Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * nbtsort.c
4 : : * Build a btree from sorted input by loading leaf pages sequentially.
5 : : *
6 : : * NOTES
7 : : *
8 : : * We use tuplesort.c to sort the given index tuples into order.
9 : : * Then we scan the index tuples in order and build the btree pages
10 : : * for each level. We load source tuples into leaf-level pages.
11 : : * Whenever we fill a page at one level, we add a link to it to its
12 : : * parent level (starting a new parent level if necessary). When
13 : : * done, we write out each final page on each level, adding it to
14 : : * its parent level. When we have only one page on a level, it must be
15 : : * the root -- it can be attached to the btree metapage and we are done.
16 : : *
17 : : * It is not wise to pack the pages entirely full, since then *any*
18 : : * insertion would cause a split (and not only of the leaf page; the need
19 : : * for a split would cascade right up the tree). The steady-state load
20 : : * factor for btrees is usually estimated at 70%. We choose to pack leaf
21 : : * pages to the user-controllable fill factor (default 90%) while upper pages
22 : : * are always packed to 70%. This gives us reasonable density (there aren't
23 : : * many upper pages if the keys are reasonable-size) without risking a lot of
24 : : * cascading splits during early insertions.
25 : : *
26 : : * We use the bulk smgr loading facility to bypass the buffer cache and
27 : : * WAL-log the pages efficiently.
28 : : *
29 : : * This code isn't concerned about the FSM at all. The caller is responsible
30 : : * for initializing that.
31 : : *
32 : : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
33 : : * Portions Copyright (c) 1994, Regents of the University of California
34 : : *
35 : : * IDENTIFICATION
36 : : * src/backend/access/nbtree/nbtsort.c
37 : : *
38 : : *-------------------------------------------------------------------------
39 : : */
40 : :
41 : : #include "postgres.h"
42 : :
43 : : #include "access/nbtree.h"
44 : : #include "access/parallel.h"
45 : : #include "access/relscan.h"
46 : : #include "access/table.h"
47 : : #include "access/xact.h"
48 : : #include "access/xloginsert.h"
49 : : #include "catalog/index.h"
50 : : #include "commands/progress.h"
51 : : #include "executor/instrument.h"
52 : : #include "miscadmin.h"
53 : : #include "pgstat.h"
54 : : #include "storage/bulk_write.h"
55 : : #include "tcop/tcopprot.h" /* pgrminclude ignore */
56 : : #include "utils/rel.h"
57 : : #include "utils/sortsupport.h"
58 : : #include "utils/tuplesort.h"
59 : :
60 : :
61 : : /* Magic numbers for parallel state sharing */
62 : : #define PARALLEL_KEY_BTREE_SHARED UINT64CONST(0xA000000000000001)
63 : : #define PARALLEL_KEY_TUPLESORT UINT64CONST(0xA000000000000002)
64 : : #define PARALLEL_KEY_TUPLESORT_SPOOL2 UINT64CONST(0xA000000000000003)
65 : : #define PARALLEL_KEY_QUERY_TEXT UINT64CONST(0xA000000000000004)
66 : : #define PARALLEL_KEY_WAL_USAGE UINT64CONST(0xA000000000000005)
67 : : #define PARALLEL_KEY_BUFFER_USAGE UINT64CONST(0xA000000000000006)
68 : :
69 : : /*
70 : : * DISABLE_LEADER_PARTICIPATION disables the leader's participation in
71 : : * parallel index builds. This may be useful as a debugging aid.
72 : : #undef DISABLE_LEADER_PARTICIPATION
73 : : */
74 : :
75 : : /*
76 : : * Status record for spooling/sorting phase. (Note we may have two of
77 : : * these due to the special requirements for uniqueness-checking with
78 : : * dead tuples.)
79 : : */
80 : : typedef struct BTSpool
81 : : {
82 : : Tuplesortstate *sortstate; /* state data for tuplesort.c */
83 : : Relation heap;
84 : : Relation index;
85 : : bool isunique;
86 : : bool nulls_not_distinct;
87 : : } BTSpool;
88 : :
89 : : /*
90 : : * Status for index builds performed in parallel. This is allocated in a
91 : : * dynamic shared memory segment. Note that there is a separate tuplesort TOC
92 : : * entry, private to tuplesort.c but allocated by this module on its behalf.
93 : : */
94 : : typedef struct BTShared
95 : : {
96 : : /*
97 : : * These fields are not modified during the sort. They primarily exist
98 : : * for the benefit of worker processes that need to create BTSpool state
99 : : * corresponding to that used by the leader.
100 : : */
101 : : Oid heaprelid;
102 : : Oid indexrelid;
103 : : bool isunique;
104 : : bool nulls_not_distinct;
105 : : bool isconcurrent;
106 : : int scantuplesortstates;
107 : :
108 : : /*
109 : : * workersdonecv is used to monitor the progress of workers. All parallel
110 : : * participants must indicate that they are done before leader can use
111 : : * mutable state that workers maintain during scan (and before leader can
112 : : * proceed to tuplesort_performsort()).
113 : : */
114 : : ConditionVariable workersdonecv;
115 : :
116 : : /*
117 : : * mutex protects all fields before heapdesc.
118 : : *
119 : : * These fields contain status information of interest to B-Tree index
120 : : * builds that must work just the same when an index is built in parallel.
121 : : */
122 : : slock_t mutex;
123 : :
124 : : /*
125 : : * Mutable state that is maintained by workers, and reported back to
126 : : * leader at end of parallel scan.
127 : : *
128 : : * nparticipantsdone is number of worker processes finished.
129 : : *
130 : : * reltuples is the total number of input heap tuples.
131 : : *
132 : : * havedead indicates if RECENTLY_DEAD tuples were encountered during
133 : : * build.
134 : : *
135 : : * indtuples is the total number of tuples that made it into the index.
136 : : *
137 : : * brokenhotchain indicates if any worker detected a broken HOT chain
138 : : * during build.
139 : : */
140 : : int nparticipantsdone;
141 : : double reltuples;
142 : : bool havedead;
143 : : double indtuples;
144 : : bool brokenhotchain;
145 : :
146 : : /*
147 : : * ParallelTableScanDescData data follows. Can't directly embed here, as
148 : : * implementations of the parallel table scan desc interface might need
149 : : * stronger alignment.
150 : : */
151 : : } BTShared;
152 : :
153 : : /*
154 : : * Return pointer to a BTShared's parallel table scan.
155 : : *
156 : : * c.f. shm_toc_allocate as to why BUFFERALIGN is used, rather than just
157 : : * MAXALIGN.
158 : : */
159 : : #define ParallelTableScanFromBTShared(shared) \
160 : : (ParallelTableScanDesc) ((char *) (shared) + BUFFERALIGN(sizeof(BTShared)))
161 : :
162 : : /*
163 : : * Status for leader in parallel index build.
164 : : */
165 : : typedef struct BTLeader
166 : : {
167 : : /* parallel context itself */
168 : : ParallelContext *pcxt;
169 : :
170 : : /*
171 : : * nparticipanttuplesorts is the exact number of worker processes
172 : : * successfully launched, plus one leader process if it participates as a
173 : : * worker (only DISABLE_LEADER_PARTICIPATION builds avoid leader
174 : : * participating as a worker).
175 : : */
176 : : int nparticipanttuplesorts;
177 : :
178 : : /*
179 : : * Leader process convenience pointers to shared state (leader avoids TOC
180 : : * lookups).
181 : : *
182 : : * btshared is the shared state for entire build. sharedsort is the
183 : : * shared, tuplesort-managed state passed to each process tuplesort.
184 : : * sharedsort2 is the corresponding btspool2 shared state, used only when
185 : : * building unique indexes. snapshot is the snapshot used by the scan iff
186 : : * an MVCC snapshot is required.
187 : : */
188 : : BTShared *btshared;
189 : : Sharedsort *sharedsort;
190 : : Sharedsort *sharedsort2;
191 : : Snapshot snapshot;
192 : : WalUsage *walusage;
193 : : BufferUsage *bufferusage;
194 : : } BTLeader;
195 : :
196 : : /*
197 : : * Working state for btbuild and its callback.
198 : : *
199 : : * When parallel CREATE INDEX is used, there is a BTBuildState for each
200 : : * participant.
201 : : */
202 : : typedef struct BTBuildState
203 : : {
204 : : bool isunique;
205 : : bool nulls_not_distinct;
206 : : bool havedead;
207 : : Relation heap;
208 : : BTSpool *spool;
209 : :
210 : : /*
211 : : * spool2 is needed only when the index is a unique index. Dead tuples are
212 : : * put into spool2 instead of spool in order to avoid uniqueness check.
213 : : */
214 : : BTSpool *spool2;
215 : : double indtuples;
216 : :
217 : : /*
218 : : * btleader is only present when a parallel index build is performed, and
219 : : * only in the leader process. (Actually, only the leader has a
220 : : * BTBuildState. Workers have their own spool and spool2, though.)
221 : : */
222 : : BTLeader *btleader;
223 : : } BTBuildState;
224 : :
225 : : /*
226 : : * Status record for a btree page being built. We have one of these
227 : : * for each active tree level.
228 : : */
229 : : typedef struct BTPageState
230 : : {
231 : : BulkWriteBuffer btps_buf; /* workspace for page building */
232 : : BlockNumber btps_blkno; /* block # to write this page at */
233 : : IndexTuple btps_lowkey; /* page's strict lower bound pivot tuple */
234 : : OffsetNumber btps_lastoff; /* last item offset loaded */
235 : : Size btps_lastextra; /* last item's extra posting list space */
236 : : uint32 btps_level; /* tree level (0 = leaf) */
237 : : Size btps_full; /* "full" if less than this much free space */
238 : : struct BTPageState *btps_next; /* link to parent level, if any */
239 : : } BTPageState;
240 : :
241 : : /*
242 : : * Overall status record for index writing phase.
243 : : */
244 : : typedef struct BTWriteState
245 : : {
246 : : Relation heap;
247 : : Relation index;
248 : : BulkWriteState *bulkstate;
249 : : BTScanInsert inskey; /* generic insertion scankey */
250 : : BlockNumber btws_pages_alloced; /* # pages allocated */
251 : : } BTWriteState;
252 : :
253 : :
254 : : static double _bt_spools_heapscan(Relation heap, Relation index,
255 : : BTBuildState *buildstate, IndexInfo *indexInfo);
256 : : static void _bt_spooldestroy(BTSpool *btspool);
257 : : static void _bt_spool(BTSpool *btspool, ItemPointer self,
258 : : Datum *values, bool *isnull);
259 : : static void _bt_leafbuild(BTSpool *btspool, BTSpool *btspool2);
260 : : static void _bt_build_callback(Relation index, ItemPointer tid, Datum *values,
261 : : bool *isnull, bool tupleIsAlive, void *state);
262 : : static BulkWriteBuffer _bt_blnewpage(BTWriteState *wstate, uint32 level);
263 : : static BTPageState *_bt_pagestate(BTWriteState *wstate, uint32 level);
264 : : static void _bt_slideleft(Page rightmostpage);
265 : : static void _bt_sortaddtup(Page page, Size itemsize,
266 : : IndexTuple itup, OffsetNumber itup_off,
267 : : bool newfirstdataitem);
268 : : static void _bt_buildadd(BTWriteState *wstate, BTPageState *state,
269 : : IndexTuple itup, Size truncextra);
270 : : static void _bt_sort_dedup_finish_pending(BTWriteState *wstate,
271 : : BTPageState *state,
272 : : BTDedupState dstate);
273 : : static void _bt_uppershutdown(BTWriteState *wstate, BTPageState *state);
274 : : static void _bt_load(BTWriteState *wstate,
275 : : BTSpool *btspool, BTSpool *btspool2);
276 : : static void _bt_begin_parallel(BTBuildState *buildstate, bool isconcurrent,
277 : : int request);
278 : : static void _bt_end_parallel(BTLeader *btleader);
279 : : static Size _bt_parallel_estimate_shared(Relation heap, Snapshot snapshot);
280 : : static double _bt_parallel_heapscan(BTBuildState *buildstate,
281 : : bool *brokenhotchain);
282 : : static void _bt_leader_participate_as_worker(BTBuildState *buildstate);
283 : : static void _bt_parallel_scan_and_sort(BTSpool *btspool, BTSpool *btspool2,
284 : : BTShared *btshared, Sharedsort *sharedsort,
285 : : Sharedsort *sharedsort2, int sortmem,
286 : : bool progress);
287 : :
288 : :
289 : : /*
290 : : * btbuild() -- build a new btree index.
291 : : */
292 : : IndexBuildResult *
2263 rhaas@postgresql.org 293 :CBC 22628 : btbuild(Relation heap, Relation index, IndexInfo *indexInfo)
294 : : {
295 : : IndexBuildResult *result;
296 : : BTBuildState buildstate;
297 : : double reltuples;
298 : :
299 : : #ifdef BTREE_BUILD_STATS
300 : : if (log_btree_build_stats)
301 : : ResetUsage();
302 : : #endif /* BTREE_BUILD_STATS */
303 : :
304 : 22628 : buildstate.isunique = indexInfo->ii_Unique;
801 peter@eisentraut.org 305 : 22628 : buildstate.nulls_not_distinct = indexInfo->ii_NullsNotDistinct;
2263 rhaas@postgresql.org 306 : 22628 : buildstate.havedead = false;
307 : 22628 : buildstate.heap = heap;
308 : 22628 : buildstate.spool = NULL;
309 : 22628 : buildstate.spool2 = NULL;
310 : 22628 : buildstate.indtuples = 0;
311 : 22628 : buildstate.btleader = NULL;
312 : :
313 : : /*
314 : : * We expect to be called exactly once for any index relation. If that's
315 : : * not the case, big trouble's what we have.
316 : : */
317 [ - + ]: 22628 : if (RelationGetNumberOfBlocks(index) != 0)
2263 rhaas@postgresql.org 318 [ # # ]:UBC 0 : elog(ERROR, "index \"%s\" already contains data",
319 : : RelationGetRelationName(index));
320 : :
2263 rhaas@postgresql.org 321 :CBC 22628 : reltuples = _bt_spools_heapscan(heap, index, &buildstate, indexInfo);
322 : :
323 : : /*
324 : : * Finish the build by (1) completing the sort of the spool file, (2)
325 : : * inserting the sorted tuples into btree pages and (3) building the upper
326 : : * levels. Finally, it may also be necessary to end use of parallelism.
327 : : */
328 : 22622 : _bt_leafbuild(buildstate.spool, buildstate.spool2);
329 : 22580 : _bt_spooldestroy(buildstate.spool);
330 [ + + ]: 22580 : if (buildstate.spool2)
331 : 11 : _bt_spooldestroy(buildstate.spool2);
332 [ + + ]: 22580 : if (buildstate.btleader)
333 : 71 : _bt_end_parallel(buildstate.btleader);
334 : :
335 : 22580 : result = (IndexBuildResult *) palloc(sizeof(IndexBuildResult));
336 : :
337 : 22580 : result->heap_tuples = reltuples;
338 : 22580 : result->index_tuples = buildstate.indtuples;
339 : :
340 : : #ifdef BTREE_BUILD_STATS
341 : : if (log_btree_build_stats)
342 : : {
343 : : ShowUsage("BTREE BUILD STATS");
344 : : ResetUsage();
345 : : }
346 : : #endif /* BTREE_BUILD_STATS */
347 : :
348 : 22580 : return result;
349 : : }
350 : :
351 : : /*
352 : : * Create and initialize one or two spool structures, and save them in caller's
353 : : * buildstate argument. May also fill-in fields within indexInfo used by index
354 : : * builds.
355 : : *
356 : : * Scans the heap, possibly in parallel, filling spools with IndexTuples. This
357 : : * routine encapsulates all aspects of managing parallelism. Caller need only
358 : : * call _bt_end_parallel() in parallel case after it is done with spool/spool2.
359 : : *
360 : : * Returns the total number of heap tuples scanned.
361 : : */
362 : : static double
363 : 22628 : _bt_spools_heapscan(Relation heap, Relation index, BTBuildState *buildstate,
364 : : IndexInfo *indexInfo)
365 : : {
7823 bruce@momjian.us 366 : 22628 : BTSpool *btspool = (BTSpool *) palloc0(sizeof(BTSpool));
2263 rhaas@postgresql.org 367 : 22628 : SortCoordinate coordinate = NULL;
368 : 22628 : double reltuples = 0;
369 : :
370 : : /*
371 : : * We size the sort area as maintenance_work_mem rather than work_mem to
372 : : * speed index creation. This should be OK since a single backend can't
373 : : * run multiple index creations in parallel (see also: notes on
374 : : * parallelism and maintenance_work_mem below).
375 : : */
4093 tgl@sss.pgh.pa.us 376 : 22628 : btspool->heap = heap;
8946 377 : 22628 : btspool->index = index;
2263 rhaas@postgresql.org 378 : 22628 : btspool->isunique = indexInfo->ii_Unique;
801 peter@eisentraut.org 379 : 22628 : btspool->nulls_not_distinct = indexInfo->ii_NullsNotDistinct;
380 : :
381 : : /* Save as primary spool */
2263 rhaas@postgresql.org 382 : 22628 : buildstate->spool = btspool;
383 : :
384 : : /* Report table scan phase started */
1839 alvherre@alvh.no-ip. 385 : 22628 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
386 : : PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN);
387 : :
388 : : /* Attempt to launch parallel worker scan when required */
2263 rhaas@postgresql.org 389 [ + + ]: 22628 : if (indexInfo->ii_ParallelWorkers > 0)
390 : 71 : _bt_begin_parallel(buildstate, indexInfo->ii_Concurrent,
391 : : indexInfo->ii_ParallelWorkers);
392 : :
393 : : /*
394 : : * If parallel build requested and at least one worker process was
395 : : * successfully launched, set up coordination state
396 : : */
397 [ + + ]: 22628 : if (buildstate->btleader)
398 : : {
399 : 71 : coordinate = (SortCoordinate) palloc0(sizeof(SortCoordinateData));
400 : 71 : coordinate->isWorker = false;
401 : 71 : coordinate->nParticipants =
402 : 71 : buildstate->btleader->nparticipanttuplesorts;
403 : 71 : coordinate->sharedsort = buildstate->btleader->sharedsort;
404 : : }
405 : :
406 : : /*
407 : : * Begin serial/leader tuplesort.
408 : : *
409 : : * In cases where parallelism is involved, the leader receives the same
410 : : * share of maintenance_work_mem as a serial sort (it is generally treated
411 : : * in the same way as a serial sort once we return). Parallel worker
412 : : * Tuplesortstates will have received only a fraction of
413 : : * maintenance_work_mem, though.
414 : : *
415 : : * We rely on the lifetime of the Leader Tuplesortstate almost not
416 : : * overlapping with any worker Tuplesortstate's lifetime. There may be
417 : : * some small overlap, but that's okay because we rely on leader
418 : : * Tuplesortstate only allocating a small, fixed amount of memory here.
419 : : * When its tuplesort_performsort() is called (by our caller), and
420 : : * significant amounts of memory are likely to be used, all workers must
421 : : * have already freed almost all memory held by their Tuplesortstates
422 : : * (they are about to go away completely, too). The overall effect is
423 : : * that maintenance_work_mem always represents an absolute high watermark
424 : : * on the amount of memory used by a CREATE INDEX operation, regardless of
425 : : * the use of parallelism or any other factor.
426 : : */
427 : 45256 : buildstate->spool->sortstate =
428 : 22628 : tuplesort_begin_index_btree(heap, index, buildstate->isunique,
801 peter@eisentraut.org 429 : 22628 : buildstate->nulls_not_distinct,
430 : : maintenance_work_mem, coordinate,
431 : : TUPLESORT_NONE);
432 : :
433 : : /*
434 : : * If building a unique index, put dead tuples in a second spool to keep
435 : : * them out of the uniqueness check. We expect that the second spool (for
436 : : * dead tuples) won't get very full, so we give it only work_mem.
437 : : */
2263 rhaas@postgresql.org 438 [ + + ]: 22628 : if (indexInfo->ii_Unique)
439 : : {
440 : 18334 : BTSpool *btspool2 = (BTSpool *) palloc0(sizeof(BTSpool));
441 : 18334 : SortCoordinate coordinate2 = NULL;
442 : :
443 : : /* Initialize secondary spool */
444 : 18334 : btspool2->heap = heap;
445 : 18334 : btspool2->index = index;
446 : 18334 : btspool2->isunique = false;
447 : : /* Save as secondary spool */
448 : 18334 : buildstate->spool2 = btspool2;
449 : :
450 [ + + ]: 18334 : if (buildstate->btleader)
451 : : {
452 : : /*
453 : : * Set up non-private state that is passed to
454 : : * tuplesort_begin_index_btree() about the basic high level
455 : : * coordination of a parallel sort.
456 : : */
457 : 32 : coordinate2 = (SortCoordinate) palloc0(sizeof(SortCoordinateData));
458 : 32 : coordinate2->isWorker = false;
459 : 32 : coordinate2->nParticipants =
460 : 32 : buildstate->btleader->nparticipanttuplesorts;
461 : 32 : coordinate2->sharedsort = buildstate->btleader->sharedsort2;
462 : : }
463 : :
464 : : /*
465 : : * We expect that the second one (for dead tuples) won't get very
466 : : * full, so we give it only work_mem
467 : : */
468 : 18334 : buildstate->spool2->sortstate =
801 peter@eisentraut.org 469 : 18334 : tuplesort_begin_index_btree(heap, index, false, false, work_mem,
470 : : coordinate2, TUPLESORT_NONE);
471 : : }
472 : :
473 : : /* Fill spool using either serial or parallel heap scan */
2263 rhaas@postgresql.org 474 [ + + ]: 22628 : if (!buildstate->btleader)
1839 alvherre@alvh.no-ip. 475 : 22557 : reltuples = table_index_build_scan(heap, index, indexInfo, true, true,
476 : : _bt_build_callback, (void *) buildstate,
477 : : NULL);
478 : : else
2263 rhaas@postgresql.org 479 : 71 : reltuples = _bt_parallel_heapscan(buildstate,
480 : : &indexInfo->ii_BrokenHotChain);
481 : :
482 : : /*
483 : : * Set the progress target for the next phase. Reset the block number
484 : : * values set by table_index_build_scan
485 : : */
486 : : {
1167 peter@eisentraut.org 487 : 22622 : const int progress_index[] = {
488 : : PROGRESS_CREATEIDX_TUPLES_TOTAL,
489 : : PROGRESS_SCAN_BLOCKS_TOTAL,
490 : : PROGRESS_SCAN_BLOCKS_DONE
491 : : };
492 : 22622 : const int64 progress_vals[] = {
1839 alvherre@alvh.no-ip. 493 : 22622 : buildstate->indtuples,
494 : : 0, 0
495 : : };
496 : :
1167 peter@eisentraut.org 497 : 22622 : pgstat_progress_update_multi_param(3, progress_index, progress_vals);
498 : : }
499 : :
500 : : /* okay, all heap tuples are spooled */
2263 rhaas@postgresql.org 501 [ + + + + ]: 22622 : if (buildstate->spool2 && !buildstate->havedead)
502 : : {
503 : : /* spool2 turns out to be unnecessary */
504 : 18323 : _bt_spooldestroy(buildstate->spool2);
505 : 18323 : buildstate->spool2 = NULL;
506 : : }
507 : :
508 : 22622 : return reltuples;
509 : : }
510 : :
511 : : /*
512 : : * clean up a spool structure and its substructures.
513 : : */
514 : : static void
8946 tgl@sss.pgh.pa.us 515 : 40914 : _bt_spooldestroy(BTSpool *btspool)
516 : : {
517 : 40914 : tuplesort_end(btspool->sortstate);
7139 neilc@samurai.com 518 : 40914 : pfree(btspool);
10141 scrappy@hub.org 519 : 40914 : }
520 : :
521 : : /*
522 : : * spool an index entry into the sort file.
523 : : */
524 : : static void
3575 rhaas@postgresql.org 525 : 5959699 : _bt_spool(BTSpool *btspool, ItemPointer self, Datum *values, bool *isnull)
526 : : {
527 : 5959699 : tuplesort_putindextuplevalues(btspool->sortstate, btspool->index,
528 : : self, values, isnull);
10141 scrappy@hub.org 529 : 5959699 : }
530 : :
531 : : /*
532 : : * given a spool loaded by successive calls to _bt_spool,
533 : : * create an entire btree.
534 : : */
535 : : static void
8648 inoue@tpf.co.jp 536 : 22622 : _bt_leafbuild(BTSpool *btspool, BTSpool *btspool2)
537 : : {
538 : : BTWriteState wstate;
539 : :
540 : : #ifdef BTREE_BUILD_STATS
541 : : if (log_btree_build_stats)
542 : : {
543 : : ShowUsage("BTREE BUILD (Spool) STATISTICS");
544 : : ResetUsage();
545 : : }
546 : : #endif /* BTREE_BUILD_STATS */
547 : :
548 : : /* Execute the sort */
1839 alvherre@alvh.no-ip. 549 : 22622 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
550 : : PROGRESS_BTREE_PHASE_PERFORMSORT_1);
7722 tgl@sss.pgh.pa.us 551 : 22622 : tuplesort_performsort(btspool->sortstate);
8648 inoue@tpf.co.jp 552 [ + + ]: 22580 : if (btspool2)
553 : : {
1839 alvherre@alvh.no-ip. 554 : 11 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
555 : : PROGRESS_BTREE_PHASE_PERFORMSORT_2);
8648 inoue@tpf.co.jp 556 : 11 : tuplesort_performsort(btspool2->sortstate);
557 : : }
558 : :
4093 tgl@sss.pgh.pa.us 559 : 22580 : wstate.heap = btspool->heap;
7256 560 : 22580 : wstate.index = btspool->index;
309 pg@bowt.ie 561 : 22580 : wstate.inskey = _bt_mkscankey(wstate.index, NULL);
562 : : /* _bt_mkscankey() won't set allequalimage without metapage */
1509 563 : 22580 : wstate.inskey->allequalimage = _bt_allequalimage(wstate.index, true);
564 : :
565 : : /* reserve the metapage */
7256 tgl@sss.pgh.pa.us 566 : 22580 : wstate.btws_pages_alloced = BTREE_METAPAGE + 1;
567 : :
1839 alvherre@alvh.no-ip. 568 : 22580 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
569 : : PROGRESS_BTREE_PHASE_LEAF_LOAD);
7256 tgl@sss.pgh.pa.us 570 : 22580 : _bt_load(&wstate, btspool, btspool2);
10141 scrappy@hub.org 571 : 22580 : }
572 : :
573 : : /*
574 : : * Per-tuple callback for table_index_build_scan
575 : : */
576 : : static void
2263 rhaas@postgresql.org 577 : 5959699 : _bt_build_callback(Relation index,
578 : : ItemPointer tid,
579 : : Datum *values,
580 : : bool *isnull,
581 : : bool tupleIsAlive,
582 : : void *state)
583 : : {
584 : 5959699 : BTBuildState *buildstate = (BTBuildState *) state;
585 : :
586 : : /*
587 : : * insert the index tuple into the appropriate spool file for subsequent
588 : : * processing
589 : : */
590 [ + + + + ]: 5959699 : if (tupleIsAlive || buildstate->spool2 == NULL)
1619 andres@anarazel.de 591 : 5959469 : _bt_spool(buildstate->spool, tid, values, isnull);
592 : : else
593 : : {
594 : : /* dead tuples are put into spool2 */
2263 rhaas@postgresql.org 595 : 230 : buildstate->havedead = true;
1619 andres@anarazel.de 596 : 230 : _bt_spool(buildstate->spool2, tid, values, isnull);
597 : : }
598 : :
2263 rhaas@postgresql.org 599 : 5959699 : buildstate->indtuples += 1;
600 : 5959699 : }
601 : :
602 : : /*
603 : : * allocate workspace for a new, clean btree page, not linked to any siblings.
604 : : */
605 : : static BulkWriteBuffer
51 heikki.linnakangas@i 606 :GNC 24231 : _bt_blnewpage(BTWriteState *wstate, uint32 level)
607 : : {
608 : : BulkWriteBuffer buf;
609 : : Page page;
610 : : BTPageOpaque opaque;
611 : :
612 : 24231 : buf = smgr_bulk_get_buf(wstate->bulkstate);
613 : 24231 : page = (Page) buf;
614 : :
615 : : /* Zero the page and set up standard page header info */
7256 tgl@sss.pgh.pa.us 616 :CBC 24231 : _bt_pageinit(page, BLCKSZ);
617 : :
618 : : /* Initialize BT opaque state */
744 michael@paquier.xyz 619 : 24231 : opaque = BTPageGetOpaque(page);
9716 bruce@momjian.us 620 : 24231 : opaque->btpo_prev = opaque->btpo_next = P_NONE;
1145 pg@bowt.ie 621 : 24231 : opaque->btpo_level = level;
7723 tgl@sss.pgh.pa.us 622 : 24231 : opaque->btpo_flags = (level > 0) ? 0 : BTP_LEAF;
6551 623 : 24231 : opaque->btpo_cycleid = 0;
624 : :
625 : : /* Make the P_HIKEY line pointer appear allocated */
7256 626 : 24231 : ((PageHeader) page)->pd_lower += sizeof(ItemIdData);
627 : :
51 heikki.linnakangas@i 628 :GNC 24231 : return buf;
629 : : }
630 : :
631 : : /*
632 : : * emit a completed btree page, and release the working storage.
633 : : */
634 : : static void
635 : 46811 : _bt_blwritepage(BTWriteState *wstate, BulkWriteBuffer buf, BlockNumber blkno)
636 : : {
637 : 46811 : smgr_bulk_write(wstate->bulkstate, blkno, buf, true);
638 : : /* smgr_bulk_write took ownership of 'buf' */
7723 tgl@sss.pgh.pa.us 639 :CBC 46811 : }
640 : :
641 : : /*
642 : : * allocate and initialize a new BTPageState. the returned structure
643 : : * is suitable for immediate use by _bt_buildadd.
644 : : */
645 : : static BTPageState *
7256 646 : 5527 : _bt_pagestate(BTWriteState *wstate, uint32 level)
647 : : {
7823 bruce@momjian.us 648 : 5527 : BTPageState *state = (BTPageState *) palloc0(sizeof(BTPageState));
649 : :
650 : : /* create initial page for level */
51 heikki.linnakangas@i 651 :GNC 5527 : state->btps_buf = _bt_blnewpage(wstate, level);
652 : :
653 : : /* and assign it a page position */
7256 tgl@sss.pgh.pa.us 654 :CBC 5527 : state->btps_blkno = wstate->btws_pages_alloced++;
655 : :
1620 pg@bowt.ie 656 : 5527 : state->btps_lowkey = NULL;
657 : : /* initialize lastoff so first item goes into P_FIRSTKEY */
8668 tgl@sss.pgh.pa.us 658 : 5527 : state->btps_lastoff = P_HIKEY;
1509 pg@bowt.ie 659 : 5527 : state->btps_lastextra = 0;
8668 tgl@sss.pgh.pa.us 660 : 5527 : state->btps_level = level;
661 : : /* set "full" threshold based on level. See notes at head of file. */
6495 662 [ + + ]: 5527 : if (level > 0)
6487 663 : 1114 : state->btps_full = (BLCKSZ * (100 - BTREE_NONLEAF_FILLFACTOR) / 100);
664 : : else
1602 michael@paquier.xyz 665 [ + - - + : 4413 : state->btps_full = BTGetTargetPageFreeSpace(wstate->index);
+ + ]
666 : :
667 : : /* no parent level, yet */
7403 neilc@samurai.com 668 : 5527 : state->btps_next = NULL;
669 : :
8668 tgl@sss.pgh.pa.us 670 : 5527 : return state;
671 : : }
672 : :
673 : : /*
674 : : * Slide the array of ItemIds from the page back one slot (from P_FIRSTKEY to
675 : : * P_HIKEY, overwriting P_HIKEY).
676 : : *
677 : : * _bt_blnewpage() makes the P_HIKEY line pointer appear allocated, but the
678 : : * rightmost page on its level is not supposed to get a high key. Now that
679 : : * it's clear that this page is a rightmost page, remove the unneeded empty
680 : : * P_HIKEY line pointer space.
681 : : */
682 : : static void
1378 pg@bowt.ie 683 : 5527 : _bt_slideleft(Page rightmostpage)
684 : : {
685 : : OffsetNumber off;
686 : : OffsetNumber maxoff;
687 : : ItemId previi;
688 : :
689 : 5527 : maxoff = PageGetMaxOffsetNumber(rightmostpage);
690 [ - + ]: 5527 : Assert(maxoff >= P_FIRSTKEY);
691 : 5527 : previi = PageGetItemId(rightmostpage, P_HIKEY);
692 [ + + ]: 319790 : for (off = P_FIRSTKEY; off <= maxoff; off = OffsetNumberNext(off))
693 : : {
694 : 314263 : ItemId thisii = PageGetItemId(rightmostpage, off);
695 : :
696 : 314263 : *previi = *thisii;
697 : 314263 : previi = thisii;
698 : : }
699 : 5527 : ((PageHeader) rightmostpage)->pd_lower -= sizeof(ItemIdData);
10141 scrappy@hub.org 700 : 5527 : }
701 : :
702 : : /*
703 : : * Add an item to a page being built.
704 : : *
705 : : * This is very similar to nbtinsert.c's _bt_pgaddtup(), but this variant
706 : : * raises an error directly.
707 : : *
708 : : * Note that our nbtsort.c caller does not know yet if the page will be
709 : : * rightmost. Offset P_FIRSTKEY is always assumed to be the first data key by
710 : : * caller. Page that turns out to be the rightmost on its level is fixed by
711 : : * calling _bt_slideleft().
712 : : */
713 : : static void
8668 tgl@sss.pgh.pa.us 714 : 5347041 : _bt_sortaddtup(Page page,
715 : : Size itemsize,
716 : : IndexTuple itup,
717 : : OffsetNumber itup_off,
718 : : bool newfirstdataitem)
719 : : {
720 : : IndexTupleData trunctuple;
721 : :
1462 pg@bowt.ie 722 [ + + ]: 5347041 : if (newfirstdataitem)
723 : : {
6654 tgl@sss.pgh.pa.us 724 : 1176 : trunctuple = *itup;
725 : 1176 : trunctuple.t_info = sizeof(IndexTupleData);
1468 pg@bowt.ie 726 : 1176 : BTreeTupleSetNAtts(&trunctuple, 0, false);
6654 tgl@sss.pgh.pa.us 727 : 1176 : itup = &trunctuple;
728 : 1176 : itemsize = sizeof(IndexTupleData);
729 : : }
730 : :
731 [ - + ]: 5347041 : if (PageAddItem(page, (Item) itup, itemsize, itup_off,
732 : : false, false) == InvalidOffsetNumber)
7573 tgl@sss.pgh.pa.us 733 [ # # ]:UBC 0 : elog(ERROR, "failed to add item to the index page");
9923 scrappy@hub.org 734 :CBC 5347041 : }
735 : :
736 : : /*----------
737 : : * Add an item to a disk page from the sort output (or add a posting list
738 : : * item formed from the sort output).
739 : : *
740 : : * We must be careful to observe the page layout conventions of nbtsearch.c:
741 : : * - rightmost pages start data items at P_HIKEY instead of at P_FIRSTKEY.
742 : : * - on non-leaf pages, the key portion of the first item need not be
743 : : * stored, we should store only the link.
744 : : *
745 : : * A leaf page being built looks like:
746 : : *
747 : : * +----------------+---------------------------------+
748 : : * | PageHeaderData | linp0 linp1 linp2 ... |
749 : : * +-----------+----+---------------------------------+
750 : : * | ... linpN | |
751 : : * +-----------+--------------------------------------+
752 : : * | ^ last |
753 : : * | |
754 : : * +-------------+------------------------------------+
755 : : * | | itemN ... |
756 : : * +-------------+------------------+-----------------+
757 : : * | ... item3 item2 item1 | "special space" |
758 : : * +--------------------------------+-----------------+
759 : : *
760 : : * Contrast this with the diagram in bufpage.h; note the mismatch
761 : : * between linps and items. This is because we reserve linp0 as a
762 : : * placeholder for the pointer to the "high key" item; when we have
763 : : * filled up the page, we will set linp0 to point to itemN and clear
764 : : * linpN. On the other hand, if we find this is the last (rightmost)
765 : : * page, we leave the items alone and slide the linp array over. If
766 : : * the high key is to be truncated, offset 1 is deleted, and we insert
767 : : * the truncated high key at offset 1.
768 : : *
769 : : * 'last' pointer indicates the last offset added to the page.
770 : : *
771 : : * 'truncextra' is the size of the posting list in itup, if any. This
772 : : * information is stashed for the next call here, when we may benefit
773 : : * from considering the impact of truncating away the posting list on
774 : : * the page before deciding to finish the page off. Posting lists are
775 : : * often relatively large, so it is worth going to the trouble of
776 : : * accounting for the saving from truncating away the posting list of
777 : : * the tuple that becomes the high key (that may be the only way to
778 : : * get close to target free space on the page). Note that this is
779 : : * only used for the soft fillfactor-wise limit, not the critical hard
780 : : * limit.
781 : : *----------
782 : : */
783 : : static void
1509 pg@bowt.ie 784 : 5328337 : _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup,
785 : : Size truncextra)
786 : : {
787 : : BulkWriteBuffer nbuf;
788 : : Page npage;
789 : : BlockNumber nblkno;
790 : : OffsetNumber last_off;
791 : : Size last_truncextra;
792 : : Size pgspc;
793 : : Size itupsz;
794 : : bool isleaf;
795 : :
796 : : /*
797 : : * This is a handy place to check for cancel interrupts during the btree
798 : : * load phase of index creation.
799 : : */
6610 tgl@sss.pgh.pa.us 800 [ + + ]: 5328337 : CHECK_FOR_INTERRUPTS();
801 : :
51 heikki.linnakangas@i 802 :GNC 5328337 : nbuf = state->btps_buf;
803 : 5328337 : npage = (Page) nbuf;
7256 tgl@sss.pgh.pa.us 804 :CBC 5328337 : nblkno = state->btps_blkno;
9716 bruce@momjian.us 805 : 5328337 : last_off = state->btps_lastoff;
1509 pg@bowt.ie 806 : 5328337 : last_truncextra = state->btps_lastextra;
807 : 5328337 : state->btps_lastextra = truncextra;
808 : :
9716 bruce@momjian.us 809 : 5328337 : pgspc = PageGetFreeSpace(npage);
2237 tgl@sss.pgh.pa.us 810 : 5328337 : itupsz = IndexTupleSize(itup);
6654 811 : 5328337 : itupsz = MAXALIGN(itupsz);
812 : : /* Leaf case has slightly different rules due to suffix truncation */
1809 pg@bowt.ie 813 : 5328337 : isleaf = (state->btps_level == 0);
814 : :
815 : : /*
816 : : * Check whether the new item can fit on a btree page on current level at
817 : : * all.
818 : : *
819 : : * Every newly built index will treat heap TID as part of the keyspace,
820 : : * which imposes the requirement that new high keys must occasionally have
821 : : * a heap TID appended within _bt_truncate(). That may leave a new pivot
822 : : * tuple one or two MAXALIGN() quantums larger than the original
823 : : * firstright tuple it's derived from. v4 deals with the problem by
824 : : * decreasing the limit on the size of tuples inserted on the leaf level
825 : : * by the same small amount. Enforce the new v4+ limit on the leaf level,
826 : : * and the old limit on internal levels, since pivot tuples may need to
827 : : * make use of the reserved space. This should never fail on internal
828 : : * pages.
829 : : */
1852 830 [ + + ]: 5328337 : if (unlikely(itupsz > BTMaxItemSize(npage)))
1809 831 : 132 : _bt_check_third_page(wstate->index, wstate->heap, isleaf, npage,
832 : : itup);
833 : :
834 : : /*
835 : : * Check to see if current page will fit new item, with space left over to
836 : : * append a heap TID during suffix truncation when page is a leaf page.
837 : : *
838 : : * It is guaranteed that we can fit at least 2 non-pivot tuples plus a
839 : : * high key with heap TID when finishing off a leaf page, since we rely on
840 : : * _bt_check_third_page() rejecting oversized non-pivot tuples. On
841 : : * internal pages we can always fit 3 pivot tuples with larger internal
842 : : * page tuple limit (includes page high key).
843 : : *
844 : : * Most of the time, a page is only "full" in the sense that the soft
845 : : * fillfactor-wise limit has been exceeded. However, we must always leave
846 : : * at least two items plus a high key on each page before starting a new
847 : : * page. Disregard fillfactor and insert on "full" current page if we
848 : : * don't have the minimum number of items yet. (Note that we deliberately
849 : : * assume that suffix truncation neither enlarges nor shrinks new high key
850 : : * when applying soft limit, except when last tuple has a posting list.)
851 : : */
1475 852 [ + + - + ]: 5328337 : Assert(last_truncextra == 0 || isleaf);
1809 853 [ + + + + ]: 5328337 : if (pgspc < itupsz + (isleaf ? MAXALIGN(sizeof(ItemPointerData)) : 0) ||
1509 854 [ + + + - ]: 5327763 : (pgspc + last_truncextra < state->btps_full && last_off > P_FIRSTKEY))
855 : : {
856 : : /*
857 : : * Finish off the page and write it out.
858 : : */
51 heikki.linnakangas@i 859 :GNC 18704 : BulkWriteBuffer obuf = nbuf;
9715 bruce@momjian.us 860 :CBC 18704 : Page opage = npage;
7168 861 : 18704 : BlockNumber oblkno = nblkno;
862 : : ItemId ii;
863 : : ItemId hii;
864 : : IndexTuple oitup;
865 : :
866 : : /* Create new page of same level */
51 heikki.linnakangas@i 867 :GNC 18704 : nbuf = _bt_blnewpage(wstate, state->btps_level);
868 : 18704 : npage = (Page) nbuf;
869 : :
870 : : /* and assign it a page position */
7256 tgl@sss.pgh.pa.us 871 :CBC 18704 : nblkno = wstate->btws_pages_alloced++;
872 : :
873 : : /*
874 : : * We copy the last item on the page into the new page, and then
875 : : * rearrange the old page so that the 'last item' becomes its high key
876 : : * rather than a true data item. There had better be at least two
877 : : * items on the page already, else the page would be empty of useful
878 : : * data.
879 : : */
8668 880 [ - + ]: 18704 : Assert(last_off > P_FIRSTKEY);
881 : 18704 : ii = PageGetItemId(opage, last_off);
6654 882 : 18704 : oitup = (IndexTuple) PageGetItem(opage, ii);
1462 pg@bowt.ie 883 : 18704 : _bt_sortaddtup(npage, ItemIdGetLength(ii), oitup, P_FIRSTKEY,
884 : 18704 : !isleaf);
885 : :
886 : : /*
887 : : * Move 'last' into the high key position on opage. _bt_blnewpage()
888 : : * allocated empty space for a line pointer when opage was first
889 : : * created, so this is a matter of rearranging already-allocated space
890 : : * on page, and initializing high key line pointer. (Actually, leaf
891 : : * pages must also swap oitup with a truncated version of oitup, which
892 : : * is sometimes larger than oitup, though never by more than the space
893 : : * needed to append a heap TID.)
894 : : */
9716 bruce@momjian.us 895 : 18704 : hii = PageGetItemId(opage, P_HIKEY);
896 : 18704 : *hii = *ii;
6059 tgl@sss.pgh.pa.us 897 : 18704 : ItemIdSetUnused(ii); /* redundant */
9716 bruce@momjian.us 898 : 18704 : ((PageHeader) opage)->pd_lower -= sizeof(ItemIdData);
899 : :
1809 pg@bowt.ie 900 [ + + ]: 18704 : if (isleaf)
901 : : {
902 : : IndexTuple lastleft;
903 : : IndexTuple truncated;
904 : :
905 : : /*
906 : : * Truncate away any unneeded attributes from high key on leaf
907 : : * level. This is only done at the leaf level because downlinks
908 : : * in internal pages are either negative infinity items, or get
909 : : * their contents from copying from one level down. See also:
910 : : * _bt_split().
911 : : *
912 : : * We don't try to bias our choice of split point to make it more
913 : : * likely that _bt_truncate() can truncate away more attributes,
914 : : * whereas the split point used within _bt_split() is chosen much
915 : : * more delicately. Even still, the lastleft and firstright
916 : : * tuples passed to _bt_truncate() here are at least not fully
917 : : * equal to each other when deduplication is used, unless there is
918 : : * a large group of duplicates (also, unique index builds usually
919 : : * have few or no spool2 duplicates). When the split point is
920 : : * between two unequal tuples, _bt_truncate() will avoid including
921 : : * a heap TID in the new high key, which is the most important
922 : : * benefit of suffix truncation.
923 : : *
924 : : * Overwrite the old item with new truncated high key directly.
925 : : * oitup is already located at the physical beginning of tuple
926 : : * space, so this should directly reuse the existing tuple space.
927 : : */
1852 928 : 18642 : ii = PageGetItemId(opage, OffsetNumberPrev(last_off));
929 : 18642 : lastleft = (IndexTuple) PageGetItem(opage, ii);
930 : :
1475 931 [ - + ]: 18642 : Assert(IndexTupleSize(oitup) > last_truncextra);
1852 932 : 18642 : truncated = _bt_truncate(wstate->index, lastleft, oitup,
933 : : wstate->inskey);
1706 934 [ - + ]: 18642 : if (!PageIndexTupleOverwrite(opage, P_HIKEY, (Item) truncated,
935 : 18642 : IndexTupleSize(truncated)))
1706 pg@bowt.ie 936 [ # # ]:UBC 0 : elog(ERROR, "failed to add high key to the index page");
2187 teodor@sigaev.ru 937 :CBC 18642 : pfree(truncated);
938 : :
939 : : /* oitup should continue to point to the page's high key */
940 : 18642 : hii = PageGetItemId(opage, P_HIKEY);
941 : 18642 : oitup = (IndexTuple) PageGetItem(opage, hii);
942 : : }
943 : :
944 : : /*
945 : : * Link the old page into its parent, using its low key. If we don't
946 : : * have a parent, we have to create one; this adds a new btree level.
947 : : */
7403 neilc@samurai.com 948 [ + + ]: 18704 : if (state->btps_next == NULL)
7256 tgl@sss.pgh.pa.us 949 : 1114 : state->btps_next = _bt_pagestate(wstate, state->btps_level + 1);
950 : :
1620 pg@bowt.ie 951 [ + - + - : 18704 : Assert((BTreeTupleGetNAtts(state->btps_lowkey, wstate->index) <=
+ - + + -
- - + ]
952 : : IndexRelationGetNumberOfKeyAttributes(wstate->index) &&
953 : : BTreeTupleGetNAtts(state->btps_lowkey, wstate->index) > 0) ||
954 : : P_LEFTMOST(BTPageGetOpaque(opage)));
955 [ + - + + : 18704 : Assert(BTreeTupleGetNAtts(state->btps_lowkey, wstate->index) == 0 ||
- + ]
956 : : !P_LEFTMOST(BTPageGetOpaque(opage)));
1581 957 : 18704 : BTreeTupleSetDownLink(state->btps_lowkey, oblkno);
1509 958 : 18704 : _bt_buildadd(wstate, state->btps_next, state->btps_lowkey, 0);
1620 959 : 18704 : pfree(state->btps_lowkey);
960 : :
961 : : /*
962 : : * Save a copy of the high key from the old page. It is also the low
963 : : * key for the new page.
964 : : */
965 : 18704 : state->btps_lowkey = CopyIndexTuple(oitup);
966 : :
967 : : /*
968 : : * Set the sibling links for both pages.
969 : : */
970 : : {
744 michael@paquier.xyz 971 : 18704 : BTPageOpaque oopaque = BTPageGetOpaque(opage);
972 : 18704 : BTPageOpaque nopaque = BTPageGetOpaque(npage);
973 : :
7256 tgl@sss.pgh.pa.us 974 : 18704 : oopaque->btpo_next = nblkno;
975 : 18704 : nopaque->btpo_prev = oblkno;
2489 976 : 18704 : nopaque->btpo_next = P_NONE; /* redundant */
977 : : }
978 : :
979 : : /*
980 : : * Write out the old page. _bt_blwritepage takes ownership of the
981 : : * 'opage' buffer.
982 : : */
51 heikki.linnakangas@i 983 :GNC 18704 : _bt_blwritepage(wstate, obuf, oblkno);
984 : :
985 : : /*
986 : : * Reset last_off to point to new page
987 : : */
8668 tgl@sss.pgh.pa.us 988 :CBC 18704 : last_off = P_FIRSTKEY;
989 : : }
990 : :
991 : : /*
992 : : * By here, either original page is still the current page, or a new page
993 : : * was created that became the current page. Either way, the current page
994 : : * definitely has space for new item.
995 : : *
996 : : * If the new item is the first for its page, it must also be the first
997 : : * item on its entire level. On later same-level pages, a low key for a
998 : : * page will be copied from the prior page in the code above. Generate a
999 : : * minus infinity low key here instead.
1000 : : */
1001 [ + + ]: 5328337 : if (last_off == P_HIKEY)
1002 : : {
1620 pg@bowt.ie 1003 [ - + ]: 5527 : Assert(state->btps_lowkey == NULL);
1004 : 5527 : state->btps_lowkey = palloc0(sizeof(IndexTupleData));
1005 : 5527 : state->btps_lowkey->t_info = sizeof(IndexTupleData);
1468 1006 : 5527 : BTreeTupleSetNAtts(state->btps_lowkey, 0, false);
1007 : : }
1008 : :
1009 : : /*
1010 : : * Add the new item into the current page.
1011 : : */
8668 tgl@sss.pgh.pa.us 1012 : 5328337 : last_off = OffsetNumberNext(last_off);
1462 pg@bowt.ie 1013 : 5328337 : _bt_sortaddtup(npage, itupsz, itup, last_off,
1014 [ + + + + ]: 5328337 : !isleaf && last_off == P_FIRSTKEY);
1015 : :
51 heikki.linnakangas@i 1016 :GNC 5328337 : state->btps_buf = nbuf;
7256 tgl@sss.pgh.pa.us 1017 :CBC 5328337 : state->btps_blkno = nblkno;
9716 bruce@momjian.us 1018 : 5328337 : state->btps_lastoff = last_off;
9923 scrappy@hub.org 1019 : 5328337 : }
1020 : :
1021 : : /*
1022 : : * Finalize pending posting list tuple, and add it to the index. Final tuple
1023 : : * is based on saved base tuple, and saved list of heap TIDs.
1024 : : *
1025 : : * This is almost like _bt_dedup_finish_pending(), but it adds a new tuple
1026 : : * using _bt_buildadd().
1027 : : */
1028 : : static void
1509 pg@bowt.ie 1029 : 2338909 : _bt_sort_dedup_finish_pending(BTWriteState *wstate, BTPageState *state,
1030 : : BTDedupState dstate)
1031 : : {
1032 [ - + ]: 2338909 : Assert(dstate->nitems > 0);
1033 : :
1034 [ + + ]: 2338909 : if (dstate->nitems == 1)
1035 : 2320532 : _bt_buildadd(wstate, state, dstate->base, 0);
1036 : : else
1037 : : {
1038 : : IndexTuple postingtuple;
1039 : : Size truncextra;
1040 : :
1041 : : /* form a tuple with a posting list */
1042 : 18377 : postingtuple = _bt_form_posting(dstate->base,
1043 : : dstate->htids,
1044 : : dstate->nhtids);
1045 : : /* Calculate posting list overhead */
1046 : 36754 : truncextra = IndexTupleSize(postingtuple) -
1047 : 18377 : BTreeTupleGetPostingOffset(postingtuple);
1048 : :
1049 : 18377 : _bt_buildadd(wstate, state, postingtuple, truncextra);
1050 : 18377 : pfree(postingtuple);
1051 : : }
1052 : :
1395 1053 : 2338909 : dstate->nmaxitems = 0;
1509 1054 : 2338909 : dstate->nhtids = 0;
1055 : 2338909 : dstate->nitems = 0;
1056 : 2338909 : dstate->phystupsize = 0;
1057 : 2338909 : }
1058 : :
1059 : : /*
1060 : : * Finish writing out the completed btree.
1061 : : */
1062 : : static void
7256 tgl@sss.pgh.pa.us 1063 : 22580 : _bt_uppershutdown(BTWriteState *wstate, BTPageState *state)
1064 : : {
1065 : : BTPageState *s;
7168 bruce@momjian.us 1066 : 22580 : BlockNumber rootblkno = P_NONE;
7503 tgl@sss.pgh.pa.us 1067 : 22580 : uint32 rootlevel = 0;
1068 : : BulkWriteBuffer metabuf;
1069 : :
1070 : : /*
1071 : : * Each iteration of this loop completes one more level of the tree.
1072 : : */
7403 neilc@samurai.com 1073 [ + + ]: 28107 : for (s = state; s != NULL; s = s->btps_next)
1074 : : {
1075 : : BlockNumber blkno;
1076 : : BTPageOpaque opaque;
1077 : :
7256 tgl@sss.pgh.pa.us 1078 : 5527 : blkno = s->btps_blkno;
51 heikki.linnakangas@i 1079 :GNC 5527 : opaque = BTPageGetOpaque((Page) s->btps_buf);
1080 : :
1081 : : /*
1082 : : * We have to link the last page on this level to somewhere.
1083 : : *
1084 : : * If we're at the top, it's the root, so attach it to the metapage.
1085 : : * Otherwise, add an entry for it to its parent using its low key.
1086 : : * This may cause the last page of the parent level to split, but
1087 : : * that's not a problem -- we haven't gotten to it yet.
1088 : : */
7403 neilc@samurai.com 1089 [ + + ]:CBC 5527 : if (s->btps_next == NULL)
1090 : : {
8668 tgl@sss.pgh.pa.us 1091 : 4413 : opaque->btpo_flags |= BTP_ROOT;
7503 1092 : 4413 : rootblkno = blkno;
1093 : 4413 : rootlevel = s->btps_level;
1094 : : }
1095 : : else
1096 : : {
1620 pg@bowt.ie 1097 [ + - + - : 1114 : Assert((BTreeTupleGetNAtts(s->btps_lowkey, wstate->index) <=
+ - - + -
- - - ]
1098 : : IndexRelationGetNumberOfKeyAttributes(wstate->index) &&
1099 : : BTreeTupleGetNAtts(s->btps_lowkey, wstate->index) > 0) ||
1100 : : P_LEFTMOST(opaque));
1101 [ + - + - : 1114 : Assert(BTreeTupleGetNAtts(s->btps_lowkey, wstate->index) == 0 ||
- + ]
1102 : : !P_LEFTMOST(opaque));
1581 1103 : 1114 : BTreeTupleSetDownLink(s->btps_lowkey, blkno);
1509 1104 : 1114 : _bt_buildadd(wstate, s->btps_next, s->btps_lowkey, 0);
1620 1105 : 1114 : pfree(s->btps_lowkey);
1106 : 1114 : s->btps_lowkey = NULL;
1107 : : }
1108 : :
1109 : : /*
1110 : : * This is the rightmost page, so the ItemId array needs to be slid
1111 : : * back one slot. Then we can dump out the page.
1112 : : */
51 heikki.linnakangas@i 1113 :GNC 5527 : _bt_slideleft((Page) s->btps_buf);
1114 : 5527 : _bt_blwritepage(wstate, s->btps_buf, s->btps_blkno);
1115 : 5527 : s->btps_buf = NULL; /* writepage took ownership of the buffer */
1116 : : }
1117 : :
1118 : : /*
1119 : : * As the last step in the process, construct the metapage and make it
1120 : : * point to the new root (unless we had no data at all, in which case it's
1121 : : * set to point to "P_NONE"). This changes the index to the "valid" state
1122 : : * by filling in a valid magic number in the metapage.
1123 : : */
1124 : 22580 : metabuf = smgr_bulk_get_buf(wstate->bulkstate);
1125 : 22580 : _bt_initmetapage((Page) metabuf, rootblkno, rootlevel,
1509 pg@bowt.ie 1126 :CBC 22580 : wstate->inskey->allequalimage);
51 heikki.linnakangas@i 1127 :GNC 22580 : _bt_blwritepage(wstate, metabuf, BTREE_METAPAGE);
10141 scrappy@hub.org 1128 :CBC 22580 : }
1129 : :
1130 : : /*
1131 : : * Read tuples in correct sort order from tuplesort, and load them into
1132 : : * btree leaves.
1133 : : */
1134 : : static void
7256 tgl@sss.pgh.pa.us 1135 : 22580 : _bt_load(BTWriteState *wstate, BTSpool *btspool, BTSpool *btspool2)
1136 : : {
8668 1137 : 22580 : BTPageState *state = NULL;
8648 inoue@tpf.co.jp 1138 : 22580 : bool merge = (btspool2 != NULL);
1139 : : IndexTuple itup,
6654 tgl@sss.pgh.pa.us 1140 : 22580 : itup2 = NULL;
1141 : : bool load1;
7256 1142 : 22580 : TupleDesc tupdes = RelationGetDescr(wstate->index);
1143 : : int i,
2199 teodor@sigaev.ru 1144 : 22580 : keysz = IndexRelationGetNumberOfKeyAttributes(wstate->index);
1145 : : SortSupport sortKeys;
1812 alvherre@alvh.no-ip. 1146 : 22580 : int64 tuples_done = 0;
1147 : : bool deduplicate;
1148 : :
51 heikki.linnakangas@i 1149 :GNC 22580 : wstate->bulkstate = smgr_bulk_start_rel(wstate->index, MAIN_FORKNUM);
1150 : :
1367 pg@bowt.ie 1151 [ + + + + :CBC 26751 : deduplicate = wstate->inskey->allequalimage && !btspool->isunique &&
+ - ]
1509 1152 [ + - - + : 4171 : BTGetDeduplicateItems(wstate->index);
+ + + - ]
1153 : :
8648 inoue@tpf.co.jp 1154 [ + + ]: 22580 : if (merge)
1155 : : {
1156 : : /*
1157 : : * Another BTSpool for dead tuples exists. Now we have to merge
1158 : : * btspool and btspool2.
1159 : : */
1160 : :
1161 : : /* the preparation of merge */
2680 rhaas@postgresql.org 1162 : 11 : itup = tuplesort_getindextuple(btspool->sortstate, true);
1163 : 11 : itup2 = tuplesort_getindextuple(btspool2->sortstate, true);
1164 : :
1165 : : /* Prepare SortSupport data for each column */
3446 1166 : 11 : sortKeys = (SortSupport) palloc0(keysz * sizeof(SortSupportData));
1167 : :
1168 [ + + ]: 23 : for (i = 0; i < keysz; i++)
1169 : : {
1170 : 12 : SortSupport sortKey = sortKeys + i;
1852 pg@bowt.ie 1171 : 12 : ScanKey scanKey = wstate->inskey->scankeys + i;
1172 : : int16 strategy;
1173 : :
3446 rhaas@postgresql.org 1174 : 12 : sortKey->ssup_cxt = CurrentMemoryContext;
1175 : 12 : sortKey->ssup_collation = scanKey->sk_collation;
1176 : 12 : sortKey->ssup_nulls_first =
1177 : 12 : (scanKey->sk_flags & SK_BT_NULLS_FIRST) != 0;
1178 : 12 : sortKey->ssup_attno = scanKey->sk_attno;
1179 : : /* Abbreviation is not supported here */
3373 1180 : 12 : sortKey->abbreviate = false;
1181 : :
534 peter@eisentraut.org 1182 [ - + ]: 12 : Assert(sortKey->ssup_attno != 0);
1183 : :
3446 rhaas@postgresql.org 1184 [ - + ]: 12 : strategy = (scanKey->sk_flags & SK_BT_DESC) != 0 ?
1185 : : BTGreaterStrategyNumber : BTLessStrategyNumber;
1186 : :
1187 : 12 : PrepareSortSupportFromIndexRel(wstate->index, strategy, sortKey);
1188 : : }
1189 : :
1190 : : for (;;)
1191 : : {
8424 bruce@momjian.us 1192 : 1641 : load1 = true; /* load BTSpool next ? */
6654 tgl@sss.pgh.pa.us 1193 [ + + ]: 1641 : if (itup2 == NULL)
1194 : : {
1195 [ + + ]: 77 : if (itup == NULL)
8648 inoue@tpf.co.jp 1196 : 11 : break;
1197 : : }
6654 tgl@sss.pgh.pa.us 1198 [ + + ]: 1564 : else if (itup != NULL)
1199 : : {
1852 pg@bowt.ie 1200 : 1477 : int32 compare = 0;
1201 : :
8648 inoue@tpf.co.jp 1202 [ + + ]: 1592 : for (i = 1; i <= keysz; i++)
1203 : : {
1204 : : SortSupport entry;
1205 : : Datum attrDatum1,
1206 : : attrDatum2;
1207 : : bool isNull1,
1208 : : isNull2;
1209 : :
3446 rhaas@postgresql.org 1210 : 1501 : entry = sortKeys + i - 1;
6305 tgl@sss.pgh.pa.us 1211 : 1501 : attrDatum1 = index_getattr(itup, i, tupdes, &isNull1);
1212 : 1501 : attrDatum2 = index_getattr(itup2, i, tupdes, &isNull2);
1213 : :
3446 rhaas@postgresql.org 1214 : 1501 : compare = ApplySortComparator(attrDatum1, isNull1,
1215 : : attrDatum2, isNull2,
1216 : : entry);
6305 tgl@sss.pgh.pa.us 1217 [ + + ]: 1501 : if (compare > 0)
1218 : : {
1219 : 128 : load1 = false;
1220 : 1386 : break;
1221 : : }
1222 [ + + ]: 1373 : else if (compare < 0)
1223 : 1258 : break;
1224 : : }
1225 : :
1226 : : /*
1227 : : * If key values are equal, we sort on ItemPointer. This is
1228 : : * required for btree indexes, since heap TID is treated as an
1229 : : * implicit last key attribute in order to ensure that all
1230 : : * keys in the index are physically unique.
1231 : : */
1852 pg@bowt.ie 1232 [ + + ]: 1477 : if (compare == 0)
1233 : : {
1234 : 91 : compare = ItemPointerCompare(&itup->t_tid, &itup2->t_tid);
1235 [ - + ]: 91 : Assert(compare != 0);
1236 [ + + ]: 91 : if (compare > 0)
1237 : 15 : load1 = false;
1238 : : }
1239 : : }
1240 : : else
8648 inoue@tpf.co.jp 1241 : 87 : load1 = false;
1242 : :
1243 : : /* When we see first tuple, create first index page */
1244 [ + + ]: 1630 : if (state == NULL)
7256 tgl@sss.pgh.pa.us 1245 : 11 : state = _bt_pagestate(wstate, 0);
1246 : :
8648 inoue@tpf.co.jp 1247 [ + + ]: 1630 : if (load1)
1248 : : {
1509 pg@bowt.ie 1249 : 1400 : _bt_buildadd(wstate, state, itup, 0);
2680 rhaas@postgresql.org 1250 : 1400 : itup = tuplesort_getindextuple(btspool->sortstate, true);
1251 : : }
1252 : : else
1253 : : {
1509 pg@bowt.ie 1254 : 230 : _bt_buildadd(wstate, state, itup2, 0);
2680 rhaas@postgresql.org 1255 : 230 : itup2 = tuplesort_getindextuple(btspool2->sortstate, true);
1256 : : }
1257 : :
1258 : : /* Report progress */
1839 alvherre@alvh.no-ip. 1259 : 1630 : pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_DONE,
1260 : : ++tuples_done);
1261 : : }
3446 rhaas@postgresql.org 1262 : 11 : pfree(sortKeys);
1263 : : }
1509 pg@bowt.ie 1264 [ + + ]: 22569 : else if (deduplicate)
1265 : : {
1266 : : /* merge is unnecessary, deduplicate into posting lists */
1267 : : BTDedupState dstate;
1268 : :
1269 : 4171 : dstate = (BTDedupState) palloc(sizeof(BTDedupStateData));
1270 : 4171 : dstate->deduplicate = true; /* unused */
1395 1271 : 4171 : dstate->nmaxitems = 0; /* unused */
1509 1272 : 4171 : dstate->maxpostingsize = 0; /* set later */
1273 : : /* Metadata about base tuple of current pending posting list */
1274 : 4171 : dstate->base = NULL;
1275 : 4171 : dstate->baseoff = InvalidOffsetNumber; /* unused */
1276 : 4171 : dstate->basetupsize = 0;
1277 : : /* Metadata about current pending posting list TIDs */
1278 : 4171 : dstate->htids = NULL;
1279 : 4171 : dstate->nhtids = 0;
1280 : 4171 : dstate->nitems = 0;
1281 : 4171 : dstate->phystupsize = 0; /* unused */
1282 : 4171 : dstate->nintervals = 0; /* unused */
1283 : :
1284 : 2994071 : while ((itup = tuplesort_getindextuple(btspool->sortstate,
1285 [ + + ]: 2994071 : true)) != NULL)
1286 : : {
1287 : : /* When we see first tuple, create first index page */
1288 [ + + ]: 2989900 : if (state == NULL)
1289 : : {
1290 : 1147 : state = _bt_pagestate(wstate, 0);
1291 : :
1292 : : /*
1293 : : * Limit size of posting list tuples to 1/10 space we want to
1294 : : * leave behind on the page, plus space for final item's line
1295 : : * pointer. This is equal to the space that we'd like to
1296 : : * leave behind on each leaf page when fillfactor is 90,
1297 : : * allowing us to get close to fillfactor% space utilization
1298 : : * when there happen to be a great many duplicates. (This
1299 : : * makes higher leaf fillfactor settings ineffective when
1300 : : * building indexes that have many duplicates, but packing
1301 : : * leaf pages full with few very large tuples doesn't seem
1302 : : * like a useful goal.)
1303 : : */
1304 : 1147 : dstate->maxpostingsize = MAXALIGN_DOWN((BLCKSZ * 10 / 100)) -
1305 : : sizeof(ItemIdData);
51 heikki.linnakangas@i 1306 [ + - - + ]:GNC 1147 : Assert(dstate->maxpostingsize <= BTMaxItemSize((Page) state->btps_buf) &&
1307 : : dstate->maxpostingsize <= INDEX_SIZE_MASK);
1509 pg@bowt.ie 1308 :CBC 1147 : dstate->htids = palloc(dstate->maxpostingsize);
1309 : :
1310 : : /* start new pending posting list with itup copy */
1311 : 1147 : _bt_dedup_start_pending(dstate, CopyIndexTuple(itup),
1312 : : InvalidOffsetNumber);
1313 : : }
1314 [ + + ]: 2988753 : else if (_bt_keep_natts_fast(wstate->index, dstate->base,
1315 [ + + ]: 654844 : itup) > keysz &&
1316 : 654844 : _bt_dedup_save_htid(dstate, itup))
1317 : : {
1318 : : /*
1319 : : * Tuple is equal to base tuple of pending posting list. Heap
1320 : : * TID from itup has been saved in state.
1321 : : */
1322 : : }
1323 : : else
1324 : : {
1325 : : /*
1326 : : * Tuple is not equal to pending posting list tuple, or
1327 : : * _bt_dedup_save_htid() opted to not merge current item into
1328 : : * pending posting list.
1329 : : */
1330 : 2337762 : _bt_sort_dedup_finish_pending(wstate, state, dstate);
1331 : 2337762 : pfree(dstate->base);
1332 : :
1333 : : /* start new pending posting list with itup copy */
1334 : 2337762 : _bt_dedup_start_pending(dstate, CopyIndexTuple(itup),
1335 : : InvalidOffsetNumber);
1336 : : }
1337 : :
1338 : : /* Report progress */
1339 : 2989900 : pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_DONE,
1340 : : ++tuples_done);
1341 : : }
1342 : :
1343 [ + + ]: 4171 : if (state)
1344 : : {
1345 : : /*
1346 : : * Handle the last item (there must be a last item when the
1347 : : * tuplesort returned one or more tuples)
1348 : : */
1349 : 1147 : _bt_sort_dedup_finish_pending(wstate, state, dstate);
1350 : 1147 : pfree(dstate->base);
1351 : 1147 : pfree(dstate->htids);
1352 : : }
1353 : :
1354 : 4171 : pfree(dstate);
1355 : : }
1356 : : else
1357 : : {
1358 : : /* merging and deduplication are both unnecessary */
6654 tgl@sss.pgh.pa.us 1359 : 2986378 : while ((itup = tuplesort_getindextuple(btspool->sortstate,
2680 rhaas@postgresql.org 1360 [ + + ]: 2986378 : true)) != NULL)
1361 : : {
1362 : : /* When we see first tuple, create first index page */
8648 inoue@tpf.co.jp 1363 [ + + ]: 2967980 : if (state == NULL)
7256 tgl@sss.pgh.pa.us 1364 : 3255 : state = _bt_pagestate(wstate, 0);
1365 : :
1509 pg@bowt.ie 1366 : 2967980 : _bt_buildadd(wstate, state, itup, 0);
1367 : :
1368 : : /* Report progress */
1839 alvherre@alvh.no-ip. 1369 : 2967980 : pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_DONE,
1370 : : ++tuples_done);
1371 : : }
1372 : : }
1373 : :
1374 : : /* Close down final pages and write the metapage */
7256 tgl@sss.pgh.pa.us 1375 : 22580 : _bt_uppershutdown(wstate, state);
51 heikki.linnakangas@i 1376 :GNC 22580 : smgr_bulk_finish(wstate->bulkstate);
10141 scrappy@hub.org 1377 :CBC 22580 : }
1378 : :
1379 : : /*
1380 : : * Create parallel context, and launch workers for leader.
1381 : : *
1382 : : * buildstate argument should be initialized (with the exception of the
1383 : : * tuplesort state in spools, which may later be created based on shared
1384 : : * state initially set up here).
1385 : : *
1386 : : * isconcurrent indicates if operation is CREATE INDEX CONCURRENTLY.
1387 : : *
1388 : : * request is the target number of parallel worker processes to launch.
1389 : : *
1390 : : * Sets buildstate's BTLeader, which caller must use to shut down parallel
1391 : : * mode by passing it to _bt_end_parallel() at the very end of its index
1392 : : * build. If not even a single worker process can be launched, this is
1393 : : * never set, and caller should proceed with a serial index build.
1394 : : */
1395 : : static void
2263 rhaas@postgresql.org 1396 : 71 : _bt_begin_parallel(BTBuildState *buildstate, bool isconcurrent, int request)
1397 : : {
1398 : : ParallelContext *pcxt;
1399 : : int scantuplesortstates;
1400 : : Snapshot snapshot;
1401 : : Size estbtshared;
1402 : : Size estsort;
1403 : : BTShared *btshared;
1404 : : Sharedsort *sharedsort;
1405 : : Sharedsort *sharedsort2;
1406 : 71 : BTSpool *btspool = buildstate->spool;
1407 : 71 : BTLeader *btleader = (BTLeader *) palloc0(sizeof(BTLeader));
1408 : : WalUsage *walusage;
1409 : : BufferUsage *bufferusage;
1410 : 71 : bool leaderparticipates = true;
1411 : : int querylen;
1412 : :
1413 : : #ifdef DISABLE_LEADER_PARTICIPATION
1414 : : leaderparticipates = false;
1415 : : #endif
1416 : :
1417 : : /*
1418 : : * Enter parallel mode, and create context for parallel build of btree
1419 : : * index
1420 : : */
1421 : 71 : EnterParallelMode();
1422 [ - + ]: 71 : Assert(request > 0);
1423 : 71 : pcxt = CreateParallelContext("postgres", "_bt_parallel_build_main",
1424 : : request);
1425 : :
1426 [ + - ]: 71 : scantuplesortstates = leaderparticipates ? request + 1 : request;
1427 : :
1428 : : /*
1429 : : * Prepare for scan of the base relation. In a normal index build, we use
1430 : : * SnapshotAny because we must retrieve all tuples and do our own time
1431 : : * qual checks (because we have to index RECENTLY_DEAD tuples). In a
1432 : : * concurrent build, we take a regular MVCC snapshot and index whatever's
1433 : : * live according to that.
1434 : : */
1435 [ + - ]: 71 : if (!isconcurrent)
1436 : 71 : snapshot = SnapshotAny;
1437 : : else
2263 rhaas@postgresql.org 1438 :UBC 0 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1439 : :
1440 : : /*
1441 : : * Estimate size for our own PARALLEL_KEY_BTREE_SHARED workspace, and
1442 : : * PARALLEL_KEY_TUPLESORT tuplesort workspace
1443 : : */
1861 andres@anarazel.de 1444 :CBC 71 : estbtshared = _bt_parallel_estimate_shared(btspool->heap, snapshot);
2263 rhaas@postgresql.org 1445 : 71 : shm_toc_estimate_chunk(&pcxt->estimator, estbtshared);
1446 : 71 : estsort = tuplesort_estimate_shared(scantuplesortstates);
1447 : 71 : shm_toc_estimate_chunk(&pcxt->estimator, estsort);
1448 : :
1449 : : /*
1450 : : * Unique case requires a second spool, and so we may have to account for
1451 : : * another shared workspace for that -- PARALLEL_KEY_TUPLESORT_SPOOL2
1452 : : */
1453 [ + + ]: 71 : if (!btspool->isunique)
1454 : 39 : shm_toc_estimate_keys(&pcxt->estimator, 2);
1455 : : else
1456 : : {
1457 : 32 : shm_toc_estimate_chunk(&pcxt->estimator, estsort);
1458 : 32 : shm_toc_estimate_keys(&pcxt->estimator, 3);
1459 : : }
1460 : :
1461 : : /*
1462 : : * Estimate space for WalUsage and BufferUsage -- PARALLEL_KEY_WAL_USAGE
1463 : : * and PARALLEL_KEY_BUFFER_USAGE.
1464 : : *
1465 : : * If there are no extensions loaded that care, we could skip this. We
1466 : : * have no way of knowing whether anyone's looking at pgWalUsage or
1467 : : * pgBufferUsage, so do it unconditionally.
1468 : : */
1471 akapila@postgresql.o 1469 : 71 : shm_toc_estimate_chunk(&pcxt->estimator,
1470 : : mul_size(sizeof(WalUsage), pcxt->nworkers));
1471 : 71 : shm_toc_estimate_keys(&pcxt->estimator, 1);
1466 1472 : 71 : shm_toc_estimate_chunk(&pcxt->estimator,
1473 : : mul_size(sizeof(BufferUsage), pcxt->nworkers));
1474 : 71 : shm_toc_estimate_keys(&pcxt->estimator, 1);
1475 : :
1476 : : /* Finally, estimate PARALLEL_KEY_QUERY_TEXT space */
1261 noah@leadboat.com 1477 [ + - ]: 71 : if (debug_query_string)
1478 : : {
1479 : 71 : querylen = strlen(debug_query_string);
1480 : 71 : shm_toc_estimate_chunk(&pcxt->estimator, querylen + 1);
1481 : 71 : shm_toc_estimate_keys(&pcxt->estimator, 1);
1482 : : }
1483 : : else
1261 noah@leadboat.com 1484 :UBC 0 : querylen = 0; /* keep compiler quiet */
1485 : :
1486 : : /* Everyone's had a chance to ask for space, so now create the DSM */
2263 rhaas@postgresql.org 1487 :CBC 71 : InitializeParallelDSM(pcxt);
1488 : :
1489 : : /* If no DSM segment was available, back out (do serial build) */
1530 tmunro@postgresql.or 1490 [ - + ]: 71 : if (pcxt->seg == NULL)
1491 : : {
1530 tmunro@postgresql.or 1492 [ # # # # ]:UBC 0 : if (IsMVCCSnapshot(snapshot))
1493 : 0 : UnregisterSnapshot(snapshot);
1494 : 0 : DestroyParallelContext(pcxt);
1495 : 0 : ExitParallelMode();
1496 : 0 : return;
1497 : : }
1498 : :
1499 : : /* Store shared build state, for which we reserved space */
2263 rhaas@postgresql.org 1500 :CBC 71 : btshared = (BTShared *) shm_toc_allocate(pcxt->toc, estbtshared);
1501 : : /* Initialize immutable state */
1502 : 71 : btshared->heaprelid = RelationGetRelid(btspool->heap);
1503 : 71 : btshared->indexrelid = RelationGetRelid(btspool->index);
1504 : 71 : btshared->isunique = btspool->isunique;
801 peter@eisentraut.org 1505 : 71 : btshared->nulls_not_distinct = btspool->nulls_not_distinct;
2263 rhaas@postgresql.org 1506 : 71 : btshared->isconcurrent = isconcurrent;
1507 : 71 : btshared->scantuplesortstates = scantuplesortstates;
1508 : 71 : ConditionVariableInit(&btshared->workersdonecv);
1509 : 71 : SpinLockInit(&btshared->mutex);
1510 : : /* Initialize mutable state */
1511 : 71 : btshared->nparticipantsdone = 0;
1512 : 71 : btshared->reltuples = 0.0;
1513 : 71 : btshared->havedead = false;
1514 : 71 : btshared->indtuples = 0.0;
1515 : 71 : btshared->brokenhotchain = false;
1861 andres@anarazel.de 1516 : 71 : table_parallelscan_initialize(btspool->heap,
1517 : : ParallelTableScanFromBTShared(btshared),
1518 : : snapshot);
1519 : :
1520 : : /*
1521 : : * Store shared tuplesort-private state, for which we reserved space.
1522 : : * Then, initialize opaque state using tuplesort routine.
1523 : : */
2263 rhaas@postgresql.org 1524 : 71 : sharedsort = (Sharedsort *) shm_toc_allocate(pcxt->toc, estsort);
1525 : 71 : tuplesort_initialize_shared(sharedsort, scantuplesortstates,
1526 : : pcxt->seg);
1527 : :
1528 : 71 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_BTREE_SHARED, btshared);
1529 : 71 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_TUPLESORT, sharedsort);
1530 : :
1531 : : /* Unique case requires a second spool, and associated shared state */
1532 [ + + ]: 71 : if (!btspool->isunique)
1533 : 39 : sharedsort2 = NULL;
1534 : : else
1535 : : {
1536 : : /*
1537 : : * Store additional shared tuplesort-private state, for which we
1538 : : * reserved space. Then, initialize opaque state using tuplesort
1539 : : * routine.
1540 : : */
1541 : 32 : sharedsort2 = (Sharedsort *) shm_toc_allocate(pcxt->toc, estsort);
1542 : 32 : tuplesort_initialize_shared(sharedsort2, scantuplesortstates,
1543 : : pcxt->seg);
1544 : :
1545 : 32 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_TUPLESORT_SPOOL2, sharedsort2);
1546 : : }
1547 : :
1548 : : /* Store query string for workers */
1261 noah@leadboat.com 1549 [ + - ]: 71 : if (debug_query_string)
1550 : : {
1551 : : char *sharedquery;
1552 : :
1553 : 71 : sharedquery = (char *) shm_toc_allocate(pcxt->toc, querylen + 1);
1554 : 71 : memcpy(sharedquery, debug_query_string, querylen + 1);
1555 : 71 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_QUERY_TEXT, sharedquery);
1556 : : }
1557 : :
1558 : : /*
1559 : : * Allocate space for each worker's WalUsage and BufferUsage; no need to
1560 : : * initialize.
1561 : : */
1471 akapila@postgresql.o 1562 : 71 : walusage = shm_toc_allocate(pcxt->toc,
1563 : 71 : mul_size(sizeof(WalUsage), pcxt->nworkers));
1564 : 71 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_WAL_USAGE, walusage);
1466 1565 : 71 : bufferusage = shm_toc_allocate(pcxt->toc,
1566 : 71 : mul_size(sizeof(BufferUsage), pcxt->nworkers));
1567 : 71 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_BUFFER_USAGE, bufferusage);
1568 : :
1569 : : /* Launch workers, saving status for leader/caller */
2263 rhaas@postgresql.org 1570 : 71 : LaunchParallelWorkers(pcxt);
1571 : 71 : btleader->pcxt = pcxt;
1572 : 71 : btleader->nparticipanttuplesorts = pcxt->nworkers_launched;
1573 [ + - ]: 71 : if (leaderparticipates)
1574 : 71 : btleader->nparticipanttuplesorts++;
1575 : 71 : btleader->btshared = btshared;
1576 : 71 : btleader->sharedsort = sharedsort;
1577 : 71 : btleader->sharedsort2 = sharedsort2;
1578 : 71 : btleader->snapshot = snapshot;
1471 akapila@postgresql.o 1579 : 71 : btleader->walusage = walusage;
1466 1580 : 71 : btleader->bufferusage = bufferusage;
1581 : :
1582 : : /* If no workers were successfully launched, back out (do serial build) */
2263 rhaas@postgresql.org 1583 [ - + ]: 71 : if (pcxt->nworkers_launched == 0)
1584 : : {
2263 rhaas@postgresql.org 1585 :UBC 0 : _bt_end_parallel(btleader);
1586 : 0 : return;
1587 : : }
1588 : :
1589 : : /* Save leader state now that it's clear build will be parallel */
2263 rhaas@postgresql.org 1590 :CBC 71 : buildstate->btleader = btleader;
1591 : :
1592 : : /* Join heap scan ourselves */
1593 [ + - ]: 71 : if (leaderparticipates)
1594 : 71 : _bt_leader_participate_as_worker(buildstate);
1595 : :
1596 : : /*
1597 : : * Caller needs to wait for all launched workers when we return. Make
1598 : : * sure that the failure-to-start case will not hang forever.
1599 : : */
1600 : 71 : WaitForParallelWorkersToAttach(pcxt);
1601 : : }
1602 : :
1603 : : /*
1604 : : * Shut down workers, destroy parallel context, and end parallel mode.
1605 : : */
1606 : : static void
1607 : 71 : _bt_end_parallel(BTLeader *btleader)
1608 : : {
1609 : : int i;
1610 : :
1611 : : /* Shutdown worker processes */
1612 : 71 : WaitForParallelWorkersToFinish(btleader->pcxt);
1613 : :
1614 : : /*
1615 : : * Next, accumulate WAL usage. (This must wait for the workers to finish,
1616 : : * or we might get incomplete data.)
1617 : : */
1471 akapila@postgresql.o 1618 [ + + ]: 142 : for (i = 0; i < btleader->pcxt->nworkers_launched; i++)
1466 1619 : 71 : InstrAccumParallelQuery(&btleader->bufferusage[i], &btleader->walusage[i]);
1620 : :
1621 : : /* Free last reference to MVCC snapshot, if one was used */
2263 rhaas@postgresql.org 1622 [ + - - + ]: 71 : if (IsMVCCSnapshot(btleader->snapshot))
2263 rhaas@postgresql.org 1623 :UBC 0 : UnregisterSnapshot(btleader->snapshot);
2263 rhaas@postgresql.org 1624 :CBC 71 : DestroyParallelContext(btleader->pcxt);
1625 : 71 : ExitParallelMode();
1626 : 71 : }
1627 : :
1628 : : /*
1629 : : * Returns size of shared memory required to store state for a parallel
1630 : : * btree index build based on the snapshot its parallel scan will use.
1631 : : */
1632 : : static Size
1861 andres@anarazel.de 1633 : 71 : _bt_parallel_estimate_shared(Relation heap, Snapshot snapshot)
1634 : : {
1635 : : /* c.f. shm_toc_allocate as to why BUFFERALIGN is used */
1636 : 71 : return add_size(BUFFERALIGN(sizeof(BTShared)),
1637 : : table_parallelscan_estimate(heap, snapshot));
1638 : : }
1639 : :
1640 : : /*
1641 : : * Within leader, wait for end of heap scan.
1642 : : *
1643 : : * When called, parallel heap scan started by _bt_begin_parallel() will
1644 : : * already be underway within worker processes (when leader participates
1645 : : * as a worker, we should end up here just as workers are finishing).
1646 : : *
1647 : : * Fills in fields needed for ambuild statistics, and lets caller set
1648 : : * field indicating that some worker encountered a broken HOT chain.
1649 : : *
1650 : : * Returns the total number of heap tuples scanned.
1651 : : */
1652 : : static double
2263 rhaas@postgresql.org 1653 : 71 : _bt_parallel_heapscan(BTBuildState *buildstate, bool *brokenhotchain)
1654 : : {
1655 : 71 : BTShared *btshared = buildstate->btleader->btshared;
1656 : : int nparticipanttuplesorts;
1657 : : double reltuples;
1658 : :
1659 : 71 : nparticipanttuplesorts = buildstate->btleader->nparticipanttuplesorts;
1660 : : for (;;)
1661 : : {
1662 [ - + ]: 187 : SpinLockAcquire(&btshared->mutex);
1663 [ + + ]: 187 : if (btshared->nparticipantsdone == nparticipanttuplesorts)
1664 : : {
1665 : 71 : buildstate->havedead = btshared->havedead;
1666 : 71 : buildstate->indtuples = btshared->indtuples;
1667 : 71 : *brokenhotchain = btshared->brokenhotchain;
1668 : 71 : reltuples = btshared->reltuples;
1669 : 71 : SpinLockRelease(&btshared->mutex);
1670 : 71 : break;
1671 : : }
1672 : 116 : SpinLockRelease(&btshared->mutex);
1673 : :
1674 : 116 : ConditionVariableSleep(&btshared->workersdonecv,
1675 : : WAIT_EVENT_PARALLEL_CREATE_INDEX_SCAN);
1676 : : }
1677 : :
1678 : 71 : ConditionVariableCancelSleep();
1679 : :
1680 : 71 : return reltuples;
1681 : : }
1682 : :
1683 : : /*
1684 : : * Within leader, participate as a parallel worker.
1685 : : */
1686 : : static void
1687 : 71 : _bt_leader_participate_as_worker(BTBuildState *buildstate)
1688 : : {
1689 : 71 : BTLeader *btleader = buildstate->btleader;
1690 : : BTSpool *leaderworker;
1691 : : BTSpool *leaderworker2;
1692 : : int sortmem;
1693 : :
1694 : : /* Allocate memory and initialize private spool */
1695 : 71 : leaderworker = (BTSpool *) palloc0(sizeof(BTSpool));
1696 : 71 : leaderworker->heap = buildstate->spool->heap;
1697 : 71 : leaderworker->index = buildstate->spool->index;
1698 : 71 : leaderworker->isunique = buildstate->spool->isunique;
801 peter@eisentraut.org 1699 : 71 : leaderworker->nulls_not_distinct = buildstate->spool->nulls_not_distinct;
1700 : :
1701 : : /* Initialize second spool, if required */
2263 rhaas@postgresql.org 1702 [ + + ]: 71 : if (!btleader->btshared->isunique)
1703 : 39 : leaderworker2 = NULL;
1704 : : else
1705 : : {
1706 : : /* Allocate memory for worker's own private secondary spool */
1707 : 32 : leaderworker2 = (BTSpool *) palloc0(sizeof(BTSpool));
1708 : :
1709 : : /* Initialize worker's own secondary spool */
1710 : 32 : leaderworker2->heap = leaderworker->heap;
1711 : 32 : leaderworker2->index = leaderworker->index;
1712 : 32 : leaderworker2->isunique = false;
1713 : : }
1714 : :
1715 : : /*
1716 : : * Might as well use reliable figure when doling out maintenance_work_mem
1717 : : * (when requested number of workers were not launched, this will be
1718 : : * somewhat higher than it is for other workers).
1719 : : */
1720 : 71 : sortmem = maintenance_work_mem / btleader->nparticipanttuplesorts;
1721 : :
1722 : : /* Perform work common to all participants */
1723 : 71 : _bt_parallel_scan_and_sort(leaderworker, leaderworker2, btleader->btshared,
1724 : : btleader->sharedsort, btleader->sharedsort2,
1725 : : sortmem, true);
1726 : :
1727 : : #ifdef BTREE_BUILD_STATS
1728 : : if (log_btree_build_stats)
1729 : : {
1730 : : ShowUsage("BTREE BUILD (Leader Partial Spool) STATISTICS");
1731 : : ResetUsage();
1732 : : }
1733 : : #endif /* BTREE_BUILD_STATS */
1734 : 71 : }
1735 : :
1736 : : /*
1737 : : * Perform work within a launched parallel process.
1738 : : */
1739 : : void
1740 : 71 : _bt_parallel_build_main(dsm_segment *seg, shm_toc *toc)
1741 : : {
1742 : : char *sharedquery;
1743 : : BTSpool *btspool;
1744 : : BTSpool *btspool2;
1745 : : BTShared *btshared;
1746 : : Sharedsort *sharedsort;
1747 : : Sharedsort *sharedsort2;
1748 : : Relation heapRel;
1749 : : Relation indexRel;
1750 : : LOCKMODE heapLockmode;
1751 : : LOCKMODE indexLockmode;
1752 : : WalUsage *walusage;
1753 : : BufferUsage *bufferusage;
1754 : : int sortmem;
1755 : :
1756 : : #ifdef BTREE_BUILD_STATS
1757 : : if (log_btree_build_stats)
1758 : : ResetUsage();
1759 : : #endif /* BTREE_BUILD_STATS */
1760 : :
1761 : : /*
1762 : : * The only possible status flag that can be set to the parallel worker is
1763 : : * PROC_IN_SAFE_IC.
1764 : : */
877 akapila@postgresql.o 1765 [ - + - - ]: 71 : Assert((MyProc->statusFlags == 0) ||
1766 : : (MyProc->statusFlags == PROC_IN_SAFE_IC));
1767 : :
1768 : : /* Set debug_query_string for individual workers first */
1261 noah@leadboat.com 1769 : 71 : sharedquery = shm_toc_lookup(toc, PARALLEL_KEY_QUERY_TEXT, true);
2215 rhaas@postgresql.org 1770 : 71 : debug_query_string = sharedquery;
1771 : :
1772 : : /* Report the query string from leader */
1773 : 71 : pgstat_report_activity(STATE_RUNNING, debug_query_string);
1774 : :
1775 : : /* Look up nbtree shared state */
2263 1776 : 71 : btshared = shm_toc_lookup(toc, PARALLEL_KEY_BTREE_SHARED, false);
1777 : :
1778 : : /* Open relations using lock modes known to be obtained by index.c */
1779 [ + - ]: 71 : if (!btshared->isconcurrent)
1780 : : {
1781 : 71 : heapLockmode = ShareLock;
1782 : 71 : indexLockmode = AccessExclusiveLock;
1783 : : }
1784 : : else
1785 : : {
2263 rhaas@postgresql.org 1786 :UBC 0 : heapLockmode = ShareUpdateExclusiveLock;
1787 : 0 : indexLockmode = RowExclusiveLock;
1788 : : }
1789 : :
1790 : : /* Open relations within worker */
1910 andres@anarazel.de 1791 :CBC 71 : heapRel = table_open(btshared->heaprelid, heapLockmode);
2263 rhaas@postgresql.org 1792 : 71 : indexRel = index_open(btshared->indexrelid, indexLockmode);
1793 : :
1794 : : /* Initialize worker's own spool */
1795 : 71 : btspool = (BTSpool *) palloc0(sizeof(BTSpool));
1796 : 71 : btspool->heap = heapRel;
1797 : 71 : btspool->index = indexRel;
1798 : 71 : btspool->isunique = btshared->isunique;
801 peter@eisentraut.org 1799 : 71 : btspool->nulls_not_distinct = btshared->nulls_not_distinct;
1800 : :
1801 : : /* Look up shared state private to tuplesort.c */
2263 rhaas@postgresql.org 1802 : 71 : sharedsort = shm_toc_lookup(toc, PARALLEL_KEY_TUPLESORT, false);
1803 : 71 : tuplesort_attach_shared(sharedsort, seg);
1804 [ + + ]: 71 : if (!btshared->isunique)
1805 : : {
1806 : 39 : btspool2 = NULL;
1807 : 39 : sharedsort2 = NULL;
1808 : : }
1809 : : else
1810 : : {
1811 : : /* Allocate memory for worker's own private secondary spool */
1812 : 32 : btspool2 = (BTSpool *) palloc0(sizeof(BTSpool));
1813 : :
1814 : : /* Initialize worker's own secondary spool */
1815 : 32 : btspool2->heap = btspool->heap;
1816 : 32 : btspool2->index = btspool->index;
1817 : 32 : btspool2->isunique = false;
1818 : : /* Look up shared state private to tuplesort.c */
1819 : 32 : sharedsort2 = shm_toc_lookup(toc, PARALLEL_KEY_TUPLESORT_SPOOL2, false);
1820 : 32 : tuplesort_attach_shared(sharedsort2, seg);
1821 : : }
1822 : :
1823 : : /* Prepare to track buffer usage during parallel execution */
1471 akapila@postgresql.o 1824 : 71 : InstrStartParallelQuery();
1825 : :
1826 : : /* Perform sorting of spool, and possibly a spool2 */
2263 rhaas@postgresql.org 1827 : 71 : sortmem = maintenance_work_mem / btshared->scantuplesortstates;
1828 : 71 : _bt_parallel_scan_and_sort(btspool, btspool2, btshared, sharedsort,
1829 : : sharedsort2, sortmem, false);
1830 : :
1831 : : /* Report WAL/buffer usage during parallel execution */
1466 akapila@postgresql.o 1832 : 71 : bufferusage = shm_toc_lookup(toc, PARALLEL_KEY_BUFFER_USAGE, false);
1471 1833 : 71 : walusage = shm_toc_lookup(toc, PARALLEL_KEY_WAL_USAGE, false);
1466 1834 : 71 : InstrEndParallelQuery(&bufferusage[ParallelWorkerNumber],
1835 : 71 : &walusage[ParallelWorkerNumber]);
1836 : :
1837 : : #ifdef BTREE_BUILD_STATS
1838 : : if (log_btree_build_stats)
1839 : : {
1840 : : ShowUsage("BTREE BUILD (Worker Partial Spool) STATISTICS");
1841 : : ResetUsage();
1842 : : }
1843 : : #endif /* BTREE_BUILD_STATS */
1844 : :
2263 rhaas@postgresql.org 1845 : 71 : index_close(indexRel, indexLockmode);
1910 andres@anarazel.de 1846 : 71 : table_close(heapRel, heapLockmode);
2263 rhaas@postgresql.org 1847 : 71 : }
1848 : :
1849 : : /*
1850 : : * Perform a worker's portion of a parallel sort.
1851 : : *
1852 : : * This generates a tuplesort for passed btspool, and a second tuplesort
1853 : : * state if a second btspool is need (i.e. for unique index builds). All
1854 : : * other spool fields should already be set when this is called.
1855 : : *
1856 : : * sortmem is the amount of working memory to use within each worker,
1857 : : * expressed in KBs.
1858 : : *
1859 : : * When this returns, workers are done, and need only release resources.
1860 : : */
1861 : : static void
1862 : 142 : _bt_parallel_scan_and_sort(BTSpool *btspool, BTSpool *btspool2,
1863 : : BTShared *btshared, Sharedsort *sharedsort,
1864 : : Sharedsort *sharedsort2, int sortmem, bool progress)
1865 : : {
1866 : : SortCoordinate coordinate;
1867 : : BTBuildState buildstate;
1868 : : TableScanDesc scan;
1869 : : double reltuples;
1870 : : IndexInfo *indexInfo;
1871 : :
1872 : : /* Initialize local tuplesort coordination state */
1873 : 142 : coordinate = palloc0(sizeof(SortCoordinateData));
1874 : 142 : coordinate->isWorker = true;
1875 : 142 : coordinate->nParticipants = -1;
1876 : 142 : coordinate->sharedsort = sharedsort;
1877 : :
1878 : : /* Begin "partial" tuplesort */
1879 : 284 : btspool->sortstate = tuplesort_begin_index_btree(btspool->heap,
1880 : : btspool->index,
1881 : 142 : btspool->isunique,
801 peter@eisentraut.org 1882 : 142 : btspool->nulls_not_distinct,
1883 : : sortmem, coordinate,
1884 : : TUPLESORT_NONE);
1885 : :
1886 : : /*
1887 : : * Just as with serial case, there may be a second spool. If so, a
1888 : : * second, dedicated spool2 partial tuplesort is required.
1889 : : */
2263 rhaas@postgresql.org 1890 [ + + ]: 142 : if (btspool2)
1891 : : {
1892 : : SortCoordinate coordinate2;
1893 : :
1894 : : /*
1895 : : * We expect that the second one (for dead tuples) won't get very
1896 : : * full, so we give it only work_mem (unless sortmem is less for
1897 : : * worker). Worker processes are generally permitted to allocate
1898 : : * work_mem independently.
1899 : : */
1900 : 64 : coordinate2 = palloc0(sizeof(SortCoordinateData));
1901 : 64 : coordinate2->isWorker = true;
1902 : 64 : coordinate2->nParticipants = -1;
1903 : 64 : coordinate2->sharedsort = sharedsort2;
1904 : 64 : btspool2->sortstate =
801 peter@eisentraut.org 1905 : 64 : tuplesort_begin_index_btree(btspool->heap, btspool->index, false, false,
1906 : : Min(sortmem, work_mem), coordinate2,
1907 : : false);
1908 : : }
1909 : :
1910 : : /* Fill in buildstate for _bt_build_callback() */
2263 rhaas@postgresql.org 1911 : 142 : buildstate.isunique = btshared->isunique;
801 peter@eisentraut.org 1912 : 142 : buildstate.nulls_not_distinct = btshared->nulls_not_distinct;
2263 rhaas@postgresql.org 1913 : 142 : buildstate.havedead = false;
1914 : 142 : buildstate.heap = btspool->heap;
1915 : 142 : buildstate.spool = btspool;
1916 : 142 : buildstate.spool2 = btspool2;
1917 : 142 : buildstate.indtuples = 0;
1918 : 142 : buildstate.btleader = NULL;
1919 : :
1920 : : /* Join parallel scan */
1921 : 142 : indexInfo = BuildIndexInfo(btspool->index);
1922 : 142 : indexInfo->ii_Concurrent = btshared->isconcurrent;
1839 alvherre@alvh.no-ip. 1923 : 142 : scan = table_beginscan_parallel(btspool->heap,
1924 : : ParallelTableScanFromBTShared(btshared));
1845 andres@anarazel.de 1925 : 142 : reltuples = table_index_build_scan(btspool->heap, btspool->index, indexInfo,
1926 : : true, progress, _bt_build_callback,
1927 : : (void *) &buildstate, scan);
1928 : :
1929 : : /* Execute this worker's part of the sort */
1038 alvherre@alvh.no-ip. 1930 [ + + ]: 142 : if (progress)
1931 : 71 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
1932 : : PROGRESS_BTREE_PHASE_PERFORMSORT_1);
2263 rhaas@postgresql.org 1933 : 142 : tuplesort_performsort(btspool->sortstate);
1934 [ + + ]: 142 : if (btspool2)
1935 : : {
1038 alvherre@alvh.no-ip. 1936 [ + + ]: 64 : if (progress)
1937 : 32 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
1938 : : PROGRESS_BTREE_PHASE_PERFORMSORT_2);
2263 rhaas@postgresql.org 1939 : 64 : tuplesort_performsort(btspool2->sortstate);
1940 : : }
1941 : :
1942 : : /*
1943 : : * Done. Record ambuild statistics, and whether we encountered a broken
1944 : : * HOT chain.
1945 : : */
1946 [ - + ]: 142 : SpinLockAcquire(&btshared->mutex);
1947 : 142 : btshared->nparticipantsdone++;
1948 : 142 : btshared->reltuples += reltuples;
1949 [ - + ]: 142 : if (buildstate.havedead)
2263 rhaas@postgresql.org 1950 :UBC 0 : btshared->havedead = true;
2263 rhaas@postgresql.org 1951 :CBC 142 : btshared->indtuples += buildstate.indtuples;
1952 [ - + ]: 142 : if (indexInfo->ii_BrokenHotChain)
2263 rhaas@postgresql.org 1953 :UBC 0 : btshared->brokenhotchain = true;
2263 rhaas@postgresql.org 1954 :CBC 142 : SpinLockRelease(&btshared->mutex);
1955 : :
1956 : : /* Notify leader */
1957 : 142 : ConditionVariableSignal(&btshared->workersdonecv);
1958 : :
1959 : : /* We can end tuplesorts immediately */
1960 : 142 : tuplesort_end(btspool->sortstate);
1961 [ + + ]: 142 : if (btspool2)
1962 : 64 : tuplesort_end(btspool2->sortstate);
1963 : 142 : }
|
