Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * xlogrecovery.c
4 : : * Functions for WAL recovery, standby mode
5 : : *
6 : : * This source file contains functions controlling WAL recovery.
7 : : * InitWalRecovery() initializes the system for crash or archive recovery,
8 : : * or standby mode, depending on configuration options and the state of
9 : : * the control file and possible backup label file. PerformWalRecovery()
10 : : * performs the actual WAL replay, calling the rmgr-specific redo routines.
11 : : * FinishWalRecovery() performs end-of-recovery checks and cleanup actions,
12 : : * and prepares information needed to initialize the WAL for writes. In
13 : : * addition to these three main functions, there are a bunch of functions
14 : : * for interrogating recovery state and controlling the recovery process.
15 : : *
16 : : *
17 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
18 : : * Portions Copyright (c) 1994, Regents of the University of California
19 : : *
20 : : * src/backend/access/transam/xlogrecovery.c
21 : : *
22 : : *-------------------------------------------------------------------------
23 : : */
24 : :
25 : : #include "postgres.h"
26 : :
27 : : #include <ctype.h>
28 : : #include <time.h>
29 : : #include <sys/stat.h>
30 : : #include <sys/time.h>
31 : : #include <unistd.h>
32 : :
33 : : #include "access/timeline.h"
34 : : #include "access/transam.h"
35 : : #include "access/xact.h"
36 : : #include "access/xlog_internal.h"
37 : : #include "access/xlogarchive.h"
38 : : #include "access/xlogprefetcher.h"
39 : : #include "access/xlogreader.h"
40 : : #include "access/xlogrecovery.h"
41 : : #include "access/xlogutils.h"
42 : : #include "access/xlogwait.h"
43 : : #include "backup/basebackup.h"
44 : : #include "catalog/pg_control.h"
45 : : #include "commands/tablespace.h"
46 : : #include "common/file_utils.h"
47 : : #include "miscadmin.h"
48 : : #include "nodes/miscnodes.h"
49 : : #include "pgstat.h"
50 : : #include "postmaster/bgwriter.h"
51 : : #include "postmaster/startup.h"
52 : : #include "replication/slot.h"
53 : : #include "replication/slotsync.h"
54 : : #include "replication/walreceiver.h"
55 : : #include "storage/fd.h"
56 : : #include "storage/ipc.h"
57 : : #include "storage/latch.h"
58 : : #include "storage/pmsignal.h"
59 : : #include "storage/procarray.h"
60 : : #include "storage/spin.h"
61 : : #include "storage/subsystems.h"
62 : : #include "utils/datetime.h"
63 : : #include "utils/fmgrprotos.h"
64 : : #include "utils/guc_hooks.h"
65 : : #include "utils/pgstat_internal.h"
66 : : #include "utils/pg_lsn.h"
67 : : #include "utils/ps_status.h"
68 : : #include "utils/pg_rusage.h"
69 : : #include "utils/wait_event.h"
70 : :
71 : : /* Unsupported old recovery command file names (relative to $PGDATA) */
72 : : #define RECOVERY_COMMAND_FILE "recovery.conf"
73 : : #define RECOVERY_COMMAND_DONE "recovery.done"
74 : :
75 : : /*
76 : : * GUC support
77 : : */
78 : : const struct config_enum_entry recovery_target_action_options[] = {
79 : : {"pause", RECOVERY_TARGET_ACTION_PAUSE, false},
80 : : {"promote", RECOVERY_TARGET_ACTION_PROMOTE, false},
81 : : {"shutdown", RECOVERY_TARGET_ACTION_SHUTDOWN, false},
82 : : {NULL, 0, false}
83 : : };
84 : :
85 : : /* options formerly taken from recovery.conf for archive recovery */
86 : : char *recoveryRestoreCommand = NULL;
87 : : char *recoveryEndCommand = NULL;
88 : : char *archiveCleanupCommand = NULL;
89 : : RecoveryTargetType recoveryTarget = RECOVERY_TARGET_UNSET;
90 : : bool recoveryTargetInclusive = true;
91 : : int recoveryTargetAction = RECOVERY_TARGET_ACTION_PAUSE;
92 : : TransactionId recoveryTargetXid;
93 : : char *recovery_target_time_string;
94 : : TimestampTz recoveryTargetTime;
95 : : const char *recoveryTargetName;
96 : : XLogRecPtr recoveryTargetLSN;
97 : : int recovery_min_apply_delay = 0;
98 : :
99 : : /* options formerly taken from recovery.conf for XLOG streaming */
100 : : char *PrimaryConnInfo = NULL;
101 : : char *PrimarySlotName = NULL;
102 : : bool wal_receiver_create_temp_slot = false;
103 : :
104 : : /*
105 : : * recoveryTargetTimeLineGoal: what the user requested, if any
106 : : *
107 : : * recoveryTargetTLIRequested: numeric value of requested timeline, if constant
108 : : *
109 : : * recoveryTargetTLI: the currently understood target timeline; changes
110 : : *
111 : : * expectedTLEs: a list of TimeLineHistoryEntries for recoveryTargetTLI and
112 : : * the timelines of its known parents, newest first (so recoveryTargetTLI is
113 : : * always the first list member). Only these TLIs are expected to be seen in
114 : : * the WAL segments we read, and indeed only these TLIs will be considered as
115 : : * candidate WAL files to open at all.
116 : : *
117 : : * curFileTLI: the TLI appearing in the name of the current input WAL file.
118 : : * (This is not necessarily the same as the timeline from which we are
119 : : * replaying WAL, which StartupXLOG calls replayTLI, because we could be
120 : : * scanning data that was copied from an ancestor timeline when the current
121 : : * file was created.) During a sequential scan we do not allow this value
122 : : * to decrease.
123 : : */
124 : : RecoveryTargetTimeLineGoal recoveryTargetTimeLineGoal = RECOVERY_TARGET_TIMELINE_LATEST;
125 : : TimeLineID recoveryTargetTLIRequested = 0;
126 : : TimeLineID recoveryTargetTLI = 0;
127 : : static List *expectedTLEs;
128 : : static TimeLineID curFileTLI;
129 : :
130 : : /*
131 : : * When ArchiveRecoveryRequested is set, archive recovery was requested,
132 : : * ie. signal files were present. When InArchiveRecovery is set, we are
133 : : * currently recovering using offline XLOG archives. These variables are only
134 : : * valid in the startup process.
135 : : *
136 : : * When ArchiveRecoveryRequested is true, but InArchiveRecovery is false, we're
137 : : * currently performing crash recovery using only XLOG files in pg_wal, but
138 : : * will switch to using offline XLOG archives as soon as we reach the end of
139 : : * WAL in pg_wal.
140 : : */
141 : : bool ArchiveRecoveryRequested = false;
142 : : bool InArchiveRecovery = false;
143 : :
144 : : /*
145 : : * When StandbyModeRequested is set, standby mode was requested, i.e.
146 : : * standby.signal file was present. When StandbyMode is set, we are currently
147 : : * in standby mode. These variables are only valid in the startup process.
148 : : * They work similarly to ArchiveRecoveryRequested and InArchiveRecovery.
149 : : */
150 : : static bool StandbyModeRequested = false;
151 : : bool StandbyMode = false;
152 : :
153 : : /* was a signal file present at startup? */
154 : : static bool standby_signal_file_found = false;
155 : : static bool recovery_signal_file_found = false;
156 : :
157 : : /*
158 : : * CheckPointLoc is the position of the checkpoint record that determines
159 : : * where to start the replay. It comes from the backup label file or the
160 : : * control file.
161 : : *
162 : : * RedoStartLSN is the checkpoint's REDO location, also from the backup label
163 : : * file or the control file. In standby mode, XLOG streaming usually starts
164 : : * from the position where an invalid record was found. But if we fail to
165 : : * read even the initial checkpoint record, we use the REDO location instead
166 : : * of the checkpoint location as the start position of XLOG streaming.
167 : : * Otherwise we would have to jump backwards to the REDO location after
168 : : * reading the checkpoint record, because the REDO record can precede the
169 : : * checkpoint record.
170 : : */
171 : : static XLogRecPtr CheckPointLoc = InvalidXLogRecPtr;
172 : : static TimeLineID CheckPointTLI = 0;
173 : : static XLogRecPtr RedoStartLSN = InvalidXLogRecPtr;
174 : : static TimeLineID RedoStartTLI = 0;
175 : :
176 : : /*
177 : : * Local copy of SharedHotStandbyActive variable. False actually means "not
178 : : * known, need to check the shared state".
179 : : */
180 : : static bool LocalHotStandbyActive = false;
181 : :
182 : : /*
183 : : * Local copy of SharedPromoteIsTriggered variable. False actually means "not
184 : : * known, need to check the shared state".
185 : : */
186 : : static bool LocalPromoteIsTriggered = false;
187 : :
188 : : /* Has the recovery code requested a walreceiver wakeup? */
189 : : static bool doRequestWalReceiverReply;
190 : :
191 : : /* XLogReader object used to parse the WAL records */
192 : : static XLogReaderState *xlogreader = NULL;
193 : :
194 : : /* XLogPrefetcher object used to consume WAL records with read-ahead */
195 : : static XLogPrefetcher *xlogprefetcher = NULL;
196 : :
197 : : /* Parameters passed down from ReadRecord to the XLogPageRead callback. */
198 : : typedef struct XLogPageReadPrivate
199 : : {
200 : : int emode;
201 : : bool fetching_ckpt; /* are we fetching a checkpoint record? */
202 : : bool randAccess;
203 : : TimeLineID replayTLI;
204 : : } XLogPageReadPrivate;
205 : :
206 : : /* flag to tell XLogPageRead that we have started replaying */
207 : : static bool InRedo = false;
208 : :
209 : : /*
210 : : * Codes indicating where we got a WAL file from during recovery, or where
211 : : * to attempt to get one.
212 : : */
213 : : typedef enum
214 : : {
215 : : XLOG_FROM_ANY = 0, /* request to read WAL from any source */
216 : : XLOG_FROM_ARCHIVE, /* restored using restore_command */
217 : : XLOG_FROM_PG_WAL, /* existing file in pg_wal */
218 : : XLOG_FROM_STREAM, /* streamed from primary */
219 : : } XLogSource;
220 : :
221 : : /* human-readable names for XLogSources, for debugging output */
222 : : static const char *const xlogSourceNames[] = {"any", "archive", "pg_wal", "stream"};
223 : :
224 : : /*
225 : : * readFile is -1 or a kernel FD for the log file segment that's currently
226 : : * open for reading. readSegNo identifies the segment. readOff is the offset
227 : : * of the page just read, readLen indicates how much of it has been read into
228 : : * readBuf, and readSource indicates where we got the currently open file from.
229 : : *
230 : : * Note: we could use Reserve/ReleaseExternalFD to track consumption of this
231 : : * FD too (like for openLogFile in xlog.c); but it doesn't currently seem
232 : : * worthwhile, since the XLOG is not read by general-purpose sessions.
233 : : */
234 : : static int readFile = -1;
235 : : static XLogSegNo readSegNo = 0;
236 : : static uint32 readOff = 0;
237 : : static uint32 readLen = 0;
238 : : static XLogSource readSource = XLOG_FROM_ANY;
239 : :
240 : : /*
241 : : * Keeps track of which source we're currently reading from. This is
242 : : * different from readSource in that this is always set, even when we don't
243 : : * currently have a WAL file open. If lastSourceFailed is set, our last
244 : : * attempt to read from currentSource failed, and we should try another source
245 : : * next.
246 : : *
247 : : * pendingWalRcvRestart is set when a config change occurs that requires a
248 : : * walreceiver restart. This is only valid in XLOG_FROM_STREAM state.
249 : : */
250 : : static XLogSource currentSource = XLOG_FROM_ANY;
251 : : static bool lastSourceFailed = false;
252 : : static bool pendingWalRcvRestart = false;
253 : :
254 : : /*
255 : : * These variables track when we last obtained some WAL data to process,
256 : : * and where we got it from. (XLogReceiptSource is initially the same as
257 : : * readSource, but readSource gets reset to zero when we don't have data
258 : : * to process right now. It is also different from currentSource, which
259 : : * also changes when we try to read from a source and fail, while
260 : : * XLogReceiptSource tracks where we last successfully read some WAL.)
261 : : */
262 : : static TimestampTz XLogReceiptTime = 0;
263 : : static XLogSource XLogReceiptSource = XLOG_FROM_ANY;
264 : :
265 : : /* Local copy of WalRcv->flushedUpto */
266 : : static XLogRecPtr flushedUpto = InvalidXLogRecPtr;
267 : : static TimeLineID receiveTLI = 0;
268 : :
269 : : /*
270 : : * Copy of minRecoveryPoint and backupEndPoint from the control file.
271 : : *
272 : : * In order to reach consistency, we must replay the WAL up to
273 : : * minRecoveryPoint. If backupEndRequired is true, we must also reach
274 : : * backupEndPoint, or if it's invalid, an end-of-backup record corresponding
275 : : * to backupStartPoint.
276 : : *
277 : : * Note: In archive recovery, after consistency has been reached, the
278 : : * functions in xlog.c will start updating minRecoveryPoint in the control
279 : : * file. But this copy of minRecoveryPoint variable reflects the value at the
280 : : * beginning of recovery, and is *not* updated after consistency is reached.
281 : : */
282 : : static XLogRecPtr minRecoveryPoint;
283 : : static TimeLineID minRecoveryPointTLI;
284 : :
285 : : static XLogRecPtr backupStartPoint;
286 : : static XLogRecPtr backupEndPoint;
287 : : static bool backupEndRequired = false;
288 : :
289 : : /*
290 : : * Have we reached a consistent database state? In crash recovery, we have
291 : : * to replay all the WAL, so reachedConsistency is never set. During archive
292 : : * recovery, the database is consistent once minRecoveryPoint is reached.
293 : : *
294 : : * Consistent state means that the system is internally consistent, all
295 : : * the WAL has been replayed up to a certain point, and importantly, there
296 : : * is no trace of later actions on disk.
297 : : *
298 : : * This flag is used only by the startup process and postmaster. When
299 : : * minRecoveryPoint is reached, the startup process sets it to true and
300 : : * sends a PMSIGNAL_RECOVERY_CONSISTENT signal to the postmaster,
301 : : * which then sets it to true upon receiving the signal.
302 : : */
303 : : bool reachedConsistency = false;
304 : :
305 : : /* Buffers dedicated to consistency checks of size BLCKSZ */
306 : : static char *replay_image_masked = NULL;
307 : : static char *primary_image_masked = NULL;
308 : :
309 : : XLogRecoveryCtlData *XLogRecoveryCtl = NULL;
310 : :
311 : : static void XLogRecoveryShmemRequest(void *arg);
312 : : static void XLogRecoveryShmemInit(void *arg);
313 : :
314 : : const ShmemCallbacks XLogRecoveryShmemCallbacks = {
315 : : .request_fn = XLogRecoveryShmemRequest,
316 : : .init_fn = XLogRecoveryShmemInit,
317 : : };
318 : :
319 : : /*
320 : : * abortedRecPtr is the start pointer of a broken record at end of WAL when
321 : : * recovery completes; missingContrecPtr is the location of the first
322 : : * contrecord that went missing. See CreateOverwriteContrecordRecord for
323 : : * details.
324 : : */
325 : : static XLogRecPtr abortedRecPtr;
326 : : static XLogRecPtr missingContrecPtr;
327 : :
328 : : /*
329 : : * if recoveryStopsBefore/After returns true, it saves information of the stop
330 : : * point here
331 : : */
332 : : static TransactionId recoveryStopXid;
333 : : static TimestampTz recoveryStopTime;
334 : : static XLogRecPtr recoveryStopLSN;
335 : : static char recoveryStopName[MAXFNAMELEN];
336 : : static bool recoveryStopAfter;
337 : :
338 : : /* prototypes for local functions */
339 : : static void ApplyWalRecord(XLogReaderState *xlogreader, XLogRecord *record, TimeLineID *replayTLI);
340 : :
341 : : static void EnableStandbyMode(void);
342 : : static void readRecoverySignalFile(void);
343 : : static void validateRecoveryParameters(void);
344 : : static bool read_backup_label(XLogRecPtr *checkPointLoc,
345 : : TimeLineID *backupLabelTLI,
346 : : bool *backupEndRequired, bool *backupFromStandby);
347 : : static bool read_tablespace_map(List **tablespaces);
348 : :
349 : : static void xlogrecovery_redo(XLogReaderState *record, TimeLineID replayTLI);
350 : : static void CheckRecoveryConsistency(void);
351 : : static void rm_redo_error_callback(void *arg);
352 : : #ifdef WAL_DEBUG
353 : : static void xlog_outrec(StringInfo buf, XLogReaderState *record);
354 : : #endif
355 : : static void xlog_block_info(StringInfo buf, XLogReaderState *record);
356 : : static void checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI,
357 : : TimeLineID prevTLI, TimeLineID replayTLI);
358 : : static bool getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime);
359 : : static void verifyBackupPageConsistency(XLogReaderState *record);
360 : :
361 : : static bool recoveryStopsBefore(XLogReaderState *record);
362 : : static bool recoveryStopsAfter(XLogReaderState *record);
363 : : static char *getRecoveryStopReason(void);
364 : : static void recoveryPausesHere(bool endOfRecovery);
365 : : static bool recoveryApplyDelay(XLogReaderState *record);
366 : : static void ConfirmRecoveryPaused(void);
367 : :
368 : : static XLogRecord *ReadRecord(XLogPrefetcher *xlogprefetcher,
369 : : int emode, bool fetching_ckpt,
370 : : TimeLineID replayTLI);
371 : :
372 : : static int XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr,
373 : : int reqLen, XLogRecPtr targetRecPtr, char *readBuf);
374 : : static XLogPageReadResult WaitForWALToBecomeAvailable(XLogRecPtr RecPtr,
375 : : bool randAccess,
376 : : bool fetching_ckpt,
377 : : XLogRecPtr tliRecPtr,
378 : : TimeLineID replayTLI,
379 : : XLogRecPtr replayLSN,
380 : : bool nonblocking);
381 : : static int emode_for_corrupt_record(int emode, XLogRecPtr RecPtr);
382 : : static XLogRecord *ReadCheckpointRecord(XLogPrefetcher *xlogprefetcher,
383 : : XLogRecPtr RecPtr, TimeLineID replayTLI);
384 : : static bool rescanLatestTimeLine(TimeLineID replayTLI, XLogRecPtr replayLSN);
385 : : static int XLogFileRead(XLogSegNo segno, TimeLineID tli,
386 : : XLogSource source, bool notfoundOk);
387 : : static int XLogFileReadAnyTLI(XLogSegNo segno, XLogSource source);
388 : :
389 : : static bool CheckForStandbyTrigger(void);
390 : : static void SetPromoteIsTriggered(void);
391 : : static bool HotStandbyActiveInReplay(void);
392 : :
393 : : static void SetCurrentChunkStartTime(TimestampTz xtime);
394 : : static void SetLatestXTime(TimestampTz xtime);
395 : :
396 : : /*
397 : : * Register shared memory for WAL recovery
398 : : */
399 : : static void
400 : 1249 : XLogRecoveryShmemRequest(void *arg)
401 : : {
402 : 1249 : ShmemRequestStruct(.name = "XLOG Recovery Ctl",
403 : : .size = sizeof(XLogRecoveryCtlData),
404 : : .ptr = (void **) &XLogRecoveryCtl,
405 : : );
406 : 1249 : }
407 : :
408 : : static void
409 : 1246 : XLogRecoveryShmemInit(void *arg)
410 : : {
411 : 1246 : memset(XLogRecoveryCtl, 0, sizeof(XLogRecoveryCtlData));
412 : :
413 : 1246 : SpinLockInit(&XLogRecoveryCtl->info_lck);
414 : 1246 : InitSharedLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
415 : 1246 : ConditionVariableInit(&XLogRecoveryCtl->recoveryNotPausedCV);
416 : 1246 : }
417 : :
418 : : /*
419 : : * A thin wrapper to enable StandbyMode and do other preparatory work as
420 : : * needed.
421 : : */
422 : : static void
423 : 122 : EnableStandbyMode(void)
424 : : {
425 : 122 : StandbyMode = true;
426 : :
427 : : /*
428 : : * To avoid server log bloat, we don't report recovery progress in a
429 : : * standby as it will always be in recovery unless promoted. We disable
430 : : * startup progress timeout in standby mode to avoid calling
431 : : * startup_progress_timeout_handler() unnecessarily.
432 : : */
433 : 122 : disable_startup_progress_timeout();
434 : 122 : }
435 : :
436 : : /*
437 : : * Prepare the system for WAL recovery, if needed.
438 : : *
439 : : * This is called by StartupXLOG() which coordinates the server startup
440 : : * sequence. This function analyzes the control file and the backup label
441 : : * file, if any, and figures out whether we need to perform crash recovery or
442 : : * archive recovery, and how far we need to replay the WAL to reach a
443 : : * consistent state.
444 : : *
445 : : * This doesn't yet change the on-disk state, except for creating the symlinks
446 : : * from table space map file if any, and for fetching WAL files needed to find
447 : : * the checkpoint record. On entry, the caller has already read the control
448 : : * file into memory, and passes it as argument. This function updates it to
449 : : * reflect the recovery state, and the caller is expected to write it back to
450 : : * disk after initializing other subsystems, but before calling
451 : : * PerformWalRecovery().
452 : : *
453 : : * This initializes some global variables like ArchiveRecoveryRequested, and
454 : : * StandbyModeRequested and InRecovery.
455 : : */
456 : : void
457 : 1089 : InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr,
458 : : bool *haveBackupLabel_ptr, bool *haveTblspcMap_ptr)
459 : : {
460 : : XLogPageReadPrivate *private;
461 : : struct stat st;
462 : : bool wasShutdown;
463 : : XLogRecord *record;
464 : : DBState dbstate_at_startup;
465 : 1089 : bool haveTblspcMap = false;
466 : 1089 : bool haveBackupLabel = false;
467 : : CheckPoint checkPoint;
468 : 1089 : bool backupFromStandby = false;
469 : :
470 : 1089 : dbstate_at_startup = ControlFile->state;
471 : :
472 : : /*
473 : : * Initialize on the assumption we want to recover to the latest timeline
474 : : * that's active according to pg_control.
475 : : */
476 : 1089 : if (ControlFile->minRecoveryPointTLI >
477 [ + + ]: 1089 : ControlFile->checkPointCopy.ThisTimeLineID)
478 : 2 : recoveryTargetTLI = ControlFile->minRecoveryPointTLI;
479 : : else
480 : 1087 : recoveryTargetTLI = ControlFile->checkPointCopy.ThisTimeLineID;
481 : :
482 : : /*
483 : : * Check for signal files, and if so set up state for offline recovery
484 : : */
485 : 1089 : readRecoverySignalFile();
486 : 1089 : validateRecoveryParameters();
487 : :
488 : : /*
489 : : * Take ownership of the wakeup latch if we're going to sleep during
490 : : * recovery, if required.
491 : : */
492 [ + + ]: 1089 : if (ArchiveRecoveryRequested)
493 : 127 : OwnLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
494 : :
495 : : /*
496 : : * Set the WAL reading processor now, as it will be needed when reading
497 : : * the checkpoint record required (backup_label or not).
498 : : */
499 : 1089 : private = palloc0_object(XLogPageReadPrivate);
500 : 1089 : xlogreader =
501 : 1089 : XLogReaderAllocate(wal_segment_size, NULL,
502 : 1089 : XL_ROUTINE(.page_read = &XLogPageRead,
503 : : .segment_open = NULL,
504 : : .segment_close = wal_segment_close),
505 : : private);
506 [ - + ]: 1089 : if (!xlogreader)
507 [ # # ]: 0 : ereport(ERROR,
508 : : (errcode(ERRCODE_OUT_OF_MEMORY),
509 : : errmsg("out of memory"),
510 : : errdetail("Failed while allocating a WAL reading processor.")));
511 : 1089 : xlogreader->system_identifier = ControlFile->system_identifier;
512 : :
513 : : /*
514 : : * Set the WAL decode buffer size. This limits how far ahead we can read
515 : : * in the WAL.
516 : : */
517 : 1089 : XLogReaderSetDecodeBuffer(xlogreader, NULL, wal_decode_buffer_size);
518 : :
519 : : /* Create a WAL prefetcher. */
520 : 1089 : xlogprefetcher = XLogPrefetcherAllocate(xlogreader);
521 : :
522 : : /*
523 : : * Allocate two page buffers dedicated to WAL consistency checks. We do
524 : : * it this way, rather than just making static arrays, for two reasons:
525 : : * (1) no need to waste the storage in most instantiations of the backend;
526 : : * (2) a static char array isn't guaranteed to have any particular
527 : : * alignment, whereas palloc() will provide MAXALIGN'd storage.
528 : : */
529 : 1089 : replay_image_masked = (char *) palloc(BLCKSZ);
530 : 1089 : primary_image_masked = (char *) palloc(BLCKSZ);
531 : :
532 : : /*
533 : : * Read the backup_label file. We want to run this part of the recovery
534 : : * process after checking for signal files and after performing validation
535 : : * of the recovery parameters.
536 : : */
537 [ + + ]: 1089 : if (read_backup_label(&CheckPointLoc, &CheckPointTLI, &backupEndRequired,
538 : : &backupFromStandby))
539 : : {
540 : 89 : List *tablespaces = NIL;
541 : :
542 : : /*
543 : : * Archive recovery was requested, and thanks to the backup label
544 : : * file, we know how far we need to replay to reach consistency. Enter
545 : : * archive recovery directly.
546 : : */
547 : 89 : InArchiveRecovery = true;
548 [ + + ]: 89 : if (StandbyModeRequested)
549 : 77 : EnableStandbyMode();
550 : :
551 : : /*
552 : : * Omitting backup_label when creating a new replica, PITR node etc.
553 : : * unfortunately is a common cause of corruption. Logging that
554 : : * backup_label was used makes it a bit easier to exclude that as the
555 : : * cause of observed corruption.
556 : : *
557 : : * Do so before we try to read the checkpoint record (which can fail),
558 : : * as otherwise it can be hard to understand why a checkpoint other
559 : : * than ControlFile->checkPoint is used.
560 : : */
561 [ + - ]: 89 : ereport(LOG,
562 : : errmsg("starting backup recovery with redo LSN %X/%08X, checkpoint LSN %X/%08X, on timeline ID %u",
563 : : LSN_FORMAT_ARGS(RedoStartLSN),
564 : : LSN_FORMAT_ARGS(CheckPointLoc),
565 : : CheckPointTLI));
566 : :
567 : : /*
568 : : * When a backup_label file is present, we want to roll forward from
569 : : * the checkpoint it identifies, rather than using pg_control.
570 : : */
571 : 89 : record = ReadCheckpointRecord(xlogprefetcher, CheckPointLoc,
572 : : CheckPointTLI);
573 [ + - ]: 89 : if (record != NULL)
574 : : {
575 : 89 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
576 : 89 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
577 [ + + ]: 89 : ereport(DEBUG1,
578 : : errmsg_internal("checkpoint record is at %X/%08X",
579 : : LSN_FORMAT_ARGS(CheckPointLoc)));
580 : 89 : InRecovery = true; /* force recovery even if SHUTDOWNED */
581 : :
582 : : /*
583 : : * Make sure that REDO location exists. This may not be the case
584 : : * if there was a crash during an online backup, which left a
585 : : * backup_label around that references a WAL segment that's
586 : : * already been archived.
587 : : */
588 [ + - ]: 89 : if (checkPoint.redo < CheckPointLoc)
589 : : {
590 : 89 : XLogPrefetcherBeginRead(xlogprefetcher, checkPoint.redo);
591 [ - + ]: 89 : if (!ReadRecord(xlogprefetcher, LOG, false,
592 : : checkPoint.ThisTimeLineID))
593 [ # # ]: 0 : ereport(FATAL,
594 : : errmsg("could not find redo location %X/%08X referenced by checkpoint record at %X/%08X",
595 : : LSN_FORMAT_ARGS(checkPoint.redo), LSN_FORMAT_ARGS(CheckPointLoc)),
596 : : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" or \"%s/standby.signal\" and add required recovery options.\n"
597 : : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
598 : : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
599 : : DataDir, DataDir, DataDir, DataDir));
600 : : }
601 : : }
602 : : else
603 : : {
604 [ # # ]: 0 : ereport(FATAL,
605 : : errmsg("could not locate required checkpoint record at %X/%08X",
606 : : LSN_FORMAT_ARGS(CheckPointLoc)),
607 : : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" or \"%s/standby.signal\" and add required recovery options.\n"
608 : : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
609 : : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
610 : : DataDir, DataDir, DataDir, DataDir));
611 : : wasShutdown = false; /* keep compiler quiet */
612 : : }
613 : :
614 : : /* Read the tablespace_map file if present and create symlinks. */
615 [ + + ]: 89 : if (read_tablespace_map(&tablespaces))
616 : : {
617 : : ListCell *lc;
618 : :
619 [ + - + + : 4 : foreach(lc, tablespaces)
+ + ]
620 : : {
621 : 2 : tablespaceinfo *ti = lfirst(lc);
622 : : char *linkloc;
623 : :
624 : 2 : linkloc = psprintf("%s/%u", PG_TBLSPC_DIR, ti->oid);
625 : :
626 : : /*
627 : : * Remove the existing symlink if any and Create the symlink
628 : : * under PGDATA.
629 : : */
630 : 2 : remove_tablespace_symlink(linkloc);
631 : :
632 [ - + ]: 2 : if (symlink(ti->path, linkloc) < 0)
633 [ # # ]: 0 : ereport(ERROR,
634 : : (errcode_for_file_access(),
635 : : errmsg("could not create symbolic link \"%s\": %m",
636 : : linkloc)));
637 : :
638 : 2 : pfree(ti->path);
639 : 2 : pfree(ti);
640 : : }
641 : :
642 : : /* tell the caller to delete it later */
643 : 2 : haveTblspcMap = true;
644 : : }
645 : :
646 : : /* tell the caller to delete it later */
647 : 89 : haveBackupLabel = true;
648 : : }
649 : : else
650 : : {
651 : : /* No backup_label file has been found if we are here. */
652 : :
653 : : /*
654 : : * If tablespace_map file is present without backup_label file, there
655 : : * is no use of such file. There is no harm in retaining it, but it
656 : : * is better to get rid of the map file so that we don't have any
657 : : * redundant file in data directory and it will avoid any sort of
658 : : * confusion. It seems prudent though to just rename the file out of
659 : : * the way rather than delete it completely, also we ignore any error
660 : : * that occurs in rename operation as even if map file is present
661 : : * without backup_label file, it is harmless.
662 : : */
663 [ + + ]: 1000 : if (stat(TABLESPACE_MAP, &st) == 0)
664 : : {
665 : 1 : unlink(TABLESPACE_MAP_OLD);
666 [ + - ]: 1 : if (durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, DEBUG1) == 0)
667 [ + - ]: 1 : ereport(LOG,
668 : : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
669 : : TABLESPACE_MAP, BACKUP_LABEL_FILE),
670 : : errdetail("File \"%s\" was renamed to \"%s\".",
671 : : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
672 : : else
673 [ # # ]: 0 : ereport(LOG,
674 : : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
675 : : TABLESPACE_MAP, BACKUP_LABEL_FILE),
676 : : errdetail("Could not rename file \"%s\" to \"%s\": %m.",
677 : : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
678 : : }
679 : :
680 : : /*
681 : : * It's possible that archive recovery was requested, but we don't
682 : : * know how far we need to replay the WAL before we reach consistency.
683 : : * This can happen for example if a base backup is taken from a
684 : : * running server using an atomic filesystem snapshot, without calling
685 : : * pg_backup_start/stop. Or if you just kill a running primary server
686 : : * and put it into archive recovery by creating a recovery signal
687 : : * file.
688 : : *
689 : : * Our strategy in that case is to perform crash recovery first,
690 : : * replaying all the WAL present in pg_wal, and only enter archive
691 : : * recovery after that.
692 : : *
693 : : * But usually we already know how far we need to replay the WAL (up
694 : : * to minRecoveryPoint, up to backupEndPoint, or until we see an
695 : : * end-of-backup record), and we can enter archive recovery directly.
696 : : */
697 [ + + ]: 1000 : if (ArchiveRecoveryRequested &&
698 [ + + ]: 45 : (XLogRecPtrIsValid(ControlFile->minRecoveryPoint) ||
699 [ + - ]: 9 : ControlFile->backupEndRequired ||
700 [ + - ]: 9 : XLogRecPtrIsValid(ControlFile->backupEndPoint) ||
701 [ + + ]: 9 : ControlFile->state == DB_SHUTDOWNED))
702 : : {
703 : 44 : InArchiveRecovery = true;
704 [ + - ]: 44 : if (StandbyModeRequested)
705 : 44 : EnableStandbyMode();
706 : : }
707 : :
708 : : /*
709 : : * For the same reason as when starting up with backup_label present,
710 : : * emit a log message when we continue initializing from a base
711 : : * backup.
712 : : */
713 [ - + ]: 1000 : if (XLogRecPtrIsValid(ControlFile->backupStartPoint))
714 [ # # ]: 0 : ereport(LOG,
715 : : errmsg("restarting backup recovery with redo LSN %X/%08X",
716 : : LSN_FORMAT_ARGS(ControlFile->backupStartPoint)));
717 : :
718 : : /* Get the last valid checkpoint record. */
719 : 1000 : CheckPointLoc = ControlFile->checkPoint;
720 : 1000 : CheckPointTLI = ControlFile->checkPointCopy.ThisTimeLineID;
721 : 1000 : RedoStartLSN = ControlFile->checkPointCopy.redo;
722 : 1000 : RedoStartTLI = ControlFile->checkPointCopy.ThisTimeLineID;
723 : 1000 : record = ReadCheckpointRecord(xlogprefetcher, CheckPointLoc,
724 : : CheckPointTLI);
725 [ + + ]: 1000 : if (record != NULL)
726 : : {
727 [ + + ]: 999 : ereport(DEBUG1,
728 : : errmsg_internal("checkpoint record is at %X/%08X",
729 : : LSN_FORMAT_ARGS(CheckPointLoc)));
730 : : }
731 : : else
732 : : {
733 : : /*
734 : : * We used to attempt to go back to a secondary checkpoint record
735 : : * here, but only when not in standby mode. We now just fail if we
736 : : * can't read the last checkpoint because this allows us to
737 : : * simplify processing around checkpoints.
738 : : */
739 [ + - ]: 1 : ereport(FATAL,
740 : : errmsg("could not locate a valid checkpoint record at %X/%08X",
741 : : LSN_FORMAT_ARGS(CheckPointLoc)));
742 : : }
743 : 999 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
744 : 999 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
745 : :
746 : : /* Make sure that REDO location exists. */
747 [ + + ]: 999 : if (checkPoint.redo < CheckPointLoc)
748 : : {
749 : 45 : XLogPrefetcherBeginRead(xlogprefetcher, checkPoint.redo);
750 [ + + ]: 45 : if (!ReadRecord(xlogprefetcher, LOG, false, checkPoint.ThisTimeLineID))
751 [ + - ]: 1 : ereport(FATAL,
752 : : errmsg("could not find redo location %X/%08X referenced by checkpoint record at %X/%08X",
753 : : LSN_FORMAT_ARGS(checkPoint.redo), LSN_FORMAT_ARGS(CheckPointLoc)));
754 : : }
755 : : }
756 : :
757 [ + + ]: 1087 : if (ArchiveRecoveryRequested)
758 : : {
759 [ + + ]: 127 : if (StandbyModeRequested)
760 [ + - ]: 122 : ereport(LOG,
761 : : (errmsg("entering standby mode")));
762 [ - + ]: 5 : else if (recoveryTarget == RECOVERY_TARGET_XID)
763 [ # # ]: 0 : ereport(LOG,
764 : : (errmsg("starting point-in-time recovery to XID %u",
765 : : recoveryTargetXid)));
766 [ - + ]: 5 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
767 [ # # ]: 0 : ereport(LOG,
768 : : (errmsg("starting point-in-time recovery to %s",
769 : : timestamptz_to_str(recoveryTargetTime))));
770 [ + + ]: 5 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
771 [ + - ]: 3 : ereport(LOG,
772 : : (errmsg("starting point-in-time recovery to \"%s\"",
773 : : recoveryTargetName)));
774 [ - + ]: 2 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
775 [ # # ]: 0 : ereport(LOG,
776 : : errmsg("starting point-in-time recovery to WAL location (LSN) \"%X/%08X\"",
777 : : LSN_FORMAT_ARGS(recoveryTargetLSN)));
778 [ - + ]: 2 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
779 [ # # ]: 0 : ereport(LOG,
780 : : (errmsg("starting point-in-time recovery to earliest consistent point")));
781 : : else
782 [ + - ]: 2 : ereport(LOG,
783 : : (errmsg("starting archive recovery")));
784 : : }
785 : :
786 : : /*
787 : : * If the location of the checkpoint record is not on the expected
788 : : * timeline in the history of the requested timeline, we cannot proceed:
789 : : * the backup is not part of the history of the requested timeline.
790 : : */
791 : : Assert(expectedTLEs); /* was initialized by reading checkpoint
792 : : * record */
793 [ - + ]: 1087 : if (tliOfPointInHistory(CheckPointLoc, expectedTLEs) !=
794 : : CheckPointTLI)
795 : : {
796 : : XLogRecPtr switchpoint;
797 : :
798 : : /*
799 : : * tliSwitchPoint will throw an error if the checkpoint's timeline is
800 : : * not in expectedTLEs at all.
801 : : */
802 : 0 : switchpoint = tliSwitchPoint(CheckPointTLI, expectedTLEs, NULL);
803 [ # # # # ]: 0 : ereport(FATAL,
804 : : (errmsg("requested timeline %u is not a child of this server's history",
805 : : recoveryTargetTLI),
806 : : /* translator: %s is a backup_label file or a pg_control file */
807 : : errdetail("Latest checkpoint in file \"%s\" is at %X/%08X on timeline %u, but in the history of the requested timeline, the server forked off from that timeline at %X/%08X.",
808 : : haveBackupLabel ? "backup_label" : "pg_control",
809 : : LSN_FORMAT_ARGS(CheckPointLoc),
810 : : CheckPointTLI,
811 : : LSN_FORMAT_ARGS(switchpoint))));
812 : : }
813 : :
814 : : /*
815 : : * The min recovery point should be part of the requested timeline's
816 : : * history, too.
817 : : */
818 [ + + ]: 1087 : if (XLogRecPtrIsValid(ControlFile->minRecoveryPoint) &&
819 : 45 : tliOfPointInHistory(ControlFile->minRecoveryPoint - 1, expectedTLEs) !=
820 [ - + ]: 45 : ControlFile->minRecoveryPointTLI)
821 [ # # ]: 0 : ereport(FATAL,
822 : : errmsg("requested timeline %u does not contain minimum recovery point %X/%08X on timeline %u",
823 : : recoveryTargetTLI,
824 : : LSN_FORMAT_ARGS(ControlFile->minRecoveryPoint),
825 : : ControlFile->minRecoveryPointTLI));
826 : :
827 [ + + + + ]: 1087 : ereport(DEBUG1,
828 : : errmsg_internal("redo record is at %X/%08X; shutdown %s",
829 : : LSN_FORMAT_ARGS(checkPoint.redo),
830 : : wasShutdown ? "true" : "false"));
831 [ + + ]: 1087 : ereport(DEBUG1,
832 : : (errmsg_internal("next transaction ID: " UINT64_FORMAT "; next OID: %u",
833 : : U64FromFullTransactionId(checkPoint.nextXid),
834 : : checkPoint.nextOid)));
835 [ + + ]: 1087 : ereport(DEBUG1,
836 : : (errmsg_internal("next MultiXactId: %u; next MultiXactOffset: %" PRIu64,
837 : : checkPoint.nextMulti, checkPoint.nextMultiOffset)));
838 [ + + ]: 1087 : ereport(DEBUG1,
839 : : (errmsg_internal("oldest unfrozen transaction ID: %u, in database %u",
840 : : checkPoint.oldestXid, checkPoint.oldestXidDB)));
841 [ + + ]: 1087 : ereport(DEBUG1,
842 : : (errmsg_internal("oldest MultiXactId: %u, in database %u",
843 : : checkPoint.oldestMulti, checkPoint.oldestMultiDB)));
844 [ + + ]: 1087 : ereport(DEBUG1,
845 : : (errmsg_internal("commit timestamp Xid oldest/newest: %u/%u",
846 : : checkPoint.oldestCommitTsXid,
847 : : checkPoint.newestCommitTsXid)));
848 [ - + ]: 1087 : if (!TransactionIdIsNormal(XidFromFullTransactionId(checkPoint.nextXid)))
849 [ # # ]: 0 : ereport(PANIC,
850 : : (errmsg("invalid next transaction ID")));
851 : :
852 : : /* sanity check */
853 [ - + ]: 1087 : if (checkPoint.redo > CheckPointLoc)
854 [ # # ]: 0 : ereport(PANIC,
855 : : (errmsg("invalid redo in checkpoint record")));
856 : :
857 : : /*
858 : : * Check whether we need to force recovery from WAL. If it appears to
859 : : * have been a clean shutdown and we did not have a recovery signal file,
860 : : * then assume no recovery needed.
861 : : */
862 [ + + ]: 1087 : if (checkPoint.redo < CheckPointLoc)
863 : : {
864 [ - + ]: 133 : if (wasShutdown)
865 [ # # ]: 0 : ereport(PANIC,
866 : : (errmsg("invalid redo record in shutdown checkpoint")));
867 : 133 : InRecovery = true;
868 : : }
869 [ + + ]: 954 : else if (ControlFile->state != DB_SHUTDOWNED)
870 : 94 : InRecovery = true;
871 [ + + ]: 860 : else if (ArchiveRecoveryRequested)
872 : : {
873 : : /* force recovery due to presence of recovery signal file */
874 : 8 : InRecovery = true;
875 : : }
876 : :
877 : : /*
878 : : * If recovery is needed, update our in-memory copy of pg_control to show
879 : : * that we are recovering and to show the selected checkpoint as the place
880 : : * we are starting from. We also mark pg_control with any minimum recovery
881 : : * stop point obtained from a backup history file.
882 : : *
883 : : * We don't write the changes to disk yet, though. Only do that after
884 : : * initializing various subsystems.
885 : : */
886 [ + + ]: 1087 : if (InRecovery)
887 : : {
888 [ + + ]: 235 : if (InArchiveRecovery)
889 : : {
890 : 133 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
891 : : }
892 : : else
893 : : {
894 [ + - ]: 102 : ereport(LOG,
895 : : (errmsg("database system was not properly shut down; "
896 : : "automatic recovery in progress")));
897 [ + + ]: 102 : if (recoveryTargetTLI > ControlFile->checkPointCopy.ThisTimeLineID)
898 [ + - ]: 2 : ereport(LOG,
899 : : (errmsg("crash recovery starts in timeline %u "
900 : : "and has target timeline %u",
901 : : ControlFile->checkPointCopy.ThisTimeLineID,
902 : : recoveryTargetTLI)));
903 : 102 : ControlFile->state = DB_IN_CRASH_RECOVERY;
904 : : }
905 : 235 : ControlFile->checkPoint = CheckPointLoc;
906 : 235 : ControlFile->checkPointCopy = checkPoint;
907 [ + + ]: 235 : if (InArchiveRecovery)
908 : : {
909 : : /* initialize minRecoveryPoint if not set yet */
910 [ + + ]: 133 : if (ControlFile->minRecoveryPoint < checkPoint.redo)
911 : : {
912 : 90 : ControlFile->minRecoveryPoint = checkPoint.redo;
913 : 90 : ControlFile->minRecoveryPointTLI = checkPoint.ThisTimeLineID;
914 : : }
915 : : }
916 : :
917 : : /*
918 : : * Set backupStartPoint if we're starting recovery from a base backup.
919 : : *
920 : : * Also set backupEndPoint and use minRecoveryPoint as the backup end
921 : : * location if we're starting recovery from a base backup which was
922 : : * taken from a standby. In this case, the database system status in
923 : : * pg_control must indicate that the database was already in recovery.
924 : : * Usually that will be DB_IN_ARCHIVE_RECOVERY but also can be
925 : : * DB_SHUTDOWNED_IN_RECOVERY if recovery previously was interrupted
926 : : * before reaching this point; e.g. because restore_command or
927 : : * primary_conninfo were faulty.
928 : : *
929 : : * Any other state indicates that the backup somehow became corrupted
930 : : * and we can't sensibly continue with recovery.
931 : : */
932 [ + + ]: 235 : if (haveBackupLabel)
933 : : {
934 : 89 : ControlFile->backupStartPoint = checkPoint.redo;
935 : 89 : ControlFile->backupEndRequired = backupEndRequired;
936 : :
937 [ + + ]: 89 : if (backupFromStandby)
938 : : {
939 [ - + - - ]: 7 : if (dbstate_at_startup != DB_IN_ARCHIVE_RECOVERY &&
940 : : dbstate_at_startup != DB_SHUTDOWNED_IN_RECOVERY)
941 [ # # ]: 0 : ereport(FATAL,
942 : : (errmsg("backup_label contains data inconsistent with control file"),
943 : : errhint("This means that the backup is corrupted and you will "
944 : : "have to use another backup for recovery.")));
945 : 7 : ControlFile->backupEndPoint = ControlFile->minRecoveryPoint;
946 : : }
947 : : }
948 : : }
949 : :
950 : : /* remember these, so that we know when we have reached consistency */
951 : 1087 : backupStartPoint = ControlFile->backupStartPoint;
952 : 1087 : backupEndRequired = ControlFile->backupEndRequired;
953 : 1087 : backupEndPoint = ControlFile->backupEndPoint;
954 [ + + ]: 1087 : if (InArchiveRecovery)
955 : : {
956 : 133 : minRecoveryPoint = ControlFile->minRecoveryPoint;
957 : 133 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
958 : : }
959 : : else
960 : : {
961 : 954 : minRecoveryPoint = InvalidXLogRecPtr;
962 : 954 : minRecoveryPointTLI = 0;
963 : : }
964 : :
965 : : /*
966 : : * Start recovery assuming that the final record isn't lost.
967 : : */
968 : 1087 : abortedRecPtr = InvalidXLogRecPtr;
969 : 1087 : missingContrecPtr = InvalidXLogRecPtr;
970 : :
971 : 1087 : *wasShutdown_ptr = wasShutdown;
972 : 1087 : *haveBackupLabel_ptr = haveBackupLabel;
973 : 1087 : *haveTblspcMap_ptr = haveTblspcMap;
974 : 1087 : }
975 : :
976 : : /*
977 : : * See if there are any recovery signal files and if so, set state for
978 : : * recovery.
979 : : *
980 : : * See if there is a recovery command file (recovery.conf), and if so
981 : : * throw an ERROR since as of PG12 we no longer recognize that.
982 : : */
983 : : static void
984 : 1089 : readRecoverySignalFile(void)
985 : : {
986 : : struct stat stat_buf;
987 : :
988 [ + + ]: 1089 : if (IsBootstrapProcessingMode())
989 : 962 : return;
990 : :
991 : : /*
992 : : * Check for old recovery API file: recovery.conf
993 : : */
994 [ - + ]: 1032 : if (stat(RECOVERY_COMMAND_FILE, &stat_buf) == 0)
995 [ # # ]: 0 : ereport(FATAL,
996 : : (errcode_for_file_access(),
997 : : errmsg("using recovery command file \"%s\" is not supported",
998 : : RECOVERY_COMMAND_FILE)));
999 : :
1000 : : /*
1001 : : * Remove unused .done file, if present. Ignore if absent.
1002 : : */
1003 : 1032 : unlink(RECOVERY_COMMAND_DONE);
1004 : :
1005 : : /*
1006 : : * Check for recovery signal files and if found, fsync them since they
1007 : : * represent server state information. We don't sweat too much about the
1008 : : * possibility of fsync failure, however.
1009 : : */
1010 [ + + ]: 1032 : if (stat(STANDBY_SIGNAL_FILE, &stat_buf) == 0)
1011 : : {
1012 : : int fd;
1013 : :
1014 : 122 : fd = BasicOpenFilePerm(STANDBY_SIGNAL_FILE, O_RDWR | PG_BINARY,
1015 : : S_IRUSR | S_IWUSR);
1016 [ + - ]: 122 : if (fd >= 0)
1017 : : {
1018 : 122 : (void) pg_fsync(fd);
1019 : 122 : close(fd);
1020 : : }
1021 : 122 : standby_signal_file_found = true;
1022 : : }
1023 : :
1024 [ + + ]: 1032 : if (stat(RECOVERY_SIGNAL_FILE, &stat_buf) == 0)
1025 : : {
1026 : : int fd;
1027 : :
1028 : 6 : fd = BasicOpenFilePerm(RECOVERY_SIGNAL_FILE, O_RDWR | PG_BINARY,
1029 : : S_IRUSR | S_IWUSR);
1030 [ + - ]: 6 : if (fd >= 0)
1031 : : {
1032 : 6 : (void) pg_fsync(fd);
1033 : 6 : close(fd);
1034 : : }
1035 : 6 : recovery_signal_file_found = true;
1036 : : }
1037 : :
1038 : : /*
1039 : : * If both signal files are present, standby signal file takes precedence.
1040 : : * If neither is present then we won't enter archive recovery.
1041 : : */
1042 : 1032 : StandbyModeRequested = false;
1043 : 1032 : ArchiveRecoveryRequested = false;
1044 [ + + ]: 1032 : if (standby_signal_file_found)
1045 : : {
1046 : 122 : StandbyModeRequested = true;
1047 : 122 : ArchiveRecoveryRequested = true;
1048 : : }
1049 [ + + ]: 910 : else if (recovery_signal_file_found)
1050 : : {
1051 : 5 : StandbyModeRequested = false;
1052 : 5 : ArchiveRecoveryRequested = true;
1053 : : }
1054 : : else
1055 : 905 : return;
1056 : :
1057 : : /*
1058 : : * We don't support standby mode in standalone backends; that requires
1059 : : * other processes such as the WAL receiver to be alive.
1060 : : */
1061 [ + + - + ]: 127 : if (StandbyModeRequested && !IsUnderPostmaster)
1062 [ # # ]: 0 : ereport(FATAL,
1063 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1064 : : errmsg("standby mode is not supported by single-user servers")));
1065 : : }
1066 : :
1067 : : static void
1068 : 1089 : validateRecoveryParameters(void)
1069 : : {
1070 [ + + ]: 1089 : if (!ArchiveRecoveryRequested)
1071 : 962 : return;
1072 : :
1073 : : /*
1074 : : * Check for compulsory parameters
1075 : : */
1076 [ + + ]: 127 : if (StandbyModeRequested)
1077 : : {
1078 [ + - + + ]: 122 : if ((PrimaryConnInfo == NULL || strcmp(PrimaryConnInfo, "") == 0) &&
1079 [ + - + + ]: 12 : (recoveryRestoreCommand == NULL || strcmp(recoveryRestoreCommand, "") == 0))
1080 [ + - ]: 2 : ereport(WARNING,
1081 : : (errmsg("specified neither \"primary_conninfo\" nor \"restore_command\""),
1082 : : errhint("The database server will regularly poll the pg_wal subdirectory to check for files placed there.")));
1083 : : }
1084 : : else
1085 : : {
1086 [ + - ]: 5 : if (recoveryRestoreCommand == NULL ||
1087 [ - + ]: 5 : strcmp(recoveryRestoreCommand, "") == 0)
1088 [ # # ]: 0 : ereport(FATAL,
1089 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1090 : : errmsg("must specify \"restore_command\" when standby mode is not enabled")));
1091 : : }
1092 : :
1093 : : /*
1094 : : * Override any inconsistent requests. Note that this is a change of
1095 : : * behaviour in 9.5; prior to this we simply ignored a request to pause if
1096 : : * hot_standby = off, which was surprising behaviour.
1097 : : */
1098 [ + + ]: 127 : if (recoveryTargetAction == RECOVERY_TARGET_ACTION_PAUSE &&
1099 [ + + ]: 120 : !EnableHotStandby)
1100 : 3 : recoveryTargetAction = RECOVERY_TARGET_ACTION_SHUTDOWN;
1101 : :
1102 : : /*
1103 : : * Final parsing of recovery_target_time string; see also
1104 : : * check_recovery_target_time().
1105 : : */
1106 [ - + ]: 127 : if (recoveryTarget == RECOVERY_TARGET_TIME)
1107 : : {
1108 : 0 : recoveryTargetTime = DatumGetTimestampTz(DirectFunctionCall3(timestamptz_in,
1109 : : CStringGetDatum(recovery_target_time_string),
1110 : : ObjectIdGetDatum(InvalidOid),
1111 : : Int32GetDatum(-1)));
1112 : : }
1113 : :
1114 : : /*
1115 : : * If user specified recovery_target_timeline, validate it or compute the
1116 : : * "latest" value. We can't do this until after we've gotten the restore
1117 : : * command and set InArchiveRecovery, because we need to fetch timeline
1118 : : * history files from the archive.
1119 : : */
1120 [ - + ]: 127 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
1121 : : {
1122 : 0 : TimeLineID rtli = recoveryTargetTLIRequested;
1123 : :
1124 : : /* Timeline 1 does not have a history file, all else should */
1125 [ # # # # ]: 0 : if (rtli != 1 && !existsTimeLineHistory(rtli))
1126 [ # # ]: 0 : ereport(FATAL,
1127 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1128 : : errmsg("recovery target timeline %u does not exist",
1129 : : rtli)));
1130 : 0 : recoveryTargetTLI = rtli;
1131 : : }
1132 [ + - ]: 127 : else if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
1133 : : {
1134 : : /* We start the "latest" search from pg_control's timeline */
1135 : 127 : recoveryTargetTLI = findNewestTimeLine(recoveryTargetTLI);
1136 : : }
1137 : : else
1138 : : {
1139 : : /*
1140 : : * else we just use the recoveryTargetTLI as already read from
1141 : : * ControlFile
1142 : : */
1143 : : Assert(recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_CONTROLFILE);
1144 : : }
1145 : : }
1146 : :
1147 : : /*
1148 : : * read_backup_label: check to see if a backup_label file is present
1149 : : *
1150 : : * If we see a backup_label during recovery, we assume that we are recovering
1151 : : * from a backup dump file, and we therefore roll forward from the checkpoint
1152 : : * identified by the label file, NOT what pg_control says. This avoids the
1153 : : * problem that pg_control might have been archived one or more checkpoints
1154 : : * later than the start of the dump, and so if we rely on it as the start
1155 : : * point, we will fail to restore a consistent database state.
1156 : : *
1157 : : * Returns true if a backup_label was found (and fills the checkpoint
1158 : : * location and TLI into *checkPointLoc and *backupLabelTLI, respectively);
1159 : : * returns false if not. If this backup_label came from a streamed backup,
1160 : : * *backupEndRequired is set to true. If this backup_label was created during
1161 : : * recovery, *backupFromStandby is set to true.
1162 : : *
1163 : : * Also sets the global variables RedoStartLSN and RedoStartTLI with the LSN
1164 : : * and TLI read from the backup file.
1165 : : */
1166 : : static bool
1167 : 1089 : read_backup_label(XLogRecPtr *checkPointLoc, TimeLineID *backupLabelTLI,
1168 : : bool *backupEndRequired, bool *backupFromStandby)
1169 : : {
1170 : : char startxlogfilename[MAXFNAMELEN];
1171 : : TimeLineID tli_from_walseg,
1172 : : tli_from_file;
1173 : : FILE *lfp;
1174 : : char ch;
1175 : : char backuptype[20];
1176 : : char backupfrom[20];
1177 : : char backuplabel[MAXPGPATH];
1178 : : char backuptime[128];
1179 : : uint32 hi,
1180 : : lo;
1181 : :
1182 : : /* suppress possible uninitialized-variable warnings */
1183 : 1089 : *checkPointLoc = InvalidXLogRecPtr;
1184 : 1089 : *backupLabelTLI = 0;
1185 : 1089 : *backupEndRequired = false;
1186 : 1089 : *backupFromStandby = false;
1187 : :
1188 : : /*
1189 : : * See if label file is present
1190 : : */
1191 : 1089 : lfp = AllocateFile(BACKUP_LABEL_FILE, "r");
1192 [ + + ]: 1089 : if (!lfp)
1193 : : {
1194 [ - + ]: 1000 : if (errno != ENOENT)
1195 [ # # ]: 0 : ereport(FATAL,
1196 : : (errcode_for_file_access(),
1197 : : errmsg("could not read file \"%s\": %m",
1198 : : BACKUP_LABEL_FILE)));
1199 : 1000 : return false; /* it's not there, all is fine */
1200 : : }
1201 : :
1202 : : /*
1203 : : * Read and parse the START WAL LOCATION and CHECKPOINT lines (this code
1204 : : * is pretty crude, but we are not expecting any variability in the file
1205 : : * format).
1206 : : */
1207 [ + - ]: 89 : if (fscanf(lfp, "START WAL LOCATION: %X/%08X (file %08X%16s)%c",
1208 [ - + ]: 89 : &hi, &lo, &tli_from_walseg, startxlogfilename, &ch) != 5 || ch != '\n')
1209 [ # # ]: 0 : ereport(FATAL,
1210 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1211 : : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
1212 : 89 : RedoStartLSN = ((uint64) hi) << 32 | lo;
1213 : 89 : RedoStartTLI = tli_from_walseg;
1214 [ + - ]: 89 : if (fscanf(lfp, "CHECKPOINT LOCATION: %X/%08X%c",
1215 [ - + ]: 89 : &hi, &lo, &ch) != 3 || ch != '\n')
1216 [ # # ]: 0 : ereport(FATAL,
1217 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1218 : : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE)));
1219 : 89 : *checkPointLoc = ((uint64) hi) << 32 | lo;
1220 : 89 : *backupLabelTLI = tli_from_walseg;
1221 : :
1222 : : /*
1223 : : * BACKUP METHOD lets us know if this was a typical backup ("streamed",
1224 : : * which could mean either pg_basebackup or the pg_backup_start/stop
1225 : : * method was used) or if this label came from somewhere else (the only
1226 : : * other option today being from pg_rewind). If this was a streamed
1227 : : * backup then we know that we need to play through until we get to the
1228 : : * end of the WAL which was generated during the backup (at which point we
1229 : : * will have reached consistency and backupEndRequired will be reset to be
1230 : : * false).
1231 : : */
1232 [ + - ]: 89 : if (fscanf(lfp, "BACKUP METHOD: %19s\n", backuptype) == 1)
1233 : : {
1234 [ + + ]: 89 : if (strcmp(backuptype, "streamed") == 0)
1235 : 88 : *backupEndRequired = true;
1236 : : }
1237 : :
1238 : : /*
1239 : : * BACKUP FROM lets us know if this was from a primary or a standby. If
1240 : : * it was from a standby, we'll double-check that the control file state
1241 : : * matches that of a standby.
1242 : : */
1243 [ + - ]: 89 : if (fscanf(lfp, "BACKUP FROM: %19s\n", backupfrom) == 1)
1244 : : {
1245 [ + + ]: 89 : if (strcmp(backupfrom, "standby") == 0)
1246 : 7 : *backupFromStandby = true;
1247 : : }
1248 : :
1249 : : /*
1250 : : * Parse START TIME and LABEL. Those are not mandatory fields for recovery
1251 : : * but checking for their presence is useful for debugging and the next
1252 : : * sanity checks. Cope also with the fact that the result buffers have a
1253 : : * pre-allocated size, hence if the backup_label file has been generated
1254 : : * with strings longer than the maximum assumed here an incorrect parsing
1255 : : * happens. That's fine as only minor consistency checks are done
1256 : : * afterwards.
1257 : : */
1258 [ + - ]: 89 : if (fscanf(lfp, "START TIME: %127[^\n]\n", backuptime) == 1)
1259 [ + + ]: 89 : ereport(DEBUG1,
1260 : : (errmsg_internal("backup time %s in file \"%s\"",
1261 : : backuptime, BACKUP_LABEL_FILE)));
1262 : :
1263 [ + + ]: 89 : if (fscanf(lfp, "LABEL: %1023[^\n]\n", backuplabel) == 1)
1264 [ + + ]: 88 : ereport(DEBUG1,
1265 : : (errmsg_internal("backup label %s in file \"%s\"",
1266 : : backuplabel, BACKUP_LABEL_FILE)));
1267 : :
1268 : : /*
1269 : : * START TIMELINE is new as of 11. Its parsing is not mandatory, still use
1270 : : * it as a sanity check if present.
1271 : : */
1272 [ + + ]: 89 : if (fscanf(lfp, "START TIMELINE: %u\n", &tli_from_file) == 1)
1273 : : {
1274 [ - + ]: 88 : if (tli_from_walseg != tli_from_file)
1275 [ # # ]: 0 : ereport(FATAL,
1276 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1277 : : errmsg("invalid data in file \"%s\"", BACKUP_LABEL_FILE),
1278 : : errdetail("Timeline ID parsed is %u, but expected %u.",
1279 : : tli_from_file, tli_from_walseg)));
1280 : :
1281 [ + + ]: 88 : ereport(DEBUG1,
1282 : : (errmsg_internal("backup timeline %u in file \"%s\"",
1283 : : tli_from_file, BACKUP_LABEL_FILE)));
1284 : : }
1285 : :
1286 [ - + ]: 89 : if (fscanf(lfp, "INCREMENTAL FROM LSN: %X/%08X\n", &hi, &lo) > 0)
1287 [ # # ]: 0 : ereport(FATAL,
1288 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1289 : : errmsg("this is an incremental backup, not a data directory"),
1290 : : errhint("Use pg_combinebackup to reconstruct a valid data directory.")));
1291 : :
1292 [ + - - + ]: 89 : if (ferror(lfp) || FreeFile(lfp))
1293 [ # # ]: 0 : ereport(FATAL,
1294 : : (errcode_for_file_access(),
1295 : : errmsg("could not read file \"%s\": %m",
1296 : : BACKUP_LABEL_FILE)));
1297 : :
1298 : 89 : return true;
1299 : : }
1300 : :
1301 : : /*
1302 : : * read_tablespace_map: check to see if a tablespace_map file is present
1303 : : *
1304 : : * If we see a tablespace_map file during recovery, we assume that we are
1305 : : * recovering from a backup dump file, and we therefore need to create symlinks
1306 : : * as per the information present in tablespace_map file.
1307 : : *
1308 : : * Returns true if a tablespace_map file was found (and fills *tablespaces
1309 : : * with a tablespaceinfo struct for each tablespace listed in the file);
1310 : : * returns false if not.
1311 : : */
1312 : : static bool
1313 : 89 : read_tablespace_map(List **tablespaces)
1314 : : {
1315 : : tablespaceinfo *ti;
1316 : : FILE *lfp;
1317 : : char str[MAXPGPATH];
1318 : : int ch,
1319 : : i,
1320 : : n;
1321 : : bool was_backslash;
1322 : :
1323 : : /*
1324 : : * See if tablespace_map file is present
1325 : : */
1326 : 89 : lfp = AllocateFile(TABLESPACE_MAP, "r");
1327 [ + + ]: 89 : if (!lfp)
1328 : : {
1329 [ - + ]: 87 : if (errno != ENOENT)
1330 [ # # ]: 0 : ereport(FATAL,
1331 : : (errcode_for_file_access(),
1332 : : errmsg("could not read file \"%s\": %m",
1333 : : TABLESPACE_MAP)));
1334 : 87 : return false; /* it's not there, all is fine */
1335 : : }
1336 : :
1337 : : /*
1338 : : * Read and parse the link name and path lines from tablespace_map file
1339 : : * (this code is pretty crude, but we are not expecting any variability in
1340 : : * the file format). De-escape any backslashes that were inserted.
1341 : : */
1342 : 2 : i = 0;
1343 : 2 : was_backslash = false;
1344 [ + + ]: 77 : while ((ch = fgetc(lfp)) != EOF)
1345 : : {
1346 [ + - + + : 75 : if (!was_backslash && (ch == '\n' || ch == '\r'))
- + ]
1347 : 2 : {
1348 : : char *endp;
1349 : :
1350 [ - + ]: 2 : if (i == 0)
1351 : 0 : continue; /* \r immediately followed by \n */
1352 : :
1353 : : /*
1354 : : * The de-escaped line should contain an OID followed by exactly
1355 : : * one space followed by a path. The path might start with
1356 : : * spaces, so don't be too liberal about parsing.
1357 : : */
1358 : 2 : str[i] = '\0';
1359 : 2 : n = 0;
1360 [ + - + + ]: 12 : while (str[n] && str[n] != ' ')
1361 : 10 : n++;
1362 [ + - - + ]: 2 : if (n < 1 || n >= i - 1)
1363 [ # # ]: 0 : ereport(FATAL,
1364 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1365 : : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1366 : 2 : str[n++] = '\0';
1367 : :
1368 : 2 : ti = palloc0_object(tablespaceinfo);
1369 : 2 : errno = 0;
1370 : 2 : ti->oid = strtoul(str, &endp, 10);
1371 [ + - + - : 2 : if (*endp != '\0' || errno == EINVAL || errno == ERANGE)
- + ]
1372 [ # # ]: 0 : ereport(FATAL,
1373 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1374 : : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1375 : 2 : ti->path = pstrdup(str + n);
1376 : 2 : *tablespaces = lappend(*tablespaces, ti);
1377 : :
1378 : 2 : i = 0;
1379 : 2 : continue;
1380 : : }
1381 [ + - - + ]: 73 : else if (!was_backslash && ch == '\\')
1382 : 0 : was_backslash = true;
1383 : : else
1384 : : {
1385 [ + - ]: 73 : if (i < sizeof(str) - 1)
1386 : 73 : str[i++] = ch;
1387 : 73 : was_backslash = false;
1388 : : }
1389 : : }
1390 : :
1391 [ + - - + ]: 2 : if (i != 0 || was_backslash) /* last line not terminated? */
1392 [ # # ]: 0 : ereport(FATAL,
1393 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1394 : : errmsg("invalid data in file \"%s\"", TABLESPACE_MAP)));
1395 : :
1396 [ + - - + ]: 2 : if (ferror(lfp) || FreeFile(lfp))
1397 [ # # ]: 0 : ereport(FATAL,
1398 : : (errcode_for_file_access(),
1399 : : errmsg("could not read file \"%s\": %m",
1400 : : TABLESPACE_MAP)));
1401 : :
1402 : 2 : return true;
1403 : : }
1404 : :
1405 : : /*
1406 : : * Finish WAL recovery.
1407 : : *
1408 : : * This does not close the 'xlogreader' yet, because in some cases the caller
1409 : : * still wants to re-read the last checkpoint record by calling
1410 : : * ReadCheckpointRecord().
1411 : : *
1412 : : * Returns the position of the last valid or applied record, after which new
1413 : : * WAL should be appended, information about why recovery was ended, and some
1414 : : * other things. See the EndOfWalRecoveryInfo struct for details.
1415 : : */
1416 : : EndOfWalRecoveryInfo *
1417 : 1019 : FinishWalRecovery(void)
1418 : : {
1419 : 1019 : EndOfWalRecoveryInfo *result = palloc_object(EndOfWalRecoveryInfo);
1420 : : XLogRecPtr lastRec;
1421 : : TimeLineID lastRecTLI;
1422 : : XLogRecPtr endOfLog;
1423 : :
1424 : : /*
1425 : : * Kill WAL receiver, if it's still running, before we continue to write
1426 : : * the startup checkpoint and aborted-contrecord records. It will trump
1427 : : * over these records and subsequent ones if it's still alive when we
1428 : : * start writing WAL.
1429 : : */
1430 : 1019 : XLogShutdownWalRcv();
1431 : :
1432 : : /*
1433 : : * Shutdown the slot sync worker to drop any temporary slots acquired by
1434 : : * it and to prevent it from keep trying to fetch the failover slots.
1435 : : *
1436 : : * We do not update the 'synced' column in 'pg_replication_slots' system
1437 : : * view from true to false here, as any failed update could leave 'synced'
1438 : : * column false for some slots. This could cause issues during slot sync
1439 : : * after restarting the server as a standby. While updating the 'synced'
1440 : : * column after switching to the new timeline is an option, it does not
1441 : : * simplify the handling for the 'synced' column. Therefore, we retain the
1442 : : * 'synced' column as true after promotion as it may provide useful
1443 : : * information about the slot origin.
1444 : : */
1445 : 1019 : ShutDownSlotSync();
1446 : :
1447 : : /*
1448 : : * We are now done reading the xlog from stream. Turn off streaming
1449 : : * recovery to force fetching the files (which would be required at end of
1450 : : * recovery, e.g., timeline history file) from archive or pg_wal.
1451 : : *
1452 : : * Note that standby mode must be turned off after killing WAL receiver,
1453 : : * i.e., calling XLogShutdownWalRcv().
1454 : : */
1455 : : Assert(!WalRcvStreaming());
1456 : 1019 : StandbyMode = false;
1457 : :
1458 : : /*
1459 : : * Determine where to start writing WAL next.
1460 : : *
1461 : : * Re-fetch the last valid or last applied record, so we can identify the
1462 : : * exact endpoint of what we consider the valid portion of WAL. There may
1463 : : * be an incomplete continuation record after that, in which case
1464 : : * 'abortedRecPtr' and 'missingContrecPtr' are set and the caller will
1465 : : * write a special OVERWRITE_CONTRECORD message to mark that the rest of
1466 : : * it is intentionally missing. See CreateOverwriteContrecordRecord().
1467 : : *
1468 : : * An important side-effect of this is to load the last page into
1469 : : * xlogreader. The caller uses it to initialize the WAL for writing.
1470 : : */
1471 [ + + ]: 1019 : if (!InRecovery)
1472 : : {
1473 : 851 : lastRec = CheckPointLoc;
1474 : 851 : lastRecTLI = CheckPointTLI;
1475 : : }
1476 : : else
1477 : : {
1478 : 168 : lastRec = XLogRecoveryCtl->lastReplayedReadRecPtr;
1479 : 168 : lastRecTLI = XLogRecoveryCtl->lastReplayedTLI;
1480 : : }
1481 : 1019 : XLogPrefetcherBeginRead(xlogprefetcher, lastRec);
1482 : 1019 : (void) ReadRecord(xlogprefetcher, PANIC, false, lastRecTLI);
1483 : 1019 : endOfLog = xlogreader->EndRecPtr;
1484 : :
1485 : : /*
1486 : : * Remember the TLI in the filename of the XLOG segment containing the
1487 : : * end-of-log. It could be different from the timeline that endOfLog
1488 : : * nominally belongs to, if there was a timeline switch in that segment,
1489 : : * and we were reading the old WAL from a segment belonging to a higher
1490 : : * timeline.
1491 : : */
1492 : 1019 : result->endOfLogTLI = xlogreader->seg.ws_tli;
1493 : :
1494 [ + + ]: 1019 : if (ArchiveRecoveryRequested)
1495 : : {
1496 : : /*
1497 : : * We are no longer in archive recovery state.
1498 : : *
1499 : : * We are now done reading the old WAL. Turn off archive fetching if
1500 : : * it was active.
1501 : : */
1502 : : Assert(InArchiveRecovery);
1503 : 60 : InArchiveRecovery = false;
1504 : :
1505 : : /*
1506 : : * If the ending log segment is still open, close it (to avoid
1507 : : * problems on Windows with trying to rename or delete an open file).
1508 : : */
1509 [ + - ]: 60 : if (readFile >= 0)
1510 : : {
1511 : 60 : close(readFile);
1512 : 60 : readFile = -1;
1513 : : }
1514 : : }
1515 : :
1516 : : /*
1517 : : * Copy the last partial block to the caller, for initializing the WAL
1518 : : * buffer for appending new WAL.
1519 : : */
1520 [ + + ]: 1019 : if (endOfLog % XLOG_BLCKSZ != 0)
1521 : : {
1522 : : char *page;
1523 : : int len;
1524 : : XLogRecPtr pageBeginPtr;
1525 : :
1526 : 997 : pageBeginPtr = endOfLog - (endOfLog % XLOG_BLCKSZ);
1527 : : Assert(readOff == XLogSegmentOffset(pageBeginPtr, wal_segment_size));
1528 : :
1529 : : /* Copy the valid part of the last block */
1530 : 997 : len = endOfLog % XLOG_BLCKSZ;
1531 : 997 : page = palloc(len);
1532 : 997 : memcpy(page, xlogreader->readBuf, len);
1533 : :
1534 : 997 : result->lastPageBeginPtr = pageBeginPtr;
1535 : 997 : result->lastPage = page;
1536 : : }
1537 : : else
1538 : : {
1539 : : /* There is no partial block to copy. */
1540 : 22 : result->lastPageBeginPtr = endOfLog;
1541 : 22 : result->lastPage = NULL;
1542 : : }
1543 : :
1544 : : /*
1545 : : * Create a comment for the history file to explain why and where timeline
1546 : : * changed.
1547 : : */
1548 : 1019 : result->recoveryStopReason = getRecoveryStopReason();
1549 : :
1550 : 1019 : result->lastRec = lastRec;
1551 : 1019 : result->lastRecTLI = lastRecTLI;
1552 : 1019 : result->endOfLog = endOfLog;
1553 : :
1554 : 1019 : result->abortedRecPtr = abortedRecPtr;
1555 : 1019 : result->missingContrecPtr = missingContrecPtr;
1556 : :
1557 : 1019 : result->standby_signal_file_found = standby_signal_file_found;
1558 : 1019 : result->recovery_signal_file_found = recovery_signal_file_found;
1559 : :
1560 : 1019 : return result;
1561 : : }
1562 : :
1563 : : /*
1564 : : * Clean up the WAL reader and leftovers from restoring WAL from archive
1565 : : */
1566 : : void
1567 : 1019 : ShutdownWalRecovery(void)
1568 : : {
1569 : : char recoveryPath[MAXPGPATH];
1570 : :
1571 : : /* Final update of pg_stat_recovery_prefetch. */
1572 : 1019 : XLogPrefetcherComputeStats(xlogprefetcher);
1573 : :
1574 : : /* Shut down xlogreader */
1575 [ + + ]: 1019 : if (readFile >= 0)
1576 : : {
1577 : 959 : close(readFile);
1578 : 959 : readFile = -1;
1579 : : }
1580 : 1019 : pfree(xlogreader->private_data);
1581 : 1019 : XLogReaderFree(xlogreader);
1582 : 1019 : XLogPrefetcherFree(xlogprefetcher);
1583 : :
1584 [ + + ]: 1019 : if (ArchiveRecoveryRequested)
1585 : : {
1586 : : /*
1587 : : * Since there might be a partial WAL segment named RECOVERYXLOG, get
1588 : : * rid of it.
1589 : : */
1590 : 60 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYXLOG");
1591 : 60 : unlink(recoveryPath); /* ignore any error */
1592 : :
1593 : : /* Get rid of any remaining recovered timeline-history file, too */
1594 : 60 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYHISTORY");
1595 : 60 : unlink(recoveryPath); /* ignore any error */
1596 : : }
1597 : :
1598 : : /*
1599 : : * We don't need the latch anymore. It's not strictly necessary to disown
1600 : : * it, but let's do it for the sake of tidiness.
1601 : : */
1602 [ + + ]: 1019 : if (ArchiveRecoveryRequested)
1603 : 60 : DisownLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
1604 : 1019 : }
1605 : :
1606 : : /*
1607 : : * Perform WAL recovery.
1608 : : *
1609 : : * If the system was shut down cleanly, this is never called.
1610 : : */
1611 : : void
1612 : 234 : PerformWalRecovery(void)
1613 : : {
1614 : : XLogRecord *record;
1615 : 234 : bool reachedRecoveryTarget = false;
1616 : : TimeLineID replayTLI;
1617 : :
1618 : : /*
1619 : : * Initialize shared variables for tracking progress of WAL replay, as if
1620 : : * we had just replayed the record before the REDO location (or the
1621 : : * checkpoint record itself, if it's a shutdown checkpoint).
1622 : : */
1623 : 234 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
1624 [ + + ]: 234 : if (RedoStartLSN < CheckPointLoc)
1625 : : {
1626 : 132 : XLogRecoveryCtl->lastReplayedReadRecPtr = InvalidXLogRecPtr;
1627 : 132 : XLogRecoveryCtl->lastReplayedEndRecPtr = RedoStartLSN;
1628 : 132 : XLogRecoveryCtl->lastReplayedTLI = RedoStartTLI;
1629 : : }
1630 : : else
1631 : : {
1632 : 102 : XLogRecoveryCtl->lastReplayedReadRecPtr = xlogreader->ReadRecPtr;
1633 : 102 : XLogRecoveryCtl->lastReplayedEndRecPtr = xlogreader->EndRecPtr;
1634 : 102 : XLogRecoveryCtl->lastReplayedTLI = CheckPointTLI;
1635 : : }
1636 : 234 : XLogRecoveryCtl->replayEndRecPtr = XLogRecoveryCtl->lastReplayedEndRecPtr;
1637 : 234 : XLogRecoveryCtl->replayEndTLI = XLogRecoveryCtl->lastReplayedTLI;
1638 : 234 : XLogRecoveryCtl->recoveryLastXTime = 0;
1639 : 234 : XLogRecoveryCtl->currentChunkStartTime = 0;
1640 : 234 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_NOT_PAUSED;
1641 : 234 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
1642 : :
1643 : : /* Also ensure XLogReceiptTime has a sane value */
1644 : 234 : XLogReceiptTime = GetCurrentTimestamp();
1645 : :
1646 : : /*
1647 : : * Let postmaster know we've started redo now, so that it can launch the
1648 : : * archiver if necessary.
1649 : : */
1650 [ + + ]: 234 : if (IsUnderPostmaster)
1651 : 225 : SendPostmasterSignal(PMSIGNAL_RECOVERY_STARTED);
1652 : :
1653 : : /*
1654 : : * Allow read-only connections immediately if we're consistent already.
1655 : : */
1656 : 234 : CheckRecoveryConsistency();
1657 : :
1658 : : /*
1659 : : * Find the first record that logically follows the checkpoint --- it
1660 : : * might physically precede it, though.
1661 : : */
1662 [ + + ]: 234 : if (RedoStartLSN < CheckPointLoc)
1663 : : {
1664 : : /* back up to find the record */
1665 : 132 : replayTLI = RedoStartTLI;
1666 : 132 : XLogPrefetcherBeginRead(xlogprefetcher, RedoStartLSN);
1667 : 132 : record = ReadRecord(xlogprefetcher, PANIC, false, replayTLI);
1668 : :
1669 : : /*
1670 : : * If a checkpoint record's redo pointer points back to an earlier
1671 : : * LSN, the record at that LSN should be an XLOG_CHECKPOINT_REDO
1672 : : * record.
1673 : : */
1674 [ + - ]: 132 : if (record->xl_rmid != RM_XLOG_ID ||
1675 [ - + ]: 132 : (record->xl_info & ~XLR_INFO_MASK) != XLOG_CHECKPOINT_REDO)
1676 [ # # ]: 0 : ereport(FATAL,
1677 : : errmsg("unexpected record type found at redo point %X/%08X",
1678 : : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr)));
1679 : : }
1680 : : else
1681 : : {
1682 : : /* just have to read next record after CheckPoint */
1683 : : Assert(xlogreader->ReadRecPtr == CheckPointLoc);
1684 : 102 : replayTLI = CheckPointTLI;
1685 : 102 : record = ReadRecord(xlogprefetcher, LOG, false, replayTLI);
1686 : : }
1687 : :
1688 [ + + ]: 234 : if (record != NULL)
1689 : : {
1690 : : TimestampTz xtime;
1691 : : PGRUsage ru0;
1692 : :
1693 : 225 : pg_rusage_init(&ru0);
1694 : :
1695 : 225 : InRedo = true;
1696 : :
1697 : 225 : RmgrStartup();
1698 : :
1699 [ + - ]: 225 : ereport(LOG,
1700 : : errmsg("redo starts at %X/%08X",
1701 : : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr)));
1702 : :
1703 : : /* Prepare to report progress of the redo phase. */
1704 [ + + ]: 225 : if (!StandbyMode)
1705 : 108 : begin_startup_progress_phase();
1706 : :
1707 : : /*
1708 : : * main redo apply loop
1709 : : */
1710 : : do
1711 : : {
1712 [ + + ]: 2964629 : if (!StandbyMode)
1713 [ - + - - ]: 327477 : ereport_startup_progress("redo in progress, elapsed time: %ld.%02d s, current LSN: %X/%08X",
1714 : : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr));
1715 : :
1716 : : #ifdef WAL_DEBUG
1717 : : if (XLOG_DEBUG)
1718 : : {
1719 : : StringInfoData buf;
1720 : :
1721 : : initStringInfo(&buf);
1722 : : appendStringInfo(&buf, "REDO @ %X/%08X; LSN %X/%08X: ",
1723 : : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr),
1724 : : LSN_FORMAT_ARGS(xlogreader->EndRecPtr));
1725 : : xlog_outrec(&buf, xlogreader);
1726 : : appendStringInfoString(&buf, " - ");
1727 : : xlog_outdesc(&buf, xlogreader);
1728 : : elog(LOG, "%s", buf.data);
1729 : : pfree(buf.data);
1730 : : }
1731 : : #endif
1732 : :
1733 : : /* Handle interrupt signals of startup process */
1734 : 2964629 : ProcessStartupProcInterrupts();
1735 : :
1736 : : /*
1737 : : * Pause WAL replay, if requested by a hot-standby session via
1738 : : * SetRecoveryPause().
1739 : : *
1740 : : * Note that we intentionally don't take the info_lck spinlock
1741 : : * here. We might therefore read a slightly stale value of the
1742 : : * recoveryPause flag, but it can't be very stale (no worse than
1743 : : * the last spinlock we did acquire). Since a pause request is a
1744 : : * pretty asynchronous thing anyway, possibly responding to it one
1745 : : * WAL record later than we otherwise would is a minor issue, so
1746 : : * it doesn't seem worth adding another spinlock cycle to prevent
1747 : : * that.
1748 : : */
1749 [ - + ]: 2964629 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
1750 : : RECOVERY_NOT_PAUSED)
1751 : 0 : recoveryPausesHere(false);
1752 : :
1753 : : /*
1754 : : * Have we reached our recovery target?
1755 : : */
1756 [ + + ]: 2964629 : if (recoveryStopsBefore(xlogreader))
1757 : : {
1758 : 2 : reachedRecoveryTarget = true;
1759 : 2 : break;
1760 : : }
1761 : :
1762 : : /*
1763 : : * If we've been asked to lag the primary, wait on latch until
1764 : : * enough time has passed.
1765 : : */
1766 [ + + ]: 2964627 : if (recoveryApplyDelay(xlogreader))
1767 : : {
1768 : : /*
1769 : : * We test for paused recovery again here. If user sets
1770 : : * delayed apply, it may be because they expect to pause
1771 : : * recovery in case of problems, so we must test again here
1772 : : * otherwise pausing during the delay-wait wouldn't work.
1773 : : */
1774 [ + + ]: 21 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
1775 : : RECOVERY_NOT_PAUSED)
1776 : 1 : recoveryPausesHere(false);
1777 : : }
1778 : :
1779 : : /*
1780 : : * Apply the record
1781 : : */
1782 : 2964627 : ApplyWalRecord(xlogreader, record, &replayTLI);
1783 : :
1784 : : /*
1785 : : * Wake up processes waiting for standby replay, write, or flush
1786 : : * LSN to reach current replay position. Replay implies that the
1787 : : * WAL was already written and flushed to disk, so write and flush
1788 : : * waiters can be woken at the replay position too.
1789 : : */
1790 : 2964625 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_REPLAY,
1791 : 2964625 : XLogRecoveryCtl->lastReplayedEndRecPtr);
1792 : 2964625 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_WRITE,
1793 : 2964625 : XLogRecoveryCtl->lastReplayedEndRecPtr);
1794 : 2964625 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_FLUSH,
1795 : 2964625 : XLogRecoveryCtl->lastReplayedEndRecPtr);
1796 : :
1797 : : /* Exit loop if we reached inclusive recovery target */
1798 [ + + ]: 2964625 : if (recoveryStopsAfter(xlogreader))
1799 : : {
1800 : 5 : reachedRecoveryTarget = true;
1801 : 5 : break;
1802 : : }
1803 : :
1804 : : /* Else, try to fetch the next WAL record */
1805 : 2964620 : record = ReadRecord(xlogprefetcher, LOG, false, replayTLI);
1806 [ + + ]: 2964557 : } while (record != NULL);
1807 : :
1808 : : /*
1809 : : * end of main redo apply loop
1810 : : */
1811 : :
1812 [ + + ]: 160 : if (reachedRecoveryTarget)
1813 : : {
1814 [ - + ]: 7 : if (!reachedConsistency)
1815 [ # # ]: 0 : ereport(FATAL,
1816 : : (errmsg("requested recovery stop point is before consistent recovery point")));
1817 : :
1818 : : /*
1819 : : * This is the last point where we can restart recovery with a new
1820 : : * recovery target, if we shutdown and begin again. After this,
1821 : : * Resource Managers may choose to do permanent corrective actions
1822 : : * at end of recovery.
1823 : : */
1824 [ - + + - ]: 7 : switch (recoveryTargetAction)
1825 : : {
1826 : 0 : case RECOVERY_TARGET_ACTION_SHUTDOWN:
1827 : :
1828 : : /*
1829 : : * exit with special return code to request shutdown of
1830 : : * postmaster. Log messages issued from postmaster.
1831 : : */
1832 : 0 : proc_exit(3);
1833 : :
1834 : 1 : case RECOVERY_TARGET_ACTION_PAUSE:
1835 : 1 : SetRecoveryPause(true);
1836 : 1 : recoveryPausesHere(true);
1837 : :
1838 : : /* drop into promote */
1839 : : pg_fallthrough;
1840 : :
1841 : 7 : case RECOVERY_TARGET_ACTION_PROMOTE:
1842 : 7 : break;
1843 : : }
1844 : : }
1845 : :
1846 : 160 : RmgrCleanup();
1847 : :
1848 [ + - ]: 160 : ereport(LOG,
1849 : : errmsg("redo done at %X/%08X system usage: %s",
1850 : : LSN_FORMAT_ARGS(xlogreader->ReadRecPtr),
1851 : : pg_rusage_show(&ru0)));
1852 : 160 : xtime = GetLatestXTime();
1853 [ + + ]: 160 : if (xtime)
1854 [ + - ]: 43 : ereport(LOG,
1855 : : (errmsg("last completed transaction was at log time %s",
1856 : : timestamptz_to_str(xtime))));
1857 : :
1858 : 160 : InRedo = false;
1859 : : }
1860 : : else
1861 : : {
1862 : : /* there are no WAL records following the checkpoint */
1863 [ + - ]: 9 : ereport(LOG,
1864 : : (errmsg("redo is not required")));
1865 : : }
1866 : :
1867 : : /*
1868 : : * This check is intentionally after the above log messages that indicate
1869 : : * how far recovery went.
1870 : : */
1871 [ + + ]: 169 : if (ArchiveRecoveryRequested &&
1872 [ + + ]: 61 : recoveryTarget != RECOVERY_TARGET_UNSET &&
1873 [ + + ]: 8 : !reachedRecoveryTarget)
1874 [ + - ]: 1 : ereport(FATAL,
1875 : : (errcode(ERRCODE_CONFIG_FILE_ERROR),
1876 : : errmsg("recovery ended before configured recovery target was reached")));
1877 : 168 : }
1878 : :
1879 : : /*
1880 : : * Subroutine of PerformWalRecovery, to apply one WAL record.
1881 : : */
1882 : : static void
1883 : 2964627 : ApplyWalRecord(XLogReaderState *xlogreader, XLogRecord *record, TimeLineID *replayTLI)
1884 : : {
1885 : : ErrorContextCallback errcallback;
1886 : 2964627 : bool switchedTLI = false;
1887 : :
1888 : : /* Setup error traceback support for ereport() */
1889 : 2964627 : errcallback.callback = rm_redo_error_callback;
1890 : 2964627 : errcallback.arg = xlogreader;
1891 : 2964627 : errcallback.previous = error_context_stack;
1892 : 2964627 : error_context_stack = &errcallback;
1893 : :
1894 : : /*
1895 : : * TransamVariables->nextXid must be beyond record's xid.
1896 : : */
1897 : 2964627 : AdvanceNextFullTransactionIdPastXid(record->xl_xid);
1898 : :
1899 : : /*
1900 : : * Before replaying this record, check if this record causes the current
1901 : : * timeline to change. The record is already considered to be part of the
1902 : : * new timeline, so we update replayTLI before replaying it. That's
1903 : : * important so that replayEndTLI, which is recorded as the minimum
1904 : : * recovery point's TLI if recovery stops after this record, is set
1905 : : * correctly.
1906 : : */
1907 [ + + ]: 2964627 : if (record->xl_rmid == RM_XLOG_ID)
1908 : : {
1909 : 118476 : TimeLineID newReplayTLI = *replayTLI;
1910 : 118476 : TimeLineID prevReplayTLI = *replayTLI;
1911 : 118476 : uint8 info = record->xl_info & ~XLR_INFO_MASK;
1912 : :
1913 [ + + ]: 118476 : if (info == XLOG_CHECKPOINT_SHUTDOWN)
1914 : : {
1915 : : CheckPoint checkPoint;
1916 : :
1917 : 44 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
1918 : 44 : newReplayTLI = checkPoint.ThisTimeLineID;
1919 : 44 : prevReplayTLI = checkPoint.PrevTimeLineID;
1920 : : }
1921 [ + + ]: 118432 : else if (info == XLOG_END_OF_RECOVERY)
1922 : : {
1923 : : xl_end_of_recovery xlrec;
1924 : :
1925 : 13 : memcpy(&xlrec, XLogRecGetData(xlogreader), sizeof(xl_end_of_recovery));
1926 : 13 : newReplayTLI = xlrec.ThisTimeLineID;
1927 : 13 : prevReplayTLI = xlrec.PrevTimeLineID;
1928 : : }
1929 : :
1930 [ + + ]: 118476 : if (newReplayTLI != *replayTLI)
1931 : : {
1932 : : /* Check that it's OK to switch to this TLI */
1933 : 14 : checkTimeLineSwitch(xlogreader->EndRecPtr,
1934 : : newReplayTLI, prevReplayTLI, *replayTLI);
1935 : :
1936 : : /* Following WAL records should be run with new TLI */
1937 : 14 : *replayTLI = newReplayTLI;
1938 : 14 : switchedTLI = true;
1939 : : }
1940 : : }
1941 : :
1942 : : /*
1943 : : * Update shared replayEndRecPtr before replaying this record, so that
1944 : : * XLogFlush will update minRecoveryPoint correctly.
1945 : : */
1946 : 2964627 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
1947 : 2964627 : XLogRecoveryCtl->replayEndRecPtr = xlogreader->EndRecPtr;
1948 : 2964627 : XLogRecoveryCtl->replayEndTLI = *replayTLI;
1949 : 2964627 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
1950 : :
1951 : : /*
1952 : : * If we are attempting to enter Hot Standby mode, process XIDs we see
1953 : : */
1954 [ + + ]: 2964627 : if (standbyState >= STANDBY_INITIALIZED &&
1955 [ + + ]: 2656800 : TransactionIdIsValid(record->xl_xid))
1956 : 2594138 : RecordKnownAssignedTransactionIds(record->xl_xid);
1957 : :
1958 : : /*
1959 : : * Some XLOG record types that are related to recovery are processed
1960 : : * directly here, rather than in xlog_redo()
1961 : : */
1962 [ + + ]: 2964627 : if (record->xl_rmid == RM_XLOG_ID)
1963 : 118476 : xlogrecovery_redo(xlogreader, *replayTLI);
1964 : :
1965 : : /* Now apply the WAL record itself */
1966 : 2964627 : GetRmgr(record->xl_rmid).rm_redo(xlogreader);
1967 : :
1968 : : /*
1969 : : * After redo, check whether the backup pages associated with the WAL
1970 : : * record are consistent with the existing pages. This check is done only
1971 : : * if consistency check is enabled for this record.
1972 : : */
1973 [ + + ]: 2964625 : if ((record->xl_info & XLR_CHECK_CONSISTENCY) != 0)
1974 : 2275675 : verifyBackupPageConsistency(xlogreader);
1975 : :
1976 : : /* Pop the error context stack */
1977 : 2964625 : error_context_stack = errcallback.previous;
1978 : :
1979 : : /*
1980 : : * Update lastReplayedEndRecPtr after this record has been successfully
1981 : : * replayed.
1982 : : */
1983 : 2964625 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
1984 : 2964625 : XLogRecoveryCtl->lastReplayedReadRecPtr = xlogreader->ReadRecPtr;
1985 : 2964625 : XLogRecoveryCtl->lastReplayedEndRecPtr = xlogreader->EndRecPtr;
1986 : 2964625 : XLogRecoveryCtl->lastReplayedTLI = *replayTLI;
1987 : 2964625 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
1988 : :
1989 : : /* ------
1990 : : * Wakeup walsenders:
1991 : : *
1992 : : * On the standby, the WAL is flushed first (which will only wake up
1993 : : * physical walsenders) and then applied, which will only wake up logical
1994 : : * walsenders.
1995 : : *
1996 : : * Indeed, logical walsenders on standby can't decode and send data until
1997 : : * it's been applied.
1998 : : *
1999 : : * Physical walsenders don't need to be woken up during replay unless
2000 : : * cascading replication is allowed and time line change occurred (so that
2001 : : * they can notice that they are on a new time line).
2002 : : *
2003 : : * That's why the wake up conditions are for:
2004 : : *
2005 : : * - physical walsenders in case of new time line and cascade
2006 : : * replication is allowed
2007 : : * - logical walsenders in case cascade replication is allowed (could not
2008 : : * be created otherwise)
2009 : : * ------
2010 : : */
2011 [ + + + + ]: 2964625 : if (AllowCascadeReplication())
2012 : 2712739 : WalSndWakeup(switchedTLI, true);
2013 : :
2014 : : /*
2015 : : * If rm_redo called XLogRequestWalReceiverReply, then we wake up the
2016 : : * receiver so that it notices the updated lastReplayedEndRecPtr and sends
2017 : : * a reply to the primary.
2018 : : */
2019 [ + + ]: 2964625 : if (doRequestWalReceiverReply)
2020 : : {
2021 : 2 : doRequestWalReceiverReply = false;
2022 : 2 : WalRcvRequestApplyReply();
2023 : : }
2024 : :
2025 : : /* Allow read-only connections if we're consistent now */
2026 : 2964625 : CheckRecoveryConsistency();
2027 : :
2028 : : /* Is this a timeline switch? */
2029 [ + + ]: 2964625 : if (switchedTLI)
2030 : : {
2031 : : /*
2032 : : * Before we continue on the new timeline, clean up any (possibly
2033 : : * bogus) future WAL segments on the old timeline.
2034 : : */
2035 : 14 : RemoveNonParentXlogFiles(xlogreader->EndRecPtr, *replayTLI);
2036 : :
2037 : : /* Reset the prefetcher. */
2038 : 14 : XLogPrefetchReconfigure();
2039 : : }
2040 : 2964625 : }
2041 : :
2042 : : /*
2043 : : * Some XLOG RM record types that are directly related to WAL recovery are
2044 : : * handled here rather than in the xlog_redo()
2045 : : */
2046 : : static void
2047 : 118476 : xlogrecovery_redo(XLogReaderState *record, TimeLineID replayTLI)
2048 : : {
2049 : 118476 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2050 : 118476 : XLogRecPtr lsn = record->EndRecPtr;
2051 : :
2052 : : Assert(XLogRecGetRmid(record) == RM_XLOG_ID);
2053 : :
2054 [ + + ]: 118476 : if (info == XLOG_OVERWRITE_CONTRECORD)
2055 : : {
2056 : : /* Verify the payload of a XLOG_OVERWRITE_CONTRECORD record. */
2057 : : xl_overwrite_contrecord xlrec;
2058 : :
2059 : 1 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_overwrite_contrecord));
2060 [ - + ]: 1 : if (xlrec.overwritten_lsn != record->overwrittenRecPtr)
2061 [ # # ]: 0 : elog(FATAL, "mismatching overwritten LSN %X/%08X -> %X/%08X",
2062 : : LSN_FORMAT_ARGS(xlrec.overwritten_lsn),
2063 : : LSN_FORMAT_ARGS(record->overwrittenRecPtr));
2064 : :
2065 : : /* We have safely skipped the aborted record */
2066 : 1 : abortedRecPtr = InvalidXLogRecPtr;
2067 : 1 : missingContrecPtr = InvalidXLogRecPtr;
2068 : :
2069 [ + - ]: 1 : ereport(LOG,
2070 : : errmsg("successfully skipped missing contrecord at %X/%08X, overwritten at %s",
2071 : : LSN_FORMAT_ARGS(xlrec.overwritten_lsn),
2072 : : timestamptz_to_str(xlrec.overwrite_time)));
2073 : :
2074 : : /* Verifying the record should only happen once */
2075 : 1 : record->overwrittenRecPtr = InvalidXLogRecPtr;
2076 : : }
2077 [ + + ]: 118475 : else if (info == XLOG_BACKUP_END)
2078 : : {
2079 : : XLogRecPtr startpoint;
2080 : :
2081 : 104 : memcpy(&startpoint, XLogRecGetData(record), sizeof(startpoint));
2082 : :
2083 [ + + ]: 104 : if (backupStartPoint == startpoint)
2084 : : {
2085 : : /*
2086 : : * We have reached the end of base backup, the point where
2087 : : * pg_backup_stop() was done. The data on disk is now consistent
2088 : : * (assuming we have also reached minRecoveryPoint). Set
2089 : : * backupEndPoint to the current LSN, so that the next call to
2090 : : * CheckRecoveryConsistency() will notice it and do the
2091 : : * end-of-backup processing.
2092 : : */
2093 [ + + ]: 86 : elog(DEBUG1, "end of backup record reached");
2094 : :
2095 : 86 : backupEndPoint = lsn;
2096 : : }
2097 : : else
2098 [ + + ]: 18 : elog(DEBUG1, "saw end-of-backup record for backup starting at %X/%08X, waiting for %X/%08X",
2099 : : LSN_FORMAT_ARGS(startpoint), LSN_FORMAT_ARGS(backupStartPoint));
2100 : : }
2101 : 118476 : }
2102 : :
2103 : : /*
2104 : : * Verify that, in non-test mode, ./pg_tblspc doesn't contain any real
2105 : : * directories.
2106 : : *
2107 : : * Replay of database creation XLOG records for databases that were later
2108 : : * dropped can create fake directories in pg_tblspc. By the time consistency
2109 : : * is reached these directories should have been removed; here we verify
2110 : : * that this did indeed happen. This is to be called at the point where
2111 : : * consistent state is reached.
2112 : : *
2113 : : * allow_in_place_tablespaces turns the PANIC into a WARNING, which is
2114 : : * useful for testing purposes, and also allows for an escape hatch in case
2115 : : * things go south.
2116 : : */
2117 : : static void
2118 : 133 : CheckTablespaceDirectory(void)
2119 : : {
2120 : : DIR *dir;
2121 : : struct dirent *de;
2122 : :
2123 : 133 : dir = AllocateDir(PG_TBLSPC_DIR);
2124 [ + + ]: 406 : while ((de = ReadDir(dir, PG_TBLSPC_DIR)) != NULL)
2125 : : {
2126 : : char path[MAXPGPATH + sizeof(PG_TBLSPC_DIR)];
2127 : :
2128 : : /* Skip entries of non-oid names */
2129 [ + + ]: 273 : if (strspn(de->d_name, "0123456789") != strlen(de->d_name))
2130 : 266 : continue;
2131 : :
2132 : 7 : snprintf(path, sizeof(path), "%s/%s", PG_TBLSPC_DIR, de->d_name);
2133 : :
2134 [ + + ]: 7 : if (get_dirent_type(path, de, false, ERROR) != PGFILETYPE_LNK)
2135 [ + - + - ]: 4 : ereport(allow_in_place_tablespaces ? WARNING : PANIC,
2136 : : (errcode(ERRCODE_DATA_CORRUPTED),
2137 : : errmsg("unexpected directory entry \"%s\" found in %s",
2138 : : de->d_name, PG_TBLSPC_DIR),
2139 : : errdetail("All directory entries in %s/ should be symbolic links.",
2140 : : PG_TBLSPC_DIR),
2141 : : errhint("Remove those directories, or set \"allow_in_place_tablespaces\" to ON transiently to let recovery complete.")));
2142 : : }
2143 : 133 : }
2144 : :
2145 : : /*
2146 : : * Checks if recovery has reached a consistent state. When consistency is
2147 : : * reached and we have a valid starting standby snapshot, tell postmaster
2148 : : * that it can start accepting read-only connections.
2149 : : */
2150 : : static void
2151 : 2964860 : CheckRecoveryConsistency(void)
2152 : : {
2153 : : XLogRecPtr lastReplayedEndRecPtr;
2154 : : TimeLineID lastReplayedTLI;
2155 : :
2156 : : /*
2157 : : * During crash recovery, we don't reach a consistent state until we've
2158 : : * replayed all the WAL.
2159 : : */
2160 [ + + ]: 2964860 : if (!XLogRecPtrIsValid(minRecoveryPoint))
2161 : 322367 : return;
2162 : :
2163 : : Assert(InArchiveRecovery);
2164 : :
2165 : : /*
2166 : : * assume that we are called in the startup process, and hence don't need
2167 : : * a lock to read lastReplayedEndRecPtr
2168 : : */
2169 : 2642493 : lastReplayedEndRecPtr = XLogRecoveryCtl->lastReplayedEndRecPtr;
2170 : 2642493 : lastReplayedTLI = XLogRecoveryCtl->lastReplayedTLI;
2171 : :
2172 : : /*
2173 : : * Have we reached the point where our base backup was completed?
2174 : : */
2175 [ + + ]: 2642493 : if (XLogRecPtrIsValid(backupEndPoint) &&
2176 [ + + ]: 1555 : backupEndPoint <= lastReplayedEndRecPtr)
2177 : : {
2178 : 89 : XLogRecPtr saveBackupStartPoint = backupStartPoint;
2179 : 89 : XLogRecPtr saveBackupEndPoint = backupEndPoint;
2180 : :
2181 [ + + ]: 89 : elog(DEBUG1, "end of backup reached");
2182 : :
2183 : : /*
2184 : : * We have reached the end of base backup, as indicated by pg_control.
2185 : : * Update the control file accordingly.
2186 : : */
2187 : 89 : ReachedEndOfBackup(lastReplayedEndRecPtr, lastReplayedTLI);
2188 : 89 : backupStartPoint = InvalidXLogRecPtr;
2189 : 89 : backupEndPoint = InvalidXLogRecPtr;
2190 : 89 : backupEndRequired = false;
2191 : :
2192 [ + - ]: 89 : ereport(LOG,
2193 : : errmsg("completed backup recovery with redo LSN %X/%08X and end LSN %X/%08X",
2194 : : LSN_FORMAT_ARGS(saveBackupStartPoint),
2195 : : LSN_FORMAT_ARGS(saveBackupEndPoint)));
2196 : : }
2197 : :
2198 : : /*
2199 : : * Have we passed our safe starting point? Note that minRecoveryPoint is
2200 : : * known to be incorrectly set if recovering from a backup, until the
2201 : : * XLOG_BACKUP_END arrives to advise us of the correct minRecoveryPoint.
2202 : : * All we know prior to that is that we're not consistent yet.
2203 : : */
2204 [ + + + + ]: 2642493 : if (!reachedConsistency && !backupEndRequired &&
2205 [ + + ]: 7978 : minRecoveryPoint <= lastReplayedEndRecPtr)
2206 : : {
2207 : : /*
2208 : : * Check to see if the XLOG sequence contained any unresolved
2209 : : * references to uninitialized pages.
2210 : : */
2211 : 133 : XLogCheckInvalidPages();
2212 : :
2213 : : /*
2214 : : * Check that pg_tblspc doesn't contain any real directories. Replay
2215 : : * of Database/CREATE_* records may have created fictitious tablespace
2216 : : * directories that should have been removed by the time consistency
2217 : : * was reached.
2218 : : */
2219 : 133 : CheckTablespaceDirectory();
2220 : :
2221 : 133 : reachedConsistency = true;
2222 : 133 : SendPostmasterSignal(PMSIGNAL_RECOVERY_CONSISTENT);
2223 [ + - ]: 133 : ereport(LOG,
2224 : : errmsg("consistent recovery state reached at %X/%08X",
2225 : : LSN_FORMAT_ARGS(lastReplayedEndRecPtr)));
2226 : : }
2227 : :
2228 : : /*
2229 : : * Have we got a valid starting snapshot that will allow queries to be
2230 : : * run? If so, we can tell postmaster that the database is consistent now,
2231 : : * enabling connections.
2232 : : */
2233 [ + + ]: 2642493 : if (standbyState == STANDBY_SNAPSHOT_READY &&
2234 [ + + + + ]: 2642232 : !LocalHotStandbyActive &&
2235 [ + - ]: 124 : reachedConsistency &&
2236 : : IsUnderPostmaster)
2237 : : {
2238 : 124 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
2239 : 124 : XLogRecoveryCtl->SharedHotStandbyActive = true;
2240 : 124 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
2241 : :
2242 : 124 : LocalHotStandbyActive = true;
2243 : :
2244 : 124 : SendPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY);
2245 : : }
2246 : : }
2247 : :
2248 : : /*
2249 : : * Error context callback for errors occurring during rm_redo().
2250 : : */
2251 : : static void
2252 : 173 : rm_redo_error_callback(void *arg)
2253 : : {
2254 : 173 : XLogReaderState *record = (XLogReaderState *) arg;
2255 : : StringInfoData buf;
2256 : :
2257 : 173 : initStringInfo(&buf);
2258 : 173 : xlog_outdesc(&buf, record);
2259 : 173 : xlog_block_info(&buf, record);
2260 : :
2261 : : /* translator: %s is a WAL record description */
2262 : 173 : errcontext("WAL redo at %X/%08X for %s",
2263 : 173 : LSN_FORMAT_ARGS(record->ReadRecPtr),
2264 : : buf.data);
2265 : :
2266 : 173 : pfree(buf.data);
2267 : 173 : }
2268 : :
2269 : : /*
2270 : : * Returns a string describing an XLogRecord, consisting of its identity
2271 : : * optionally followed by a colon, a space, and a further description.
2272 : : */
2273 : : void
2274 : 173 : xlog_outdesc(StringInfo buf, XLogReaderState *record)
2275 : : {
2276 : 173 : RmgrData rmgr = GetRmgr(XLogRecGetRmid(record));
2277 : 173 : uint8 info = XLogRecGetInfo(record);
2278 : : const char *id;
2279 : :
2280 : 173 : appendStringInfoString(buf, rmgr.rm_name);
2281 : 173 : appendStringInfoChar(buf, '/');
2282 : :
2283 : 173 : id = rmgr.rm_identify(info);
2284 [ - + ]: 173 : if (id == NULL)
2285 : 0 : appendStringInfo(buf, "UNKNOWN (%X): ", info & ~XLR_INFO_MASK);
2286 : : else
2287 : 173 : appendStringInfo(buf, "%s: ", id);
2288 : :
2289 : 173 : rmgr.rm_desc(buf, record);
2290 : 173 : }
2291 : :
2292 : : #ifdef WAL_DEBUG
2293 : :
2294 : : static void
2295 : : xlog_outrec(StringInfo buf, XLogReaderState *record)
2296 : : {
2297 : : appendStringInfo(buf, "prev %X/%08X; xid %u",
2298 : : LSN_FORMAT_ARGS(XLogRecGetPrev(record)),
2299 : : XLogRecGetXid(record));
2300 : :
2301 : : appendStringInfo(buf, "; len %u",
2302 : : XLogRecGetDataLen(record));
2303 : :
2304 : : xlog_block_info(buf, record);
2305 : : }
2306 : : #endif /* WAL_DEBUG */
2307 : :
2308 : : /*
2309 : : * Returns a string giving information about all the blocks in an
2310 : : * XLogRecord.
2311 : : */
2312 : : static void
2313 : 173 : xlog_block_info(StringInfo buf, XLogReaderState *record)
2314 : : {
2315 : : int block_id;
2316 : :
2317 : : /* decode block references */
2318 [ + + ]: 244 : for (block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
2319 : : {
2320 : : RelFileLocator rlocator;
2321 : : ForkNumber forknum;
2322 : : BlockNumber blk;
2323 : :
2324 [ - + ]: 71 : if (!XLogRecGetBlockTagExtended(record, block_id,
2325 : : &rlocator, &forknum, &blk, NULL))
2326 : 0 : continue;
2327 : :
2328 [ + + ]: 71 : if (forknum != MAIN_FORKNUM)
2329 : 9 : appendStringInfo(buf, "; blkref #%d: rel %u/%u/%u, fork %u, blk %u",
2330 : : block_id,
2331 : : rlocator.spcOid, rlocator.dbOid,
2332 : : rlocator.relNumber,
2333 : : forknum,
2334 : : blk);
2335 : : else
2336 : 62 : appendStringInfo(buf, "; blkref #%d: rel %u/%u/%u, blk %u",
2337 : : block_id,
2338 : : rlocator.spcOid, rlocator.dbOid,
2339 : : rlocator.relNumber,
2340 : : blk);
2341 [ + + ]: 71 : if (XLogRecHasBlockImage(record, block_id))
2342 : 51 : appendStringInfoString(buf, " FPW");
2343 : : }
2344 : 173 : }
2345 : :
2346 : :
2347 : : /*
2348 : : * Check that it's OK to switch to new timeline during recovery.
2349 : : *
2350 : : * 'lsn' is the address of the shutdown checkpoint record we're about to
2351 : : * replay. (Currently, timeline can only change at a shutdown checkpoint).
2352 : : */
2353 : : static void
2354 : 14 : checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI, TimeLineID prevTLI,
2355 : : TimeLineID replayTLI)
2356 : : {
2357 : : /* Check that the record agrees on what the current (old) timeline is */
2358 [ - + ]: 14 : if (prevTLI != replayTLI)
2359 [ # # ]: 0 : ereport(PANIC,
2360 : : (errmsg("unexpected previous timeline ID %u (current timeline ID %u) in checkpoint record",
2361 : : prevTLI, replayTLI)));
2362 : :
2363 : : /*
2364 : : * The new timeline better be in the list of timelines we expect to see,
2365 : : * according to the timeline history. It should also not decrease.
2366 : : */
2367 [ + - - + ]: 14 : if (newTLI < replayTLI || !tliInHistory(newTLI, expectedTLEs))
2368 [ # # ]: 0 : ereport(PANIC,
2369 : : (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
2370 : : newTLI, replayTLI)));
2371 : :
2372 : : /*
2373 : : * If we have not yet reached min recovery point, and we're about to
2374 : : * switch to a timeline greater than the timeline of the min recovery
2375 : : * point: trouble. After switching to the new timeline, we could not
2376 : : * possibly visit the min recovery point on the correct timeline anymore.
2377 : : * This can happen if there is a newer timeline in the archive that
2378 : : * branched before the timeline the min recovery point is on, and you
2379 : : * attempt to do PITR to the new timeline.
2380 : : */
2381 [ + + ]: 14 : if (XLogRecPtrIsValid(minRecoveryPoint) &&
2382 [ + + ]: 12 : lsn < minRecoveryPoint &&
2383 [ - + ]: 1 : newTLI > minRecoveryPointTLI)
2384 [ # # ]: 0 : ereport(PANIC,
2385 : : errmsg("unexpected timeline ID %u in checkpoint record, before reaching minimum recovery point %X/%08X on timeline %u",
2386 : : newTLI,
2387 : : LSN_FORMAT_ARGS(minRecoveryPoint),
2388 : : minRecoveryPointTLI));
2389 : :
2390 : : /* Looks good */
2391 : 14 : }
2392 : :
2393 : :
2394 : : /*
2395 : : * Extract timestamp from WAL record.
2396 : : *
2397 : : * If the record contains a timestamp, returns true, and saves the timestamp
2398 : : * in *recordXtime. If the record type has no timestamp, returns false.
2399 : : * Currently, only transaction commit/abort records and restore points contain
2400 : : * timestamps.
2401 : : */
2402 : : static bool
2403 : 47920 : getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime)
2404 : : {
2405 : 47920 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2406 : 47920 : uint8 xact_info = info & XLOG_XACT_OPMASK;
2407 : 47920 : uint8 rmid = XLogRecGetRmid(record);
2408 : :
2409 [ + + + - ]: 47920 : if (rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
2410 : : {
2411 : 2 : *recordXtime = ((xl_restore_point *) XLogRecGetData(record))->rp_time;
2412 : 2 : return true;
2413 : : }
2414 [ + - + + : 47918 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_COMMIT ||
+ + ]
2415 : : xact_info == XLOG_XACT_COMMIT_PREPARED))
2416 : : {
2417 : 43882 : *recordXtime = ((xl_xact_commit *) XLogRecGetData(record))->xact_time;
2418 : 43882 : return true;
2419 : : }
2420 [ + - + + : 4036 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_ABORT ||
+ - ]
2421 : : xact_info == XLOG_XACT_ABORT_PREPARED))
2422 : : {
2423 : 4036 : *recordXtime = ((xl_xact_abort *) XLogRecGetData(record))->xact_time;
2424 : 4036 : return true;
2425 : : }
2426 : 0 : return false;
2427 : : }
2428 : :
2429 : : /*
2430 : : * Checks whether the current buffer page and backup page stored in the
2431 : : * WAL record are consistent or not. Before comparing the two pages, a
2432 : : * masking can be applied to the pages to ignore certain areas like hint bits,
2433 : : * unused space between pd_lower and pd_upper among other things. This
2434 : : * function should be called once WAL replay has been completed for a
2435 : : * given record.
2436 : : */
2437 : : static void
2438 : 2275675 : verifyBackupPageConsistency(XLogReaderState *record)
2439 : : {
2440 : 2275675 : RmgrData rmgr = GetRmgr(XLogRecGetRmid(record));
2441 : : RelFileLocator rlocator;
2442 : : ForkNumber forknum;
2443 : : BlockNumber blkno;
2444 : : int block_id;
2445 : :
2446 : : /* Records with no backup blocks have no need for consistency checks. */
2447 [ + + ]: 2275675 : if (!XLogRecHasAnyBlockRefs(record))
2448 : 75 : return;
2449 : :
2450 : : Assert((XLogRecGetInfo(record) & XLR_CHECK_CONSISTENCY) != 0);
2451 : :
2452 [ + + ]: 4723087 : for (block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
2453 : : {
2454 : : Buffer buf;
2455 : : Page page;
2456 : :
2457 [ + + ]: 2447487 : if (!XLogRecGetBlockTagExtended(record, block_id,
2458 : : &rlocator, &forknum, &blkno, NULL))
2459 : : {
2460 : : /*
2461 : : * WAL record doesn't contain a block reference with the given id.
2462 : : * Do nothing.
2463 : : */
2464 : 2090 : continue;
2465 : : }
2466 : :
2467 : : Assert(XLogRecHasBlockImage(record, block_id));
2468 : :
2469 [ + + ]: 2445397 : if (XLogRecBlockImageApply(record, block_id))
2470 : : {
2471 : : /*
2472 : : * WAL record has already applied the page, so bypass the
2473 : : * consistency check as that would result in comparing the full
2474 : : * page stored in the record with itself.
2475 : : */
2476 : 27046 : continue;
2477 : : }
2478 : :
2479 : : /*
2480 : : * Read the contents from the current buffer and store it in a
2481 : : * temporary page.
2482 : : */
2483 : 2418351 : buf = XLogReadBufferExtended(rlocator, forknum, blkno,
2484 : : RBM_NORMAL_NO_LOG,
2485 : : InvalidBuffer);
2486 [ - + ]: 2418351 : if (!BufferIsValid(buf))
2487 : 0 : continue;
2488 : :
2489 : 2418351 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
2490 : 2418351 : page = BufferGetPage(buf);
2491 : :
2492 : : /*
2493 : : * Take a copy of the local page where WAL has been applied to have a
2494 : : * comparison base before masking it...
2495 : : */
2496 : 2418351 : memcpy(replay_image_masked, page, BLCKSZ);
2497 : :
2498 : : /* No need for this page anymore now that a copy is in. */
2499 : 2418351 : UnlockReleaseBuffer(buf);
2500 : :
2501 : : /*
2502 : : * If the block LSN is already ahead of this WAL record, we can't
2503 : : * expect contents to match. This can happen if recovery is
2504 : : * restarted.
2505 : : */
2506 [ - + ]: 2418351 : if (PageGetLSN(replay_image_masked) > record->EndRecPtr)
2507 : 0 : continue;
2508 : :
2509 : : /*
2510 : : * Read the contents from the backup copy, stored in WAL record and
2511 : : * store it in a temporary page. There is no need to allocate a new
2512 : : * page here, a local buffer is fine to hold its contents and a mask
2513 : : * can be directly applied on it.
2514 : : */
2515 [ - + ]: 2418351 : if (!RestoreBlockImage(record, block_id, primary_image_masked))
2516 [ # # ]: 0 : ereport(ERROR,
2517 : : (errcode(ERRCODE_INTERNAL_ERROR),
2518 : : errmsg_internal("%s", record->errormsg_buf)));
2519 : :
2520 : : /*
2521 : : * If masking function is defined, mask both the primary and replay
2522 : : * images
2523 : : */
2524 [ + - ]: 2418351 : if (rmgr.rm_mask != NULL)
2525 : : {
2526 : 2418351 : rmgr.rm_mask(replay_image_masked, blkno);
2527 : 2418351 : rmgr.rm_mask(primary_image_masked, blkno);
2528 : : }
2529 : :
2530 : : /* Time to compare the primary and replay images. */
2531 [ - + ]: 2418351 : if (memcmp(replay_image_masked, primary_image_masked, BLCKSZ) != 0)
2532 : : {
2533 [ # # ]: 0 : elog(FATAL,
2534 : : "inconsistent page found, rel %u/%u/%u, forknum %u, blkno %u",
2535 : : rlocator.spcOid, rlocator.dbOid, rlocator.relNumber,
2536 : : forknum, blkno);
2537 : : }
2538 : : }
2539 : : }
2540 : :
2541 : : /*
2542 : : * For point-in-time recovery, this function decides whether we want to
2543 : : * stop applying the XLOG before the current record.
2544 : : *
2545 : : * Returns true if we are stopping, false otherwise. If stopping, some
2546 : : * information is saved in recoveryStopXid et al for use in annotating the
2547 : : * new timeline's history file.
2548 : : */
2549 : : static bool
2550 : 2964629 : recoveryStopsBefore(XLogReaderState *record)
2551 : : {
2552 : 2964629 : bool stopsHere = false;
2553 : : uint8 xact_info;
2554 : : bool isCommit;
2555 : 2964629 : TimestampTz recordXtime = 0;
2556 : : TransactionId recordXid;
2557 : :
2558 : : /*
2559 : : * Ignore recovery target settings when not in archive recovery (meaning
2560 : : * we are in crash recovery).
2561 : : */
2562 [ + + ]: 2964629 : if (!ArchiveRecoveryRequested)
2563 : 307813 : return false;
2564 : :
2565 : : /* Check if we should stop as soon as reaching consistency */
2566 [ - + - - ]: 2656816 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
2567 : : {
2568 [ # # ]: 0 : ereport(LOG,
2569 : : (errmsg("recovery stopping after reaching consistency")));
2570 : :
2571 : 0 : recoveryStopAfter = false;
2572 : 0 : recoveryStopXid = InvalidTransactionId;
2573 : 0 : recoveryStopLSN = InvalidXLogRecPtr;
2574 : 0 : recoveryStopTime = 0;
2575 : 0 : recoveryStopName[0] = '\0';
2576 : 0 : return true;
2577 : : }
2578 : :
2579 : : /* Check if target LSN has been reached */
2580 [ + + ]: 2656816 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
2581 [ + + ]: 8554 : !recoveryTargetInclusive &&
2582 [ + + ]: 482 : record->ReadRecPtr >= recoveryTargetLSN)
2583 : : {
2584 : 2 : recoveryStopAfter = false;
2585 : 2 : recoveryStopXid = InvalidTransactionId;
2586 : 2 : recoveryStopLSN = record->ReadRecPtr;
2587 : 2 : recoveryStopTime = 0;
2588 : 2 : recoveryStopName[0] = '\0';
2589 [ + - ]: 2 : ereport(LOG,
2590 : : errmsg("recovery stopping before WAL location (LSN) \"%X/%08X\"",
2591 : : LSN_FORMAT_ARGS(recoveryStopLSN)));
2592 : 2 : return true;
2593 : : }
2594 : :
2595 : : /* Otherwise we only consider stopping before COMMIT or ABORT records. */
2596 [ + + ]: 2656814 : if (XLogRecGetRmid(record) != RM_XACT_ID)
2597 : 2632550 : return false;
2598 : :
2599 : 24264 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
2600 : :
2601 [ + + ]: 24264 : if (xact_info == XLOG_XACT_COMMIT)
2602 : : {
2603 : 21901 : isCommit = true;
2604 : 21901 : recordXid = XLogRecGetXid(record);
2605 : : }
2606 [ + + ]: 2363 : else if (xact_info == XLOG_XACT_COMMIT_PREPARED)
2607 : : {
2608 : 26 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
2609 : : xl_xact_parsed_commit parsed;
2610 : :
2611 : 26 : isCommit = true;
2612 : 26 : ParseCommitRecord(XLogRecGetInfo(record),
2613 : : xlrec,
2614 : : &parsed);
2615 : 26 : recordXid = parsed.twophase_xid;
2616 : : }
2617 [ + + ]: 2337 : else if (xact_info == XLOG_XACT_ABORT)
2618 : : {
2619 : 2003 : isCommit = false;
2620 : 2003 : recordXid = XLogRecGetXid(record);
2621 : : }
2622 [ + + ]: 334 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
2623 : : {
2624 : 15 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
2625 : : xl_xact_parsed_abort parsed;
2626 : :
2627 : 15 : isCommit = false;
2628 : 15 : ParseAbortRecord(XLogRecGetInfo(record),
2629 : : xlrec,
2630 : : &parsed);
2631 : 15 : recordXid = parsed.twophase_xid;
2632 : : }
2633 : : else
2634 : 319 : return false;
2635 : :
2636 [ - + - - ]: 23945 : if (recoveryTarget == RECOVERY_TARGET_XID && !recoveryTargetInclusive)
2637 : : {
2638 : : /*
2639 : : * There can be only one transaction end record with this exact
2640 : : * transactionid
2641 : : *
2642 : : * when testing for an xid, we MUST test for equality only, since
2643 : : * transactions are numbered in the order they start, not the order
2644 : : * they complete. A higher numbered xid will complete before you about
2645 : : * 50% of the time...
2646 : : */
2647 : 0 : stopsHere = (recordXid == recoveryTargetXid);
2648 : : }
2649 : :
2650 : : /*
2651 : : * Note: we must fetch recordXtime regardless of recoveryTarget setting.
2652 : : * We don't expect getRecordTimestamp ever to fail, since we already know
2653 : : * this is a commit or abort record; but test its result anyway.
2654 : : */
2655 [ + - ]: 23945 : if (getRecordTimestamp(record, &recordXtime) &&
2656 [ - + ]: 23945 : recoveryTarget == RECOVERY_TARGET_TIME)
2657 : : {
2658 : : /*
2659 : : * There can be many transactions that share the same commit time, so
2660 : : * we stop after the last one, if we are inclusive, or stop at the
2661 : : * first one if we are exclusive
2662 : : */
2663 [ # # ]: 0 : if (recoveryTargetInclusive)
2664 : 0 : stopsHere = (recordXtime > recoveryTargetTime);
2665 : : else
2666 : 0 : stopsHere = (recordXtime >= recoveryTargetTime);
2667 : : }
2668 : :
2669 [ - + ]: 23945 : if (stopsHere)
2670 : : {
2671 : 0 : recoveryStopAfter = false;
2672 : 0 : recoveryStopXid = recordXid;
2673 : 0 : recoveryStopTime = recordXtime;
2674 : 0 : recoveryStopLSN = InvalidXLogRecPtr;
2675 : 0 : recoveryStopName[0] = '\0';
2676 : :
2677 [ # # ]: 0 : if (isCommit)
2678 : : {
2679 [ # # ]: 0 : ereport(LOG,
2680 : : (errmsg("recovery stopping before commit of transaction %u, time %s",
2681 : : recoveryStopXid,
2682 : : timestamptz_to_str(recoveryStopTime))));
2683 : : }
2684 : : else
2685 : : {
2686 [ # # ]: 0 : ereport(LOG,
2687 : : (errmsg("recovery stopping before abort of transaction %u, time %s",
2688 : : recoveryStopXid,
2689 : : timestamptz_to_str(recoveryStopTime))));
2690 : : }
2691 : : }
2692 : :
2693 : 23945 : return stopsHere;
2694 : : }
2695 : :
2696 : : /*
2697 : : * Same as recoveryStopsBefore, but called after applying the record.
2698 : : *
2699 : : * We also track the timestamp of the latest applied COMMIT/ABORT
2700 : : * record in XLogRecoveryCtl->recoveryLastXTime.
2701 : : */
2702 : : static bool
2703 : 2964625 : recoveryStopsAfter(XLogReaderState *record)
2704 : : {
2705 : : uint8 info;
2706 : : uint8 xact_info;
2707 : : uint8 rmid;
2708 : 2964625 : TimestampTz recordXtime = 0;
2709 : :
2710 : : /*
2711 : : * Ignore recovery target settings when not in archive recovery (meaning
2712 : : * we are in crash recovery).
2713 : : */
2714 [ + + ]: 2964625 : if (!ArchiveRecoveryRequested)
2715 : 307813 : return false;
2716 : :
2717 : 2656812 : info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2718 : 2656812 : rmid = XLogRecGetRmid(record);
2719 : :
2720 : : /*
2721 : : * There can be many restore points that share the same name; we stop at
2722 : : * the first one.
2723 : : */
2724 [ + + + + ]: 2656812 : if (recoveryTarget == RECOVERY_TARGET_NAME &&
2725 [ + + ]: 20 : rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
2726 : : {
2727 : : xl_restore_point *recordRestorePointData;
2728 : :
2729 : 3 : recordRestorePointData = (xl_restore_point *) XLogRecGetData(record);
2730 : :
2731 [ + + ]: 3 : if (strcmp(recordRestorePointData->rp_name, recoveryTargetName) == 0)
2732 : : {
2733 : 2 : recoveryStopAfter = true;
2734 : 2 : recoveryStopXid = InvalidTransactionId;
2735 : 2 : recoveryStopLSN = InvalidXLogRecPtr;
2736 : 2 : (void) getRecordTimestamp(record, &recoveryStopTime);
2737 : 2 : strlcpy(recoveryStopName, recordRestorePointData->rp_name, MAXFNAMELEN);
2738 : :
2739 [ + - ]: 2 : ereport(LOG,
2740 : : (errmsg("recovery stopping at restore point \"%s\", time %s",
2741 : : recoveryStopName,
2742 : : timestamptz_to_str(recoveryStopTime))));
2743 : 2 : return true;
2744 : : }
2745 : : }
2746 : :
2747 : : /* Check if the target LSN has been reached */
2748 [ + + + + ]: 2656810 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
2749 : 8072 : recoveryTargetInclusive &&
2750 [ + + ]: 8072 : record->ReadRecPtr >= recoveryTargetLSN)
2751 : : {
2752 : 3 : recoveryStopAfter = true;
2753 : 3 : recoveryStopXid = InvalidTransactionId;
2754 : 3 : recoveryStopLSN = record->ReadRecPtr;
2755 : 3 : recoveryStopTime = 0;
2756 : 3 : recoveryStopName[0] = '\0';
2757 [ + - ]: 3 : ereport(LOG,
2758 : : errmsg("recovery stopping after WAL location (LSN) \"%X/%08X\"",
2759 : : LSN_FORMAT_ARGS(recoveryStopLSN)));
2760 : 3 : return true;
2761 : : }
2762 : :
2763 [ + + ]: 2656807 : if (rmid != RM_XACT_ID)
2764 : 2632545 : return false;
2765 : :
2766 : 24262 : xact_info = info & XLOG_XACT_OPMASK;
2767 : :
2768 [ + + + + ]: 24262 : if (xact_info == XLOG_XACT_COMMIT ||
2769 [ + + ]: 2337 : xact_info == XLOG_XACT_COMMIT_PREPARED ||
2770 [ + + ]: 334 : xact_info == XLOG_XACT_ABORT ||
2771 : : xact_info == XLOG_XACT_ABORT_PREPARED)
2772 : : {
2773 : : TransactionId recordXid;
2774 : :
2775 : : /* Update the last applied transaction timestamp */
2776 [ + - ]: 23943 : if (getRecordTimestamp(record, &recordXtime))
2777 : 23943 : SetLatestXTime(recordXtime);
2778 : :
2779 : : /* Extract the XID of the committed/aborted transaction */
2780 [ + + ]: 23943 : if (xact_info == XLOG_XACT_COMMIT_PREPARED)
2781 : : {
2782 : 26 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
2783 : : xl_xact_parsed_commit parsed;
2784 : :
2785 : 26 : ParseCommitRecord(XLogRecGetInfo(record),
2786 : : xlrec,
2787 : : &parsed);
2788 : 26 : recordXid = parsed.twophase_xid;
2789 : : }
2790 [ + + ]: 23917 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
2791 : : {
2792 : 15 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
2793 : : xl_xact_parsed_abort parsed;
2794 : :
2795 : 15 : ParseAbortRecord(XLogRecGetInfo(record),
2796 : : xlrec,
2797 : : &parsed);
2798 : 15 : recordXid = parsed.twophase_xid;
2799 : : }
2800 : : else
2801 : 23902 : recordXid = XLogRecGetXid(record);
2802 : :
2803 : : /*
2804 : : * There can be only one transaction end record with this exact
2805 : : * transactionid
2806 : : *
2807 : : * when testing for an xid, we MUST test for equality only, since
2808 : : * transactions are numbered in the order they start, not the order
2809 : : * they complete. A higher numbered xid will complete before you about
2810 : : * 50% of the time...
2811 : : */
2812 [ - + - - ]: 23943 : if (recoveryTarget == RECOVERY_TARGET_XID && recoveryTargetInclusive &&
2813 [ # # ]: 0 : recordXid == recoveryTargetXid)
2814 : : {
2815 : 0 : recoveryStopAfter = true;
2816 : 0 : recoveryStopXid = recordXid;
2817 : 0 : recoveryStopTime = recordXtime;
2818 : 0 : recoveryStopLSN = InvalidXLogRecPtr;
2819 : 0 : recoveryStopName[0] = '\0';
2820 : :
2821 [ # # # # ]: 0 : if (xact_info == XLOG_XACT_COMMIT ||
2822 : : xact_info == XLOG_XACT_COMMIT_PREPARED)
2823 : : {
2824 [ # # ]: 0 : ereport(LOG,
2825 : : (errmsg("recovery stopping after commit of transaction %u, time %s",
2826 : : recoveryStopXid,
2827 : : timestamptz_to_str(recoveryStopTime))));
2828 : : }
2829 [ # # # # ]: 0 : else if (xact_info == XLOG_XACT_ABORT ||
2830 : : xact_info == XLOG_XACT_ABORT_PREPARED)
2831 : : {
2832 [ # # ]: 0 : ereport(LOG,
2833 : : (errmsg("recovery stopping after abort of transaction %u, time %s",
2834 : : recoveryStopXid,
2835 : : timestamptz_to_str(recoveryStopTime))));
2836 : : }
2837 : 0 : return true;
2838 : : }
2839 : : }
2840 : :
2841 : : /* Check if we should stop as soon as reaching consistency */
2842 [ - + - - ]: 24262 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
2843 : : {
2844 [ # # ]: 0 : ereport(LOG,
2845 : : (errmsg("recovery stopping after reaching consistency")));
2846 : :
2847 : 0 : recoveryStopAfter = true;
2848 : 0 : recoveryStopXid = InvalidTransactionId;
2849 : 0 : recoveryStopTime = 0;
2850 : 0 : recoveryStopLSN = InvalidXLogRecPtr;
2851 : 0 : recoveryStopName[0] = '\0';
2852 : 0 : return true;
2853 : : }
2854 : :
2855 : 24262 : return false;
2856 : : }
2857 : :
2858 : : /*
2859 : : * Create a comment for the history file to explain why and where
2860 : : * timeline changed.
2861 : : */
2862 : : static char *
2863 : 1019 : getRecoveryStopReason(void)
2864 : : {
2865 : : char reason[200];
2866 : :
2867 [ - + ]: 1019 : if (recoveryTarget == RECOVERY_TARGET_XID)
2868 : 0 : snprintf(reason, sizeof(reason),
2869 : : "%s transaction %u",
2870 [ # # ]: 0 : recoveryStopAfter ? "after" : "before",
2871 : : recoveryStopXid);
2872 [ - + ]: 1019 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
2873 : 0 : snprintf(reason, sizeof(reason),
2874 : : "%s %s\n",
2875 [ # # ]: 0 : recoveryStopAfter ? "after" : "before",
2876 : : timestamptz_to_str(recoveryStopTime));
2877 [ + + ]: 1019 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
2878 : 6 : snprintf(reason, sizeof(reason),
2879 : : "%s LSN %X/%08X\n",
2880 : 6 : recoveryStopAfter ? "after" : "before",
2881 [ + + ]: 6 : LSN_FORMAT_ARGS(recoveryStopLSN));
2882 [ + + ]: 1013 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
2883 : 3 : snprintf(reason, sizeof(reason),
2884 : : "at restore point \"%s\"",
2885 : : recoveryStopName);
2886 [ - + ]: 1010 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
2887 : 0 : snprintf(reason, sizeof(reason), "reached consistency");
2888 : : else
2889 : 1010 : snprintf(reason, sizeof(reason), "no recovery target specified");
2890 : :
2891 : 1019 : return pstrdup(reason);
2892 : : }
2893 : :
2894 : : /*
2895 : : * Wait until shared recoveryPauseState is set to RECOVERY_NOT_PAUSED.
2896 : : *
2897 : : * endOfRecovery is true if the recovery target is reached and
2898 : : * the paused state starts at the end of recovery because of
2899 : : * recovery_target_action=pause, and false otherwise.
2900 : : */
2901 : : static void
2902 : 9 : recoveryPausesHere(bool endOfRecovery)
2903 : : {
2904 : : /* Don't pause unless users can connect! */
2905 [ - + ]: 9 : if (!LocalHotStandbyActive)
2906 : 0 : return;
2907 : :
2908 : : /* Don't pause after standby promotion has been triggered */
2909 [ - + ]: 9 : if (LocalPromoteIsTriggered)
2910 : 0 : return;
2911 : :
2912 [ + + ]: 9 : if (endOfRecovery)
2913 [ + - ]: 1 : ereport(LOG,
2914 : : (errmsg("pausing at the end of recovery"),
2915 : : errhint("Execute pg_wal_replay_resume() to promote.")));
2916 : : else
2917 [ + - ]: 8 : ereport(LOG,
2918 : : (errmsg("recovery has paused"),
2919 : : errhint("Execute pg_wal_replay_resume() to continue.")));
2920 : :
2921 : : /* loop until recoveryPauseState is set to RECOVERY_NOT_PAUSED */
2922 [ + + ]: 28 : while (GetRecoveryPauseState() != RECOVERY_NOT_PAUSED)
2923 : : {
2924 : 21 : ProcessStartupProcInterrupts();
2925 [ + + ]: 21 : if (CheckForStandbyTrigger())
2926 : 2 : return;
2927 : :
2928 : : /*
2929 : : * If recovery pause is requested then set it paused. While we are in
2930 : : * the loop, user might resume and pause again so set this every time.
2931 : : */
2932 : 19 : ConfirmRecoveryPaused();
2933 : :
2934 : : /*
2935 : : * We wait on a condition variable that will wake us as soon as the
2936 : : * pause ends, but we use a timeout so we can check the above exit
2937 : : * condition periodically too.
2938 : : */
2939 : 19 : ConditionVariableTimedSleep(&XLogRecoveryCtl->recoveryNotPausedCV, 1000,
2940 : : WAIT_EVENT_RECOVERY_PAUSE);
2941 : : }
2942 : 7 : ConditionVariableCancelSleep();
2943 : : }
2944 : :
2945 : : /*
2946 : : * When recovery_min_apply_delay is set, we wait long enough to make sure
2947 : : * certain record types are applied at least that interval behind the primary.
2948 : : *
2949 : : * Returns true if we waited.
2950 : : *
2951 : : * Note that the delay is calculated between the WAL record log time and
2952 : : * the current time on standby. We would prefer to keep track of when this
2953 : : * standby received each WAL record, which would allow a more consistent
2954 : : * approach and one not affected by time synchronisation issues, but that
2955 : : * is significantly more effort and complexity for little actual gain in
2956 : : * usability.
2957 : : */
2958 : : static bool
2959 : 2964627 : recoveryApplyDelay(XLogReaderState *record)
2960 : : {
2961 : : uint8 xact_info;
2962 : : TimestampTz xtime;
2963 : : TimestampTz delayUntil;
2964 : : long msecs;
2965 : :
2966 : : /* nothing to do if no delay configured */
2967 [ + + ]: 2964627 : if (recovery_min_apply_delay <= 0)
2968 : 2964473 : return false;
2969 : :
2970 : : /* no delay is applied on a database not yet consistent */
2971 [ + + ]: 154 : if (!reachedConsistency)
2972 : 4 : return false;
2973 : :
2974 : : /* nothing to do if crash recovery is requested */
2975 [ - + ]: 150 : if (!ArchiveRecoveryRequested)
2976 : 0 : return false;
2977 : :
2978 : : /*
2979 : : * Is it a COMMIT record?
2980 : : *
2981 : : * We deliberately choose not to delay aborts since they have no effect on
2982 : : * MVCC. We already allow replay of records that don't have a timestamp,
2983 : : * so there is already opportunity for issues caused by early conflicts on
2984 : : * standbys.
2985 : : */
2986 [ + + ]: 150 : if (XLogRecGetRmid(record) != RM_XACT_ID)
2987 : 120 : return false;
2988 : :
2989 : 30 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
2990 : :
2991 [ - + - - ]: 30 : if (xact_info != XLOG_XACT_COMMIT &&
2992 : : xact_info != XLOG_XACT_COMMIT_PREPARED)
2993 : 0 : return false;
2994 : :
2995 [ - + ]: 30 : if (!getRecordTimestamp(record, &xtime))
2996 : 0 : return false;
2997 : :
2998 : 30 : delayUntil = TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
2999 : :
3000 : : /*
3001 : : * Exit without arming the latch if it's already past time to apply this
3002 : : * record
3003 : : */
3004 : 30 : msecs = TimestampDifferenceMilliseconds(GetCurrentTimestamp(), delayUntil);
3005 [ + + ]: 30 : if (msecs <= 0)
3006 : 9 : return false;
3007 : :
3008 : : while (true)
3009 : : {
3010 : 82 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3011 : :
3012 : : /* This might change recovery_min_apply_delay. */
3013 : 82 : ProcessStartupProcInterrupts();
3014 : :
3015 [ - + ]: 82 : if (CheckForStandbyTrigger())
3016 : 0 : break;
3017 : :
3018 : : /*
3019 : : * Recalculate delayUntil as recovery_min_apply_delay could have
3020 : : * changed while waiting in this loop.
3021 : : */
3022 : 82 : delayUntil = TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
3023 : :
3024 : : /*
3025 : : * Wait for difference between GetCurrentTimestamp() and delayUntil.
3026 : : */
3027 : 82 : msecs = TimestampDifferenceMilliseconds(GetCurrentTimestamp(),
3028 : : delayUntil);
3029 : :
3030 [ + + ]: 82 : if (msecs <= 0)
3031 : 21 : break;
3032 : :
3033 [ - + ]: 61 : elog(DEBUG2, "recovery apply delay %ld milliseconds", msecs);
3034 : :
3035 : 61 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3036 : : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
3037 : : msecs,
3038 : : WAIT_EVENT_RECOVERY_APPLY_DELAY);
3039 : : }
3040 : 21 : return true;
3041 : : }
3042 : :
3043 : : /*
3044 : : * Get the current state of the recovery pause request.
3045 : : */
3046 : : RecoveryPauseState
3047 : 37 : GetRecoveryPauseState(void)
3048 : : {
3049 : : RecoveryPauseState state;
3050 : :
3051 : 37 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3052 : 37 : state = XLogRecoveryCtl->recoveryPauseState;
3053 : 37 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3054 : :
3055 : 37 : return state;
3056 : : }
3057 : :
3058 : : /*
3059 : : * Set the recovery pause state.
3060 : : *
3061 : : * If recovery pause is requested then sets the recovery pause state to
3062 : : * 'pause requested' if it is not already 'paused'. Otherwise, sets it
3063 : : * to 'not paused' to resume the recovery. The recovery pause will be
3064 : : * confirmed by the ConfirmRecoveryPaused.
3065 : : */
3066 : : void
3067 : 69 : SetRecoveryPause(bool recoveryPause)
3068 : : {
3069 : 69 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3070 : :
3071 [ + + ]: 69 : if (!recoveryPause)
3072 : 60 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_NOT_PAUSED;
3073 [ + - ]: 9 : else if (XLogRecoveryCtl->recoveryPauseState == RECOVERY_NOT_PAUSED)
3074 : 9 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_PAUSE_REQUESTED;
3075 : :
3076 : 69 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3077 : :
3078 [ + + ]: 69 : if (!recoveryPause)
3079 : 60 : ConditionVariableBroadcast(&XLogRecoveryCtl->recoveryNotPausedCV);
3080 : 69 : }
3081 : :
3082 : : /*
3083 : : * Confirm the recovery pause by setting the recovery pause state to
3084 : : * RECOVERY_PAUSED.
3085 : : */
3086 : : static void
3087 : 19 : ConfirmRecoveryPaused(void)
3088 : : {
3089 : : /* If recovery pause is requested then set it paused */
3090 : 19 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
3091 [ + + ]: 19 : if (XLogRecoveryCtl->recoveryPauseState == RECOVERY_PAUSE_REQUESTED)
3092 : 9 : XLogRecoveryCtl->recoveryPauseState = RECOVERY_PAUSED;
3093 : 19 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
3094 : 19 : }
3095 : :
3096 : :
3097 : : /*
3098 : : * Attempt to read the next XLOG record.
3099 : : *
3100 : : * Before first call, the reader needs to be positioned to the first record
3101 : : * by calling XLogPrefetcherBeginRead().
3102 : : *
3103 : : * If no valid record is available, returns NULL, or fails if emode is PANIC.
3104 : : * (emode must be either PANIC, LOG). In standby mode, retries until a valid
3105 : : * record is available.
3106 : : */
3107 : : static XLogRecord *
3108 : 2967096 : ReadRecord(XLogPrefetcher *xlogprefetcher, int emode,
3109 : : bool fetching_ckpt, TimeLineID replayTLI)
3110 : : {
3111 : : XLogRecord *record;
3112 : 2967096 : XLogReaderState *xlogreader = XLogPrefetcherGetReader(xlogprefetcher);
3113 : 2967096 : XLogPageReadPrivate *private = (XLogPageReadPrivate *) xlogreader->private_data;
3114 : :
3115 : : Assert(AmStartupProcess() || !IsUnderPostmaster);
3116 : :
3117 : : /* Pass through parameters to XLogPageRead */
3118 : 2967096 : private->fetching_ckpt = fetching_ckpt;
3119 : 2967096 : private->emode = emode;
3120 : 2967096 : private->randAccess = !XLogRecPtrIsValid(xlogreader->ReadRecPtr);
3121 : 2967096 : private->replayTLI = replayTLI;
3122 : :
3123 : : /* This is the first attempt to read this page. */
3124 : 2967096 : lastSourceFailed = false;
3125 : :
3126 : : for (;;)
3127 : 168 : {
3128 : : char *errormsg;
3129 : :
3130 : 2967264 : record = XLogPrefetcherReadRecord(xlogprefetcher, &errormsg);
3131 [ + + ]: 2967201 : if (record == NULL)
3132 : : {
3133 : : /*
3134 : : * When we find that WAL ends in an incomplete record, keep track
3135 : : * of that record. After recovery is done, we'll write a record
3136 : : * to indicate to downstream WAL readers that that portion is to
3137 : : * be ignored.
3138 : : *
3139 : : * However, when ArchiveRecoveryRequested = true, we're going to
3140 : : * switch to a new timeline at the end of recovery. We will only
3141 : : * copy WAL over to the new timeline up to the end of the last
3142 : : * complete record, so if we did this, we would later create an
3143 : : * overwrite contrecord in the wrong place, breaking everything.
3144 : : */
3145 [ + + ]: 332 : if (!ArchiveRecoveryRequested &&
3146 [ + + ]: 110 : XLogRecPtrIsValid(xlogreader->abortedRecPtr))
3147 : : {
3148 : 11 : abortedRecPtr = xlogreader->abortedRecPtr;
3149 : 11 : missingContrecPtr = xlogreader->missingContrecPtr;
3150 : : }
3151 : :
3152 [ + + ]: 332 : if (readFile >= 0)
3153 : : {
3154 : 305 : close(readFile);
3155 : 305 : readFile = -1;
3156 : : }
3157 : :
3158 : : /*
3159 : : * We only end up here without a message when XLogPageRead()
3160 : : * failed - in that case we already logged something. In
3161 : : * StandbyMode that only happens if we have been triggered, so we
3162 : : * shouldn't loop anymore in that case.
3163 : : */
3164 [ + + ]: 332 : if (errormsg)
3165 [ + + ]: 305 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3166 : : (errmsg_internal("%s", errormsg) /* already translated */ ));
3167 : : }
3168 : :
3169 : : /*
3170 : : * Check page TLI is one of the expected values.
3171 : : */
3172 [ - + ]: 2966869 : else if (!tliInHistory(xlogreader->latestPageTLI, expectedTLEs))
3173 : : {
3174 : : char fname[MAXFNAMELEN];
3175 : : XLogSegNo segno;
3176 : : int32 offset;
3177 : :
3178 : 0 : XLByteToSeg(xlogreader->latestPagePtr, segno, wal_segment_size);
3179 : 0 : offset = XLogSegmentOffset(xlogreader->latestPagePtr,
3180 : : wal_segment_size);
3181 : 0 : XLogFileName(fname, xlogreader->seg.ws_tli, segno,
3182 : : wal_segment_size);
3183 [ # # ]: 0 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3184 : : errmsg("unexpected timeline ID %u in WAL segment %s, LSN %X/%08X, offset %u",
3185 : : xlogreader->latestPageTLI,
3186 : : fname,
3187 : : LSN_FORMAT_ARGS(xlogreader->latestPagePtr),
3188 : : offset));
3189 : 0 : record = NULL;
3190 : : }
3191 : :
3192 [ + + ]: 2967201 : if (record)
3193 : : {
3194 : : /* Great, got a record */
3195 : 2967033 : return record;
3196 : : }
3197 : : else
3198 : : {
3199 : : /* No valid record available from this source */
3200 : 332 : lastSourceFailed = true;
3201 : :
3202 : : /*
3203 : : * If archive recovery was requested, but we were still doing
3204 : : * crash recovery, switch to archive recovery and retry using the
3205 : : * offline archive. We have now replayed all the valid WAL in
3206 : : * pg_wal, so we are presumably now consistent.
3207 : : *
3208 : : * We require that there's at least some valid WAL present in
3209 : : * pg_wal, however (!fetching_ckpt). We could recover using the
3210 : : * WAL from the archive, even if pg_wal is completely empty, but
3211 : : * we'd have no idea how far we'd have to replay to reach
3212 : : * consistency. So err on the safe side and give up.
3213 : : */
3214 [ + + + + ]: 332 : if (!InArchiveRecovery && ArchiveRecoveryRequested &&
3215 [ + - ]: 1 : !fetching_ckpt)
3216 : : {
3217 [ - + ]: 1 : ereport(DEBUG1,
3218 : : (errmsg_internal("reached end of WAL in pg_wal, entering archive recovery")));
3219 : 1 : InArchiveRecovery = true;
3220 [ + - ]: 1 : if (StandbyModeRequested)
3221 : 1 : EnableStandbyMode();
3222 : :
3223 : 1 : SwitchIntoArchiveRecovery(xlogreader->EndRecPtr, replayTLI);
3224 : 1 : minRecoveryPoint = xlogreader->EndRecPtr;
3225 : 1 : minRecoveryPointTLI = replayTLI;
3226 : :
3227 : 1 : CheckRecoveryConsistency();
3228 : :
3229 : : /*
3230 : : * Before we retry, reset lastSourceFailed and currentSource
3231 : : * so that we will check the archive next.
3232 : : */
3233 : 1 : lastSourceFailed = false;
3234 : 1 : currentSource = XLOG_FROM_ANY;
3235 : :
3236 : 168 : continue;
3237 : : }
3238 : :
3239 : : /* In standby mode, loop back to retry. Otherwise, give up. */
3240 [ + + + + ]: 331 : if (StandbyMode && !CheckForStandbyTrigger())
3241 : 167 : continue;
3242 : : else
3243 : 164 : return NULL;
3244 : : }
3245 : : }
3246 : : }
3247 : :
3248 : : /*
3249 : : * Read the XLOG page containing targetPagePtr into readBuf (if not read
3250 : : * already). Returns number of bytes read, if the page is read successfully,
3251 : : * or XLREAD_FAIL in case of errors. When errors occur, they are ereport'ed,
3252 : : * but only if they have not been previously reported.
3253 : : *
3254 : : * See XLogReaderRoutine.page_read for more details.
3255 : : *
3256 : : * While prefetching, xlogreader->nonblocking may be set. In that case,
3257 : : * returns XLREAD_WOULDBLOCK if we'd otherwise have to wait for more WAL.
3258 : : *
3259 : : * This is responsible for restoring files from archive as needed, as well
3260 : : * as for waiting for the requested WAL record to arrive in standby mode.
3261 : : *
3262 : : * xlogreader->private_data->emode specifies the log level used for reporting
3263 : : * "file not found" or "end of WAL" situations in archive recovery, or in
3264 : : * standby mode when promotion is triggered. If set to WARNING or below,
3265 : : * XLogPageRead() returns XLREAD_FAIL in those situations, on higher log
3266 : : * levels the ereport() won't return.
3267 : : *
3268 : : * In standby mode, if after a successful return of XLogPageRead() the
3269 : : * caller finds the record it's interested in to be broken, it should
3270 : : * ereport the error with the level determined by
3271 : : * emode_for_corrupt_record(), and then set lastSourceFailed
3272 : : * and call XLogPageRead() again with the same arguments. This lets
3273 : : * XLogPageRead() to try fetching the record from another source, or to
3274 : : * sleep and retry.
3275 : : */
3276 : : static int
3277 : 1519693 : XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr, int reqLen,
3278 : : XLogRecPtr targetRecPtr, char *readBuf)
3279 : : {
3280 : 1519693 : XLogPageReadPrivate *private =
3281 : : (XLogPageReadPrivate *) xlogreader->private_data;
3282 : 1519693 : int emode = private->emode;
3283 : : uint32 targetPageOff;
3284 : : XLogSegNo targetSegNo PG_USED_FOR_ASSERTS_ONLY;
3285 : : int r;
3286 : : instr_time io_start;
3287 : :
3288 : : Assert(AmStartupProcess() || !IsUnderPostmaster);
3289 : :
3290 : 1519693 : XLByteToSeg(targetPagePtr, targetSegNo, wal_segment_size);
3291 : 1519693 : targetPageOff = XLogSegmentOffset(targetPagePtr, wal_segment_size);
3292 : :
3293 : : /*
3294 : : * See if we need to switch to a new segment because the requested record
3295 : : * is not in the currently open one.
3296 : : */
3297 [ + + ]: 1519693 : if (readFile >= 0 &&
3298 [ + + ]: 1517855 : !XLByteInSeg(targetPagePtr, readSegNo, wal_segment_size))
3299 : : {
3300 : : /*
3301 : : * Request a restartpoint if we've replayed too much xlog since the
3302 : : * last one.
3303 : : */
3304 [ + + + - ]: 1475 : if (ArchiveRecoveryRequested && IsUnderPostmaster)
3305 : : {
3306 [ + + ]: 1458 : if (XLogCheckpointNeeded(readSegNo))
3307 : : {
3308 : 1315 : (void) GetRedoRecPtr();
3309 [ + + ]: 1315 : if (XLogCheckpointNeeded(readSegNo))
3310 : 1305 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
3311 : : }
3312 : : }
3313 : :
3314 : 1475 : close(readFile);
3315 : 1475 : readFile = -1;
3316 : 1475 : readSource = XLOG_FROM_ANY;
3317 : : }
3318 : :
3319 : 1519693 : XLByteToSeg(targetPagePtr, readSegNo, wal_segment_size);
3320 : :
3321 : 1519697 : retry:
3322 : : /* See if we need to retrieve more data */
3323 [ + + ]: 1519697 : if (readFile < 0 ||
3324 [ + + ]: 1516380 : (readSource == XLOG_FROM_STREAM &&
3325 [ + + ]: 1503268 : flushedUpto < targetPagePtr + reqLen))
3326 : : {
3327 [ + + ]: 29234 : if (readFile >= 0 &&
3328 [ + + ]: 25917 : xlogreader->nonblocking &&
3329 [ + - ]: 12807 : readSource == XLOG_FROM_STREAM &&
3330 [ + - ]: 12807 : flushedUpto < targetPagePtr + reqLen)
3331 : 12807 : return XLREAD_WOULDBLOCK;
3332 : :
3333 [ + + + - ]: 16364 : switch (WaitForWALToBecomeAvailable(targetPagePtr + reqLen,
3334 : 16427 : private->randAccess,
3335 : 16427 : private->fetching_ckpt,
3336 : : targetRecPtr,
3337 : : private->replayTLI,
3338 : : xlogreader->EndRecPtr,
3339 : 16427 : xlogreader->nonblocking))
3340 : : {
3341 : 395 : case XLREAD_WOULDBLOCK:
3342 : 395 : return XLREAD_WOULDBLOCK;
3343 : 52 : case XLREAD_FAIL:
3344 [ - + ]: 52 : if (readFile >= 0)
3345 : 0 : close(readFile);
3346 : 52 : readFile = -1;
3347 : 52 : readLen = 0;
3348 : 52 : readSource = XLOG_FROM_ANY;
3349 : 52 : return XLREAD_FAIL;
3350 : 15917 : case XLREAD_SUCCESS:
3351 : 15917 : break;
3352 : : }
3353 : : }
3354 : :
3355 : : /*
3356 : : * At this point, we have the right segment open and if we're streaming we
3357 : : * know the requested record is in it.
3358 : : */
3359 : : Assert(readFile != -1);
3360 : :
3361 : : /*
3362 : : * If the current segment is being streamed from the primary, calculate
3363 : : * how much of the current page we have received already. We know the
3364 : : * requested record has been received, but this is for the benefit of
3365 : : * future calls, to allow quick exit at the top of this function.
3366 : : */
3367 [ + + ]: 1506380 : if (readSource == XLOG_FROM_STREAM)
3368 : : {
3369 [ + + ]: 1491494 : if (((targetPagePtr) / XLOG_BLCKSZ) != (flushedUpto / XLOG_BLCKSZ))
3370 : 1482777 : readLen = XLOG_BLCKSZ;
3371 : : else
3372 : 8717 : readLen = XLogSegmentOffset(flushedUpto, wal_segment_size) -
3373 : : targetPageOff;
3374 : : }
3375 : : else
3376 : 14886 : readLen = XLOG_BLCKSZ;
3377 : :
3378 : : /* Read the requested page */
3379 : 1506380 : readOff = targetPageOff;
3380 : :
3381 : : /* Measure I/O timing when reading segment */
3382 : 1506380 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3383 : :
3384 : 1506380 : pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
3385 : 1506380 : r = pg_pread(readFile, readBuf, XLOG_BLCKSZ, (pgoff_t) readOff);
3386 [ - + ]: 1506380 : if (r != XLOG_BLCKSZ)
3387 : : {
3388 : : char fname[MAXFNAMELEN];
3389 : 0 : int save_errno = errno;
3390 : :
3391 : 0 : pgstat_report_wait_end();
3392 : :
3393 : : /* Count I/O stats only for successful short reads */
3394 [ # # ]: 0 : if (r > 0)
3395 : 0 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_READ,
3396 : : io_start, 1, r);
3397 : :
3398 : 0 : XLogFileName(fname, curFileTLI, readSegNo, wal_segment_size);
3399 [ # # ]: 0 : if (r < 0)
3400 : : {
3401 : 0 : errno = save_errno;
3402 [ # # ]: 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
3403 : : (errcode_for_file_access(),
3404 : : errmsg("could not read from WAL segment %s, LSN %X/%08X, offset %u: %m",
3405 : : fname, LSN_FORMAT_ARGS(targetPagePtr),
3406 : : readOff)));
3407 : : }
3408 : : else
3409 [ # # ]: 0 : ereport(emode_for_corrupt_record(emode, targetPagePtr + reqLen),
3410 : : (errcode(ERRCODE_DATA_CORRUPTED),
3411 : : errmsg("could not read from WAL segment %s, LSN %X/%08X, offset %u: read %d of %zu",
3412 : : fname, LSN_FORMAT_ARGS(targetPagePtr),
3413 : : readOff, r, (Size) XLOG_BLCKSZ)));
3414 : 0 : goto next_record_is_invalid;
3415 : : }
3416 : 1506380 : pgstat_report_wait_end();
3417 : :
3418 : 1506380 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_READ,
3419 : : io_start, 1, r);
3420 : :
3421 : : Assert(targetSegNo == readSegNo);
3422 : : Assert(targetPageOff == readOff);
3423 : : Assert(reqLen <= readLen);
3424 : :
3425 : 1506380 : xlogreader->seg.ws_tli = curFileTLI;
3426 : :
3427 : : /*
3428 : : * Check the page header immediately, so that we can retry immediately if
3429 : : * it's not valid. This may seem unnecessary, because ReadPageInternal()
3430 : : * validates the page header anyway, and would propagate the failure up to
3431 : : * ReadRecord(), which would retry. However, there's a corner case with
3432 : : * continuation records, if a record is split across two pages such that
3433 : : * we would need to read the two pages from different sources across two
3434 : : * WAL segments.
3435 : : *
3436 : : * The first page is only available locally, in pg_wal, because it's
3437 : : * already been recycled on the primary. The second page, however, is not
3438 : : * present in pg_wal, and we should stream it from the primary. There is a
3439 : : * recycled WAL segment present in pg_wal, with garbage contents, however.
3440 : : * We would read the first page from the local WAL segment, but when
3441 : : * reading the second page, we would read the bogus, recycled, WAL
3442 : : * segment. If we didn't catch that case here, we would never recover,
3443 : : * because ReadRecord() would retry reading the whole record from the
3444 : : * beginning.
3445 : : *
3446 : : * Of course, this only catches errors in the page header, which is what
3447 : : * happens in the case of a recycled WAL segment. Other kinds of errors or
3448 : : * corruption still has the same problem. But this at least fixes the
3449 : : * common case, which can happen as part of normal operation.
3450 : : *
3451 : : * Validating the page header is cheap enough that doing it twice
3452 : : * shouldn't be a big deal from a performance point of view.
3453 : : *
3454 : : * When not in standby mode, an invalid page header should cause recovery
3455 : : * to end, not retry reading the page, so we don't need to validate the
3456 : : * page header here for the retry. Instead, ReadPageInternal() is
3457 : : * responsible for the validation.
3458 : : */
3459 [ + + ]: 1506380 : if (StandbyMode &&
3460 [ + + ]: 1495912 : (targetPagePtr % wal_segment_size) == 0 &&
3461 [ + + ]: 1491 : !XLogReaderValidatePageHeader(xlogreader, targetPagePtr, readBuf))
3462 : : {
3463 : : /*
3464 : : * Emit this error right now then retry this page immediately. Use
3465 : : * errmsg_internal() because the message was already translated.
3466 : : */
3467 [ + - ]: 5 : if (xlogreader->errormsg_buf[0])
3468 [ + + ]: 5 : ereport(emode_for_corrupt_record(emode, xlogreader->EndRecPtr),
3469 : : (errmsg_internal("%s", xlogreader->errormsg_buf)));
3470 : :
3471 : : /* reset any error XLogReaderValidatePageHeader() might have set */
3472 : 5 : XLogReaderResetError(xlogreader);
3473 : 5 : goto next_record_is_invalid;
3474 : : }
3475 : :
3476 : 1506375 : return readLen;
3477 : :
3478 : 5 : next_record_is_invalid:
3479 : :
3480 : : /*
3481 : : * If we're reading ahead, give up fast. Retries and error reporting will
3482 : : * be handled by a later read when recovery catches up to this point.
3483 : : */
3484 [ + + ]: 5 : if (xlogreader->nonblocking)
3485 : 1 : return XLREAD_WOULDBLOCK;
3486 : :
3487 : 4 : lastSourceFailed = true;
3488 : :
3489 [ + - ]: 4 : if (readFile >= 0)
3490 : 4 : close(readFile);
3491 : 4 : readFile = -1;
3492 : 4 : readLen = 0;
3493 : 4 : readSource = XLOG_FROM_ANY;
3494 : :
3495 : : /* In standby-mode, keep trying */
3496 [ + - ]: 4 : if (StandbyMode)
3497 : 4 : goto retry;
3498 : : else
3499 : 0 : return XLREAD_FAIL;
3500 : : }
3501 : :
3502 : : /*
3503 : : * Open the WAL segment containing WAL location 'RecPtr'.
3504 : : *
3505 : : * The segment can be fetched via restore_command, or via walreceiver having
3506 : : * streamed the record, or it can already be present in pg_wal. Checking
3507 : : * pg_wal is mainly for crash recovery, but it will be polled in standby mode
3508 : : * too, in case someone copies a new segment directly to pg_wal. That is not
3509 : : * documented or recommended, though.
3510 : : *
3511 : : * If 'fetching_ckpt' is true, we're fetching a checkpoint record, and should
3512 : : * prepare to read WAL starting from RedoStartLSN after this.
3513 : : *
3514 : : * 'RecPtr' might not point to the beginning of the record we're interested
3515 : : * in, it might also point to the page or segment header. In that case,
3516 : : * 'tliRecPtr' is the position of the WAL record we're interested in. It is
3517 : : * used to decide which timeline to stream the requested WAL from.
3518 : : *
3519 : : * 'replayLSN' is the current replay LSN, so that if we scan for new
3520 : : * timelines, we can reject a switch to a timeline that branched off before
3521 : : * this point.
3522 : : *
3523 : : * If the record is not immediately available, the function returns XLREAD_FAIL
3524 : : * if we're not in standby mode. In standby mode, the function waits for it to
3525 : : * become available.
3526 : : *
3527 : : * When the requested record becomes available, the function opens the file
3528 : : * containing it (if not open already), and returns XLREAD_SUCCESS. When end
3529 : : * of standby mode is triggered by the user, and there is no more WAL
3530 : : * available, returns XLREAD_FAIL.
3531 : : *
3532 : : * If nonblocking is true, then give up immediately if we can't satisfy the
3533 : : * request, returning XLREAD_WOULDBLOCK instead of waiting.
3534 : : */
3535 : : static XLogPageReadResult
3536 : 16427 : WaitForWALToBecomeAvailable(XLogRecPtr RecPtr, bool randAccess,
3537 : : bool fetching_ckpt, XLogRecPtr tliRecPtr,
3538 : : TimeLineID replayTLI, XLogRecPtr replayLSN,
3539 : : bool nonblocking)
3540 : : {
3541 : : static TimestampTz last_fail_time = 0;
3542 : : TimestampTz now;
3543 : 16427 : bool streaming_reply_sent = false;
3544 : :
3545 : : /*-------
3546 : : * Standby mode is implemented by a state machine:
3547 : : *
3548 : : * 1. Read from either archive or pg_wal (XLOG_FROM_ARCHIVE), or just
3549 : : * pg_wal (XLOG_FROM_PG_WAL)
3550 : : * 2. Check for promotion trigger request
3551 : : * 3. Read from primary server via walreceiver (XLOG_FROM_STREAM)
3552 : : * 4. Rescan timelines
3553 : : * 5. Sleep wal_retrieve_retry_interval milliseconds, and loop back to 1.
3554 : : *
3555 : : * Failure to read from the current source advances the state machine to
3556 : : * the next state.
3557 : : *
3558 : : * 'currentSource' indicates the current state. There are no currentSource
3559 : : * values for "check trigger", "rescan timelines", and "sleep" states,
3560 : : * those actions are taken when reading from the previous source fails, as
3561 : : * part of advancing to the next state.
3562 : : *
3563 : : * If standby mode is turned off while reading WAL from stream, we move
3564 : : * to XLOG_FROM_ARCHIVE and reset lastSourceFailed, to force fetching
3565 : : * the files (which would be required at end of recovery, e.g., timeline
3566 : : * history file) from archive or pg_wal. We don't need to kill WAL receiver
3567 : : * here because it's already stopped when standby mode is turned off at
3568 : : * the end of recovery.
3569 : : *-------
3570 : : */
3571 [ + + ]: 16427 : if (!InArchiveRecovery)
3572 : 1079 : currentSource = XLOG_FROM_PG_WAL;
3573 [ + + ]: 15348 : else if (currentSource == XLOG_FROM_ANY ||
3574 [ + + - + ]: 15214 : (!StandbyMode && currentSource == XLOG_FROM_STREAM))
3575 : : {
3576 : 134 : lastSourceFailed = false;
3577 : 134 : currentSource = XLOG_FROM_ARCHIVE;
3578 : : }
3579 : :
3580 : : for (;;)
3581 : 14103 : {
3582 : 30530 : XLogSource oldSource = currentSource;
3583 : 30530 : bool startWalReceiver = false;
3584 : :
3585 : : /*
3586 : : * First check if we failed to read from the current source, and
3587 : : * advance the state machine if so. The failure to read might've
3588 : : * happened outside this function, e.g when a CRC check fails on a
3589 : : * record, or within this loop.
3590 : : */
3591 [ + + ]: 30530 : if (lastSourceFailed)
3592 : : {
3593 : : /*
3594 : : * Don't allow any retry loops to occur during nonblocking
3595 : : * readahead. Let the caller process everything that has been
3596 : : * decoded already first.
3597 : : */
3598 [ + + ]: 614 : if (nonblocking)
3599 : 89 : return XLREAD_WOULDBLOCK;
3600 : :
3601 [ + + - ]: 525 : switch (currentSource)
3602 : : {
3603 : 314 : case XLOG_FROM_ARCHIVE:
3604 : : case XLOG_FROM_PG_WAL:
3605 : :
3606 : : /*
3607 : : * Check to see if promotion is requested. Note that we do
3608 : : * this only after failure, so when you promote, we still
3609 : : * finish replaying as much as we can from archive and
3610 : : * pg_wal before failover.
3611 : : */
3612 [ + + + + ]: 314 : if (StandbyMode && CheckForStandbyTrigger())
3613 : : {
3614 : 22 : XLogShutdownWalRcv();
3615 : 22 : return XLREAD_FAIL;
3616 : : }
3617 : :
3618 : : /*
3619 : : * Not in standby mode, and we've now tried the archive
3620 : : * and pg_wal.
3621 : : */
3622 [ + + ]: 292 : if (!StandbyMode)
3623 : 30 : return XLREAD_FAIL;
3624 : :
3625 : : /*
3626 : : * Move to XLOG_FROM_STREAM state, and set to start a
3627 : : * walreceiver if necessary.
3628 : : */
3629 : 262 : currentSource = XLOG_FROM_STREAM;
3630 : 262 : startWalReceiver = true;
3631 : 262 : break;
3632 : :
3633 : 211 : case XLOG_FROM_STREAM:
3634 : :
3635 : : /*
3636 : : * Failure while streaming. Most likely, we got here
3637 : : * because streaming replication was terminated, or
3638 : : * promotion was triggered. But we also get here if we
3639 : : * find an invalid record in the WAL streamed from the
3640 : : * primary, in which case something is seriously wrong.
3641 : : * There's little chance that the problem will just go
3642 : : * away, but PANIC is not good for availability either,
3643 : : * especially in hot standby mode. So, we treat that the
3644 : : * same as disconnection, and retry from archive/pg_wal
3645 : : * again. The WAL in the archive should be identical to
3646 : : * what was streamed, so it's unlikely that it helps, but
3647 : : * one can hope...
3648 : : */
3649 : :
3650 : : /*
3651 : : * We should be able to move to XLOG_FROM_STREAM only in
3652 : : * standby mode.
3653 : : */
3654 : : Assert(StandbyMode);
3655 : :
3656 : : /*
3657 : : * Before we leave XLOG_FROM_STREAM state, make sure that
3658 : : * walreceiver is not active, so that it won't overwrite
3659 : : * WAL that we restore from archive.
3660 : : *
3661 : : * If walreceiver is actively streaming (or attempting to
3662 : : * connect), we must shut it down. However, if it's
3663 : : * already in WAITING state (e.g., due to timeline
3664 : : * divergence), we only need to reset the install flag to
3665 : : * allow archive restoration.
3666 : : */
3667 [ + + ]: 211 : if (WalRcvStreaming())
3668 : 36 : XLogShutdownWalRcv();
3669 : : else
3670 : : {
3671 : : /*
3672 : : * WALRCV_STOPPING state is a transient state while
3673 : : * the startup process is in ShutdownWalRcv(). It
3674 : : * should never appear here since we would be waiting
3675 : : * for the walreceiver to reach WALRCV_STOPPED in that
3676 : : * case.
3677 : : */
3678 : : Assert(WalRcvGetState() != WALRCV_STOPPING);
3679 : 175 : ResetInstallXLogFileSegmentActive();
3680 : : }
3681 : :
3682 : : /*
3683 : : * Before we sleep, re-scan for possible new timelines if
3684 : : * we were requested to recover to the latest timeline.
3685 : : */
3686 [ + - ]: 211 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
3687 : : {
3688 [ + + ]: 211 : if (rescanLatestTimeLine(replayTLI, replayLSN))
3689 : : {
3690 : 9 : currentSource = XLOG_FROM_ARCHIVE;
3691 : 9 : break;
3692 : : }
3693 : : }
3694 : :
3695 : : /*
3696 : : * XLOG_FROM_STREAM is the last state in our state
3697 : : * machine, so we've exhausted all the options for
3698 : : * obtaining the requested WAL. We're going to loop back
3699 : : * and retry from the archive, but if it hasn't been long
3700 : : * since last attempt, sleep wal_retrieve_retry_interval
3701 : : * milliseconds to avoid busy-waiting.
3702 : : */
3703 : 202 : now = GetCurrentTimestamp();
3704 [ + + ]: 202 : if (!TimestampDifferenceExceeds(last_fail_time, now,
3705 : : wal_retrieve_retry_interval))
3706 : : {
3707 : : long wait_time;
3708 : :
3709 : 220 : wait_time = wal_retrieve_retry_interval -
3710 : 110 : TimestampDifferenceMilliseconds(last_fail_time, now);
3711 : :
3712 [ + - ]: 110 : elog(LOG, "waiting for WAL to become available at %X/%08X",
3713 : : LSN_FORMAT_ARGS(RecPtr));
3714 : :
3715 : : /* Do background tasks that might benefit us later. */
3716 : 110 : KnownAssignedTransactionIdsIdleMaintenance();
3717 : :
3718 : 110 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3719 : : WL_LATCH_SET | WL_TIMEOUT |
3720 : : WL_EXIT_ON_PM_DEATH,
3721 : : wait_time,
3722 : : WAIT_EVENT_RECOVERY_RETRIEVE_RETRY_INTERVAL);
3723 : 110 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3724 : 110 : now = GetCurrentTimestamp();
3725 : :
3726 : : /* Handle interrupt signals of startup process */
3727 : 110 : ProcessStartupProcInterrupts();
3728 : : }
3729 : 188 : last_fail_time = now;
3730 : 188 : currentSource = XLOG_FROM_ARCHIVE;
3731 : 188 : break;
3732 : :
3733 : 0 : default:
3734 [ # # ]: 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
3735 : : }
3736 : : }
3737 [ + + ]: 29916 : else if (currentSource == XLOG_FROM_PG_WAL)
3738 : : {
3739 : : /*
3740 : : * We just successfully read a file in pg_wal. We prefer files in
3741 : : * the archive over ones in pg_wal, so try the next file again
3742 : : * from the archive first.
3743 : : */
3744 [ - + ]: 1073 : if (InArchiveRecovery)
3745 : 0 : currentSource = XLOG_FROM_ARCHIVE;
3746 : : }
3747 : :
3748 [ + + ]: 30375 : if (currentSource != oldSource)
3749 [ + + + - ]: 459 : elog(DEBUG2, "switched WAL source from %s to %s after %s",
3750 : : xlogSourceNames[oldSource], xlogSourceNames[currentSource],
3751 : : lastSourceFailed ? "failure" : "success");
3752 : :
3753 : : /*
3754 : : * We've now handled possible failure. Try to read from the chosen
3755 : : * source.
3756 : : */
3757 : 30375 : lastSourceFailed = false;
3758 : :
3759 [ + + - ]: 30375 : switch (currentSource)
3760 : : {
3761 : 1981 : case XLOG_FROM_ARCHIVE:
3762 : : case XLOG_FROM_PG_WAL:
3763 : :
3764 : : /*
3765 : : * WAL receiver must not be running when reading WAL from
3766 : : * archive or pg_wal.
3767 : : */
3768 : : Assert(!WalRcvStreaming());
3769 : :
3770 : : /* Close any old file we might have open. */
3771 [ + + ]: 1981 : if (readFile >= 0)
3772 : : {
3773 : 107 : close(readFile);
3774 : 107 : readFile = -1;
3775 : : }
3776 : : /* Reset curFileTLI if random fetch. */
3777 [ + + ]: 1981 : if (randAccess)
3778 : 1255 : curFileTLI = 0;
3779 : :
3780 : : /*
3781 : : * Try to restore the file from archive, or read an existing
3782 : : * file from pg_wal.
3783 : : */
3784 : 1981 : readFile = XLogFileReadAnyTLI(readSegNo,
3785 [ + + ]: 1981 : currentSource == XLOG_FROM_ARCHIVE ? XLOG_FROM_ANY :
3786 : : currentSource);
3787 [ + + ]: 1981 : if (readFile >= 0)
3788 : 1774 : return XLREAD_SUCCESS; /* success! */
3789 : :
3790 : : /*
3791 : : * Nope, not found in archive or pg_wal.
3792 : : */
3793 : 207 : lastSourceFailed = true;
3794 : 207 : break;
3795 : :
3796 : 28394 : case XLOG_FROM_STREAM:
3797 : : {
3798 : : bool havedata;
3799 : :
3800 : : /*
3801 : : * We should be able to move to XLOG_FROM_STREAM only in
3802 : : * standby mode.
3803 : : */
3804 : : Assert(StandbyMode);
3805 : :
3806 : : /*
3807 : : * First, shutdown walreceiver if its restart has been
3808 : : * requested -- but no point if we're already slated for
3809 : : * starting it.
3810 : : */
3811 [ + + + - ]: 28394 : if (pendingWalRcvRestart && !startWalReceiver)
3812 : : {
3813 : 8 : XLogShutdownWalRcv();
3814 : :
3815 : : /*
3816 : : * Re-scan for possible new timelines if we were
3817 : : * requested to recover to the latest timeline.
3818 : : */
3819 [ + - ]: 8 : if (recoveryTargetTimeLineGoal ==
3820 : : RECOVERY_TARGET_TIMELINE_LATEST)
3821 : 8 : rescanLatestTimeLine(replayTLI, replayLSN);
3822 : :
3823 : 8 : startWalReceiver = true;
3824 : : }
3825 : 28394 : pendingWalRcvRestart = false;
3826 : :
3827 : : /*
3828 : : * Launch walreceiver if needed.
3829 : : *
3830 : : * If fetching_ckpt is true, RecPtr points to the initial
3831 : : * checkpoint location. In that case, we use RedoStartLSN
3832 : : * as the streaming start position instead of RecPtr, so
3833 : : * that when we later jump backwards to start redo at
3834 : : * RedoStartLSN, we will have the logs streamed already.
3835 : : */
3836 [ + + + - ]: 28394 : if (startWalReceiver &&
3837 [ + + ]: 270 : PrimaryConnInfo && strcmp(PrimaryConnInfo, "") != 0)
3838 : : {
3839 : : XLogRecPtr ptr;
3840 : : TimeLineID tli;
3841 : :
3842 [ - + ]: 220 : if (fetching_ckpt)
3843 : : {
3844 : 0 : ptr = RedoStartLSN;
3845 : 0 : tli = RedoStartTLI;
3846 : : }
3847 : : else
3848 : : {
3849 : 220 : ptr = RecPtr;
3850 : :
3851 : : /*
3852 : : * Use the record begin position to determine the
3853 : : * TLI, rather than the position we're reading.
3854 : : */
3855 : 220 : tli = tliOfPointInHistory(tliRecPtr, expectedTLEs);
3856 : :
3857 [ + - - + ]: 220 : if (curFileTLI > 0 && tli < curFileTLI)
3858 [ # # ]: 0 : elog(ERROR, "according to history file, WAL location %X/%08X belongs to timeline %u, but previous recovered WAL file came from timeline %u",
3859 : : LSN_FORMAT_ARGS(tliRecPtr),
3860 : : tli, curFileTLI);
3861 : : }
3862 : 220 : curFileTLI = tli;
3863 : 220 : SetInstallXLogFileSegmentActive();
3864 : 220 : RequestXLogStreaming(tli, ptr, PrimaryConnInfo,
3865 : : PrimarySlotName,
3866 : : wal_receiver_create_temp_slot);
3867 : 220 : flushedUpto = InvalidXLogRecPtr;
3868 : : }
3869 : :
3870 : : /*
3871 : : * Check if WAL receiver is active or wait to start up.
3872 : : */
3873 [ + + ]: 28394 : if (!WalRcvStreaming())
3874 : : {
3875 : 175 : lastSourceFailed = true;
3876 : 175 : break;
3877 : : }
3878 : :
3879 : : /*
3880 : : * Walreceiver is active, so see if new data has arrived.
3881 : : *
3882 : : * We only advance XLogReceiptTime when we obtain fresh
3883 : : * WAL from walreceiver and observe that we had already
3884 : : * processed everything before the most recent "chunk"
3885 : : * that it flushed to disk. In steady state where we are
3886 : : * keeping up with the incoming data, XLogReceiptTime will
3887 : : * be updated on each cycle. When we are behind,
3888 : : * XLogReceiptTime will not advance, so the grace time
3889 : : * allotted to conflicting queries will decrease.
3890 : : */
3891 [ + + ]: 28219 : if (RecPtr < flushedUpto)
3892 : 1776 : havedata = true;
3893 : : else
3894 : : {
3895 : : XLogRecPtr latestChunkStart;
3896 : :
3897 : 26443 : flushedUpto = GetWalRcvFlushRecPtr(&latestChunkStart, &receiveTLI);
3898 [ + + + - ]: 26443 : if (RecPtr < flushedUpto && receiveTLI == curFileTLI)
3899 : : {
3900 : 13543 : havedata = true;
3901 [ + + ]: 13543 : if (latestChunkStart <= RecPtr)
3902 : : {
3903 : 8901 : XLogReceiptTime = GetCurrentTimestamp();
3904 : 8901 : SetCurrentChunkStartTime(XLogReceiptTime);
3905 : : }
3906 : : }
3907 : : else
3908 : 12900 : havedata = false;
3909 : : }
3910 [ + + ]: 28219 : if (havedata)
3911 : : {
3912 : : /*
3913 : : * Great, streamed far enough. Open the file if it's
3914 : : * not open already. Also read the timeline history
3915 : : * file if we haven't initialized timeline history
3916 : : * yet; it should be streamed over and present in
3917 : : * pg_wal by now. Use XLOG_FROM_STREAM so that source
3918 : : * info is set correctly and XLogReceiptTime isn't
3919 : : * changed.
3920 : : *
3921 : : * NB: We must set readTimeLineHistory based on
3922 : : * recoveryTargetTLI, not receiveTLI. Normally they'll
3923 : : * be the same, but if recovery_target_timeline is
3924 : : * 'latest' and archiving is configured, then it's
3925 : : * possible that we managed to retrieve one or more
3926 : : * new timeline history files from the archive,
3927 : : * updating recoveryTargetTLI.
3928 : : */
3929 [ + + ]: 15319 : if (readFile < 0)
3930 : : {
3931 [ - + ]: 1176 : if (!expectedTLEs)
3932 : 0 : expectedTLEs = readTimeLineHistory(recoveryTargetTLI);
3933 : 1176 : readFile = XLogFileRead(readSegNo, receiveTLI,
3934 : : XLOG_FROM_STREAM, false);
3935 : : Assert(readFile >= 0);
3936 : : }
3937 : : else
3938 : : {
3939 : : /* just make sure source info is correct... */
3940 : 14143 : readSource = XLOG_FROM_STREAM;
3941 : 14143 : XLogReceiptSource = XLOG_FROM_STREAM;
3942 : 14143 : return XLREAD_SUCCESS;
3943 : : }
3944 : 1176 : break;
3945 : : }
3946 : :
3947 : : /* In nonblocking mode, return rather than sleeping. */
3948 [ + + ]: 12900 : if (nonblocking)
3949 : 306 : return XLREAD_WOULDBLOCK;
3950 : :
3951 : : /*
3952 : : * Data not here yet. Check for trigger, then wait for
3953 : : * walreceiver to wake us up when new WAL arrives.
3954 : : */
3955 [ + + ]: 12594 : if (CheckForStandbyTrigger())
3956 : : {
3957 : : /*
3958 : : * Note that we don't return XLREAD_FAIL immediately
3959 : : * here. After being triggered, we still want to
3960 : : * replay all the WAL that was already streamed. It's
3961 : : * in pg_wal now, so we just treat this as a failure,
3962 : : * and the state machine will move on to replay the
3963 : : * streamed WAL from pg_wal, and then recheck the
3964 : : * trigger and exit replay.
3965 : : */
3966 : 36 : lastSourceFailed = true;
3967 : 36 : break;
3968 : : }
3969 : :
3970 : : /*
3971 : : * Since we have replayed everything we have received so
3972 : : * far and are about to start waiting for more WAL, let's
3973 : : * tell the upstream server our replay location now so
3974 : : * that pg_stat_replication doesn't show stale
3975 : : * information.
3976 : : */
3977 [ + + ]: 12558 : if (!streaming_reply_sent)
3978 : : {
3979 : 10698 : WalRcvRequestApplyReply();
3980 : 10698 : streaming_reply_sent = true;
3981 : : }
3982 : :
3983 : : /* Do any background tasks that might benefit us later. */
3984 : 12558 : KnownAssignedTransactionIdsIdleMaintenance();
3985 : :
3986 : : /* Update pg_stat_recovery_prefetch before sleeping. */
3987 : 12558 : XLogPrefetcherComputeStats(xlogprefetcher);
3988 : :
3989 : : /*
3990 : : * Wait for more WAL to arrive, when we will be woken
3991 : : * immediately by the WAL receiver.
3992 : : */
3993 : 12558 : (void) WaitLatch(&XLogRecoveryCtl->recoveryWakeupLatch,
3994 : : WL_LATCH_SET | WL_EXIT_ON_PM_DEATH,
3995 : : -1L,
3996 : : WAIT_EVENT_RECOVERY_WAL_STREAM);
3997 : 12558 : ResetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
3998 : 12558 : break;
3999 : : }
4000 : :
4001 : 0 : default:
4002 [ # # ]: 0 : elog(ERROR, "unexpected WAL source %d", currentSource);
4003 : : }
4004 : :
4005 : : /*
4006 : : * Check for recovery pause here so that we can confirm more quickly
4007 : : * that a requested pause has actually taken effect.
4008 : : */
4009 [ + + ]: 14152 : if (((volatile XLogRecoveryCtlData *) XLogRecoveryCtl)->recoveryPauseState !=
4010 : : RECOVERY_NOT_PAUSED)
4011 : 7 : recoveryPausesHere(false);
4012 : :
4013 : : /*
4014 : : * This possibly-long loop needs to handle interrupts of startup
4015 : : * process.
4016 : : */
4017 : 14152 : ProcessStartupProcInterrupts();
4018 : : }
4019 : :
4020 : : return XLREAD_FAIL; /* not reached */
4021 : : }
4022 : :
4023 : :
4024 : : /*
4025 : : * Determine what log level should be used to report a corrupt WAL record
4026 : : * in the current WAL page, previously read by XLogPageRead().
4027 : : *
4028 : : * 'emode' is the error mode that would be used to report a file-not-found
4029 : : * or legitimate end-of-WAL situation. Generally, we use it as-is, but if
4030 : : * we're retrying the exact same record that we've tried previously, only
4031 : : * complain the first time to keep the noise down. However, we only do when
4032 : : * reading from pg_wal, because we don't expect any invalid records in archive
4033 : : * or in records streamed from the primary. Files in the archive should be complete,
4034 : : * and we should never hit the end of WAL because we stop and wait for more WAL
4035 : : * to arrive before replaying it.
4036 : : *
4037 : : * NOTE: This function remembers the RecPtr value it was last called with,
4038 : : * to suppress repeated messages about the same record. Only call this when
4039 : : * you are about to ereport(), or you might cause a later message to be
4040 : : * erroneously suppressed.
4041 : : */
4042 : : static int
4043 : 310 : emode_for_corrupt_record(int emode, XLogRecPtr RecPtr)
4044 : : {
4045 : : static XLogRecPtr lastComplaint = InvalidXLogRecPtr;
4046 : :
4047 [ + + + - ]: 310 : if (readSource == XLOG_FROM_PG_WAL && emode == LOG)
4048 : : {
4049 [ + + ]: 307 : if (RecPtr == lastComplaint)
4050 : 93 : emode = DEBUG1;
4051 : : else
4052 : 214 : lastComplaint = RecPtr;
4053 : : }
4054 : 310 : return emode;
4055 : : }
4056 : :
4057 : :
4058 : : /*
4059 : : * Subroutine to try to fetch and validate a prior checkpoint record.
4060 : : */
4061 : : static XLogRecord *
4062 : 1089 : ReadCheckpointRecord(XLogPrefetcher *xlogprefetcher, XLogRecPtr RecPtr,
4063 : : TimeLineID replayTLI)
4064 : : {
4065 : : XLogRecord *record;
4066 : : uint8 info;
4067 : :
4068 : : Assert(xlogreader != NULL);
4069 : :
4070 [ - + ]: 1089 : if (!XRecOffIsValid(RecPtr))
4071 : : {
4072 [ # # ]: 0 : ereport(LOG,
4073 : : (errmsg("invalid checkpoint location")));
4074 : 0 : return NULL;
4075 : : }
4076 : :
4077 : 1089 : XLogPrefetcherBeginRead(xlogprefetcher, RecPtr);
4078 : 1089 : record = ReadRecord(xlogprefetcher, LOG, true, replayTLI);
4079 : :
4080 [ + + ]: 1089 : if (record == NULL)
4081 : : {
4082 [ + - ]: 1 : ereport(LOG,
4083 : : (errmsg("invalid checkpoint record")));
4084 : 1 : return NULL;
4085 : : }
4086 [ - + ]: 1088 : if (record->xl_rmid != RM_XLOG_ID)
4087 : : {
4088 [ # # ]: 0 : ereport(LOG,
4089 : : (errmsg("invalid resource manager ID in checkpoint record")));
4090 : 0 : return NULL;
4091 : : }
4092 : 1088 : info = record->xl_info & ~XLR_INFO_MASK;
4093 [ + + - + ]: 1088 : if (info != XLOG_CHECKPOINT_SHUTDOWN &&
4094 : : info != XLOG_CHECKPOINT_ONLINE)
4095 : : {
4096 [ # # ]: 0 : ereport(LOG,
4097 : : (errmsg("invalid xl_info in checkpoint record")));
4098 : 0 : return NULL;
4099 : : }
4100 [ - + ]: 1088 : if (record->xl_tot_len != SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(CheckPoint))
4101 : : {
4102 [ # # ]: 0 : ereport(LOG,
4103 : : (errmsg("invalid length of checkpoint record")));
4104 : 0 : return NULL;
4105 : : }
4106 : 1088 : return record;
4107 : : }
4108 : :
4109 : : /*
4110 : : * Scan for new timelines that might have appeared in the archive since we
4111 : : * started recovery.
4112 : : *
4113 : : * If there are any, the function changes recovery target TLI to the latest
4114 : : * one and returns 'true'.
4115 : : */
4116 : : static bool
4117 : 219 : rescanLatestTimeLine(TimeLineID replayTLI, XLogRecPtr replayLSN)
4118 : : {
4119 : : List *newExpectedTLEs;
4120 : : bool found;
4121 : : ListCell *cell;
4122 : : TimeLineID newtarget;
4123 : 219 : TimeLineID oldtarget = recoveryTargetTLI;
4124 : 219 : TimeLineHistoryEntry *currentTle = NULL;
4125 : :
4126 : 219 : newtarget = findNewestTimeLine(recoveryTargetTLI);
4127 [ + + ]: 219 : if (newtarget == recoveryTargetTLI)
4128 : : {
4129 : : /* No new timelines found */
4130 : 210 : return false;
4131 : : }
4132 : :
4133 : : /*
4134 : : * Determine the list of expected TLIs for the new TLI
4135 : : */
4136 : :
4137 : 9 : newExpectedTLEs = readTimeLineHistory(newtarget);
4138 : :
4139 : : /*
4140 : : * If the current timeline is not part of the history of the new timeline,
4141 : : * we cannot proceed to it.
4142 : : */
4143 : 9 : found = false;
4144 [ + - + - : 18 : foreach(cell, newExpectedTLEs)
+ - ]
4145 : : {
4146 : 18 : currentTle = (TimeLineHistoryEntry *) lfirst(cell);
4147 : :
4148 [ + + ]: 18 : if (currentTle->tli == recoveryTargetTLI)
4149 : : {
4150 : 9 : found = true;
4151 : 9 : break;
4152 : : }
4153 : : }
4154 [ - + ]: 9 : if (!found)
4155 : : {
4156 [ # # ]: 0 : ereport(LOG,
4157 : : (errmsg("new timeline %u is not a child of database system timeline %u",
4158 : : newtarget,
4159 : : replayTLI)));
4160 : 0 : return false;
4161 : : }
4162 : :
4163 : : /*
4164 : : * The current timeline was found in the history file, but check that the
4165 : : * next timeline was forked off from it *after* the current recovery
4166 : : * location.
4167 : : */
4168 [ - + ]: 9 : if (currentTle->end < replayLSN)
4169 : : {
4170 [ # # ]: 0 : ereport(LOG,
4171 : : errmsg("new timeline %u forked off current database system timeline %u before current recovery point %X/%08X",
4172 : : newtarget,
4173 : : replayTLI,
4174 : : LSN_FORMAT_ARGS(replayLSN)));
4175 : 0 : return false;
4176 : : }
4177 : :
4178 : : /* The new timeline history seems valid. Switch target */
4179 : 9 : recoveryTargetTLI = newtarget;
4180 : 9 : list_free_deep(expectedTLEs);
4181 : 9 : expectedTLEs = newExpectedTLEs;
4182 : :
4183 : : /*
4184 : : * As in StartupXLOG(), try to ensure we have all the history files
4185 : : * between the old target and new target in pg_wal.
4186 : : */
4187 : 9 : restoreTimeLineHistoryFiles(oldtarget + 1, newtarget);
4188 : :
4189 [ + - ]: 9 : ereport(LOG,
4190 : : (errmsg("new target timeline is %u",
4191 : : recoveryTargetTLI)));
4192 : :
4193 : 9 : return true;
4194 : : }
4195 : :
4196 : :
4197 : : /*
4198 : : * Open a logfile segment for reading (during recovery).
4199 : : *
4200 : : * If source == XLOG_FROM_ARCHIVE, the segment is retrieved from archive.
4201 : : * Otherwise, it's assumed to be already available in pg_wal.
4202 : : */
4203 : : static int
4204 : 3726 : XLogFileRead(XLogSegNo segno, TimeLineID tli,
4205 : : XLogSource source, bool notfoundOk)
4206 : : {
4207 : : char xlogfname[MAXFNAMELEN];
4208 : : char activitymsg[MAXFNAMELEN + 16];
4209 : : char path[MAXPGPATH];
4210 : : int fd;
4211 : :
4212 : 3726 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
4213 : :
4214 [ + + - ]: 3726 : switch (source)
4215 : : {
4216 : 921 : case XLOG_FROM_ARCHIVE:
4217 : : /* Report recovery progress in PS display */
4218 : 921 : snprintf(activitymsg, sizeof(activitymsg), "waiting for %s",
4219 : : xlogfname);
4220 : 921 : set_ps_display(activitymsg);
4221 : :
4222 [ + + ]: 921 : if (!RestoreArchivedFile(path, xlogfname,
4223 : : "RECOVERYXLOG",
4224 : : wal_segment_size,
4225 : : InRedo))
4226 : 556 : return -1;
4227 : 365 : break;
4228 : :
4229 : 2805 : case XLOG_FROM_PG_WAL:
4230 : : case XLOG_FROM_STREAM:
4231 : 2805 : XLogFilePath(path, tli, segno, wal_segment_size);
4232 : 2805 : break;
4233 : :
4234 : 0 : default:
4235 [ # # ]: 0 : elog(ERROR, "invalid XLogFileRead source %d", source);
4236 : : }
4237 : :
4238 : : /*
4239 : : * If the segment was fetched from archival storage, replace the existing
4240 : : * xlog segment (if any) with the archival version.
4241 : : */
4242 [ + + ]: 3170 : if (source == XLOG_FROM_ARCHIVE)
4243 : : {
4244 : : Assert(!IsInstallXLogFileSegmentActive());
4245 : 365 : KeepFileRestoredFromArchive(path, xlogfname);
4246 : :
4247 : : /*
4248 : : * Set path to point at the new file in pg_wal.
4249 : : */
4250 : 365 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlogfname);
4251 : : }
4252 : :
4253 : 3170 : fd = BasicOpenFile(path, O_RDONLY | PG_BINARY);
4254 [ + + ]: 3170 : if (fd >= 0)
4255 : : {
4256 : : /* Success! */
4257 : 2950 : curFileTLI = tli;
4258 : :
4259 : : /* Report recovery progress in PS display */
4260 : 2950 : snprintf(activitymsg, sizeof(activitymsg), "recovering %s",
4261 : : xlogfname);
4262 : 2950 : set_ps_display(activitymsg);
4263 : :
4264 : : /* Track source of data in assorted state variables */
4265 : 2950 : readSource = source;
4266 : 2950 : XLogReceiptSource = source;
4267 : : /* In FROM_STREAM case, caller tracks receipt time, not me */
4268 [ + + ]: 2950 : if (source != XLOG_FROM_STREAM)
4269 : 1774 : XLogReceiptTime = GetCurrentTimestamp();
4270 : :
4271 : 2950 : return fd;
4272 : : }
4273 [ + - - + ]: 220 : if (errno != ENOENT || !notfoundOk) /* unexpected failure? */
4274 [ # # ]: 0 : ereport(PANIC,
4275 : : (errcode_for_file_access(),
4276 : : errmsg("could not open file \"%s\": %m", path)));
4277 : 220 : return -1;
4278 : : }
4279 : :
4280 : : /*
4281 : : * Open a logfile segment for reading (during recovery).
4282 : : *
4283 : : * This version searches for the segment with any TLI listed in expectedTLEs.
4284 : : */
4285 : : static int
4286 : 1981 : XLogFileReadAnyTLI(XLogSegNo segno, XLogSource source)
4287 : : {
4288 : : char path[MAXPGPATH];
4289 : : ListCell *cell;
4290 : : int fd;
4291 : : List *tles;
4292 : :
4293 : : /*
4294 : : * Loop looking for a suitable timeline ID: we might need to read any of
4295 : : * the timelines listed in expectedTLEs.
4296 : : *
4297 : : * We expect curFileTLI on entry to be the TLI of the preceding file in
4298 : : * sequence, or 0 if there was no predecessor. We do not allow curFileTLI
4299 : : * to go backwards; this prevents us from picking up the wrong file when a
4300 : : * parent timeline extends to higher segment numbers than the child we
4301 : : * want to read.
4302 : : *
4303 : : * If we haven't read the timeline history file yet, read it now, so that
4304 : : * we know which TLIs to scan. We don't save the list in expectedTLEs,
4305 : : * however, unless we actually find a valid segment. That way if there is
4306 : : * neither a timeline history file nor a WAL segment in the archive, and
4307 : : * streaming replication is set up, we'll read the timeline history file
4308 : : * streamed from the primary when we start streaming, instead of
4309 : : * recovering with a dummy history generated here.
4310 : : */
4311 [ + + ]: 1981 : if (expectedTLEs)
4312 : 892 : tles = expectedTLEs;
4313 : : else
4314 : 1089 : tles = readTimeLineHistory(recoveryTargetTLI);
4315 : :
4316 [ + - + + : 2208 : foreach(cell, tles)
+ + ]
4317 : : {
4318 : 2007 : TimeLineHistoryEntry *hent = (TimeLineHistoryEntry *) lfirst(cell);
4319 : 2007 : TimeLineID tli = hent->tli;
4320 : :
4321 [ + + ]: 2007 : if (tli < curFileTLI)
4322 : 6 : break; /* don't bother looking at too-old TLIs */
4323 : :
4324 : : /*
4325 : : * Skip scanning the timeline ID that the logfile segment to read
4326 : : * doesn't belong to
4327 : : */
4328 [ + + ]: 2001 : if (XLogRecPtrIsValid(hent->begin))
4329 : : {
4330 : 85 : XLogSegNo beginseg = 0;
4331 : :
4332 : 85 : XLByteToSeg(hent->begin, beginseg, wal_segment_size);
4333 : :
4334 : : /*
4335 : : * The logfile segment that doesn't belong to the timeline is
4336 : : * older or newer than the segment that the timeline started or
4337 : : * ended at, respectively. It's sufficient to check only the
4338 : : * starting segment of the timeline here. Since the timelines are
4339 : : * scanned in descending order in this loop, any segments newer
4340 : : * than the ending segment should belong to newer timeline and
4341 : : * have already been read before. So it's not necessary to check
4342 : : * the ending segment of the timeline here.
4343 : : */
4344 [ + + ]: 85 : if (segno < beginseg)
4345 : 7 : continue;
4346 : : }
4347 : :
4348 [ + + - + ]: 1994 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_ARCHIVE)
4349 : : {
4350 : 921 : fd = XLogFileRead(segno, tli, XLOG_FROM_ARCHIVE, true);
4351 [ + + ]: 921 : if (fd != -1)
4352 : : {
4353 [ - + ]: 365 : elog(DEBUG1, "got WAL segment from archive");
4354 [ + + ]: 365 : if (!expectedTLEs)
4355 : 19 : expectedTLEs = tles;
4356 : 1774 : return fd;
4357 : : }
4358 : : }
4359 : :
4360 [ + + + - ]: 1629 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_PG_WAL)
4361 : : {
4362 : 1629 : fd = XLogFileRead(segno, tli, XLOG_FROM_PG_WAL, true);
4363 [ + + ]: 1629 : if (fd != -1)
4364 : : {
4365 [ + + ]: 1409 : if (!expectedTLEs)
4366 : 1069 : expectedTLEs = tles;
4367 : 1409 : return fd;
4368 : : }
4369 : : }
4370 : : }
4371 : :
4372 : : /* Couldn't find it. For simplicity, complain about front timeline */
4373 : 207 : XLogFilePath(path, recoveryTargetTLI, segno, wal_segment_size);
4374 : 207 : errno = ENOENT;
4375 [ + + ]: 207 : ereport(DEBUG2,
4376 : : (errcode_for_file_access(),
4377 : : errmsg("could not open file \"%s\": %m", path)));
4378 : 207 : return -1;
4379 : : }
4380 : :
4381 : : /*
4382 : : * Set flag to signal the walreceiver to restart. (The startup process calls
4383 : : * this on noticing a relevant configuration change.)
4384 : : */
4385 : : void
4386 : 13 : StartupRequestWalReceiverRestart(void)
4387 : : {
4388 [ + - + + ]: 13 : if (currentSource == XLOG_FROM_STREAM && WalRcvRunning())
4389 : : {
4390 [ + - ]: 8 : ereport(LOG,
4391 : : (errmsg("WAL receiver process shutdown requested")));
4392 : :
4393 : 8 : pendingWalRcvRestart = true;
4394 : : }
4395 : 13 : }
4396 : :
4397 : :
4398 : : /*
4399 : : * Has a standby promotion already been triggered?
4400 : : *
4401 : : * Unlike CheckForStandbyTrigger(), this works in any process
4402 : : * that's connected to shared memory.
4403 : : */
4404 : : bool
4405 : 85 : PromoteIsTriggered(void)
4406 : : {
4407 : : /*
4408 : : * We check shared state each time only until a standby promotion is
4409 : : * triggered. We can't trigger a promotion again, so there's no need to
4410 : : * keep checking after the shared variable has once been seen true.
4411 : : */
4412 [ + + ]: 85 : if (LocalPromoteIsTriggered)
4413 : 56 : return true;
4414 : :
4415 : 29 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4416 : 29 : LocalPromoteIsTriggered = XLogRecoveryCtl->SharedPromoteIsTriggered;
4417 : 29 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4418 : :
4419 : 29 : return LocalPromoteIsTriggered;
4420 : : }
4421 : :
4422 : : static void
4423 : 53 : SetPromoteIsTriggered(void)
4424 : : {
4425 : 53 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4426 : 53 : XLogRecoveryCtl->SharedPromoteIsTriggered = true;
4427 : 53 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4428 : :
4429 : : /*
4430 : : * Mark the recovery pause state as 'not paused' because the paused state
4431 : : * ends and promotion continues if a promotion is triggered while recovery
4432 : : * is paused. Otherwise pg_get_wal_replay_pause_state() can mistakenly
4433 : : * return 'paused' while a promotion is ongoing.
4434 : : */
4435 : 53 : SetRecoveryPause(false);
4436 : :
4437 : 53 : LocalPromoteIsTriggered = true;
4438 : 53 : }
4439 : :
4440 : : /*
4441 : : * Check whether a promote request has arrived.
4442 : : */
4443 : : static bool
4444 : 13200 : CheckForStandbyTrigger(void)
4445 : : {
4446 [ + + ]: 13200 : if (LocalPromoteIsTriggered)
4447 : 59 : return true;
4448 : :
4449 [ + + + - ]: 13141 : if (IsPromoteSignaled() && CheckPromoteSignal())
4450 : : {
4451 [ + - ]: 53 : ereport(LOG, (errmsg("received promote request")));
4452 : 53 : RemovePromoteSignalFiles();
4453 : 53 : ResetPromoteSignaled();
4454 : 53 : SetPromoteIsTriggered();
4455 : 53 : return true;
4456 : : }
4457 : :
4458 : 13088 : return false;
4459 : : }
4460 : :
4461 : : /*
4462 : : * Remove the files signaling a standby promotion request.
4463 : : */
4464 : : void
4465 : 1047 : RemovePromoteSignalFiles(void)
4466 : : {
4467 : 1047 : unlink(PROMOTE_SIGNAL_FILE);
4468 : 1047 : }
4469 : :
4470 : : /*
4471 : : * Check to see if a promote request has arrived.
4472 : : */
4473 : : bool
4474 : 793 : CheckPromoteSignal(void)
4475 : : {
4476 : : struct stat stat_buf;
4477 : :
4478 [ + + ]: 793 : if (stat(PROMOTE_SIGNAL_FILE, &stat_buf) == 0)
4479 : 106 : return true;
4480 : :
4481 : 687 : return false;
4482 : : }
4483 : :
4484 : : /*
4485 : : * Wake up startup process to replay newly arrived WAL, or to notice that
4486 : : * failover has been requested.
4487 : : */
4488 : : void
4489 : 48939 : WakeupRecovery(void)
4490 : : {
4491 : 48939 : SetLatch(&XLogRecoveryCtl->recoveryWakeupLatch);
4492 : 48939 : }
4493 : :
4494 : : /*
4495 : : * Schedule a walreceiver wakeup in the main recovery loop.
4496 : : */
4497 : : void
4498 : 2 : XLogRequestWalReceiverReply(void)
4499 : : {
4500 : 2 : doRequestWalReceiverReply = true;
4501 : 2 : }
4502 : :
4503 : : /*
4504 : : * Is HotStandby active yet? This is only important in special backends
4505 : : * since normal backends won't ever be able to connect until this returns
4506 : : * true. Postmaster knows this by way of signal, not via shared memory.
4507 : : *
4508 : : * Unlike testing standbyState, this works in any process that's connected to
4509 : : * shared memory. (And note that standbyState alone doesn't tell the truth
4510 : : * anyway.)
4511 : : */
4512 : : bool
4513 : 170 : HotStandbyActive(void)
4514 : : {
4515 : : /*
4516 : : * We check shared state each time only until Hot Standby is active. We
4517 : : * can't de-activate Hot Standby, so there's no need to keep checking
4518 : : * after the shared variable has once been seen true.
4519 : : */
4520 [ + + ]: 170 : if (LocalHotStandbyActive)
4521 : 22 : return true;
4522 : : else
4523 : : {
4524 : : /* spinlock is essential on machines with weak memory ordering! */
4525 : 148 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4526 : 148 : LocalHotStandbyActive = XLogRecoveryCtl->SharedHotStandbyActive;
4527 : 148 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4528 : :
4529 : 148 : return LocalHotStandbyActive;
4530 : : }
4531 : : }
4532 : :
4533 : : /*
4534 : : * Like HotStandbyActive(), but to be used only in WAL replay code,
4535 : : * where we don't need to ask any other process what the state is.
4536 : : */
4537 : : static bool
4538 : 0 : HotStandbyActiveInReplay(void)
4539 : : {
4540 : : Assert(AmStartupProcess() || !IsPostmasterEnvironment);
4541 : 0 : return LocalHotStandbyActive;
4542 : : }
4543 : :
4544 : : /*
4545 : : * Get latest redo apply position.
4546 : : *
4547 : : * Exported to allow WALReceiver to read the pointer directly.
4548 : : */
4549 : : XLogRecPtr
4550 : 119238 : GetXLogReplayRecPtr(TimeLineID *replayTLI)
4551 : : {
4552 : : XLogRecPtr recptr;
4553 : : TimeLineID tli;
4554 : :
4555 : 119238 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4556 : 119238 : recptr = XLogRecoveryCtl->lastReplayedEndRecPtr;
4557 : 119238 : tli = XLogRecoveryCtl->lastReplayedTLI;
4558 : 119238 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4559 : :
4560 [ + + ]: 119238 : if (replayTLI)
4561 : 2868 : *replayTLI = tli;
4562 : 119238 : return recptr;
4563 : : }
4564 : :
4565 : :
4566 : : /*
4567 : : * Get position of last applied, or the record being applied.
4568 : : *
4569 : : * This is different from GetXLogReplayRecPtr() in that if a WAL
4570 : : * record is currently being applied, this includes that record.
4571 : : */
4572 : : XLogRecPtr
4573 : 6876 : GetCurrentReplayRecPtr(TimeLineID *replayEndTLI)
4574 : : {
4575 : : XLogRecPtr recptr;
4576 : : TimeLineID tli;
4577 : :
4578 : 6876 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4579 : 6876 : recptr = XLogRecoveryCtl->replayEndRecPtr;
4580 : 6876 : tli = XLogRecoveryCtl->replayEndTLI;
4581 : 6876 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4582 : :
4583 [ + - ]: 6876 : if (replayEndTLI)
4584 : 6876 : *replayEndTLI = tli;
4585 : 6876 : return recptr;
4586 : : }
4587 : :
4588 : : /*
4589 : : * Save timestamp of latest processed commit/abort record.
4590 : : *
4591 : : * We keep this in XLogRecoveryCtl, not a simple static variable, so that it can be
4592 : : * seen by processes other than the startup process. Note in particular
4593 : : * that CreateRestartPoint is executed in the checkpointer.
4594 : : */
4595 : : static void
4596 : 23943 : SetLatestXTime(TimestampTz xtime)
4597 : : {
4598 : 23943 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4599 : 23943 : XLogRecoveryCtl->recoveryLastXTime = xtime;
4600 : 23943 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4601 : 23943 : }
4602 : :
4603 : : /*
4604 : : * Fetch timestamp of latest processed commit/abort record.
4605 : : */
4606 : : TimestampTz
4607 : 373 : GetLatestXTime(void)
4608 : : {
4609 : : TimestampTz xtime;
4610 : :
4611 : 373 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4612 : 373 : xtime = XLogRecoveryCtl->recoveryLastXTime;
4613 : 373 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4614 : :
4615 : 373 : return xtime;
4616 : : }
4617 : :
4618 : : /*
4619 : : * Save timestamp of the next chunk of WAL records to apply.
4620 : : *
4621 : : * We keep this in XLogRecoveryCtl, not a simple static variable, so that it can be
4622 : : * seen by all backends.
4623 : : */
4624 : : static void
4625 : 8901 : SetCurrentChunkStartTime(TimestampTz xtime)
4626 : : {
4627 : 8901 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4628 : 8901 : XLogRecoveryCtl->currentChunkStartTime = xtime;
4629 : 8901 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4630 : 8901 : }
4631 : :
4632 : : /*
4633 : : * Fetch timestamp of latest processed commit/abort record.
4634 : : * Startup process maintains an accurate local copy in XLogReceiptTime
4635 : : */
4636 : : TimestampTz
4637 : 268 : GetCurrentChunkReplayStartTime(void)
4638 : : {
4639 : : TimestampTz xtime;
4640 : :
4641 : 268 : SpinLockAcquire(&XLogRecoveryCtl->info_lck);
4642 : 268 : xtime = XLogRecoveryCtl->currentChunkStartTime;
4643 : 268 : SpinLockRelease(&XLogRecoveryCtl->info_lck);
4644 : :
4645 : 268 : return xtime;
4646 : : }
4647 : :
4648 : : /*
4649 : : * Returns time of receipt of current chunk of XLOG data, as well as
4650 : : * whether it was received from streaming replication or from archives.
4651 : : */
4652 : : void
4653 : 28 : GetXLogReceiptTime(TimestampTz *rtime, bool *fromStream)
4654 : : {
4655 : : /*
4656 : : * This must be executed in the startup process, since we don't export the
4657 : : * relevant state to shared memory.
4658 : : */
4659 : : Assert(InRecovery);
4660 : :
4661 : 28 : *rtime = XLogReceiptTime;
4662 : 28 : *fromStream = (XLogReceiptSource == XLOG_FROM_STREAM);
4663 : 28 : }
4664 : :
4665 : : /*
4666 : : * Note that text field supplied is a parameter name and does not require
4667 : : * translation
4668 : : */
4669 : : void
4670 : 730 : RecoveryRequiresIntParameter(const char *param_name, int currValue, int minValue)
4671 : : {
4672 [ - + ]: 730 : if (currValue < minValue)
4673 : : {
4674 [ # # ]: 0 : if (HotStandbyActiveInReplay())
4675 : : {
4676 : 0 : bool warned_for_promote = false;
4677 : :
4678 [ # # ]: 0 : ereport(WARNING,
4679 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4680 : : errmsg("hot standby is not possible because of insufficient parameter settings"),
4681 : : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4682 : : param_name,
4683 : : currValue,
4684 : : minValue)));
4685 : :
4686 : 0 : SetRecoveryPause(true);
4687 : :
4688 [ # # ]: 0 : ereport(LOG,
4689 : : (errmsg("recovery has paused"),
4690 : : errdetail("If recovery is unpaused, the server will shut down."),
4691 : : errhint("You can then restart the server after making the necessary configuration changes.")));
4692 : :
4693 [ # # ]: 0 : while (GetRecoveryPauseState() != RECOVERY_NOT_PAUSED)
4694 : : {
4695 : 0 : ProcessStartupProcInterrupts();
4696 : :
4697 [ # # ]: 0 : if (CheckForStandbyTrigger())
4698 : : {
4699 [ # # ]: 0 : if (!warned_for_promote)
4700 [ # # ]: 0 : ereport(WARNING,
4701 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4702 : : errmsg("promotion is not possible because of insufficient parameter settings"),
4703 : :
4704 : : /*
4705 : : * Repeat the detail from above so it's easy to find
4706 : : * in the log.
4707 : : */
4708 : : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4709 : : param_name,
4710 : : currValue,
4711 : : minValue),
4712 : : errhint("Restart the server after making the necessary configuration changes.")));
4713 : 0 : warned_for_promote = true;
4714 : : }
4715 : :
4716 : : /*
4717 : : * If recovery pause is requested then set it paused. While
4718 : : * we are in the loop, user might resume and pause again so
4719 : : * set this every time.
4720 : : */
4721 : 0 : ConfirmRecoveryPaused();
4722 : :
4723 : : /*
4724 : : * We wait on a condition variable that will wake us as soon
4725 : : * as the pause ends, but we use a timeout so we can check the
4726 : : * above conditions periodically too.
4727 : : */
4728 : 0 : ConditionVariableTimedSleep(&XLogRecoveryCtl->recoveryNotPausedCV, 1000,
4729 : : WAIT_EVENT_RECOVERY_PAUSE);
4730 : : }
4731 : 0 : ConditionVariableCancelSleep();
4732 : : }
4733 : :
4734 [ # # ]: 0 : ereport(FATAL,
4735 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4736 : : errmsg("recovery aborted because of insufficient parameter settings"),
4737 : : /* Repeat the detail from above so it's easy to find in the log. */
4738 : : errdetail("%s = %d is a lower setting than on the primary server, where its value was %d.",
4739 : : param_name,
4740 : : currValue,
4741 : : minValue),
4742 : : errhint("You can restart the server after making the necessary configuration changes.")));
4743 : : }
4744 : 730 : }
4745 : :
4746 : :
4747 : : /*
4748 : : * GUC check_hook for primary_slot_name
4749 : : */
4750 : : bool
4751 : 1479 : check_primary_slot_name(char **newval, void **extra, GucSource source)
4752 : : {
4753 : : int err_code;
4754 : 1479 : char *err_msg = NULL;
4755 : 1479 : char *err_hint = NULL;
4756 : :
4757 [ + - + + ]: 1479 : if (*newval && strcmp(*newval, "") != 0 &&
4758 [ - + ]: 188 : !ReplicationSlotValidateNameInternal(*newval, false, &err_code,
4759 : : &err_msg, &err_hint))
4760 : : {
4761 : 0 : GUC_check_errcode(err_code);
4762 : 0 : GUC_check_errdetail("%s", err_msg);
4763 [ # # ]: 0 : if (err_hint != NULL)
4764 : 0 : GUC_check_errhint("%s", err_hint);
4765 : 0 : return false;
4766 : : }
4767 : :
4768 : 1479 : return true;
4769 : : }
4770 : :
4771 : : /*
4772 : : * Recovery target settings: Only one of the several recovery_target* settings
4773 : : * may be set. Setting a second one results in an error. The global variable
4774 : : * recoveryTarget tracks which kind of recovery target was chosen. Other
4775 : : * variables store the actual target value (for example a string or a xid).
4776 : : * The assign functions of the parameters check whether a competing parameter
4777 : : * was already set. But we want to allow setting the same parameter multiple
4778 : : * times. We also want to allow unsetting a parameter and setting a different
4779 : : * one, so we unset recoveryTarget when the parameter is set to an empty
4780 : : * string.
4781 : : *
4782 : : * XXX this code is broken by design. Throwing an error from a GUC assign
4783 : : * hook breaks fundamental assumptions of guc.c. So long as all the variables
4784 : : * for which this can happen are PGC_POSTMASTER, the consequences are limited,
4785 : : * since we'd just abort postmaster startup anyway. Nonetheless it's likely
4786 : : * that we have odd behaviors such as unexpected GUC ordering dependencies.
4787 : : */
4788 : :
4789 : : pg_noreturn static void
4790 : 1 : error_multiple_recovery_targets(void)
4791 : : {
4792 [ + - ]: 1 : ereport(ERROR,
4793 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4794 : : errmsg("multiple recovery targets specified"),
4795 : : errdetail("At most one of \"recovery_target\", \"recovery_target_lsn\", \"recovery_target_name\", \"recovery_target_time\", \"recovery_target_xid\" may be set.")));
4796 : : }
4797 : :
4798 : : /*
4799 : : * GUC check_hook for recovery_target
4800 : : */
4801 : : bool
4802 : 1292 : check_recovery_target(char **newval, void **extra, GucSource source)
4803 : : {
4804 [ + + - + ]: 1292 : if (strcmp(*newval, "immediate") != 0 && strcmp(*newval, "") != 0)
4805 : : {
4806 : 0 : GUC_check_errdetail("The only allowed value is \"immediate\".");
4807 : 0 : return false;
4808 : : }
4809 : 1292 : return true;
4810 : : }
4811 : :
4812 : : /*
4813 : : * GUC assign_hook for recovery_target
4814 : : */
4815 : : void
4816 : 1292 : assign_recovery_target(const char *newval, void *extra)
4817 : : {
4818 [ - + ]: 1292 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4819 [ # # ]: 0 : recoveryTarget != RECOVERY_TARGET_IMMEDIATE)
4820 : 0 : error_multiple_recovery_targets();
4821 : :
4822 [ + - + + ]: 1292 : if (newval && strcmp(newval, "") != 0)
4823 : 1 : recoveryTarget = RECOVERY_TARGET_IMMEDIATE;
4824 : : else
4825 : 1291 : recoveryTarget = RECOVERY_TARGET_UNSET;
4826 : 1292 : }
4827 : :
4828 : : /*
4829 : : * GUC check_hook for recovery_target_lsn
4830 : : */
4831 : : bool
4832 : 1298 : check_recovery_target_lsn(char **newval, void **extra, GucSource source)
4833 : : {
4834 [ + + ]: 1298 : if (strcmp(*newval, "") != 0)
4835 : : {
4836 : : XLogRecPtr lsn;
4837 : : XLogRecPtr *myextra;
4838 : 8 : ErrorSaveContext escontext = {T_ErrorSaveContext};
4839 : :
4840 : 8 : lsn = pg_lsn_in_safe(*newval, (Node *) &escontext);
4841 [ - + ]: 8 : if (escontext.error_occurred)
4842 : 0 : return false;
4843 : :
4844 : 8 : myextra = (XLogRecPtr *) guc_malloc(LOG, sizeof(XLogRecPtr));
4845 [ - + ]: 8 : if (!myextra)
4846 : 0 : return false;
4847 : 8 : *myextra = lsn;
4848 : 8 : *extra = myextra;
4849 : : }
4850 : 1298 : return true;
4851 : : }
4852 : :
4853 : : /*
4854 : : * GUC assign_hook for recovery_target_lsn
4855 : : */
4856 : : void
4857 : 1298 : assign_recovery_target_lsn(const char *newval, void *extra)
4858 : : {
4859 [ - + ]: 1298 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4860 [ # # ]: 0 : recoveryTarget != RECOVERY_TARGET_LSN)
4861 : 0 : error_multiple_recovery_targets();
4862 : :
4863 [ + - + + ]: 1298 : if (newval && strcmp(newval, "") != 0)
4864 : : {
4865 : 8 : recoveryTarget = RECOVERY_TARGET_LSN;
4866 : 8 : recoveryTargetLSN = *((XLogRecPtr *) extra);
4867 : : }
4868 : : else
4869 : 1290 : recoveryTarget = RECOVERY_TARGET_UNSET;
4870 : 1298 : }
4871 : :
4872 : : /*
4873 : : * GUC check_hook for recovery_target_name
4874 : : */
4875 : : bool
4876 : 1298 : check_recovery_target_name(char **newval, void **extra, GucSource source)
4877 : : {
4878 : : /* Use the value of newval directly */
4879 [ - + ]: 1298 : if (strlen(*newval) >= MAXFNAMELEN)
4880 : : {
4881 : 0 : GUC_check_errdetail("\"%s\" is too long (maximum %d characters).",
4882 : : "recovery_target_name", MAXFNAMELEN - 1);
4883 : 0 : return false;
4884 : : }
4885 : 1298 : return true;
4886 : : }
4887 : :
4888 : : /*
4889 : : * GUC assign_hook for recovery_target_name
4890 : : */
4891 : : void
4892 : 1298 : assign_recovery_target_name(const char *newval, void *extra)
4893 : : {
4894 [ - + ]: 1298 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4895 [ # # ]: 0 : recoveryTarget != RECOVERY_TARGET_NAME)
4896 : 0 : error_multiple_recovery_targets();
4897 : :
4898 [ + - + + ]: 1298 : if (newval && strcmp(newval, "") != 0)
4899 : : {
4900 : 6 : recoveryTarget = RECOVERY_TARGET_NAME;
4901 : 6 : recoveryTargetName = newval;
4902 : : }
4903 : : else
4904 : 1292 : recoveryTarget = RECOVERY_TARGET_UNSET;
4905 : 1298 : }
4906 : :
4907 : : /*
4908 : : * GUC check_hook for recovery_target_time
4909 : : *
4910 : : * The interpretation of the recovery_target_time string can depend on the
4911 : : * time zone setting, so we need to wait until after all GUC processing is
4912 : : * done before we can do the final parsing of the string. This check function
4913 : : * only does a parsing pass to catch syntax errors, but we store the string
4914 : : * and parse it again when we need to use it.
4915 : : */
4916 : : bool
4917 : 1294 : check_recovery_target_time(char **newval, void **extra, GucSource source)
4918 : : {
4919 [ + + ]: 1294 : if (strcmp(*newval, "") != 0)
4920 : : {
4921 : : /* reject some special values */
4922 [ + - ]: 3 : if (strcmp(*newval, "now") == 0 ||
4923 [ + - ]: 3 : strcmp(*newval, "today") == 0 ||
4924 [ + - ]: 3 : strcmp(*newval, "tomorrow") == 0 ||
4925 [ - + ]: 3 : strcmp(*newval, "yesterday") == 0)
4926 : : {
4927 : 0 : return false;
4928 : : }
4929 : :
4930 : : /*
4931 : : * parse timestamp value (see also timestamptz_in())
4932 : : */
4933 : : {
4934 : 3 : char *str = *newval;
4935 : : fsec_t fsec;
4936 : : struct pg_tm tt,
4937 : 3 : *tm = &tt;
4938 : : int tz;
4939 : : int dtype;
4940 : : int nf;
4941 : : int dterr;
4942 : : char *field[MAXDATEFIELDS];
4943 : : int ftype[MAXDATEFIELDS];
4944 : : char workbuf[MAXDATELEN + MAXDATEFIELDS];
4945 : : DateTimeErrorExtra dtextra;
4946 : : TimestampTz timestamp;
4947 : :
4948 : 3 : dterr = ParseDateTime(str, workbuf, sizeof(workbuf),
4949 : : field, ftype, MAXDATEFIELDS, &nf);
4950 [ + - ]: 3 : if (dterr == 0)
4951 : 3 : dterr = DecodeDateTime(field, ftype, nf,
4952 : : &dtype, tm, &fsec, &tz, &dtextra);
4953 [ - + ]: 3 : if (dterr != 0)
4954 : 0 : return false;
4955 [ - + ]: 3 : if (dtype != DTK_DATE)
4956 : 0 : return false;
4957 : :
4958 [ - + ]: 3 : if (tm2timestamp(tm, fsec, &tz, ×tamp) != 0)
4959 : : {
4960 : 0 : GUC_check_errdetail("Timestamp out of range: \"%s\".", str);
4961 : 0 : return false;
4962 : : }
4963 : : }
4964 : : }
4965 : 1294 : return true;
4966 : : }
4967 : :
4968 : : /*
4969 : : * GUC assign_hook for recovery_target_time
4970 : : */
4971 : : void
4972 : 1294 : assign_recovery_target_time(const char *newval, void *extra)
4973 : : {
4974 [ + + ]: 1294 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
4975 [ + - ]: 1 : recoveryTarget != RECOVERY_TARGET_TIME)
4976 : 1 : error_multiple_recovery_targets();
4977 : :
4978 [ + - + + ]: 1293 : if (newval && strcmp(newval, "") != 0)
4979 : 2 : recoveryTarget = RECOVERY_TARGET_TIME;
4980 : : else
4981 : 1291 : recoveryTarget = RECOVERY_TARGET_UNSET;
4982 : 1293 : }
4983 : :
4984 : : /*
4985 : : * GUC check_hook for recovery_target_timeline
4986 : : */
4987 : : bool
4988 : 1295 : check_recovery_target_timeline(char **newval, void **extra, GucSource source)
4989 : : {
4990 : : RecoveryTargetTimeLineGoal rttg;
4991 : : RecoveryTargetTimeLineGoal *myextra;
4992 : :
4993 [ - + ]: 1295 : if (strcmp(*newval, "current") == 0)
4994 : 0 : rttg = RECOVERY_TARGET_TIMELINE_CONTROLFILE;
4995 [ + + ]: 1295 : else if (strcmp(*newval, "latest") == 0)
4996 : 1292 : rttg = RECOVERY_TARGET_TIMELINE_LATEST;
4997 : : else
4998 : : {
4999 : : char *endp;
5000 : : uint64 timeline;
5001 : :
5002 : 3 : rttg = RECOVERY_TARGET_TIMELINE_NUMERIC;
5003 : :
5004 : 3 : errno = 0;
5005 : 3 : timeline = strtou64(*newval, &endp, 0);
5006 : :
5007 [ + + + - : 3 : if (*endp != '\0' || errno == EINVAL || errno == ERANGE)
- + ]
5008 : : {
5009 : 1 : GUC_check_errdetail("\"%s\" is not a valid number.",
5010 : : "recovery_target_timeline");
5011 : 3 : return false;
5012 : : }
5013 : :
5014 [ + + + - ]: 2 : if (timeline < 1 || timeline > PG_UINT32_MAX)
5015 : : {
5016 : 2 : GUC_check_errdetail("\"%s\" must be between %u and %u.",
5017 : : "recovery_target_timeline", 1, PG_UINT32_MAX);
5018 : 2 : return false;
5019 : : }
5020 : : }
5021 : :
5022 : 1292 : myextra = (RecoveryTargetTimeLineGoal *) guc_malloc(LOG, sizeof(RecoveryTargetTimeLineGoal));
5023 [ - + ]: 1292 : if (!myextra)
5024 : 0 : return false;
5025 : 1292 : *myextra = rttg;
5026 : 1292 : *extra = myextra;
5027 : :
5028 : 1292 : return true;
5029 : : }
5030 : :
5031 : : /*
5032 : : * GUC assign_hook for recovery_target_timeline
5033 : : */
5034 : : void
5035 : 1292 : assign_recovery_target_timeline(const char *newval, void *extra)
5036 : : {
5037 : 1292 : recoveryTargetTimeLineGoal = *((RecoveryTargetTimeLineGoal *) extra);
5038 [ - + ]: 1292 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
5039 : 0 : recoveryTargetTLIRequested = (TimeLineID) strtoul(newval, NULL, 0);
5040 : : else
5041 : 1292 : recoveryTargetTLIRequested = 0;
5042 : 1292 : }
5043 : :
5044 : : /*
5045 : : * GUC check_hook for recovery_target_xid
5046 : : */
5047 : : bool
5048 : 1292 : check_recovery_target_xid(char **newval, void **extra, GucSource source)
5049 : : {
5050 [ + + ]: 1292 : if (strcmp(*newval, "") != 0)
5051 : : {
5052 : : TransactionId xid;
5053 : : TransactionId *myextra;
5054 : : char *endp;
5055 : : char *val;
5056 : :
5057 : 1 : errno = 0;
5058 : :
5059 : : /*
5060 : : * Consume leading whitespace to determine if number is negative
5061 : : */
5062 : 1 : val = *newval;
5063 : :
5064 [ - + ]: 1 : while (isspace((unsigned char) *val))
5065 : 0 : val++;
5066 : :
5067 : : /*
5068 : : * This cast will remove the epoch, if any
5069 : : */
5070 : 1 : xid = (TransactionId) strtou64(val, &endp, 0);
5071 : :
5072 [ + - + - : 1 : if (*endp != '\0' || errno == EINVAL || errno == ERANGE || *val == '-')
+ - - + ]
5073 : : {
5074 : 0 : GUC_check_errdetail("\"%s\" is not a valid number.",
5075 : : "recovery_target_xid");
5076 : 0 : return false;
5077 : : }
5078 : :
5079 [ - + ]: 1 : if (xid < FirstNormalTransactionId)
5080 : : {
5081 : 0 : GUC_check_errdetail("\"%s\" without epoch must be greater than or equal to %u.",
5082 : : "recovery_target_xid",
5083 : : FirstNormalTransactionId);
5084 : 0 : return false;
5085 : : }
5086 : :
5087 : 1 : myextra = (TransactionId *) guc_malloc(LOG, sizeof(TransactionId));
5088 [ - + ]: 1 : if (!myextra)
5089 : 0 : return false;
5090 : 1 : *myextra = xid;
5091 : 1 : *extra = myextra;
5092 : : }
5093 : 1292 : return true;
5094 : : }
5095 : :
5096 : : /*
5097 : : * GUC assign_hook for recovery_target_xid
5098 : : */
5099 : : void
5100 : 1292 : assign_recovery_target_xid(const char *newval, void *extra)
5101 : : {
5102 [ - + ]: 1292 : if (recoveryTarget != RECOVERY_TARGET_UNSET &&
5103 [ # # ]: 0 : recoveryTarget != RECOVERY_TARGET_XID)
5104 : 0 : error_multiple_recovery_targets();
5105 : :
5106 [ + - + + ]: 1292 : if (newval && strcmp(newval, "") != 0)
5107 : : {
5108 : 1 : recoveryTarget = RECOVERY_TARGET_XID;
5109 : 1 : recoveryTargetXid = *((TransactionId *) extra);
5110 : : }
5111 : : else
5112 : 1291 : recoveryTarget = RECOVERY_TARGET_UNSET;
5113 : 1292 : }
|
