summaryrefslogtreecommitdiffstats
path: root/contrib/pg_prewarm/autoprewarm.c
blob: 1701da9dff4dee2e78ef95cee9a33b57aebd7878 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
/*-------------------------------------------------------------------------
 *
 * autoprewarm.c
 *		Periodically dump information about the blocks present in
 *		shared_buffers, and reload them on server restart.
 *
 *		Due to locking considerations, we can't actually begin prewarming
 *		until the server reaches a consistent state.  We need the catalogs
 *		to be consistent so that we can figure out which relation to lock,
 *		and we need to lock the relations so that we don't try to prewarm
 *		pages from a relation that is in the process of being dropped.
 *
 *		While prewarming, autoprewarm will use two workers.  There's a
 *		master worker that reads and sorts the list of blocks to be
 *		prewarmed and then launches a per-database worker for each
 *		relevant database in turn.  The former keeps running after the
 *		initial prewarm is complete to update the dump file periodically.
 *
 *	Copyright (c) 2016-2020, PostgreSQL Global Development Group
 *
 *	IDENTIFICATION
 *		contrib/pg_prewarm/autoprewarm.c
 *
 *-------------------------------------------------------------------------
 */

#include "postgres.h"

#include <unistd.h>

#include "access/relation.h"
#include "access/xact.h"
#include "catalog/pg_class.h"
#include "catalog/pg_type.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "postmaster/bgworker.h"
#include "storage/buf_internals.h"
#include "storage/dsm.h"
#include "storage/ipc.h"
#include "storage/latch.h"
#include "storage/lwlock.h"
#include "storage/proc.h"
#include "storage/procsignal.h"
#include "storage/shmem.h"
#include "storage/smgr.h"
#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/guc.h"
#include "utils/memutils.h"
#include "utils/rel.h"
#include "utils/relfilenodemap.h"
#include "utils/resowner.h"

#define AUTOPREWARM_FILE "autoprewarm.blocks"

/* Metadata for each block we dump. */
typedef struct BlockInfoRecord
{
	Oid			database;
	Oid			tablespace;
	Oid			filenode;
	ForkNumber	forknum;
	BlockNumber blocknum;
} BlockInfoRecord;

/* Shared state information for autoprewarm bgworker. */
typedef struct AutoPrewarmSharedState
{
	LWLock		lock;			/* mutual exclusion */
	pid_t		bgworker_pid;	/* for main bgworker */
	pid_t		pid_using_dumpfile; /* for autoprewarm or block dump */

	/* Following items are for communication with per-database worker */
	dsm_handle	block_info_handle;
	Oid			database;
	int			prewarm_start_idx;
	int			prewarm_stop_idx;
	int			prewarmed_blocks;
} AutoPrewarmSharedState;

void		_PG_init(void);
void		autoprewarm_main(Datum main_arg);
void		autoprewarm_database_main(Datum main_arg);

PG_FUNCTION_INFO_V1(autoprewarm_start_worker);
PG_FUNCTION_INFO_V1(autoprewarm_dump_now);

static void apw_load_buffers(void);
static int	apw_dump_now(bool is_bgworker, bool dump_unlogged);
static void apw_start_master_worker(void);
static void apw_start_database_worker(void);
static bool apw_init_shmem(void);
static void apw_detach_shmem(int code, Datum arg);
static int	apw_compare_blockinfo(const void *p, const void *q);
static void apw_sigterm_handler(SIGNAL_ARGS);
static void apw_sighup_handler(SIGNAL_ARGS);

/* Flags set by signal handlers */
static volatile sig_atomic_t got_sigterm = false;
static volatile sig_atomic_t got_sighup = false;

/* Pointer to shared-memory state. */
static AutoPrewarmSharedState *apw_state = NULL;

/* GUC variables. */
static bool autoprewarm = true; /* start worker? */
static int	autoprewarm_interval;	/* dump interval */

/*
 * Module load callback.
 */
void
_PG_init(void)
{
	DefineCustomIntVariable("pg_prewarm.autoprewarm_interval",
							"Sets the interval between dumps of shared buffers",
							"If set to zero, time-based dumping is disabled.",
							&autoprewarm_interval,
							300,
							0, INT_MAX / 1000,
							PGC_SIGHUP,
							GUC_UNIT_S,
							NULL,
							NULL,
							NULL);

	if (!process_shared_preload_libraries_in_progress)
		return;

	/* can't define PGC_POSTMASTER variable after startup */
	DefineCustomBoolVariable("pg_prewarm.autoprewarm",
							 "Starts the autoprewarm worker.",
							 NULL,
							 &autoprewarm,
							 true,
							 PGC_POSTMASTER,
							 0,
							 NULL,
							 NULL,
							 NULL);

	EmitWarningsOnPlaceholders("pg_prewarm");

	RequestAddinShmemSpace(MAXALIGN(sizeof(AutoPrewarmSharedState)));

	/* Register autoprewarm worker, if enabled. */
	if (autoprewarm)
		apw_start_master_worker();
}

/*
 * Main entry point for the master autoprewarm process.  Per-database workers
 * have a separate entry point.
 */
void
autoprewarm_main(Datum main_arg)
{
	bool		first_time = true;
	bool		final_dump_allowed = true;
	TimestampTz last_dump_time = 0;

	/* Establish signal handlers; once that's done, unblock signals. */
	pqsignal(SIGTERM, apw_sigterm_handler);
	pqsignal(SIGHUP, apw_sighup_handler);
	pqsignal(SIGUSR1, procsignal_sigusr1_handler);
	BackgroundWorkerUnblockSignals();

	/* Create (if necessary) and attach to our shared memory area. */
	if (apw_init_shmem())
		first_time = false;

	/* Set on-detach hook so that our PID will be cleared on exit. */
	on_shmem_exit(apw_detach_shmem, 0);

	/*
	 * Store our PID in the shared memory area --- unless there's already
	 * another worker running, in which case just exit.
	 */
	LWLockAcquire(&apw_state->lock, LW_EXCLUSIVE);
	if (apw_state->bgworker_pid != InvalidPid)
	{
		LWLockRelease(&apw_state->lock);
		ereport(LOG,
				(errmsg("autoprewarm worker is already running under PID %lu",
						(unsigned long) apw_state->bgworker_pid)));
		return;
	}
	apw_state->bgworker_pid = MyProcPid;
	LWLockRelease(&apw_state->lock);

	/*
	 * Preload buffers from the dump file only if we just created the shared
	 * memory region.  Otherwise, it's either already been done or shouldn't
	 * be done - e.g. because the old dump file has been overwritten since the
	 * server was started.
	 *
	 * There's not much point in performing a dump immediately after we finish
	 * preloading; so, if we do end up preloading, consider the last dump time
	 * to be equal to the current time.
	 *
	 * If apw_load_buffers() is terminated early by a shutdown request,
	 * prevent dumping out our state below the loop, because we'd effectively
	 * just truncate the saved state to however much we'd managed to preload.
	 */
	if (first_time)
	{
		apw_load_buffers();
		final_dump_allowed = !got_sigterm;
		last_dump_time = GetCurrentTimestamp();
	}

	/* Periodically dump buffers until terminated. */
	while (!got_sigterm)
	{
		/* In case of a SIGHUP, just reload the configuration. */
		if (got_sighup)
		{
			got_sighup = false;
			ProcessConfigFile(PGC_SIGHUP);
		}

		if (autoprewarm_interval <= 0)
		{
			/* We're only dumping at shutdown, so just wait forever. */
			(void) WaitLatch(&MyProc->procLatch,
							 WL_LATCH_SET | WL_EXIT_ON_PM_DEATH,
							 -1L,
							 PG_WAIT_EXTENSION);
		}
		else
		{
			TimestampTz next_dump_time;
			long		delay_in_ms;

			/* Compute the next dump time. */
			next_dump_time =
				TimestampTzPlusMilliseconds(last_dump_time,
											autoprewarm_interval * 1000);
			delay_in_ms =
				TimestampDifferenceMilliseconds(GetCurrentTimestamp(),
												next_dump_time);

			/* Perform a dump if it's time. */
			if (delay_in_ms <= 0)
			{
				last_dump_time = GetCurrentTimestamp();
				apw_dump_now(true, false);
				continue;
			}

			/* Sleep until the next dump time. */
			(void) WaitLatch(&MyProc->procLatch,
							 WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
							 delay_in_ms,
							 PG_WAIT_EXTENSION);
		}

		/* Reset the latch, loop. */
		ResetLatch(&MyProc->procLatch);
	}

	/*
	 * Dump one last time.  We assume this is probably the result of a system
	 * shutdown, although it's possible that we've merely been terminated.
	 */
	if (final_dump_allowed)
		apw_dump_now(true, true);
}

/*
 * Read the dump file and launch per-database workers one at a time to
 * prewarm the buffers found there.
 */
static void
apw_load_buffers(void)
{
	FILE	   *file = NULL;
	int			num_elements,
				i;
	BlockInfoRecord *blkinfo;
	dsm_segment *seg;

	/*
	 * Skip the prewarm if the dump file is in use; otherwise, prevent any
	 * other process from writing it while we're using it.
	 */
	LWLockAcquire(&apw_state->lock, LW_EXCLUSIVE);
	if (apw_state->pid_using_dumpfile == InvalidPid)
		apw_state->pid_using_dumpfile = MyProcPid;
	else
	{
		LWLockRelease(&apw_state->lock);
		ereport(LOG,
				(errmsg("skipping prewarm because block dump file is being written by PID %lu",
						(unsigned long) apw_state->pid_using_dumpfile)));
		return;
	}
	LWLockRelease(&apw_state->lock);

	/*
	 * Open the block dump file.  Exit quietly if it doesn't exist, but report
	 * any other error.
	 */
	file = AllocateFile(AUTOPREWARM_FILE, "r");
	if (!file)
	{
		if (errno == ENOENT)
		{
			LWLockAcquire(&apw_state->lock, LW_EXCLUSIVE);
			apw_state->pid_using_dumpfile = InvalidPid;
			LWLockRelease(&apw_state->lock);
			return;				/* No file to load. */
		}
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not read file \"%s\": %m",
						AUTOPREWARM_FILE)));
	}

	/* First line of the file is a record count. */
	if (fscanf(file, "<<%d>>\n", &num_elements) != 1)
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not read from file \"%s\": %m",
						AUTOPREWARM_FILE)));

	/* Allocate a dynamic shared memory segment to store the record data. */
	seg = dsm_create(sizeof(BlockInfoRecord) * num_elements, 0);
	blkinfo = (BlockInfoRecord *) dsm_segment_address(seg);

	/* Read records, one per line. */
	for (i = 0; i < num_elements; i++)
	{
		unsigned	forknum;

		if (fscanf(file, "%u,%u,%u,%u,%u\n", &blkinfo[i].database,
				   &blkinfo[i].tablespace, &blkinfo[i].filenode,
				   &forknum, &blkinfo[i].blocknum) != 5)
			ereport(ERROR,
					(errmsg("autoprewarm block dump file is corrupted at line %d",
							i + 1)));
		blkinfo[i].forknum = forknum;
	}

	FreeFile(file);

	/* Sort the blocks to be loaded. */
	pg_qsort(blkinfo, num_elements, sizeof(BlockInfoRecord),
			 apw_compare_blockinfo);

	/* Populate shared memory state. */
	apw_state->block_info_handle = dsm_segment_handle(seg);
	apw_state->prewarm_start_idx = apw_state->prewarm_stop_idx = 0;
	apw_state->prewarmed_blocks = 0;

	/* Get the info position of the first block of the next database. */
	while (apw_state->prewarm_start_idx < num_elements)
	{
		int			j = apw_state->prewarm_start_idx;
		Oid			current_db = blkinfo[j].database;

		/*
		 * Advance the prewarm_stop_idx to the first BlockInfoRecord that does
		 * not belong to this database.
		 */
		j++;
		while (j < num_elements)
		{
			if (current_db != blkinfo[j].database)
			{
				/*
				 * Combine BlockInfoRecords for global objects with those of
				 * the database.
				 */
				if (current_db != InvalidOid)
					break;
				current_db = blkinfo[j].database;
			}

			j++;
		}

		/*
		 * If we reach this point with current_db == InvalidOid, then only
		 * BlockInfoRecords belonging to global objects exist.  We can't
		 * prewarm without a database connection, so just bail out.
		 */
		if (current_db == InvalidOid)
			break;

		/* Configure stop point and database for next per-database worker. */
		apw_state->prewarm_stop_idx = j;
		apw_state->database = current_db;
		Assert(apw_state->prewarm_start_idx < apw_state->prewarm_stop_idx);

		/* If we've run out of free buffers, don't launch another worker. */
		if (!have_free_buffer())
			break;

		/*
		 * Likewise, don't launch if we've already been told to shut down.
		 * (The launch would fail anyway, but we might as well skip it.)
		 */
		if (got_sigterm)
			break;

		/*
		 * Start a per-database worker to load blocks for this database; this
		 * function will return once the per-database worker exits.
		 */
		apw_start_database_worker();

		/* Prepare for next database. */
		apw_state->prewarm_start_idx = apw_state->prewarm_stop_idx;
	}

	/* Clean up. */
	dsm_detach(seg);
	LWLockAcquire(&apw_state->lock, LW_EXCLUSIVE);
	apw_state->block_info_handle = DSM_HANDLE_INVALID;
	apw_state->pid_using_dumpfile = InvalidPid;
	LWLockRelease(&apw_state->lock);

	/* Report our success, if we were able to finish. */
	if (!got_sigterm)
		ereport(LOG,
				(errmsg("autoprewarm successfully prewarmed %d of %d previously-loaded blocks",
						apw_state->prewarmed_blocks, num_elements)));
}

/*
 * Prewarm all blocks for one database (and possibly also global objects, if
 * those got grouped with this database).
 */
void
autoprewarm_database_main(Datum main_arg)
{
	int			pos;
	BlockInfoRecord *block_info;
	Relation	rel = NULL;
	BlockNumber nblocks = 0;
	BlockInfoRecord *old_blk = NULL;
	dsm_segment *seg;

	/* Establish signal handlers; once that's done, unblock signals. */
	pqsignal(SIGTERM, die);
	BackgroundWorkerUnblockSignals();

	/* Connect to correct database and get block information. */
	apw_init_shmem();
	seg = dsm_attach(apw_state->block_info_handle);
	if (seg == NULL)
		ereport(ERROR,
				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
				 errmsg("could not map dynamic shared memory segment")));
	BackgroundWorkerInitializeConnectionByOid(apw_state->database, InvalidOid, 0);
	block_info = (BlockInfoRecord *) dsm_segment_address(seg);
	pos = apw_state->prewarm_start_idx;

	/*
	 * Loop until we run out of blocks to prewarm or until we run out of free
	 * buffers.
	 */
	while (pos < apw_state->prewarm_stop_idx && have_free_buffer())
	{
		BlockInfoRecord *blk = &block_info[pos++];
		Buffer		buf;

		CHECK_FOR_INTERRUPTS();

		/*
		 * Quit if we've reached records for another database. If previous
		 * blocks are of some global objects, then continue pre-warming.
		 */
		if (old_blk != NULL && old_blk->database != blk->database &&
			old_blk->database != 0)
			break;

		/*
		 * As soon as we encounter a block of a new relation, close the old
		 * relation. Note that rel will be NULL if try_relation_open failed
		 * previously; in that case, there is nothing to close.
		 */
		if (old_blk != NULL && old_blk->filenode != blk->filenode &&
			rel != NULL)
		{
			relation_close(rel, AccessShareLock);
			rel = NULL;
			CommitTransactionCommand();
		}

		/*
		 * Try to open each new relation, but only once, when we first
		 * encounter it. If it's been dropped, skip the associated blocks.
		 */
		if (old_blk == NULL || old_blk->filenode != blk->filenode)
		{
			Oid			reloid;

			Assert(rel == NULL);
			StartTransactionCommand();
			reloid = RelidByRelfilenode(blk->tablespace, blk->filenode);
			if (OidIsValid(reloid))
				rel = try_relation_open(reloid, AccessShareLock);

			if (!rel)
				CommitTransactionCommand();
		}
		if (!rel)
		{
			old_blk = blk;
			continue;
		}

		/* Once per fork, check for fork existence and size. */
		if (old_blk == NULL ||
			old_blk->filenode != blk->filenode ||
			old_blk->forknum != blk->forknum)
		{
			RelationOpenSmgr(rel);

			/*
			 * smgrexists is not safe for illegal forknum, hence check whether
			 * the passed forknum is valid before using it in smgrexists.
			 */
			if (blk->forknum > InvalidForkNumber &&
				blk->forknum <= MAX_FORKNUM &&
				smgrexists(rel->rd_smgr, blk->forknum))
				nblocks = RelationGetNumberOfBlocksInFork(rel, blk->forknum);
			else
				nblocks = 0;
		}

		/* Check whether blocknum is valid and within fork file size. */
		if (blk->blocknum >= nblocks)
		{
			/* Move to next forknum. */
			old_blk = blk;
			continue;
		}

		/* Prewarm buffer. */
		buf = ReadBufferExtended(rel, blk->forknum, blk->blocknum, RBM_NORMAL,
								 NULL);
		if (BufferIsValid(buf))
		{
			apw_state->prewarmed_blocks++;
			ReleaseBuffer(buf);
		}

		old_blk = blk;
	}

	dsm_detach(seg);

	/* Release lock on previous relation. */
	if (rel)
	{
		relation_close(rel, AccessShareLock);
		CommitTransactionCommand();
	}
}

/*
 * Dump information on blocks in shared buffers.  We use a text format here
 * so that it's easy to understand and even change the file contents if
 * necessary.
 * Returns the number of blocks dumped.
 */
static int
apw_dump_now(bool is_bgworker, bool dump_unlogged)
{
	int			num_blocks;
	int			i;
	int			ret;
	BlockInfoRecord *block_info_array;
	BufferDesc *bufHdr;
	FILE	   *file;
	char		transient_dump_file_path[MAXPGPATH];
	pid_t		pid;

	LWLockAcquire(&apw_state->lock, LW_EXCLUSIVE);
	pid = apw_state->pid_using_dumpfile;
	if (apw_state->pid_using_dumpfile == InvalidPid)
		apw_state->pid_using_dumpfile = MyProcPid;
	LWLockRelease(&apw_state->lock);

	if (pid != InvalidPid)
	{
		if (!is_bgworker)
			ereport(ERROR,
					(errmsg("could not perform block dump because dump file is being used by PID %lu",
							(unsigned long) apw_state->pid_using_dumpfile)));

		ereport(LOG,
				(errmsg("skipping block dump because it is already being performed by PID %lu",
						(unsigned long) apw_state->pid_using_dumpfile)));
		return 0;
	}

	block_info_array =
		(BlockInfoRecord *) palloc(sizeof(BlockInfoRecord) * NBuffers);

	for (num_blocks = 0, i = 0; i < NBuffers; i++)
	{
		uint32		buf_state;

		CHECK_FOR_INTERRUPTS();

		bufHdr = GetBufferDescriptor(i);

		/* Lock each buffer header before inspecting. */
		buf_state = LockBufHdr(bufHdr);

		/*
		 * Unlogged tables will be automatically truncated after a crash or
		 * unclean shutdown. In such cases we need not prewarm them. Dump them
		 * only if requested by caller.
		 */
		if (buf_state & BM_TAG_VALID &&
			((buf_state & BM_PERMANENT) || dump_unlogged))
		{
			block_info_array[num_blocks].database = bufHdr->tag.rnode.dbNode;
			block_info_array[num_blocks].tablespace = bufHdr->tag.rnode.spcNode;
			block_info_array[num_blocks].filenode = bufHdr->tag.rnode.relNode;
			block_info_array[num_blocks].forknum = bufHdr->tag.forkNum;
			block_info_array[num_blocks].blocknum = bufHdr->tag.blockNum;
			++num_blocks;
		}

		UnlockBufHdr(bufHdr, buf_state);
	}

	snprintf(transient_dump_file_path, MAXPGPATH, "%s.tmp", AUTOPREWARM_FILE);
	file = AllocateFile(transient_dump_file_path, "w");
	if (!file)
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not open file \"%s\": %m",
						transient_dump_file_path)));

	ret = fprintf(file, "<<%d>>\n", num_blocks);
	if (ret < 0)
	{
		int			save_errno = errno;

		FreeFile(file);
		unlink(transient_dump_file_path);
		errno = save_errno;
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not write to file \"%s\": %m",
						transient_dump_file_path)));
	}

	for (i = 0; i < num_blocks; i++)
	{
		CHECK_FOR_INTERRUPTS();

		ret = fprintf(file, "%u,%u,%u,%u,%u\n",
					  block_info_array[i].database,
					  block_info_array[i].tablespace,
					  block_info_array[i].filenode,
					  (uint32) block_info_array[i].forknum,
					  block_info_array[i].blocknum);
		if (ret < 0)
		{
			int			save_errno = errno;

			FreeFile(file);
			unlink(transient_dump_file_path);
			errno = save_errno;
			ereport(ERROR,
					(errcode_for_file_access(),
					 errmsg("could not write to file \"%s\": %m",
							transient_dump_file_path)));
		}
	}

	pfree(block_info_array);

	/*
	 * Rename transient_dump_file_path to AUTOPREWARM_FILE to make things
	 * permanent.
	 */
	ret = FreeFile(file);
	if (ret != 0)
	{
		int			save_errno = errno;

		unlink(transient_dump_file_path);
		errno = save_errno;
		ereport(ERROR,
				(errcode_for_file_access(),
				 errmsg("could not close file \"%s\": %m",
						transient_dump_file_path)));
	}

	(void) durable_rename(transient_dump_file_path, AUTOPREWARM_FILE, ERROR);
	apw_state->pid_using_dumpfile = InvalidPid;

	ereport(DEBUG1,
			(errmsg("wrote block details for %d blocks", num_blocks)));
	return num_blocks;
}

/*
 * SQL-callable function to launch autoprewarm.
 */
Datum
autoprewarm_start_worker(PG_FUNCTION_ARGS)
{
	pid_t		pid;

	if (!autoprewarm)
		ereport(ERROR,
				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
				 errmsg("autoprewarm is disabled")));

	apw_init_shmem();
	LWLockAcquire(&apw_state->lock, LW_EXCLUSIVE);
	pid = apw_state->bgworker_pid;
	LWLockRelease(&apw_state->lock);

	if (pid != InvalidPid)
		ereport(ERROR,
				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
				 errmsg("autoprewarm worker is already running under PID %lu",
						(unsigned long) pid)));

	apw_start_master_worker();

	PG_RETURN_VOID();
}

/*
 * SQL-callable function to perform an immediate block dump.
 *
 * Note: this is declared to return int8, as insurance against some
 * very distant day when we might make NBuffers wider than int.
 */
Datum
autoprewarm_dump_now(PG_FUNCTION_ARGS)
{
	int			num_blocks;

	apw_init_shmem();

	PG_ENSURE_ERROR_CLEANUP(apw_detach_shmem, 0);
	{
		num_blocks = apw_dump_now(false, true);
	}
	PG_END_ENSURE_ERROR_CLEANUP(apw_detach_shmem, 0);

	PG_RETURN_INT64((int64) num_blocks);
}

/*
 * Allocate and initialize autoprewarm related shared memory, if not already
 * done, and set up backend-local pointer to that state.  Returns true if an
 * existing shared memory segment was found.
 */
static bool
apw_init_shmem(void)
{
	bool		found;

	LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
	apw_state = ShmemInitStruct("autoprewarm",
								sizeof(AutoPrewarmSharedState),
								&found);
	if (!found)
	{
		/* First time through ... */
		LWLockInitialize(&apw_state->lock, LWLockNewTrancheId());
		apw_state->bgworker_pid = InvalidPid;
		apw_state->pid_using_dumpfile = InvalidPid;
	}
	LWLockRelease(AddinShmemInitLock);

	LWLockRegisterTranche(apw_state->lock.tranche, "autoprewarm");

	return found;
}

/*
 * Clear our PID from autoprewarm shared state.
 */
static void
apw_detach_shmem(int code, Datum arg)
{
	LWLockAcquire(&apw_state->lock, LW_EXCLUSIVE);
	if (apw_state->pid_using_dumpfile == MyProcPid)
		apw_state->pid_using_dumpfile = InvalidPid;
	if (apw_state->bgworker_pid == MyProcPid)
		apw_state->bgworker_pid = InvalidPid;
	LWLockRelease(&apw_state->lock);
}

/*
 * Start autoprewarm master worker process.
 */
static void
apw_start_master_worker(void)
{
	BackgroundWorker worker;
	BackgroundWorkerHandle *handle;
	BgwHandleStatus status;
	pid_t		pid;

	MemSet(&worker, 0, sizeof(BackgroundWorker));
	worker.bgw_flags = BGWORKER_SHMEM_ACCESS;
	worker.bgw_start_time = BgWorkerStart_ConsistentState;
	strcpy(worker.bgw_library_name, "pg_prewarm");
	strcpy(worker.bgw_function_name, "autoprewarm_main");
	strcpy(worker.bgw_name, "autoprewarm master");
	strcpy(worker.bgw_type, "autoprewarm master");

	if (process_shared_preload_libraries_in_progress)
	{
		RegisterBackgroundWorker(&worker);
		return;
	}

	/* must set notify PID to wait for startup */
	worker.bgw_notify_pid = MyProcPid;

	if (!RegisterDynamicBackgroundWorker(&worker, &handle))
		ereport(ERROR,
				(errcode(ERRCODE_INSUFFICIENT_RESOURCES),
				 errmsg("could not register background process"),
				 errhint("You may need to increase max_worker_processes.")));

	status = WaitForBackgroundWorkerStartup(handle, &pid);
	if (status != BGWH_STARTED)
		ereport(ERROR,
				(errcode(ERRCODE_INSUFFICIENT_RESOURCES),
				 errmsg("could not start background process"),
				 errhint("More details may be available in the server log.")));
}

/*
 * Start autoprewarm per-database worker process.
 */
static void
apw_start_database_worker(void)
{
	BackgroundWorker worker;
	BackgroundWorkerHandle *handle;

	MemSet(&worker, 0, sizeof(BackgroundWorker));
	worker.bgw_flags =
		BGWORKER_SHMEM_ACCESS | BGWORKER_BACKEND_DATABASE_CONNECTION;
	worker.bgw_start_time = BgWorkerStart_ConsistentState;
	worker.bgw_restart_time = BGW_NEVER_RESTART;
	strcpy(worker.bgw_library_name, "pg_prewarm");
	strcpy(worker.bgw_function_name, "autoprewarm_database_main");
	strcpy(worker.bgw_name, "autoprewarm worker");
	strcpy(worker.bgw_type, "autoprewarm worker");

	/* must set notify PID to wait for shutdown */
	worker.bgw_notify_pid = MyProcPid;

	if (!RegisterDynamicBackgroundWorker(&worker, &handle))
		ereport(ERROR,
				(errcode(ERRCODE_INSUFFICIENT_RESOURCES),
				 errmsg("registering dynamic bgworker autoprewarm failed"),
				 errhint("Consider increasing configuration parameter \"max_worker_processes\".")));

	/*
	 * Ignore return value; if it fails, postmaster has died, but we have
	 * checks for that elsewhere.
	 */
	WaitForBackgroundWorkerShutdown(handle);
}

/* Compare member elements to check whether they are not equal. */
#define cmp_member_elem(fld)	\
do { \
	if (a->fld < b->fld)		\
		return -1;				\
	else if (a->fld > b->fld)	\
		return 1;				\
} while(0)

/*
 * apw_compare_blockinfo
 *
 * We depend on all records for a particular database being consecutive
 * in the dump file; each per-database worker will preload blocks until
 * it sees a block for some other database.  Sorting by tablespace,
 * filenode, forknum, and blocknum isn't critical for correctness, but
 * helps us get a sequential I/O pattern.
 */
static int
apw_compare_blockinfo(const void *p, const void *q)
{
	const BlockInfoRecord *a = (const BlockInfoRecord *) p;
	const BlockInfoRecord *b = (const BlockInfoRecord *) q;

	cmp_member_elem(database);
	cmp_member_elem(tablespace);
	cmp_member_elem(filenode);
	cmp_member_elem(forknum);
	cmp_member_elem(blocknum);

	return 0;
}

/*
 * Signal handler for SIGTERM
 */
static void
apw_sigterm_handler(SIGNAL_ARGS)
{
	int			save_errno = errno;

	got_sigterm = true;

	if (MyProc)
		SetLatch(&MyProc->procLatch);

	errno = save_errno;
}

/*
 * Signal handler for SIGHUP
 */
static void
apw_sighup_handler(SIGNAL_ARGS)
{
	int			save_errno = errno;

	got_sighup = true;

	if (MyProc)
		SetLatch(&MyProc->procLatch);

	errno = save_errno;
}