : (errmsg("online backup was canceled, recovery cannot continue")));
8332 : :
8333 : : /*
8334 : : * If we see a shutdown checkpoint, we know that nothing was running
8335 : : * on the primary at this point. So fake-up an empty running-xacts
8336 : : * record and use that here and now. Recover additional standby state
8337 : : * for prepared transactions.
8338 : : */
5740 simon@2ndQuadrant.co 8339 [ + + ]:CBC 30 : if (standbyState >= STANDBY_INITIALIZED)
8340 : : {
8341 : : TransactionId *xids;
8342 : : int nxids;
8343 : : TransactionId oldestActiveXID;
8344 : : TransactionId latestCompletedXid;
8345 : : RunningTransactionsData running;
8346 : :
5625 heikki.linnakangas@i 8347 : 28 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
8348 : :
8349 : : /* Update pg_subtrans entries for any prepared transactions */
436 8350 : 28 : StandbyRecoverPreparedTransactions();
8351 : :
8352 : : /*
8353 : : * Construct a RunningTransactions snapshot representing a shut
8354 : : * down server, with only prepared transactions still alive. We're
8355 : : * never overflowed at this point because all subxids are listed
8356 : : * with their parent prepared transactions.
8357 : : */
5625 8358 : 28 : running.xcnt = nxids;
4661 simon@2ndQuadrant.co 8359 : 28 : running.subxcnt = 0;
436 heikki.linnakangas@i 8360 : 28 : running.subxid_status = SUBXIDS_IN_SUBTRANS;
1852 andres@anarazel.de 8361 : 28 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5625 heikki.linnakangas@i 8362 : 28 : running.oldestRunningXid = oldestActiveXID;
1852 andres@anarazel.de 8363 : 28 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5595 simon@2ndQuadrant.co 8364 [ - + ]: 28 : TransactionIdRetreat(latestCompletedXid);
5594 8365 [ - + ]: 28 : Assert(TransactionIdIsNormal(latestCompletedXid));
5595 8366 : 28 : running.latestCompletedXid = latestCompletedXid;
5625 heikki.linnakangas@i 8367 : 28 : running.xids = xids;
8368 : :
8369 : 28 : ProcArrayApplyRecoveryInfo(&running);
8370 : : }
8371 : :
8372 : : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
1916 tmunro@postgresql.or 8373 : 30 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
1852 andres@anarazel.de 8374 : 30 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
1916 tmunro@postgresql.or 8375 : 30 : LWLockRelease(ControlFileLock);
8376 : :
8377 : : /*
8378 : : * We should've already switched to the new TLI before replaying this
8379 : : * record.
8380 : : */
1298 heikki.linnakangas@i 8381 : 30 : (void) GetCurrentReplayRecPtr(&replayTLI);
1401 rhaas@postgresql.org 8382 [ - + ]: 30 : if (checkPoint.ThisTimeLineID != replayTLI)
4656 heikki.linnakangas@i 8383 [ # # ]:UBC 0 : ereport(PANIC,
8384 : : (errmsg("unexpected timeline ID %u (should be %u) in shutdown checkpoint record",
8385 : : checkPoint.ThisTimeLineID, replayTLI)));
8386 : :
1382 rhaas@postgresql.org 8387 :CBC 30 : RecoveryRestartPoint(&checkPoint, record);
8388 : : }
8943 tgl@sss.pgh.pa.us 8389 [ + + ]: 41315 : else if (info == XLOG_CHECKPOINT_ONLINE)
8390 : : {
8391 : : CheckPoint checkPoint;
8392 : : TimeLineID replayTLI;
8393 : :
8394 : 664 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8395 : : /* In an ONLINE checkpoint, treat the XID counter as a minimum */
4961 8396 : 664 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
638 heikki.linnakangas@i 8397 [ - + ]: 664 : if (FullTransactionIdPrecedes(TransamVariables->nextXid,
8398 : : checkPoint.nextXid))
638 heikki.linnakangas@i 8399 :UBC 0 : TransamVariables->nextXid = checkPoint.nextXid;
4961 tgl@sss.pgh.pa.us 8400 :CBC 664 : LWLockRelease(XidGenLock);
8401 : :
8402 : : /*
8403 : : * We ignore the nextOid counter in an ONLINE checkpoint, preferring
8404 : : * to track OID assignment through XLOG_NEXTOID records. The nextOid
8405 : : * counter is from the start of the checkpoint and might well be stale
8406 : : * compared to later XLOG_NEXTOID records. We could try to take the
8407 : : * maximum of the nextOid counter and our latest value, but since
8408 : : * there's no particular guarantee about the speed with which the OID
8409 : : * counter wraps around, that's a risky thing to do. In any case,
8410 : : * users of the nextOid counter are required to avoid assignment of
8411 : : * duplicates, so that a somewhat out-of-date value should be safe.
8412 : : */
8413 : :
8414 : : /* Handle multixact */
7395 8415 : 664 : MultiXactAdvanceNextMXact(checkPoint.nextMulti,
8416 : : checkPoint.nextMultiOffset);
8417 : :
8418 : : /*
8419 : : * NB: This may perform multixact truncation when replaying WAL
8420 : : * generated by an older primary.
8421 : : */
3633 andres@anarazel.de 8422 : 664 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8423 : : checkPoint.oldestMultiDB);
638 heikki.linnakangas@i 8424 [ - + ]: 664 : if (TransactionIdPrecedes(TransamVariables->oldestXid,
8425 : : checkPoint.oldestXid))
5680 tgl@sss.pgh.pa.us 8426 :UBC 0 : SetTransactionIdLimit(checkPoint.oldestXid,
8427 : : checkPoint.oldestXidDB);
8428 : : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
1916 tmunro@postgresql.or 8429 :CBC 664 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
1852 andres@anarazel.de 8430 : 664 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
1916 tmunro@postgresql.or 8431 : 664 : LWLockRelease(ControlFileLock);
8432 : :
8433 : : /* TLI should not change in an on-line checkpoint */
1298 heikki.linnakangas@i 8434 : 664 : (void) GetCurrentReplayRecPtr(&replayTLI);
1401 rhaas@postgresql.org 8435 [ - + ]: 664 : if (checkPoint.ThisTimeLineID != replayTLI)
7878 tgl@sss.pgh.pa.us 8436 [ # # ]:UBC 0 : ereport(PANIC,
8437 : : (errmsg("unexpected timeline ID %u (should be %u) in online checkpoint record",
8438 : : checkPoint.ThisTimeLineID, replayTLI)));
8439 : :
1382 rhaas@postgresql.org 8440 :CBC 664 : RecoveryRestartPoint(&checkPoint, record);
8441 : : }
1438 alvherre@alvh.no-ip. 8442 [ + + ]: 40651 : else if (info == XLOG_OVERWRITE_CONTRECORD)
8443 : : {
8444 : : /* nothing to do here, handled in xlogrecovery_redo() */
8445 : : }
4603 simon@2ndQuadrant.co 8446 [ + + ]: 40650 : else if (info == XLOG_END_OF_RECOVERY)
8447 : : {
8448 : : xl_end_of_recovery xlrec;
8449 : : TimeLineID replayTLI;
8450 : :
8451 : 9 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
8452 : :
8453 : : /*
8454 : : * For Hot Standby, we could treat this like a Shutdown Checkpoint,
8455 : : * but this case is rarer and harder to test, so the benefit doesn't
8456 : : * outweigh the potential extra cost of maintenance.
8457 : : */
8458 : :
8459 : : /*
8460 : : * We should've already switched to the new TLI before replaying this
8461 : : * record.
8462 : : */
1298 heikki.linnakangas@i 8463 : 9 : (void) GetCurrentReplayRecPtr(&replayTLI);
1401 rhaas@postgresql.org 8464 [ - + ]: 9 : if (xlrec.ThisTimeLineID != replayTLI)
4603 simon@2ndQuadrant.co 8465 [ # # ]:UBC 0 : ereport(PANIC,
8466 : : (errmsg("unexpected timeline ID %u (should be %u) in end-of-recovery record",
8467 : : xlrec.ThisTimeLineID, replayTLI)));
8468 : : }
6684 tgl@sss.pgh.pa.us 8469 [ + - ]:CBC 40641 : else if (info == XLOG_NOOP)
8470 : : {
8471 : : /* nothing to do here */
8472 : : }
6971 8473 [ + + ]: 40641 : else if (info == XLOG_SWITCH)
8474 : : {
8475 : : /* nothing to do here */
8476 : : }
5324 simon@2ndQuadrant.co 8477 [ + + ]: 40209 : else if (info == XLOG_RESTORE_POINT)
8478 : : {
8479 : : /* nothing to do here, handled in xlogrecovery.c */
8480 : : }
3939 heikki.linnakangas@i 8481 [ + + + + ]: 40204 : else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
8482 : : {
8483 : : /*
8484 : : * XLOG_FPI records contain nothing else but one or more block
8485 : : * references. Every block reference must include a full-page image
8486 : : * even if full_page_writes was disabled when the record was generated
8487 : : * - otherwise there would be no point in this record.
8488 : : *
8489 : : * XLOG_FPI_FOR_HINT records are generated when a page needs to be
8490 : : * WAL-logged because of a hint bit update. They are only generated
8491 : : * when checksums and/or wal_log_hints are enabled. They may include
8492 : : * no full-page images if full_page_writes was disabled when they were
8493 : : * generated. In this case there is nothing to do here.
8494 : : *
8495 : : * No recovery conflicts are generated by these generic records - if a
8496 : : * resource manager needs to generate conflicts, it has to define a
8497 : : * separate WAL record type and redo routine.
8498 : : */
1268 tmunro@postgresql.or 8499 [ + + ]: 83637 : for (uint8 block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
8500 : : {
8501 : : Buffer buffer;
8502 : :
1508 fujii@postgresql.org 8503 [ + + ]: 44209 : if (!XLogRecHasBlockImage(record, block_id))
8504 : : {
8505 [ - + ]: 66 : if (info == XLOG_FPI)
1508 fujii@postgresql.org 8506 [ # # ]:UBC 0 : elog(ERROR, "XLOG_FPI record did not contain a full-page image");
1508 fujii@postgresql.org 8507 :CBC 66 : continue;
8508 : : }
8509 : :
2348 heikki.linnakangas@i 8510 [ - + ]: 44143 : if (XLogReadBufferForRedo(record, block_id, &buffer) != BLK_RESTORED)
2348 heikki.linnakangas@i 8511 [ # # ]:UBC 0 : elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
2348 heikki.linnakangas@i 8512 :CBC 44143 : UnlockReleaseBuffer(buffer);
8513 : : }
8514 : : }
5724 8515 [ + + ]: 776 : else if (info == XLOG_BACKUP_END)
8516 : : {
8517 : : /* nothing to do here, handled in xlogrecovery_redo() */
8518 : : }
5610 8519 [ + + ]: 695 : else if (info == XLOG_PARAMETER_CHANGE)
8520 : : {
8521 : : xl_parameter_change xlrec;
8522 : :
8523 : : /* Update our copy of the parameters in pg_control */
8524 : 30 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
8525 : :
8526 : : /*
8527 : : * Invalidate logical slots if we are in hot standby and the primary
8528 : : * does not have a WAL level sufficient for logical decoding. No need
8529 : : * to search for potentially conflicting logically slots if standby is
8530 : : * running with wal_level lower than logical, because in that case, we
8531 : : * would have either disallowed creation of logical slots or
8532 : : * invalidated existing ones.
8533 : : */
883 andres@anarazel.de 8534 [ + - + + ]: 30 : if (InRecovery && InHotStandby &&
8535 [ + + ]: 15 : xlrec.wal_level < WAL_LEVEL_LOGICAL &&
8536 [ + + ]: 5 : wal_level >= WAL_LEVEL_LOGICAL)
8537 : 3 : InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_LEVEL,
8538 : : 0, InvalidOid,
8539 : : InvalidTransactionId);
8540 : :
5605 heikki.linnakangas@i 8541 : 30 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
5610 8542 : 30 : ControlFile->MaxConnections = xlrec.MaxConnections;
4447 rhaas@postgresql.org 8543 : 30 : ControlFile->max_worker_processes = xlrec.max_worker_processes;
2398 michael@paquier.xyz 8544 : 30 : ControlFile->max_wal_senders = xlrec.max_wal_senders;
5610 heikki.linnakangas@i 8545 : 30 : ControlFile->max_prepared_xacts = xlrec.max_prepared_xacts;
8546 : 30 : ControlFile->max_locks_per_xact = xlrec.max_locks_per_xact;
8547 : 30 : ControlFile->wal_level = xlrec.wal_level;
3887 8548 : 30 : ControlFile->wal_log_hints = xlrec.wal_log_hints;
8549 : :
8550 : : /*
8551 : : * Update minRecoveryPoint to ensure that if recovery is aborted, we
8552 : : * recover back up to this point before allowing hot standby again.
8553 : : * This is important if the max_* settings are decreased, to ensure
8554 : : * you don't run queries against the WAL preceding the change. The
8555 : : * local copies cannot be updated as long as crash recovery is
8556 : : * happening and we expect all the WAL to be replayed.
8557 : : */
2620 michael@paquier.xyz 8558 [ + + ]: 30 : if (InArchiveRecovery)
8559 : : {
1298 heikki.linnakangas@i 8560 : 16 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
8561 : 16 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
8562 : : }
8563 [ + + + + ]: 30 : if (LocalMinRecoveryPoint != InvalidXLogRecPtr && LocalMinRecoveryPoint < lsn)
8564 : : {
8565 : : TimeLineID replayTLI;
8566 : :
8567 : 5 : (void) GetCurrentReplayRecPtr(&replayTLI);
5605 8568 : 5 : ControlFile->minRecoveryPoint = lsn;
1401 rhaas@postgresql.org 8569 : 5 : ControlFile->minRecoveryPointTLI = replayTLI;
8570 : : }
8571 : :
3628 alvherre@alvh.no-ip. 8572 : 30 : CommitTsParameterChange(xlrec.track_commit_timestamp,
8573 : 30 : ControlFile->track_commit_timestamp);
8574 : 30 : ControlFile->track_commit_timestamp = xlrec.track_commit_timestamp;
8575 : :
5610 heikki.linnakangas@i 8576 : 30 : UpdateControlFile();
5605 8577 : 30 : LWLockRelease(ControlFileLock);
8578 : :
8579 : : /* Check to see if any parameter change gives a problem on recovery */
5610 8580 : 30 : CheckRequiredParameterValues();
8581 : : }
4973 simon@2ndQuadrant.co 8582 [ - + ]: 665 : else if (info == XLOG_FPW_CHANGE)
8583 : : {
8584 : : bool fpw;
8585 : :
4973 simon@2ndQuadrant.co 8586 :UBC 0 : memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
8587 : :
8588 : : /*
8589 : : * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
8590 : : * do_pg_backup_start() and do_pg_backup_stop() can check whether
8591 : : * full_page_writes has been disabled during online backup.
8592 : : */
8593 [ # # ]: 0 : if (!fpw)
8594 : : {
4002 andres@anarazel.de 8595 [ # # ]: 0 : SpinLockAcquire(&XLogCtl->info_lck);
1382 rhaas@postgresql.org 8596 [ # # ]: 0 : if (XLogCtl->lastFpwDisableRecPtr < record->ReadRecPtr)
8597 : 0 : XLogCtl->lastFpwDisableRecPtr = record->ReadRecPtr;
4002 andres@anarazel.de 8598 : 0 : SpinLockRelease(&XLogCtl->info_lck);
8599 : : }
8600 : :
8601 : : /* Keep track of full_page_writes */
4973 simon@2ndQuadrant.co 8602 : 0 : lastFullPageWrites = fpw;
8603 : : }
8604 : : else if (info == XLOG_CHECKPOINT_REDO)
8605 : : {
8606 : : /* nothing to do here, just for informational purposes */
8607 : : }
9086 vadim4o@yahoo.com 8608 :CBC 41433 : }
8609 : :
8610 : : /*
8611 : : * Return the extra open flags used for opening a file, depending on the
8612 : : * value of the GUCs wal_sync_method, fsync and debug_io_direct.
8613 : : */
8614 : : static int
6324 magnus@hagander.net 8615 : 15257 : get_sync_bit(int method)
8616 : : {
5671 bruce@momjian.us 8617 : 15257 : int o_direct_flag = 0;
8618 : :
8619 : : /*
8620 : : * Use O_DIRECT if requested, except in walreceiver process. The WAL
8621 : : * written by walreceiver is normally read by the startup process soon
8622 : : * after it's written. Also, walreceiver performs unaligned writes, which
8623 : : * don't work with O_DIRECT, so it is required for correctness too.
8624 : : */
882 tmunro@postgresql.or 8625 [ + + + - ]: 15257 : if ((io_direct_flags & IO_DIRECT_WAL) && !AmWalReceiverProcess())
5678 heikki.linnakangas@i 8626 : 9 : o_direct_flag = PG_O_DIRECT;
8627 : :
8628 : : /* If fsync is disabled, never open in sync mode */
882 tmunro@postgresql.or 8629 [ + - ]: 15257 : if (!enableFsync)
8630 : 15257 : return o_direct_flag;
8631 : :
6324 magnus@hagander.net 8632 [ # # # # ]:UBC 0 : switch (method)
8633 : : {
8634 : : /*
8635 : : * enum values for all sync options are defined even if they are
8636 : : * not supported on the current platform. But if not, they are
8637 : : * not included in the enum option array, and therefore will never
8638 : : * be seen here.
8639 : : */
694 nathan@postgresql.or 8640 : 0 : case WAL_SYNC_METHOD_FSYNC:
8641 : : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8642 : : case WAL_SYNC_METHOD_FDATASYNC:
882 tmunro@postgresql.or 8643 : 0 : return o_direct_flag;
8644 : : #ifdef O_SYNC
694 nathan@postgresql.or 8645 : 0 : case WAL_SYNC_METHOD_OPEN:
1142 tmunro@postgresql.or 8646 : 0 : return O_SYNC | o_direct_flag;
8647 : : #endif
8648 : : #ifdef O_DSYNC
694 nathan@postgresql.or 8649 : 0 : case WAL_SYNC_METHOD_OPEN_DSYNC:
1142 tmunro@postgresql.or 8650 : 0 : return O_DSYNC | o_direct_flag;
8651 : : #endif
6326 magnus@hagander.net 8652 : 0 : default:
8653 : : /* can't happen (unless we are out of sync with option array) */
477 peter@eisentraut.org 8654 [ # # ]: 0 : elog(ERROR, "unrecognized \"wal_sync_method\": %d", method);
8655 : : return 0; /* silence warning */
8656 : : }
8657 : : }
8658 : :
8659 : : /*
8660 : : * GUC support
8661 : : */
8662 : : void
694 nathan@postgresql.or 8663 :CBC 1067 : assign_wal_sync_method(int new_wal_sync_method, void *extra)
8664 : : {
8665 [ - + ]: 1067 : if (wal_sync_method != new_wal_sync_method)
8666 : : {
8667 : : /*
8668 : : * To ensure that no blocks escape unsynced, force an fsync on the
8669 : : * currently open log segment (if any). Also, if the open flag is
8670 : : * changing, close the log file so it will be reopened (with new flag
8671 : : * bit) at next use.
8672 : : */
8940 tgl@sss.pgh.pa.us 8673 [ # # ]:UBC 0 : if (openLogFile >= 0)
8674 : : {
3094 rhaas@postgresql.org 8675 : 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN);
8940 tgl@sss.pgh.pa.us 8676 [ # # ]: 0 : if (pg_fsync(openLogFile) != 0)
8677 : : {
8678 : : char xlogfname[MAXFNAMELEN];
8679 : : int save_errno;
8680 : :
2104 michael@paquier.xyz 8681 : 0 : save_errno = errno;
1401 rhaas@postgresql.org 8682 : 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo,
8683 : : wal_segment_size);
2104 michael@paquier.xyz 8684 : 0 : errno = save_errno;
8083 tgl@sss.pgh.pa.us 8685 [ # # ]: 0 : ereport(PANIC,
8686 : : (errcode_for_file_access(),
8687 : : errmsg("could not fsync file \"%s\": %m", xlogfname)));
8688 : : }
8689 : :
3094 rhaas@postgresql.org 8690 : 0 : pgstat_report_wait_end();
694 nathan@postgresql.or 8691 [ # # ]: 0 : if (get_sync_bit(wal_sync_method) != get_sync_bit(new_wal_sync_method))
7023 bruce@momjian.us 8692 : 0 : XLogFileClose();
8693 : : }
8694 : : }
8940 tgl@sss.pgh.pa.us 8695 :CBC 1067 : }
8696 : :
8697 : :
8698 : : /*
8699 : : * Issue appropriate kind of fsync (if any) for an XLOG output file.
8700 : : *
8701 : : * 'fd' is a file descriptor for the XLOG file to be fsync'd.
8702 : : * 'segno' is for error reporting purposes.
8703 : : */
8704 : : void
1401 rhaas@postgresql.org 8705 : 191240 : issue_xlog_fsync(int fd, XLogSegNo segno, TimeLineID tli)
8706 : : {
2104 michael@paquier.xyz 8707 : 191240 : char *msg = NULL;
8708 : : instr_time start;
8709 : :
1401 rhaas@postgresql.org 8710 [ - + ]: 191240 : Assert(tli != 0);
8711 : :
8712 : : /*
8713 : : * Quick exit if fsync is disabled or write() has already synced the WAL
8714 : : * file.
8715 : : */
1642 fujii@postgresql.org 8716 [ - + ]: 191240 : if (!enableFsync ||
694 nathan@postgresql.or 8717 [ # # ]:UBC 0 : wal_sync_method == WAL_SYNC_METHOD_OPEN ||
8718 [ # # ]: 0 : wal_sync_method == WAL_SYNC_METHOD_OPEN_DSYNC)
1642 fujii@postgresql.org 8719 :CBC 191240 : return;
8720 : :
8721 : : /*
8722 : : * Measure I/O timing to sync the WAL file for pg_stat_io.
8723 : : */
192 michael@paquier.xyz 8724 :UBC 0 : start = pgstat_prepare_io_time(track_wal_io_timing);
8725 : :
2623 8726 : 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC);
694 nathan@postgresql.or 8727 [ # # # # ]: 0 : switch (wal_sync_method)
8728 : : {
8729 : 0 : case WAL_SYNC_METHOD_FSYNC:
5713 heikki.linnakangas@i 8730 [ # # ]: 0 : if (pg_fsync_no_writethrough(fd) != 0)
2104 michael@paquier.xyz 8731 : 0 : msg = _("could not fsync file \"%s\": %m");
8940 tgl@sss.pgh.pa.us 8732 : 0 : break;
8733 : : #ifdef HAVE_FSYNC_WRITETHROUGH
8734 : : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8735 : : if (pg_fsync_writethrough(fd) != 0)
8736 : : msg = _("could not fsync write-through file \"%s\": %m");
8737 : : break;
8738 : : #endif
694 nathan@postgresql.or 8739 : 0 : case WAL_SYNC_METHOD_FDATASYNC:
5713 heikki.linnakangas@i 8740 [ # # ]: 0 : if (pg_fdatasync(fd) != 0)
2104 michael@paquier.xyz 8741 : 0 : msg = _("could not fdatasync file \"%s\": %m");
8940 tgl@sss.pgh.pa.us 8742 : 0 : break;
694 nathan@postgresql.or 8743 : 0 : case WAL_SYNC_METHOD_OPEN:
8744 : : case WAL_SYNC_METHOD_OPEN_DSYNC:
8745 : : /* not reachable */
1642 fujii@postgresql.org 8746 : 0 : Assert(false);
8747 : : break;
8940 tgl@sss.pgh.pa.us 8748 : 0 : default:
521 dgustafsson@postgres 8749 [ # # ]: 0 : ereport(PANIC,
8750 : : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8751 : : errmsg_internal("unrecognized \"wal_sync_method\": %d", wal_sync_method));
8752 : : break;
8753 : : }
8754 : :
8755 : : /* PANIC if failed to fsync */
2104 michael@paquier.xyz 8756 [ # # ]: 0 : if (msg)
8757 : : {
8758 : : char xlogfname[MAXFNAMELEN];
8759 : 0 : int save_errno = errno;
8760 : :
1401 rhaas@postgresql.org 8761 : 0 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
2104 michael@paquier.xyz 8762 : 0 : errno = save_errno;
8763 [ # # ]: 0 : ereport(PANIC,
8764 : : (errcode_for_file_access(),
8765 : : errmsg(msg, xlogfname)));
8766 : : }
8767 : :
8768 : 0 : pgstat_report_wait_end();
8769 : :
214 8770 : 0 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_FSYNC,
8771 : : start, 1, 0);
8772 : : }
8773 : :
8774 : : /*
8775 : : * do_pg_backup_start is the workhorse of the user-visible pg_backup_start()
8776 : : * function. It creates the necessary starting checkpoint and constructs the
8777 : : * backup state and tablespace map.
8778 : : *
8779 : : * Input parameters are "state" (the backup state), "fast" (if true, we do
8780 : : * the checkpoint in fast mode), and "tablespaces" (if non-NULL, indicates a
8781 : : * list of tablespaceinfo structs describing the cluster's tablespaces.).
8782 : : *
8783 : : * The tablespace map contents are appended to passed-in parameter
8784 : : * tablespace_map and the caller is responsible for including it in the backup
8785 : : * archive as 'tablespace_map'. The tablespace_map file is required mainly for
8786 : : * tar format in windows as native windows utilities are not able to create
8787 : : * symlinks while extracting files from tar. However for consistency and
8788 : : * platform-independence, we do it the same way everywhere.
8789 : : *
8790 : : * It fills in "state" with the information required for the backup, such
8791 : : * as the minimum WAL location that must be present to restore from this
8792 : : * backup (starttli) and the corresponding timeline ID (starttli).
8793 : : *
8794 : : * Every successfully started backup must be stopped by calling
8795 : : * do_pg_backup_stop() or do_pg_abort_backup(). There can be many
8796 : : * backups active at the same time.
8797 : : *
8798 : : * It is the responsibility of the caller of this function to verify the
8799 : : * permissions of the calling user!
8800 : : */
8801 : : void
1076 michael@paquier.xyz 8802 :CBC 168 : do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces,
8803 : : BackupState *state, StringInfo tblspcmapfile)
8804 : : {
8805 : : bool backup_started_in_recovery;
8806 : :
8807 [ - + ]: 168 : Assert(state != NULL);
4973 simon@2ndQuadrant.co 8808 : 168 : backup_started_in_recovery = RecoveryInProgress();
8809 : :
8810 : : /*
8811 : : * During recovery, we don't need to check WAL level. Because, if WAL
8812 : : * level is not sufficient, it's impossible to get here during recovery.
8813 : : */
8814 [ + + - + ]: 168 : if (!backup_started_in_recovery && !XLogIsNeeded())
6555 tgl@sss.pgh.pa.us 8815 [ # # ]:UBC 0 : ereport(ERROR,
8816 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8817 : : errmsg("WAL level not sufficient for making an online backup"),
8818 : : errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
8819 : :
5332 heikki.linnakangas@i 8820 [ + + ]:CBC 168 : if (strlen(backupidstr) > MAXPGPATH)
8821 [ + - ]: 1 : ereport(ERROR,
8822 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8823 : : errmsg("backup label too long (max %d bytes)",
8824 : : MAXPGPATH)));
8825 : :
431 dgustafsson@postgres 8826 : 167 : strlcpy(state->name, backupidstr, sizeof(state->name));
8827 : :
8828 : : /*
8829 : : * Mark backup active in shared memory. We must do full-page WAL writes
8830 : : * during an on-line backup even if not doing so at other times, because
8831 : : * it's quite possible for the backup dump to obtain a "torn" (partially
8832 : : * written) copy of a database page if it reads the page concurrently with
8833 : : * our write to the same page. This can be fixed as long as the first
8834 : : * write to the page in the WAL sequence is a full-page write. Hence, we
8835 : : * increment runningBackups then force a CHECKPOINT, to ensure there are
8836 : : * no dirty pages in shared memory that might get dumped while the backup
8837 : : * is in progress without having a corresponding WAL record. (Once the
8838 : : * backup is complete, we need not force full-page writes anymore, since
8839 : : * we expect that any pages not modified during the backup interval must
8840 : : * have been correctly captured by the backup.)
8841 : : *
8842 : : * Note that forcing full-page writes has no effect during an online
8843 : : * backup from the standby.
8844 : : *
8845 : : * We must hold all the insertion locks to change the value of
8846 : : * runningBackups, to ensure adequate interlocking against
8847 : : * XLogInsertRecord().
8848 : : */
4187 heikki.linnakangas@i 8849 : 167 : WALInsertLockAcquireExclusive();
1249 sfrost@snowman.net 8850 : 167 : XLogCtl->Insert.runningBackups++;
4187 heikki.linnakangas@i 8851 : 167 : WALInsertLockRelease();
8852 : :
8853 : : /*
8854 : : * Ensure we decrement runningBackups if we fail below. NB -- for this to
8855 : : * work correctly, it is critical that sessionBackupState is only updated
8856 : : * after this block is over.
8857 : : */
32 peter@eisentraut.org 8858 [ + - ]:GNC 167 : PG_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, BoolGetDatum(true));
8859 : : {
5263 bruce@momjian.us 8860 :CBC 167 : bool gotUniqueStartpoint = false;
8861 : : DIR *tblspcdir;
8862 : : struct dirent *de;
8863 : : tablespaceinfo *ti;
8864 : : int datadirpathlen;
8865 : :
8866 : : /*
8867 : : * Force an XLOG file switch before the checkpoint, to ensure that the
8868 : : * WAL segment the checkpoint is written to doesn't contain pages with
8869 : : * old timeline IDs. That would otherwise happen if you called
8870 : : * pg_backup_start() right after restoring from a PITR archive: the
8871 : : * first WAL segment containing the startup checkpoint has pages in
8872 : : * the beginning with the old timeline ID. That can cause trouble at
8873 : : * recovery: we won't have a history file covering the old timeline if
8874 : : * pg_wal directory was not included in the base backup and the WAL
8875 : : * archive was cleared too before starting the backup.
8876 : : *
8877 : : * This also ensures that we have emitted a WAL page header that has
8878 : : * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
8879 : : * Therefore, if a WAL archiver (such as pglesslog) is trying to
8880 : : * compress out removable backup blocks, it won't remove any that
8881 : : * occur after this point.
8882 : : *
8883 : : * During recovery, we skip forcing XLOG file switch, which means that
8884 : : * the backup taken during recovery is not available for the special
8885 : : * recovery case described above.
8886 : : */
4973 simon@2ndQuadrant.co 8887 [ + + ]: 167 : if (!backup_started_in_recovery)
3180 andres@anarazel.de 8888 : 161 : RequestXLogSwitch(false);
8889 : :
8890 : : do
8891 : : {
8892 : : bool checkpointfpw;
8893 : :
8894 : : /*
8895 : : * Force a CHECKPOINT. Aside from being necessary to prevent torn
8896 : : * page problems, this guarantees that two successive backup runs
8897 : : * will have different checkpoint positions and hence different
8898 : : * history file names, even if nothing happened in between.
8899 : : *
8900 : : * During recovery, establish a restartpoint if possible. We use
8901 : : * the last restartpoint as the backup starting checkpoint. This
8902 : : * means that two successive backup runs can have same checkpoint
8903 : : * positions.
8904 : : *
8905 : : * Since the fact that we are executing do_pg_backup_start()
8906 : : * during recovery means that checkpointer is running, we can use
8907 : : * RequestCheckpoint() to establish a restartpoint.
8908 : : *
8909 : : * We use CHECKPOINT_FAST only if requested by user (via passing
8910 : : * fast = true). Otherwise this can take awhile.
8911 : : */
5283 heikki.linnakangas@i 8912 [ + + ]: 167 : RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_WAIT |
8913 : : (fast ? CHECKPOINT_FAST : 0));
8914 : :
8915 : : /*
8916 : : * Now we need to fetch the checkpoint record location, and also
8917 : : * its REDO pointer. The oldest point in WAL that would be needed
8918 : : * to restore starting from the checkpoint is precisely the REDO
8919 : : * pointer.
8920 : : */
8921 : 167 : LWLockAcquire(ControlFileLock, LW_SHARED);
1076 michael@paquier.xyz 8922 : 167 : state->checkpointloc = ControlFile->checkPoint;
8923 : 167 : state->startpoint = ControlFile->checkPointCopy.redo;
8924 : 167 : state->starttli = ControlFile->checkPointCopy.ThisTimeLineID;
4973 simon@2ndQuadrant.co 8925 : 167 : checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
5283 heikki.linnakangas@i 8926 : 167 : LWLockRelease(ControlFileLock);
8927 : :
4973 simon@2ndQuadrant.co 8928 [ + + ]: 167 : if (backup_started_in_recovery)
8929 : : {
8930 : : XLogRecPtr recptr;
8931 : :
8932 : : /*
8933 : : * Check to see if all WAL replayed during online backup
8934 : : * (i.e., since last restartpoint used as backup starting
8935 : : * checkpoint) contain full-page writes.
8936 : : */
4002 andres@anarazel.de 8937 [ - + ]: 6 : SpinLockAcquire(&XLogCtl->info_lck);
8938 : 6 : recptr = XLogCtl->lastFpwDisableRecPtr;
8939 : 6 : SpinLockRelease(&XLogCtl->info_lck);
8940 : :
1076 michael@paquier.xyz 8941 [ + - - + ]: 6 : if (!checkpointfpw || state->startpoint <= recptr)
4973 simon@2ndQuadrant.co 8942 [ # # ]:UBC 0 : ereport(ERROR,
8943 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8944 : : errmsg("WAL generated with \"full_page_writes=off\" was replayed "
8945 : : "since last restartpoint"),
8946 : : errhint("This means that the backup being taken on the standby "
8947 : : "is corrupt and should not be used. "
8948 : : "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
8949 : : "and then try an online backup again.")));
8950 : :
8951 : : /*
8952 : : * During recovery, since we don't use the end-of-backup WAL
8953 : : * record and don't write the backup history file, the
8954 : : * starting WAL location doesn't need to be unique. This means
8955 : : * that two base backups started at the same time might use
8956 : : * the same checkpoint as starting locations.
8957 : : */
4973 simon@2ndQuadrant.co 8958 :CBC 6 : gotUniqueStartpoint = true;
8959 : : }
8960 : :
8961 : : /*
8962 : : * If two base backups are started at the same time (in WAL sender
8963 : : * processes), we need to make sure that they use different
8964 : : * checkpoints as starting locations, because we use the starting
8965 : : * WAL location as a unique identifier for the base backup in the
8966 : : * end-of-backup WAL record and when we write the backup history
8967 : : * file. Perhaps it would be better generate a separate unique ID
8968 : : * for each backup instead of forcing another checkpoint, but
8969 : : * taking a checkpoint right after another is not that expensive
8970 : : * either because only few buffers have been dirtied yet.
8971 : : */
4187 heikki.linnakangas@i 8972 : 167 : WALInsertLockAcquireExclusive();
1076 michael@paquier.xyz 8973 [ + - ]: 167 : if (XLogCtl->Insert.lastBackupStart < state->startpoint)
8974 : : {
8975 : 167 : XLogCtl->Insert.lastBackupStart = state->startpoint;
5283 heikki.linnakangas@i 8976 : 167 : gotUniqueStartpoint = true;
8977 : : }
4187 8978 : 167 : WALInsertLockRelease();
5263 bruce@momjian.us 8979 [ - + ]: 167 : } while (!gotUniqueStartpoint);
8980 : :
8981 : : /*
8982 : : * Construct tablespace_map file.
8983 : : */
3770 andrew@dunslane.net 8984 : 167 : datadirpathlen = strlen(DataDir);
8985 : :
8986 : : /* Collect information about all tablespaces */
368 michael@paquier.xyz 8987 : 167 : tblspcdir = AllocateDir(PG_TBLSPC_DIR);
8988 [ + + ]: 540 : while ((de = ReadDir(tblspcdir, PG_TBLSPC_DIR)) != NULL)
8989 : : {
8990 : : char fullpath[MAXPGPATH + sizeof(PG_TBLSPC_DIR)];
8991 : : char linkpath[MAXPGPATH];
3770 andrew@dunslane.net 8992 : 373 : char *relpath = NULL;
8993 : : char *s;
8994 : : PGFileType de_type;
8995 : : char *badp;
8996 : : Oid tsoid;
8997 : :
8998 : : /*
8999 : : * Try to parse the directory name as an unsigned integer.
9000 : : *
9001 : : * Tablespace directories should be positive integers that can be
9002 : : * represented in 32 bits, with no leading zeroes or trailing
9003 : : * garbage. If we come across a name that doesn't meet those
9004 : : * criteria, skip it.
9005 : : */
684 rhaas@postgresql.org 9006 [ + + - + ]: 373 : if (de->d_name[0] < '1' || de->d_name[1] > '9')
9007 : 334 : continue;
9008 : 39 : errno = 0;
9009 : 39 : tsoid = strtoul(de->d_name, &badp, 10);
9010 [ + - + - : 39 : if (*badp != '\0' || errno == EINVAL || errno == ERANGE)
- + ]
3770 andrew@dunslane.net 9011 :UBC 0 : continue;
9012 : :
368 michael@paquier.xyz 9013 :CBC 39 : snprintf(fullpath, sizeof(fullpath), "%s/%s", PG_TBLSPC_DIR, de->d_name);
9014 : :
872 rhaas@postgresql.org 9015 : 39 : de_type = get_dirent_type(fullpath, de, false, ERROR);
9016 : :
9017 [ + + ]: 39 : if (de_type == PGFILETYPE_LNK)
9018 : : {
9019 : : StringInfoData escapedpath;
9020 : : int rllen;
9021 : :
9022 : 25 : rllen = readlink(fullpath, linkpath, sizeof(linkpath));
9023 [ - + ]: 25 : if (rllen < 0)
9024 : : {
872 rhaas@postgresql.org 9025 [ # # ]:UBC 0 : ereport(WARNING,
9026 : : (errmsg("could not read symbolic link \"%s\": %m",
9027 : : fullpath)));
9028 : 0 : continue;
9029 : : }
872 rhaas@postgresql.org 9030 [ - + ]:CBC 25 : else if (rllen >= sizeof(linkpath))
9031 : : {
872 rhaas@postgresql.org 9032 [ # # ]:UBC 0 : ereport(WARNING,
9033 : : (errmsg("symbolic link \"%s\" target is too long",
9034 : : fullpath)));
9035 : 0 : continue;
9036 : : }
872 rhaas@postgresql.org 9037 :CBC 25 : linkpath[rllen] = '\0';
9038 : :
9039 : : /*
9040 : : * Relpath holds the relative path of the tablespace directory
9041 : : * when it's located within PGDATA, or NULL if it's located
9042 : : * elsewhere.
9043 : : */
9044 [ - + ]: 25 : if (rllen > datadirpathlen &&
872 rhaas@postgresql.org 9045 [ # # ]:UBC 0 : strncmp(linkpath, DataDir, datadirpathlen) == 0 &&
841 tgl@sss.pgh.pa.us 9046 [ # # ]: 0 : IS_DIR_SEP(linkpath[datadirpathlen]))
872 rhaas@postgresql.org 9047 : 0 : relpath = pstrdup(linkpath + datadirpathlen + 1);
9048 : :
9049 : : /*
9050 : : * Add a backslash-escaped version of the link path to the
9051 : : * tablespace map file.
9052 : : */
872 rhaas@postgresql.org 9053 :CBC 25 : initStringInfo(&escapedpath);
9054 [ + + ]: 594 : for (s = linkpath; *s; s++)
9055 : : {
9056 [ + - + - : 569 : if (*s == '\n' || *s == '\r' || *s == '\\')
- + ]
872 rhaas@postgresql.org 9057 :UBC 0 : appendStringInfoChar(&escapedpath, '\\');
872 rhaas@postgresql.org 9058 :CBC 569 : appendStringInfoChar(&escapedpath, *s);
9059 : : }
9060 : 25 : appendStringInfo(tblspcmapfile, "%s %s\n",
9061 : 25 : de->d_name, escapedpath.data);
9062 : 25 : pfree(escapedpath.data);
9063 : : }
9064 [ + - ]: 14 : else if (de_type == PGFILETYPE_DIR)
9065 : : {
9066 : : /*
9067 : : * It's possible to use allow_in_place_tablespaces to create
9068 : : * directories directly under pg_tblspc, for testing purposes
9069 : : * only.
9070 : : *
9071 : : * In this case, we store a relative path rather than an
9072 : : * absolute path into the tablespaceinfo.
9073 : : */
368 michael@paquier.xyz 9074 : 14 : snprintf(linkpath, sizeof(linkpath), "%s/%s",
9075 : 14 : PG_TBLSPC_DIR, de->d_name);
872 rhaas@postgresql.org 9076 : 14 : relpath = pstrdup(linkpath);
9077 : : }
9078 : : else
9079 : : {
9080 : : /* Skip any other file type that appears here. */
872 rhaas@postgresql.org 9081 :UBC 0 : continue;
9082 : : }
9083 : :
3770 andrew@dunslane.net 9084 :CBC 39 : ti = palloc(sizeof(tablespaceinfo));
684 rhaas@postgresql.org 9085 : 39 : ti->oid = tsoid;
1634 tgl@sss.pgh.pa.us 9086 : 39 : ti->path = pstrdup(linkpath);
872 rhaas@postgresql.org 9087 : 39 : ti->rpath = relpath;
1907 9088 : 39 : ti->size = -1;
9089 : :
3759 bruce@momjian.us 9090 [ + - ]: 39 : if (tablespaces)
9091 : 39 : *tablespaces = lappend(*tablespaces, ti);
9092 : : }
2833 tgl@sss.pgh.pa.us 9093 : 167 : FreeDir(tblspcdir);
9094 : :
1076 michael@paquier.xyz 9095 : 167 : state->starttime = (pg_time_t) time(NULL);
9096 : : }
32 peter@eisentraut.org 9097 [ - + ]:GNC 167 : PG_END_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, BoolGetDatum(true));
9098 : :
1076 michael@paquier.xyz 9099 :CBC 167 : state->started_in_recovery = backup_started_in_recovery;
9100 : :
9101 : : /*
9102 : : * Mark that the start phase has correctly finished for the backup.
9103 : : */
1249 sfrost@snowman.net 9104 : 167 : sessionBackupState = SESSION_BACKUP_RUNNING;
7704 tgl@sss.pgh.pa.us 9105 : 167 : }
9106 : :
9107 : : /*
9108 : : * Utility routine to fetch the session-level status of a backup running.
9109 : : */
9110 : : SessionBackupState
3088 teodor@sigaev.ru 9111 : 188 : get_backup_status(void)
9112 : : {
9113 : 188 : return sessionBackupState;
9114 : : }
9115 : :
9116 : : /*
9117 : : * do_pg_backup_stop
9118 : : *
9119 : : * Utility function called at the end of an online backup. It creates history
9120 : : * file (if required), resets sessionBackupState and so on. It can optionally
9121 : : * wait for WAL segments to be archived.
9122 : : *
9123 : : * "state" is filled with the information necessary to restore from this
9124 : : * backup with its stop LSN (stoppoint), its timeline ID (stoptli), etc.
9125 : : *
9126 : : * It is the responsibility of the caller of this function to verify the
9127 : : * permissions of the calling user!
9128 : : */
9129 : : void
1076 michael@paquier.xyz 9130 : 161 : do_pg_backup_stop(BackupState *state, bool waitforarchive)
9131 : : {
9132 : 161 : bool backup_stopped_in_recovery = false;
9133 : : char histfilepath[MAXPGPATH];
9134 : : char lastxlogfilename[MAXFNAMELEN];
9135 : : char histfilename[MAXFNAMELEN];
9136 : : XLogSegNo _logSegNo;
9137 : : FILE *fp;
9138 : : int seconds_before_warning;
6363 bruce@momjian.us 9139 : 161 : int waits = 0;
5620 simon@2ndQuadrant.co 9140 : 161 : bool reported_waiting = false;
9141 : :
1076 michael@paquier.xyz 9142 [ - + ]: 161 : Assert(state != NULL);
9143 : :
9144 : 161 : backup_stopped_in_recovery = RecoveryInProgress();
9145 : :
9146 : : /*
9147 : : * During recovery, we don't need to check WAL level. Because, if WAL
9148 : : * level is not sufficient, it's impossible to get here during recovery.
9149 : : */
9150 [ + + - + ]: 161 : if (!backup_stopped_in_recovery && !XLogIsNeeded())
6207 tgl@sss.pgh.pa.us 9151 [ # # ]:UBC 0 : ereport(ERROR,
9152 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9153 : : errmsg("WAL level not sufficient for making an online backup"),
9154 : : errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
9155 : :
9156 : : /*
9157 : : * OK to update backup counter and session-level lock.
9158 : : *
9159 : : * Note that CHECK_FOR_INTERRUPTS() must not occur while updating them,
9160 : : * otherwise they can be updated inconsistently, which might cause
9161 : : * do_pg_abort_backup() to fail.
9162 : : */
3154 fujii@postgresql.org 9163 :CBC 161 : WALInsertLockAcquireExclusive();
9164 : :
9165 : : /*
9166 : : * It is expected that each do_pg_backup_start() call is matched by
9167 : : * exactly one do_pg_backup_stop() call.
9168 : : */
1249 sfrost@snowman.net 9169 [ - + ]: 161 : Assert(XLogCtl->Insert.runningBackups > 0);
9170 : 161 : XLogCtl->Insert.runningBackups--;
9171 : :
9172 : : /*
9173 : : * Clean up session-level lock.
9174 : : *
9175 : : * You might think that WALInsertLockRelease() can be called before
9176 : : * cleaning up session-level lock because session-level lock doesn't need
9177 : : * to be protected with WAL insertion lock. But since
9178 : : * CHECK_FOR_INTERRUPTS() can occur in it, session-level lock must be
9179 : : * cleaned up before it.
9180 : : */
3088 teodor@sigaev.ru 9181 : 161 : sessionBackupState = SESSION_BACKUP_NONE;
9182 : :
2818 fujii@postgresql.org 9183 : 161 : WALInsertLockRelease();
9184 : :
9185 : : /*
9186 : : * If we are taking an online backup from the standby, we confirm that the
9187 : : * standby has not been promoted during the backup.
9188 : : */
1076 michael@paquier.xyz 9189 [ + + - + ]: 161 : if (state->started_in_recovery && !backup_stopped_in_recovery)
4973 simon@2ndQuadrant.co 9190 [ # # ]:UBC 0 : ereport(ERROR,
9191 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9192 : : errmsg("the standby was promoted during online backup"),
9193 : : errhint("This means that the backup being taken is corrupt "
9194 : : "and should not be used. "
9195 : : "Try taking another online backup.")));
9196 : :
9197 : : /*
9198 : : * During recovery, we don't write an end-of-backup record. We assume that
9199 : : * pg_control was backed up last and its minimum recovery point can be
9200 : : * available as the backup end location. Since we don't have an
9201 : : * end-of-backup record, we use the pg_control value to check whether
9202 : : * we've reached the end of backup when starting recovery from this
9203 : : * backup. We have no way of checking if pg_control wasn't backed up last
9204 : : * however.
9205 : : *
9206 : : * We don't force a switch to new WAL file but it is still possible to
9207 : : * wait for all the required files to be archived if waitforarchive is
9208 : : * true. This is okay if we use the backup to start a standby and fetch
9209 : : * the missing WAL using streaming replication. But in the case of an
9210 : : * archive recovery, a user should set waitforarchive to true and wait for
9211 : : * them to be archived to ensure that all the required files are
9212 : : * available.
9213 : : *
9214 : : * We return the current minimum recovery point as the backup end
9215 : : * location. Note that it can be greater than the exact backup end
9216 : : * location if the minimum recovery point is updated after the backup of
9217 : : * pg_control. This is harmless for current uses.
9218 : : *
9219 : : * XXX currently a backup history file is for informational and debug
9220 : : * purposes only. It's not essential for an online backup. Furthermore,
9221 : : * even if it's created, it will not be archived during recovery because
9222 : : * an archiver is not invoked. So it doesn't seem worthwhile to write a
9223 : : * backup history file during recovery.
9224 : : */
1076 michael@paquier.xyz 9225 [ + + ]:CBC 161 : if (backup_stopped_in_recovery)
9226 : : {
9227 : : XLogRecPtr recptr;
9228 : :
9229 : : /*
9230 : : * Check to see if all WAL replayed during online backup contain
9231 : : * full-page writes.
9232 : : */
4002 andres@anarazel.de 9233 [ - + ]: 6 : SpinLockAcquire(&XLogCtl->info_lck);
9234 : 6 : recptr = XLogCtl->lastFpwDisableRecPtr;
9235 : 6 : SpinLockRelease(&XLogCtl->info_lck);
9236 : :
1076 michael@paquier.xyz 9237 [ - + ]: 6 : if (state->startpoint <= recptr)
4973 simon@2ndQuadrant.co 9238 [ # # ]:UBC 0 : ereport(ERROR,
9239 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9240 : : errmsg("WAL generated with \"full_page_writes=off\" was replayed "
9241 : : "during online backup"),
9242 : : errhint("This means that the backup being taken on the standby "
9243 : : "is corrupt and should not be used. "
9244 : : "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
9245 : : "and then try an online backup again.")));
9246 : :
9247 : :
4973 simon@2ndQuadrant.co 9248 :CBC 6 : LWLockAcquire(ControlFileLock, LW_SHARED);
1076 michael@paquier.xyz 9249 : 6 : state->stoppoint = ControlFile->minRecoveryPoint;
9250 : 6 : state->stoptli = ControlFile->minRecoveryPointTLI;
4973 simon@2ndQuadrant.co 9251 : 6 : LWLockRelease(ControlFileLock);
9252 : : }
9253 : : else
9254 : : {
9255 : : char *history_file;
9256 : :
9257 : : /*
9258 : : * Write the backup-end xlog record
9259 : : */
2954 rhaas@postgresql.org 9260 : 155 : XLogBeginInsert();
207 peter@eisentraut.org 9261 : 155 : XLogRegisterData(&state->startpoint,
9262 : : sizeof(state->startpoint));
1076 michael@paquier.xyz 9263 : 155 : state->stoppoint = XLogInsert(RM_XLOG_ID, XLOG_BACKUP_END);
9264 : :
9265 : : /*
9266 : : * Given that we're not in recovery, InsertTimeLineID is set and can't
9267 : : * change, so we can read it without a lock.
9268 : : */
9269 : 155 : state->stoptli = XLogCtl->InsertTimeLineID;
9270 : :
9271 : : /*
9272 : : * Force a switch to a new xlog segment file, so that the backup is
9273 : : * valid as soon as archiver moves out the current segment file.
9274 : : */
2954 rhaas@postgresql.org 9275 : 155 : RequestXLogSwitch(false);
9276 : :
1076 michael@paquier.xyz 9277 : 155 : state->stoptime = (pg_time_t) time(NULL);
9278 : :
9279 : : /*
9280 : : * Write the backup history file
9281 : : */
9282 : 155 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9283 : 155 : BackupHistoryFilePath(histfilepath, state->stoptli, _logSegNo,
9284 : : state->startpoint, wal_segment_size);
2954 rhaas@postgresql.org 9285 : 155 : fp = AllocateFile(histfilepath, "w");
9286 [ - + ]: 155 : if (!fp)
2954 rhaas@postgresql.org 9287 [ # # ]:UBC 0 : ereport(ERROR,
9288 : : (errcode_for_file_access(),
9289 : : errmsg("could not create file \"%s\": %m",
9290 : : histfilepath)));
9291 : :
9292 : : /* Build and save the contents of the backup history file */
1076 michael@paquier.xyz 9293 :CBC 155 : history_file = build_backup_content(state, true);
1075 9294 : 155 : fprintf(fp, "%s", history_file);
1076 9295 : 155 : pfree(history_file);
9296 : :
2954 rhaas@postgresql.org 9297 [ + - + - : 155 : if (fflush(fp) || ferror(fp) || FreeFile(fp))
- + ]
2954 rhaas@postgresql.org 9298 [ # # ]:UBC 0 : ereport(ERROR,
9299 : : (errcode_for_file_access(),
9300 : : errmsg("could not write file \"%s\": %m",
9301 : : histfilepath)));
9302 : :
9303 : : /*
9304 : : * Clean out any no-longer-needed history files. As a side effect,
9305 : : * this will post a .ready file for the newly created history file,
9306 : : * notifying the archiver that history file may be archived
9307 : : * immediately.
9308 : : */
2954 rhaas@postgresql.org 9309 :CBC 155 : CleanupBackupHistory();
9310 : : }
9311 : :
9312 : : /*
9313 : : * If archiving is enabled, wait for all the required WAL files to be
9314 : : * archived before returning. If archiving isn't enabled, the required WAL
9315 : : * needs to be transported via streaming replication (hopefully with
9316 : : * wal_keep_size set high enough), or some more exotic mechanism like
9317 : : * polling and copying files from pg_wal with script. We have no knowledge
9318 : : * of those mechanisms, so it's up to the user to ensure that he gets all
9319 : : * the required WAL.
9320 : : *
9321 : : * We wait until both the last WAL file filled during backup and the
9322 : : * history file have been archived, and assume that the alphabetic sorting
9323 : : * property of the WAL files ensures any earlier WAL files are safely
9324 : : * archived as well.
9325 : : *
9326 : : * We wait forever, since archive_command is supposed to work and we
9327 : : * assume the admin wanted his backup to work completely. If you don't
9328 : : * wish to wait, then either waitforarchive should be passed in as false,
9329 : : * or you can set statement_timeout. Also, some notices are issued to
9330 : : * clue in anyone who might be doing this interactively.
9331 : : */
9332 : :
9333 [ + + ]: 161 : if (waitforarchive &&
1076 michael@paquier.xyz 9334 [ + + + + : 10 : ((!backup_stopped_in_recovery && XLogArchivingActive()) ||
- + + + +
+ ]
9335 [ + - - + : 1 : (backup_stopped_in_recovery && XLogArchivingAlways())))
- + ]
9336 : : {
9337 : 4 : XLByteToPrevSeg(state->stoppoint, _logSegNo, wal_segment_size);
9338 : 4 : XLogFileName(lastxlogfilename, state->stoptli, _logSegNo,
9339 : : wal_segment_size);
9340 : :
9341 : 4 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9342 : 4 : BackupHistoryFileName(histfilename, state->stoptli, _logSegNo,
9343 : : state->startpoint, wal_segment_size);
9344 : :
5541 bruce@momjian.us 9345 : 4 : seconds_before_warning = 60;
9346 : 4 : waits = 0;
9347 : :
9348 [ + + - + ]: 12 : while (XLogArchiveIsBusy(lastxlogfilename) ||
9349 : 4 : XLogArchiveIsBusy(histfilename))
9350 : : {
9351 [ - + ]: 4 : CHECK_FOR_INTERRUPTS();
9352 : :
9353 [ + - - + ]: 4 : if (!reported_waiting && waits > 5)
9354 : : {
5541 bruce@momjian.us 9355 [ # # ]:UBC 0 : ereport(NOTICE,
9356 : : (errmsg("base backup done, waiting for required WAL segments to be archived")));
9357 : 0 : reported_waiting = true;
9358 : : }
9359 : :
1523 michael@paquier.xyz 9360 :CBC 4 : (void) WaitLatch(MyLatch,
9361 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
9362 : : 1000L,
9363 : : WAIT_EVENT_BACKUP_WAIT_WAL_ARCHIVE);
9364 : 4 : ResetLatch(MyLatch);
9365 : :
5541 bruce@momjian.us 9366 [ - + ]: 4 : if (++waits >= seconds_before_warning)
9367 : : {
5541 bruce@momjian.us 9368 :UBC 0 : seconds_before_warning *= 2; /* This wraps in >10 years... */
9369 [ # # ]: 0 : ereport(WARNING,
9370 : : (errmsg("still waiting for all required WAL segments to be archived (%d seconds elapsed)",
9371 : : waits),
9372 : : errhint("Check that your \"archive_command\" is executing properly. "
9373 : : "You can safely cancel this backup, "
9374 : : "but the database backup will not be usable without all the WAL segments.")));
9375 : : }
9376 : : }
9377 : :
5541 bruce@momjian.us 9378 [ + + ]:CBC 4 : ereport(NOTICE,
9379 : : (errmsg("all required WAL segments have been archived")));
9380 : : }
5323 magnus@hagander.net 9381 [ + + ]: 157 : else if (waitforarchive)
5609 tgl@sss.pgh.pa.us 9382 [ + - ]: 6 : ereport(NOTICE,
9383 : : (errmsg("WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup")));
5354 magnus@hagander.net 9384 : 161 : }
9385 : :
9386 : :
9387 : : /*
9388 : : * do_pg_abort_backup: abort a running backup
9389 : : *
9390 : : * This does just the most basic steps of do_pg_backup_stop(), by taking the
9391 : : * system out of backup mode, thus making it a lot more safe to call from
9392 : : * an error handler.
9393 : : *
9394 : : * 'arg' indicates that it's being called during backup setup; so
9395 : : * sessionBackupState has not been modified yet, but runningBackups has
9396 : : * already been incremented. When it's false, then it's invoked as a
9397 : : * before_shmem_exit handler, and therefore we must not change state
9398 : : * unless sessionBackupState indicates that a backup is actually running.
9399 : : *
9400 : : * NB: This gets used as a PG_ENSURE_ERROR_CLEANUP callback and
9401 : : * before_shmem_exit handler, hence the odd-looking signature.
9402 : : */
9403 : : void
2088 rhaas@postgresql.org 9404 : 8 : do_pg_abort_backup(int code, Datum arg)
9405 : : {
1053 alvherre@alvh.no-ip. 9406 : 8 : bool during_backup_start = DatumGetBool(arg);
9407 : :
9408 : : /* If called during backup start, there shouldn't be one already running */
1048 9409 [ - + - - ]: 8 : Assert(!during_backup_start || sessionBackupState == SESSION_BACKUP_NONE);
9410 : :
1053 9411 [ + - + + ]: 8 : if (during_backup_start || sessionBackupState != SESSION_BACKUP_NONE)
9412 : : {
9413 : 6 : WALInsertLockAcquireExclusive();
9414 [ - + ]: 6 : Assert(XLogCtl->Insert.runningBackups > 0);
9415 : 6 : XLogCtl->Insert.runningBackups--;
9416 : :
9417 : 6 : sessionBackupState = SESSION_BACKUP_NONE;
9418 : 6 : WALInsertLockRelease();
9419 : :
9420 [ + - ]: 6 : if (!during_backup_start)
9421 [ + - ]: 6 : ereport(WARNING,
9422 : : errmsg("aborting backup due to backend exiting before pg_backup_stop was called"));
9423 : : }
2088 rhaas@postgresql.org 9424 : 8 : }
9425 : :
9426 : : /*
9427 : : * Register a handler that will warn about unterminated backups at end of
9428 : : * session, unless this has already been done.
9429 : : */
9430 : : void
9431 : 4 : register_persistent_abort_backup_handler(void)
9432 : : {
9433 : : static bool already_done = false;
9434 : :
9435 [ + + ]: 4 : if (already_done)
9436 : 1 : return;
32 peter@eisentraut.org 9437 :GNC 3 : before_shmem_exit(do_pg_abort_backup, BoolGetDatum(false));
2088 rhaas@postgresql.org 9438 :CBC 3 : already_done = true;
9439 : : }
9440 : :
9441 : : /*
9442 : : * Get latest WAL insert pointer
9443 : : */
9444 : : XLogRecPtr
4987 heikki.linnakangas@i 9445 : 1940 : GetXLogInsertRecPtr(void)
9446 : : {
4002 andres@anarazel.de 9447 : 1940 : XLogCtlInsert *Insert = &XLogCtl->Insert;
9448 : : uint64 current_bytepos;
9449 : :
4443 heikki.linnakangas@i 9450 [ - + ]: 1940 : SpinLockAcquire(&Insert->insertpos_lck);
9451 : 1940 : current_bytepos = Insert->CurrBytePos;
9452 : 1940 : SpinLockRelease(&Insert->insertpos_lck);
9453 : :
9454 : 1940 : return XLogBytePosToRecPtr(current_bytepos);
9455 : : }
9456 : :
9457 : : /*
9458 : : * Get latest WAL write pointer
9459 : : */
9460 : : XLogRecPtr
1298 9461 : 1494 : GetXLogWriteRecPtr(void)
9462 : : {
521 alvherre@alvh.no-ip. 9463 : 1494 : RefreshXLogWriteResult(LogwrtResult);
9464 : :
1298 heikki.linnakangas@i 9465 : 1494 : return LogwrtResult.Write;
9466 : : }
9467 : :
9468 : : /*
9469 : : * Returns the redo pointer of the last checkpoint or restartpoint. This is
9470 : : * the oldest point in WAL that we still need, if we have to restart recovery.
9471 : : */
9472 : : void
9473 : 381 : GetOldestRestartPoint(XLogRecPtr *oldrecptr, TimeLineID *oldtli)
9474 : : {
9475 : 381 : LWLockAcquire(ControlFileLock, LW_SHARED);
9476 : 381 : *oldrecptr = ControlFile->checkPointCopy.redo;
9477 : 381 : *oldtli = ControlFile->checkPointCopy.ThisTimeLineID;
9478 : 381 : LWLockRelease(ControlFileLock);
7106 tgl@sss.pgh.pa.us 9479 : 381 : }
9480 : :
9481 : : /* Thin wrapper around ShutdownWalRcv(). */
9482 : : void
1531 noah@leadboat.com 9483 : 998 : XLogShutdownWalRcv(void)
9484 : : {
9485 : 998 : ShutdownWalRcv();
9486 : :
9487 : 998 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9488 : 998 : XLogCtl->InstallXLogFileSegmentActive = false;
9489 : 998 : LWLockRelease(ControlFileLock);
9490 : 998 : }
9491 : :
9492 : : /* Enable WAL file recycling and preallocation. */
9493 : : void
1298 heikki.linnakangas@i 9494 : 1042 : SetInstallXLogFileSegmentActive(void)
9495 : : {
9496 : 1042 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9497 : 1042 : XLogCtl->InstallXLogFileSegmentActive = true;
9498 : 1042 : LWLockRelease(ControlFileLock);
3650 fujii@postgresql.org 9499 : 1042 : }
9500 : :
9501 : : bool
1298 heikki.linnakangas@i 9502 : 358 : IsInstallXLogFileSegmentActive(void)
9503 : : {
9504 : : bool result;
9505 : :
9506 : 358 : LWLockAcquire(ControlFileLock, LW_SHARED);
9507 : 358 : result = XLogCtl->InstallXLogFileSegmentActive;
9508 : 358 : LWLockRelease(ControlFileLock);
9509 : :
9510 : 358 : return result;
9511 : : }
9512 : :
9513 : : /*
9514 : : * Update the WalWriterSleeping flag.
9515 : : */
9516 : : void
4869 tgl@sss.pgh.pa.us 9517 : 449 : SetWalWriterSleeping(bool sleeping)
9518 : : {
4002 andres@anarazel.de 9519 [ - + ]: 449 : SpinLockAcquire(&XLogCtl->info_lck);
9520 : 449 : XLogCtl->WalWriterSleeping = sleeping;
9521 : 449 : SpinLockRelease(&XLogCtl->info_lck);
4869 tgl@sss.pgh.pa.us 9522 : 449 : }
|
