summaryrefslogtreecommitdiff
path: root/btree/bt_evict.c
blob: cd4cb87bfb47302cf599402a3bc146ce72e2a983 (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
942
943
944
/*-
 * See the file LICENSE for redistribution information.
 *
 * Copyright (c) 2008-2011 WiredTiger, Inc.
 *	All rights reserved.
 *
 * $Id$
 */

#include "wt_internal.h"

static int  __wt_evict(WT_TOC *);
static int  __wt_evict_compare_lru(const void *a, const void *b);
static int  __wt_evict_compare_page(const void *a, const void *b);
static void __wt_evict_hazard_check(WT_TOC *);
static int  __wt_evict_hazard_compare(const void *a, const void *b);
static void __wt_evict_page(WT_TOC *, int);
static int  __wt_evict_page_subtrees(WT_PAGE *);
static void __wt_evict_set(WT_TOC *);
static void __wt_evict_state_check(WT_TOC *);
static int  __wt_evict_walk(WT_TOC *);
static int  __wt_evict_walk_single(WT_TOC *, IDB *, uint);
static void __wt_evict_write(WT_TOC *);

#ifdef HAVE_DIAGNOSTIC
static void __wt_evict_hazard_validate(ENV *, WT_PAGE *);
#endif

/*
 * Tuning constants -- I hesitate to call this tuning, but we should review some
 * number of pages from each file's in-memory tree for each page we evict, and
 * we should amortize the comparison of the hazard references across some number
 * of eviction candidates.
 */
#define	WT_EVICT_GROUP		10	/* Evict N pages at a time */
#define	WT_EVICT_WALK_PER_TABLE	5	/* Pages to visit per file */
#define	WT_EVICT_WALK_BASE	25	/* Pages tracked across file visits */

/*
 * WT_EVICT_FOREACH --
 *	Walk a list of eviction candidates.
 */
#define	WT_EVICT_FOREACH(cache, p, i)					\
	for ((i) = 0, (p) = (cache)->evict; (i) < WT_EVICT_GROUP; ++(i), ++(p))

/*
 * WT_EVICT_REF_CLR --
 *	Clear an eviction list entry.
 */
#define	WT_EVICT_CLR(p) do {						\
	(p)->ref = NULL;						\
	(p)->idb = WT_DEBUG_POINT;					\
} while (0)

/*
 * __wt_workq_evict_server --
 *	See if the eviction server thread needs to be awakened.
 */
void
__wt_workq_evict_server(ENV *env, int force)
{
	WT_CACHE *cache;
	uint64_t bytes_inuse, bytes_max;

	cache = env->ienv->cache;

	/* If the eviction server is running, there's nothing to do. */
	if (!cache->evict_sleeping)
		return;

	/*
	 * If we're locking out reads, or over our cache limit, or forcing the
	 * issue (when closing the environment), run the eviction server.
	 */
	bytes_inuse = __wt_cache_bytes_inuse(cache);
	bytes_max = WT_STAT(cache->stats, CACHE_BYTES_MAX);
	if (!force && !cache->read_lockout && bytes_inuse < bytes_max)
		return;

	WT_VERBOSE(env, WT_VERB_EVICT, (env,
	    "waking eviction server: force %sset, read lockout %sset, "
	    "bytes inuse %s max (%lluMB %s %lluMB), ",
	    force ? "" : "not ", cache->read_lockout ? "" : "not ",
	    bytes_inuse <= bytes_max ? "<=" : ">",
	    (unsigned long long)(bytes_inuse / WT_MEGABYTE),
	    bytes_inuse <= bytes_max ? "<=" : ">",
	    (unsigned long long)(bytes_max / WT_MEGABYTE)));

	cache->evict_sleeping = 0;
	__wt_unlock(env, cache->mtx_evict);
}

/*
 * __wt_cache_evict_server --
 *	Thread to evict pages from the cache.
 */
void *
__wt_cache_evict_server(void *arg)
{
	ENV *env;
	IENV *ienv;
	WT_CACHE *cache;
	WT_TOC *toc;
	uint64_t bytes_inuse, bytes_max;
	int ret;

	env = arg;
	ienv = env->ienv;
	cache = ienv->cache;
	ret = 0;

	/* We need a thread of control because we're reading/writing pages. */
	toc = NULL;
	WT_ERR(__wt_toc_api_set(env, "CacheReconciliation", NULL, &toc));

	/*
	 * Multiple pages are marked for eviction by the eviction server, which
	 * means nobody can read them -- but, this thread of control has to
	 * update higher pages in the tree when it writes this page, which
	 * requires reading other pages, which might themselves be marked for
	 * eviction.   Set a flag to allow this thread of control to see pages
	 * marked for eviction -- we know it's safe, because only this thread
	 * is writing pages.
	 *
	 * Reconciliation is probably running because the cache is full, which
	 * means reads are locked out -- reconciliation can read, regardless.
	 */
	F_SET(toc, WT_READ_EVICT | WT_READ_PRIORITY);

	/*
	 * Allocate memory for a copy of the hazard references -- it's a fixed
	 * size so doesn't need run-time adjustments.
	 */
	cache->hazard_elem = env->toc_size * env->hazard_size;
	WT_ERR(__wt_calloc(
	    env, cache->hazard_elem, sizeof(WT_PAGE *), &cache->hazard));
	cache->hazard_len = cache->hazard_elem * sizeof(WT_PAGE *);

	for (;;) {
		WT_VERBOSE(env,
		    WT_VERB_EVICT, (env, "eviction server sleeping"));
		cache->evict_sleeping = 1;
		__wt_lock(env, cache->mtx_evict);
		WT_VERBOSE(env,
		    WT_VERB_EVICT, (env, "eviction server waking"));

		/*
		 * Check for environment exit; do it here, instead of the top of
		 * the loop because doing it here keeps us from doing a bunch of
		 * worked when simply awakened to quit.
		 */
		if (!F_ISSET(ienv, WT_SERVER_RUN))
			break;

		for (;;) {
			/* Single-thread reconciliation. */
			__wt_lock(env, cache->mtx_reconcile);
			ret = __wt_evict(toc);
			__wt_unlock(env, cache->mtx_reconcile);
			if (ret != 0)
				goto err;

			/*
			 * If we've locked out reads, keep evicting until we
			 * get to at least 5% under the maximum cache.  Else,
			 * quit evicting as soon as we get under the maximum
			 * cache.
			 */
			bytes_inuse = __wt_cache_bytes_inuse(cache);
			bytes_max = WT_STAT(cache->stats, CACHE_BYTES_MAX);
			if (cache->read_lockout) {
				if (bytes_inuse <= bytes_max - (bytes_max / 20))
					break;
			} else if (bytes_inuse < bytes_max)
				break;
		}
	}

err:	if (cache->evict != NULL)
		__wt_free(env, cache->evict, cache->evict_len);
	if (cache->hazard != NULL)
		__wt_free(env, cache->hazard, cache->hazard_len);
	if (toc != NULL)
		WT_TRET(toc->close(toc, 0));

	if (ret != 0)
		__wt_api_env_err(env, ret, "cache eviction server error");

	WT_VERBOSE(
	    env, WT_VERB_EVICT, (env, "cache eviction server exiting"));

	return (NULL);
}

/*
 * __wt_evict --
 *	Evict pages from the cache.
 */
static int
__wt_evict(WT_TOC *toc)
{
	ENV *env;
	WT_CACHE *cache;
	WT_EVICT_LIST *evict;
	uint elem, i, j;

	env = toc->env;
	cache = env->ienv->cache;

	/* Get some more pages to consider for eviction. */
	WT_RET(__wt_evict_walk(toc));

	/*
	 * We have an array of page eviction references that may contain NULLs,
	 * as well as duplicate entries.
	 *
	 * First, sort the array by WT_REF address, then delete any duplicates.
	 * The reason is because we might evict the page but leave a duplicate
	 * entry in the "saved" area of the array, and that would be a NULL
	 * dereference on the next run.  (If someone ever tries to remove this
	 * duplicate cleanup for better performance, you can't fix it just by
	 * checking the WT_REF state -- that only works if you are discarding
	 * a page from a single level of the tree; if you are discarding a
	 * page and its parent, the duplicate of the page's WT_REF might have
	 * been free'd before a subsequent review of the eviction array.)
	 */
	evict = cache->evict;
	elem = cache->evict_elem;
	qsort(evict,
	    (size_t)elem, sizeof(WT_EVICT_LIST), __wt_evict_compare_page);
	for (i = 0; i < elem; i = j)
		for (j = i + 1; j < elem; ++j) {
			/*
			 * If the leading pointer hits a NULL, we're done, the
			 * NULLs all sorted to the top of the array.
			 */
			if (evict[j].ref == NULL)
				goto done_duplicates;

			/* Delete the second and any subsequent duplicates. */
			if (evict[i].ref == evict[j].ref)
				WT_EVICT_CLR(&evict[j]);
			else
				break;
		}
done_duplicates:

	/* Second, sort the array by LRU. */
	qsort(evict,
	    (size_t)elem, sizeof(WT_EVICT_LIST), __wt_evict_compare_lru);

	/*
	 * Discarding pages is done in 5 steps:
	 *	Set the WT_EVICT state
	 *	Check for any hazard references
	 *	Discard clean pages
	 *	Reconcile dirty pages (making them clean)
	 *	Discard clean pages
	 *
	 * The reason we release clean pages, then reconcile dirty pages, then
	 * release clean pages again is because reconciling a dirty page is a
	 * slow operation, and this releases space sooner.   (Arguably, we are
	 * going to discard all of the pages anyway, so what does it matter if
	 * we make clean pages wait for the dirty page writes?   On the other
	 * hand, it's a small change and benefits any thread waiting to read a
	 * clean page we picked for discarding, unlikely though that may be.)
	 */
	__wt_evict_set(toc);
	__wt_evict_hazard_check(toc);
	__wt_evict_state_check(toc);
	__wt_evict_page(toc, 0);
	__wt_evict_write(toc);
	__wt_evict_page(toc, 1);

	return (0);
}

/*
 * __wt_evict_walk --
 *	Fill in the array by walk the next set of pages.
 */
static int
__wt_evict_walk(WT_TOC *toc)
{
	ENV *env;
	IDB *idb;
	IENV *ienv;
	WT_CACHE *cache;
	uint elem, i;
	int ret;

	env = toc->env;
	ienv = env->ienv;
	cache = ienv->cache;

	/*
	 * Resize the array in which we're tracking pages, as necessary, then
	 * get some pages from each underlying file.  We hold a mutex for the
	 * entire time -- it's slow, but (1) how often do new files get added
	 * or removed to/from the system, and (2) it's all in-memory stuff, so
	 * it's not that slow.
	 */
	ret = 0;
	__wt_lock(env, ienv->mtx);
	elem = WT_EVICT_WALK_BASE + (ienv->dbqcnt * WT_EVICT_WALK_PER_TABLE);
	if (elem <= cache->evict_elem || (ret = __wt_realloc(env,
	    &cache->evict_len,
	    elem * sizeof(WT_EVICT_LIST), &cache->evict)) == 0) {
		cache->evict_elem = elem;

		i = WT_EVICT_WALK_BASE;
		TAILQ_FOREACH(idb, &ienv->dbqh, q) {
			if ((ret = __wt_evict_walk_single(toc, idb, i)) != 0)
				break;
			i += WT_EVICT_WALK_PER_TABLE;
		}
	}
	__wt_unlock(env, ienv->mtx);
	return (ret);
}

/*
 * __wt_evict_walk_single --
 *	Get a few page eviction candidates from a single underlying file.
 */
static int
__wt_evict_walk_single(WT_TOC *toc, IDB *idb, uint slot)
{
	WT_CACHE *cache;
	WT_EVICT_LIST *evict;
	int i, restarted_once;

	cache = toc->env->ienv->cache;

	/*
	 * Tricky little loop that restarts the walk as necessary, without
	 * resetting the count of pages retrieved.
	 */
	i = restarted_once = 0;

	/* If we haven't yet opened a tree-walk structure, do so. */
	if (idb->evict_walk.tree == NULL)
restart:	WT_RET(__wt_walk_begin(toc, &idb->root_page, &idb->evict_walk));

	/* Get the next WT_EVICT_WALK_PER_TABLE entries. */
	do {
		evict = &cache->evict[slot];
		WT_RET(__wt_walk_next(toc, &idb->evict_walk, &evict->ref));

		/*
		 * Restart the walk as necessary,  but only once (after one
		 * restart we've already acquired all of the pages, and we
		 * could loop infinitely on a tree with a single, pinned, page).
		 */
		if (evict->ref == NULL) {
			if (restarted_once++)
				break;
			goto restart;
		}

		evict->idb = idb;
		++slot;
	} while (++i < WT_EVICT_WALK_PER_TABLE);

	return (0);
}

/*
 * __wt_evict_db_clear --
 *	Remove any entries for a file from the eviction list.
 */
void
__wt_evict_db_clear(WT_TOC *toc)
{
	ENV *env;
	IDB *idb;
	IENV *ienv;
	WT_CACHE *cache;
	WT_EVICT_LIST *evict;
	uint i;

	env = toc->env;
	idb = toc->db->idb;
	ienv = env->ienv;
	cache = ienv->cache;

	/*
	 * Discard any entries in the eviction list to a file we're closing
	 * (the caller better have locked out the eviction thread).
	 */
	if (cache->evict == NULL)
		return;
	WT_EVICT_FOREACH(cache, evict, i)
		if (evict->ref != NULL && evict->idb == idb)
			WT_EVICT_CLR(evict);
}

/*
 * __wt_evict_set --
 *	Set the WT_EVICT flag on a set of pages.
 */
static void
__wt_evict_set(WT_TOC *toc)
{
	ENV *env;
	WT_CACHE *cache;
	WT_EVICT_LIST *evict;
	WT_REF *ref;
	uint i;

	env = toc->env;
	cache = env->ienv->cache;

	/*
	 * Set the entry state so readers don't try and use the pages.   Once
	 * that's done, any thread searching for a page will either see our
	 * state value, or will have already set a hazard reference to the page.
	 * We don't evict a page with a hazard reference set, so we can't race.
	 *
	 * No memory flush needed, the state field is declared volatile.
	 */
	WT_EVICT_FOREACH(cache, evict, i) {
		if ((ref = evict->ref) == NULL)
			continue;
		ref->state = WT_EVICT;
	}
}

/*
 * __wt_evict_hazard_check --
 *	Compare the list of hazard references to the list of pages to be
 *	discarded.
 */
static void
__wt_evict_hazard_check(WT_TOC *toc)
{
	ENV *env;
	IENV *ienv;
	WT_CACHE *cache;
	WT_EVICT_LIST *evict;
	WT_PAGE **hazard, **end_hazard, *page;
	WT_REF *ref;
	WT_STATS *stats;
	uint i;

	env = toc->env;
	ienv = env->ienv;
	cache = ienv->cache;
	stats = cache->stats;

	/* Sort the eviction candidates by WT_PAGE address. */
	qsort(cache->evict, (size_t)WT_EVICT_GROUP,
	    sizeof(WT_EVICT_LIST), __wt_evict_compare_page);

	/* Copy the hazard reference array and sort it by WT_PAGE address. */
	hazard = cache->hazard;
	end_hazard = hazard + cache->hazard_elem;
	memcpy(hazard, ienv->hazard, cache->hazard_elem * sizeof(WT_PAGE *));
	qsort(hazard, (size_t)cache->hazard_elem,
	    sizeof(WT_PAGE *), __wt_evict_hazard_compare);

	/* Walk the lists in parallel and look for matches. */
	WT_EVICT_FOREACH(cache, evict, i) {
		if ((ref = evict->ref) == NULL)
			continue;

		/*
		 * Look for the page in the hazard list until we reach the end
		 * of the list or find a hazard pointer larger than the page.
		 */
		for (page = ref->page;
		    hazard < end_hazard && *hazard < page; ++hazard)
			;
		if (hazard == end_hazard)
			break;

		/*
		 * If we find a matching hazard reference, the page is in use:
		 * remove it from the eviction list.
		 *
		 * No memory flush needed, the state field is declared volatile.
		 */
		if (*hazard == page) {
			WT_VERBOSE(env, WT_VERB_EVICT, (env,
			    "eviction skipped page addr %lu (hazard reference)",
			    page->addr));
			WT_STAT_INCR(stats, CACHE_EVICT_HAZARD);

			/*
			 * A page with a low LRU and a hazard reference?
			 *
			 * Set the page's LRU so we don't select it again.
			 * Return the page to service.
			 * Discard our reference.
			 */
			ref->page->read_gen = ++cache->read_gen;
			ref->state = WT_OK;
			WT_EVICT_CLR(evict);
		}
	}
}

/*
 * __wt_evict_state_check --
 *	Confirm these are pages we want to evict.
 */
static void
__wt_evict_state_check(WT_TOC *toc)
{
	ENV *env;
	WT_CACHE *cache;
	WT_EVICT_LIST *evict;
	WT_PAGE *page;
	WT_REF *ref;
	uint i;

	env = toc->env;
	cache = env->ienv->cache;

	/*
	 * We "own" the pages (we've flagged them for eviction, and there were
	 * no hazard references).   Now do checks to see if these are pages we
	 * can evict -- we have to wait until after we own the page because the
	 * page might be updated and race with us.
	 */
	WT_EVICT_FOREACH(cache, evict, i) {
		if ((ref = evict->ref) == NULL)
			continue;
		page = ref->page;

		/* Ignore pinned pages. */
		if (F_ISSET(page, WT_PINNED)) {
			WT_VERBOSE(env, WT_VERB_EVICT, (env,
			    "eviction skipped page addr %lu (pinned)",
			    page->addr));
			goto skip;
		}

		/* Ignore pages with in-memory subtrees. */
		switch (page->dsk->type) {
		case WT_PAGE_COL_INT:
		case WT_PAGE_DUP_INT:
		case WT_PAGE_ROW_INT:
		case WT_PAGE_ROW_LEAF:
			if (__wt_evict_page_subtrees(page)) {
				WT_VERBOSE(env, WT_VERB_EVICT, (env,
				    "eviction skipped page addr %lu (subtrees)",
				    page->addr));
				goto skip;
			}
			break;
		default:
			break;
		}

		continue;

skip:		/*
		 * Set the page's LRU so we don't select it again.
		 * Return the page to service.
		 * Discard our reference.
		 */
		page->read_gen = ++cache->read_gen;
		ref->state = WT_OK;
		WT_EVICT_CLR(evict);
	}
}

/*
 * __wt_evict_write --
 *	Write any modified pages.
 */
static void
__wt_evict_write(WT_TOC *toc)
{
	ENV *env;
	WT_CACHE *cache;
	WT_EVICT_LIST *evict;
	WT_PAGE *page;
	WT_REF *ref;
	uint i;

	env = toc->env;
	cache = env->ienv->cache;

	WT_EVICT_FOREACH(cache, evict, i) {
		if ((ref = evict->ref) == NULL)
			continue;
		page = ref->page;

		/* Ignore dirty pages. */
		if (!WT_PAGE_IS_MODIFIED(page))
			continue;

		/*
		 * We're using our WT_TOC handle, it needs to reference the
		 * correct DB handle.
		 *
		 * XXX
		 * This is pretty sleazy, but I'm hesitant to try and drive
		 * a separate DB/IDB handle down through the reconciliation
		 * code.
		 */
		toc->db = evict->idb->db;
		(void)__wt_page_reconcile(toc, page);
	}
}

/*
 * __wt_evict_page --
 *	Evict cache pages.
 */
static void
__wt_evict_page(WT_TOC *toc, int was_dirty)
{
	ENV *env;
	WT_CACHE *cache;
	WT_EVICT_LIST *evict;
	WT_PAGE *page;
	WT_REF *ref;
	WT_STATS *stats;
	uint i;

	env = toc->env;
	cache = env->ienv->cache;
	stats = cache->stats;

	WT_EVICT_FOREACH(cache, evict, i) {
		if ((ref = evict->ref) == NULL)
			continue;
		page = ref->page;

		/*
		 * The first time we're called, we get rid of the clean pages;
		 * the second time we're called, we get rid of the pages that
		 * were dirty but have since been cleaned.  Ignore dirty pages
		 * in all cases, it's simpler.
		 */
		if (WT_PAGE_IS_MODIFIED(page))
			continue;

		if (was_dirty)
			WT_STAT_INCR(stats, CACHE_EVICT_MODIFIED);
		else
			WT_STAT_INCR(stats, CACHE_EVICT_UNMODIFIED);

#ifdef HAVE_DIAGNOSTIC
		__wt_evict_hazard_validate(env, page);
#endif
		WT_VERBOSE(env, WT_VERB_EVICT, (env,
		    "cache evicting page addr %lu", page->addr));

		/*
		 * Copy a page reference, then make the cache entry available
		 * for re-use.
		 *
		 * No memory flush needed, the state field is declared volatile.
		 */
		ref->page = NULL;
		ref->state = WT_EMPTY;

		/* Remove the entry from the eviction list. */
		WT_EVICT_CLR(evict);

		/* We've got more space. */
		WT_CACHE_PAGE_OUT(cache, page->size);

		/* The page can no longer be found, free the memory. */
		__wt_page_discard(toc, page);
	}
}

/*
 * __wt_evict_page_subtrees --
 *	Return if a page has an in-memory subtree.
 */
static int
__wt_evict_page_subtrees(WT_PAGE *page)
{
	WT_REF *ref, **dupp;
	uint32_t i;

	/*
	 * Return if a page has an in-memory subtree -- this array search could
	 * be replaced by a reference count in the page, but (1) the eviction
	 * thread isn't where I expect performance problems, (2) I hate to lose
	 * more bytes on every page, (3) how often will an internal page be
	 * evicted anyway?
	 */
	switch (page->dsk->type) {
	case WT_PAGE_COL_INT:
	case WT_PAGE_DUP_INT:
	case WT_PAGE_ROW_INT:
		WT_REF_FOREACH(page, ref, i)
			if (ref->state != WT_EMPTY)
				return (1);
		break;
	case WT_PAGE_ROW_LEAF:
		if (WT_PAGE_DUP_TREES(page))
			WT_DUP_FOREACH(page, dupp, i)
				if (*dupp != NULL && (*dupp)->state != WT_EMPTY)
					return (1);
		break;
	default:
		break;
	}

	return (0);
}

/*
 * __wt_evict_compare_page --
 *	Qsort function: sort WT_EVICT_LIST array based on the page's address.
 */
static int
__wt_evict_compare_page(const void *a, const void *b)
{
	WT_REF *a_ref, *b_ref;
	WT_PAGE *a_page, *b_page;

	/*
	 * There may be NULL references in the array; sort them as greater than
	 * anything else so they migrate to the end of the array.
	 */
	a_ref = ((WT_EVICT_LIST *)a)->ref;
	b_ref = ((WT_EVICT_LIST *)b)->ref;
	if (a_ref == NULL)
		return (b_ref == NULL ? 0 : 1);
	if (b_ref == NULL)
		return (-1);

	/* Sort the page address in ascending order. */
	a_page = a_ref->page;
	b_page = b_ref->page;
	return (a_page > b_page ? 1 : (a_page < b_page ? -1 : 0));
}

/*
 * __wt_evict_compare_lru --
 *	Qsort function: sort WT_EVICT_LIST array based on the page's read
 *	generation.
 */
static int
__wt_evict_compare_lru(const void *a, const void *b)
{
	WT_REF *a_ref, *b_ref;
	uint32_t a_lru, b_lru;

	/*
	 * There may be NULL references in the array; sort them as greater than
	 * anything else so they migrate to the end of the array.
	 */
	a_ref = ((WT_EVICT_LIST *)a)->ref;
	b_ref = ((WT_EVICT_LIST *)b)->ref;
	if (a_ref == NULL)
		return (b_ref == NULL ? 0 : 1);
	if (b_ref == NULL)
		return (-1);

	/* Sort the LRU in ascending order. */
	a_lru = a_ref->page->read_gen;
	b_lru = b_ref->page->read_gen;
	return (a_lru > b_lru ? 1 : (a_lru < b_lru ? -1 : 0));
}

/*
 * __wt_evict_hazard_compare --
 *	Qsort function: sort hazard list based on the page's address.
 */
static int
__wt_evict_hazard_compare(const void *a, const void *b)
{
	WT_PAGE *a_page, *b_page;

	a_page = *(WT_PAGE **)a;
	b_page = *(WT_PAGE **)b;

	return (a_page > b_page ? 1 : (a_page < b_page ? -1 : 0));
}

#ifdef HAVE_DIAGNOSTIC
/*
 * __wt_evict_hazard_validate --
 *	Return if a page is or isn't on the hazard list.
 */
static void
__wt_evict_hazard_validate(ENV *env, WT_PAGE *page)
{
	IENV *ienv;
	WT_PAGE **hp;
	WT_TOC **tp, *toc;

	ienv = env->ienv;

	for (tp = ienv->toc; (toc = *tp) != NULL; ++tp)
		for (hp = toc->hazard;
		    hp < toc->hazard + toc->env->hazard_size; ++hp)
			if (*hp == page) {
				__wt_api_env_errx(env,
				    "hazard eviction check for page %lu "
				    "failed",
				    (u_long)page->addr);
				__wt_abort(env);
			}
}

/*
 * __wt_evict_dump --
 *	Display the eviction list.
 */
void
__wt_evict_dump(WT_TOC *toc)
{
	ENV *env;
	WT_CACHE *cache;
	WT_EVICT_LIST *evict;
	WT_MBUF mb;
	uint n;
	int sep;

	env = toc->env;
	cache = env->ienv->cache;

	__wt_mb_init(env, &mb);
	__wt_mb_add(&mb, "eviction list");

	for (sep = ':', n = 0; n < cache->evict_elem; ++n) {
		evict = &cache->evict[n];
		if (evict->ref == NULL)
			continue;
		__wt_mb_add(&mb, "%c %lu", sep, (u_long)evict->ref->page->addr);
		sep = ',';
	}
	__wt_mb_discard(&mb);
}

/*
 * __wt_evict_dump_cache
 *	Dump the in-memory cache.
 */
int
__wt_evict_cache_dump(WT_TOC *toc)
{
	IDB *idb;
	IENV *ienv;

	ienv = toc->env->ienv;

	TAILQ_FOREACH(idb, &ienv->dbqh, q)
		WT_RET(__wt_evict_tree_dump(toc, idb));
	return (0);
}

/*
 * __wt_evict_tree_dump
 *	Dump an in-memory tree.
 */
int
__wt_evict_tree_dump(WT_TOC *toc, IDB *idb)
{
	ENV *env;
	WT_CACHE *cache;
	WT_REF *ref;
	WT_WALK walk;
	WT_MBUF mb;
	int sep;

	env = toc->env;
	cache = env->ienv->cache;

	WT_VERBOSE(env, WT_VERB_EVICT, (env,
	    "%s: pages inuse %llu, bytes inuse (%llu), max (%llu)",
	    idb->name,
	    __wt_cache_pages_inuse(cache),
	    __wt_cache_bytes_inuse(cache),
	    WT_STAT(cache->stats, CACHE_BYTES_MAX)));

	__wt_mb_init(env, &mb);
	__wt_mb_add(&mb, "in-memory page list");

	WT_CLEAR(walk);
	WT_RET(__wt_walk_begin(toc, &idb->root_page, &walk));
	for (sep = ':';;) {
		WT_RET(__wt_walk_next(toc, &walk, &ref));
		if (ref == NULL)
			break;
		__wt_mb_add(&mb, "%c %lu", sep, (u_long)ref->page->addr);
		sep = ',';
	}
	__wt_walk_end(env, &walk);
	__wt_mb_discard(&mb);

	return (0);
}

/*
 * __wt_evict_cache_count
 *	Retrun the count of nodes in the cache.
 */
int
__wt_evict_cache_count(WT_TOC *toc, uint64_t *nodesp)
{
	IDB *idb;
	IENV *ienv;
	uint64_t nodes;

	ienv = toc->env->ienv;

	*nodesp = 0;
	TAILQ_FOREACH(idb, &ienv->dbqh, q) {
		WT_RET(__wt_evict_tree_count(toc, idb, &nodes));
		*nodesp += nodes;
	}
	return (0);
}

/*
 * __wt_evict_tree_count
 *	Return a count of nodes in the tree.
 */
int
__wt_evict_tree_count(WT_TOC *toc, IDB *idb, uint64_t *nodesp)
{
	ENV *env;
	WT_REF *ref;
	WT_WALK walk;
	uint64_t nodes;

	env = toc->env;

	WT_CLEAR(walk);
	WT_RET(__wt_walk_begin(toc, &idb->root_page, &walk));
	for (nodes = 0;;) {
		WT_RET(__wt_walk_next(toc, &walk, &ref));
		if (ref == NULL)
			break;
		++nodes;
	}
	*nodesp = nodes;
	__wt_walk_end(env, &walk);

	return (0);
}
#endif