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) 1996-2025, 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/tableam.h"
48 : : #include "access/xact.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"
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 : : /* Query ID, for report in worker processes */
109 : : int64 queryid;
110 : :
111 : : /*
112 : : * workersdonecv is used to monitor the progress of workers. All parallel
113 : : * participants must indicate that they are done before leader can use
114 : : * mutable state that workers maintain during scan (and before leader can
115 : : * proceed to tuplesort_performsort()).
116 : : */
117 : : ConditionVariable workersdonecv;
118 : :
119 : : /*
120 : : * mutex protects all fields before heapdesc.
121 : : *
122 : : * These fields contain status information of interest to B-Tree index
123 : : * builds that must work just the same when an index is built in parallel.
124 : : */
125 : : slock_t mutex;
126 : :
127 : : /*
128 : : * Mutable state that is maintained by workers, and reported back to
129 : : * leader at end of parallel scan.
130 : : *
131 : : * nparticipantsdone is number of worker processes finished.
132 : : *
133 : : * reltuples is the total number of input heap tuples.
134 : : *
135 : : * havedead indicates if RECENTLY_DEAD tuples were encountered during
136 : : * build.
137 : : *
138 : : * indtuples is the total number of tuples that made it into the index.
139 : : *
140 : : * brokenhotchain indicates if any worker detected a broken HOT chain
141 : : * during build.
142 : : */
143 : : int nparticipantsdone;
144 : : double reltuples;
145 : : bool havedead;
146 : : double indtuples;
147 : : bool brokenhotchain;
148 : :
149 : : /*
150 : : * ParallelTableScanDescData data follows. Can't directly embed here, as
151 : : * implementations of the parallel table scan desc interface might need
152 : : * stronger alignment.
153 : : */
154 : : } BTShared;
155 : :
156 : : /*
157 : : * Return pointer to a BTShared's parallel table scan.
158 : : *
159 : : * c.f. shm_toc_allocate as to why BUFFERALIGN is used, rather than just
160 : : * MAXALIGN.
161 : : */
162 : : #define ParallelTableScanFromBTShared(shared) \
163 : : (ParallelTableScanDesc) ((char *) (shared) + BUFFERALIGN(sizeof(BTShared)))
164 : :
165 : : /*
166 : : * Status for leader in parallel index build.
167 : : */
168 : : typedef struct BTLeader
169 : : {
170 : : /* parallel context itself */
171 : : ParallelContext *pcxt;
172 : :
173 : : /*
174 : : * nparticipanttuplesorts is the exact number of worker processes
175 : : * successfully launched, plus one leader process if it participates as a
176 : : * worker (only DISABLE_LEADER_PARTICIPATION builds avoid leader
177 : : * participating as a worker).
178 : : */
179 : : int nparticipanttuplesorts;
180 : :
181 : : /*
182 : : * Leader process convenience pointers to shared state (leader avoids TOC
183 : : * lookups).
184 : : *
185 : : * btshared is the shared state for entire build. sharedsort is the
186 : : * shared, tuplesort-managed state passed to each process tuplesort.
187 : : * sharedsort2 is the corresponding btspool2 shared state, used only when
188 : : * building unique indexes. snapshot is the snapshot used by the scan iff
189 : : * an MVCC snapshot is required.
190 : : */
191 : : BTShared *btshared;
192 : : Sharedsort *sharedsort;
193 : : Sharedsort *sharedsort2;
194 : : Snapshot snapshot;
195 : : WalUsage *walusage;
196 : : BufferUsage *bufferusage;
197 : : } BTLeader;
198 : :
199 : : /*
200 : : * Working state for btbuild and its callback.
201 : : *
202 : : * When parallel CREATE INDEX is used, there is a BTBuildState for each
203 : : * participant.
204 : : */
205 : : typedef struct BTBuildState
206 : : {
207 : : bool isunique;
208 : : bool nulls_not_distinct;
209 : : bool havedead;
210 : : Relation heap;
211 : : BTSpool *spool;
212 : :
213 : : /*
214 : : * spool2 is needed only when the index is a unique index. Dead tuples are
215 : : * put into spool2 instead of spool in order to avoid uniqueness check.
216 : : */
217 : : BTSpool *spool2;
218 : : double indtuples;
219 : :
220 : : /*
221 : : * btleader is only present when a parallel index build is performed, and
222 : : * only in the leader process. (Actually, only the leader has a
223 : : * BTBuildState. Workers have their own spool and spool2, though.)
224 : : */
225 : : BTLeader *btleader;
226 : : } BTBuildState;
227 : :
228 : : /*
229 : : * Status record for a btree page being built. We have one of these
230 : : * for each active tree level.
231 : : */
232 : : typedef struct BTPageState
233 : : {
234 : : BulkWriteBuffer btps_buf; /* workspace for page building */
235 : : BlockNumber btps_blkno; /* block # to write this page at */
236 : : IndexTuple btps_lowkey; /* page's strict lower bound pivot tuple */
237 : : OffsetNumber btps_lastoff; /* last item offset loaded */
238 : : Size btps_lastextra; /* last item's extra posting list space */
239 : : uint32 btps_level; /* tree level (0 = leaf) */
240 : : Size btps_full; /* "full" if less than this much free space */
241 : : struct BTPageState *btps_next; /* link to parent level, if any */
242 : : } BTPageState;
243 : :
244 : : /*
245 : : * Overall status record for index writing phase.
246 : : */
247 : : typedef struct BTWriteState
248 : : {
249 : : Relation heap;
250 : : Relation index;
251 : : BulkWriteState *bulkstate;
252 : : BTScanInsert inskey; /* generic insertion scankey */
253 : : BlockNumber btws_pages_alloced; /* # pages allocated */
254 : : } BTWriteState;
255 : :
256 : :
257 : : static double _bt_spools_heapscan(Relation heap, Relation index,
258 : : BTBuildState *buildstate, IndexInfo *indexInfo);
259 : : static void _bt_spooldestroy(BTSpool *btspool);
260 : : static void _bt_spool(BTSpool *btspool, ItemPointer self,
261 : : Datum *values, bool *isnull);
262 : : static void _bt_leafbuild(BTSpool *btspool, BTSpool *btspool2);
263 : : static void _bt_build_callback(Relation index, ItemPointer tid, Datum *values,
264 : : bool *isnull, bool tupleIsAlive, void *state);
265 : : static BulkWriteBuffer _bt_blnewpage(BTWriteState *wstate, uint32 level);
266 : : static BTPageState *_bt_pagestate(BTWriteState *wstate, uint32 level);
267 : : static void _bt_slideleft(Page rightmostpage);
268 : : static void _bt_sortaddtup(Page page, Size itemsize,
269 : : const IndexTupleData *itup, OffsetNumber itup_off,
270 : : bool newfirstdataitem);
271 : : static void _bt_buildadd(BTWriteState *wstate, BTPageState *state,
272 : : IndexTuple itup, Size truncextra);
273 : : static void _bt_sort_dedup_finish_pending(BTWriteState *wstate,
274 : : BTPageState *state,
275 : : BTDedupState dstate);
276 : : static void _bt_uppershutdown(BTWriteState *wstate, BTPageState *state);
277 : : static void _bt_load(BTWriteState *wstate,
278 : : BTSpool *btspool, BTSpool *btspool2);
279 : : static void _bt_begin_parallel(BTBuildState *buildstate, bool isconcurrent,
280 : : int request);
281 : : static void _bt_end_parallel(BTLeader *btleader);
282 : : static Size _bt_parallel_estimate_shared(Relation heap, Snapshot snapshot);
283 : : static double _bt_parallel_heapscan(BTBuildState *buildstate,
284 : : bool *brokenhotchain);
285 : : static void _bt_leader_participate_as_worker(BTBuildState *buildstate);
286 : : static void _bt_parallel_scan_and_sort(BTSpool *btspool, BTSpool *btspool2,
287 : : BTShared *btshared, Sharedsort *sharedsort,
288 : : Sharedsort *sharedsort2, int sortmem,
289 : : bool progress);
290 : :
291 : :
292 : : /*
293 : : * btbuild() -- build a new btree index.
294 : : */
295 : : IndexBuildResult *
2825 rhaas@postgresql.org 296 :CBC 25279 : btbuild(Relation heap, Relation index, IndexInfo *indexInfo)
297 : : {
298 : : IndexBuildResult *result;
299 : : BTBuildState buildstate;
300 : : double reltuples;
301 : :
302 : : #ifdef BTREE_BUILD_STATS
303 : : if (log_btree_build_stats)
304 : : ResetUsage();
305 : : #endif /* BTREE_BUILD_STATS */
306 : :
307 : 25279 : buildstate.isunique = indexInfo->ii_Unique;
1363 peter@eisentraut.org 308 : 25279 : buildstate.nulls_not_distinct = indexInfo->ii_NullsNotDistinct;
2825 rhaas@postgresql.org 309 : 25279 : buildstate.havedead = false;
310 : 25279 : buildstate.heap = heap;
311 : 25279 : buildstate.spool = NULL;
312 : 25279 : buildstate.spool2 = NULL;
313 : 25279 : buildstate.indtuples = 0;
314 : 25279 : buildstate.btleader = NULL;
315 : :
316 : : /*
317 : : * We expect to be called exactly once for any index relation. If that's
318 : : * not the case, big trouble's what we have.
319 : : */
320 [ - + ]: 25279 : if (RelationGetNumberOfBlocks(index) != 0)
2825 rhaas@postgresql.org 321 [ # # ]:UBC 0 : elog(ERROR, "index \"%s\" already contains data",
322 : : RelationGetRelationName(index));
323 : :
2825 rhaas@postgresql.org 324 :CBC 25279 : reltuples = _bt_spools_heapscan(heap, index, &buildstate, indexInfo);
325 : :
326 : : /*
327 : : * Finish the build by (1) completing the sort of the spool file, (2)
328 : : * inserting the sorted tuples into btree pages and (3) building the upper
329 : : * levels. Finally, it may also be necessary to end use of parallelism.
330 : : */
331 : 25273 : _bt_leafbuild(buildstate.spool, buildstate.spool2);
332 : 25231 : _bt_spooldestroy(buildstate.spool);
333 [ + + ]: 25231 : if (buildstate.spool2)
334 : 15 : _bt_spooldestroy(buildstate.spool2);
335 [ + + ]: 25231 : if (buildstate.btleader)
336 : 74 : _bt_end_parallel(buildstate.btleader);
337 : :
338 : 25231 : result = (IndexBuildResult *) palloc(sizeof(IndexBuildResult));
339 : :
340 : 25231 : result->heap_tuples = reltuples;
341 : 25231 : result->index_tuples = buildstate.indtuples;
342 : :
343 : : #ifdef BTREE_BUILD_STATS
344 : : if (log_btree_build_stats)
345 : : {
346 : : ShowUsage("BTREE BUILD STATS");
347 : : ResetUsage();
348 : : }
349 : : #endif /* BTREE_BUILD_STATS */
350 : :
351 : 25231 : return result;
352 : : }
353 : :
354 : : /*
355 : : * Create and initialize one or two spool structures, and save them in caller's
356 : : * buildstate argument. May also fill-in fields within indexInfo used by index
357 : : * builds.
358 : : *
359 : : * Scans the heap, possibly in parallel, filling spools with IndexTuples. This
360 : : * routine encapsulates all aspects of managing parallelism. Caller need only
361 : : * call _bt_end_parallel() in parallel case after it is done with spool/spool2.
362 : : *
363 : : * Returns the total number of heap tuples scanned.
364 : : */
365 : : static double
366 : 25279 : _bt_spools_heapscan(Relation heap, Relation index, BTBuildState *buildstate,
367 : : IndexInfo *indexInfo)
368 : : {
8385 bruce@momjian.us 369 : 25279 : BTSpool *btspool = (BTSpool *) palloc0(sizeof(BTSpool));
2825 rhaas@postgresql.org 370 : 25279 : SortCoordinate coordinate = NULL;
371 : 25279 : double reltuples = 0;
372 : :
373 : : /*
374 : : * We size the sort area as maintenance_work_mem rather than work_mem to
375 : : * speed index creation. This should be OK since a single backend can't
376 : : * run multiple index creations in parallel (see also: notes on
377 : : * parallelism and maintenance_work_mem below).
378 : : */
4655 tgl@sss.pgh.pa.us 379 : 25279 : btspool->heap = heap;
9508 380 : 25279 : btspool->index = index;
2825 rhaas@postgresql.org 381 : 25279 : btspool->isunique = indexInfo->ii_Unique;
1363 peter@eisentraut.org 382 : 25279 : btspool->nulls_not_distinct = indexInfo->ii_NullsNotDistinct;
383 : :
384 : : /* Save as primary spool */
2825 rhaas@postgresql.org 385 : 25279 : buildstate->spool = btspool;
386 : :
387 : : /* Report table scan phase started */
2401 alvherre@alvh.no-ip. 388 : 25279 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
389 : : PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN);
390 : :
391 : : /* Attempt to launch parallel worker scan when required */
2825 rhaas@postgresql.org 392 [ + + ]: 25279 : if (indexInfo->ii_ParallelWorkers > 0)
393 : 74 : _bt_begin_parallel(buildstate, indexInfo->ii_Concurrent,
394 : : indexInfo->ii_ParallelWorkers);
395 : :
396 : : /*
397 : : * If parallel build requested and at least one worker process was
398 : : * successfully launched, set up coordination state
399 : : */
400 [ + + ]: 25279 : if (buildstate->btleader)
401 : : {
402 : 74 : coordinate = (SortCoordinate) palloc0(sizeof(SortCoordinateData));
403 : 74 : coordinate->isWorker = false;
404 : 74 : coordinate->nParticipants =
405 : 74 : buildstate->btleader->nparticipanttuplesorts;
406 : 74 : coordinate->sharedsort = buildstate->btleader->sharedsort;
407 : : }
408 : :
409 : : /*
410 : : * Begin serial/leader tuplesort.
411 : : *
412 : : * In cases where parallelism is involved, the leader receives the same
413 : : * share of maintenance_work_mem as a serial sort (it is generally treated
414 : : * in the same way as a serial sort once we return). Parallel worker
415 : : * Tuplesortstates will have received only a fraction of
416 : : * maintenance_work_mem, though.
417 : : *
418 : : * We rely on the lifetime of the Leader Tuplesortstate almost not
419 : : * overlapping with any worker Tuplesortstate's lifetime. There may be
420 : : * some small overlap, but that's okay because we rely on leader
421 : : * Tuplesortstate only allocating a small, fixed amount of memory here.
422 : : * When its tuplesort_performsort() is called (by our caller), and
423 : : * significant amounts of memory are likely to be used, all workers must
424 : : * have already freed almost all memory held by their Tuplesortstates
425 : : * (they are about to go away completely, too). The overall effect is
426 : : * that maintenance_work_mem always represents an absolute high watermark
427 : : * on the amount of memory used by a CREATE INDEX operation, regardless of
428 : : * the use of parallelism or any other factor.
429 : : */
430 : 50558 : buildstate->spool->sortstate =
431 : 25279 : tuplesort_begin_index_btree(heap, index, buildstate->isunique,
1363 peter@eisentraut.org 432 : 25279 : buildstate->nulls_not_distinct,
433 : : maintenance_work_mem, coordinate,
434 : : TUPLESORT_NONE);
435 : :
436 : : /*
437 : : * If building a unique index, put dead tuples in a second spool to keep
438 : : * them out of the uniqueness check. We expect that the second spool (for
439 : : * dead tuples) won't get very full, so we give it only work_mem.
440 : : */
2825 rhaas@postgresql.org 441 [ + + ]: 25279 : if (indexInfo->ii_Unique)
442 : : {
443 : 20640 : BTSpool *btspool2 = (BTSpool *) palloc0(sizeof(BTSpool));
444 : 20640 : SortCoordinate coordinate2 = NULL;
445 : :
446 : : /* Initialize secondary spool */
447 : 20640 : btspool2->heap = heap;
448 : 20640 : btspool2->index = index;
449 : 20640 : btspool2->isunique = false;
450 : : /* Save as secondary spool */
451 : 20640 : buildstate->spool2 = btspool2;
452 : :
453 [ + + ]: 20640 : if (buildstate->btleader)
454 : : {
455 : : /*
456 : : * Set up non-private state that is passed to
457 : : * tuplesort_begin_index_btree() about the basic high level
458 : : * coordination of a parallel sort.
459 : : */
460 : 32 : coordinate2 = (SortCoordinate) palloc0(sizeof(SortCoordinateData));
461 : 32 : coordinate2->isWorker = false;
462 : 32 : coordinate2->nParticipants =
463 : 32 : buildstate->btleader->nparticipanttuplesorts;
464 : 32 : coordinate2->sharedsort = buildstate->btleader->sharedsort2;
465 : : }
466 : :
467 : : /*
468 : : * We expect that the second one (for dead tuples) won't get very
469 : : * full, so we give it only work_mem
470 : : */
471 : 20640 : buildstate->spool2->sortstate =
1363 peter@eisentraut.org 472 : 20640 : tuplesort_begin_index_btree(heap, index, false, false, work_mem,
473 : : coordinate2, TUPLESORT_NONE);
474 : : }
475 : :
476 : : /* Fill spool using either serial or parallel heap scan */
2825 rhaas@postgresql.org 477 [ + + ]: 25279 : if (!buildstate->btleader)
2401 alvherre@alvh.no-ip. 478 : 25205 : reltuples = table_index_build_scan(heap, index, indexInfo, true, true,
479 : : _bt_build_callback, buildstate,
480 : : NULL);
481 : : else
2825 rhaas@postgresql.org 482 : 74 : reltuples = _bt_parallel_heapscan(buildstate,
483 : : &indexInfo->ii_BrokenHotChain);
484 : :
485 : : /*
486 : : * Set the progress target for the next phase. Reset the block number
487 : : * values set by table_index_build_scan
488 : : */
489 : : {
1729 peter@eisentraut.org 490 : 25273 : const int progress_index[] = {
491 : : PROGRESS_CREATEIDX_TUPLES_TOTAL,
492 : : PROGRESS_SCAN_BLOCKS_TOTAL,
493 : : PROGRESS_SCAN_BLOCKS_DONE
494 : : };
495 : 25273 : const int64 progress_vals[] = {
2401 alvherre@alvh.no-ip. 496 : 25273 : buildstate->indtuples,
497 : : 0, 0
498 : : };
499 : :
1729 peter@eisentraut.org 500 : 25273 : pgstat_progress_update_multi_param(3, progress_index, progress_vals);
501 : : }
502 : :
503 : : /* okay, all heap tuples are spooled */
2825 rhaas@postgresql.org 504 [ + + + + ]: 25273 : if (buildstate->spool2 && !buildstate->havedead)
505 : : {
506 : : /* spool2 turns out to be unnecessary */
507 : 20625 : _bt_spooldestroy(buildstate->spool2);
508 : 20625 : buildstate->spool2 = NULL;
509 : : }
510 : :
511 : 25273 : return reltuples;
512 : : }
513 : :
514 : : /*
515 : : * clean up a spool structure and its substructures.
516 : : */
517 : : static void
9508 tgl@sss.pgh.pa.us 518 : 45871 : _bt_spooldestroy(BTSpool *btspool)
519 : : {
520 : 45871 : tuplesort_end(btspool->sortstate);
7701 neilc@samurai.com 521 : 45871 : pfree(btspool);
10703 scrappy@hub.org 522 : 45871 : }
523 : :
524 : : /*
525 : : * spool an index entry into the sort file.
526 : : */
527 : : static void
4137 rhaas@postgresql.org 528 : 6382984 : _bt_spool(BTSpool *btspool, ItemPointer self, Datum *values, bool *isnull)
529 : : {
530 : 6382984 : tuplesort_putindextuplevalues(btspool->sortstate, btspool->index,
531 : : self, values, isnull);
10703 scrappy@hub.org 532 : 6382984 : }
533 : :
534 : : /*
535 : : * given a spool loaded by successive calls to _bt_spool,
536 : : * create an entire btree.
537 : : */
538 : : static void
9210 inoue@tpf.co.jp 539 : 25273 : _bt_leafbuild(BTSpool *btspool, BTSpool *btspool2)
540 : : {
541 : : BTWriteState wstate;
542 : :
543 : : #ifdef BTREE_BUILD_STATS
544 : : if (log_btree_build_stats)
545 : : {
546 : : ShowUsage("BTREE BUILD (Spool) STATISTICS");
547 : : ResetUsage();
548 : : }
549 : : #endif /* BTREE_BUILD_STATS */
550 : :
551 : : /* Execute the sort */
2401 alvherre@alvh.no-ip. 552 : 25273 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
553 : : PROGRESS_BTREE_PHASE_PERFORMSORT_1);
8284 tgl@sss.pgh.pa.us 554 : 25273 : tuplesort_performsort(btspool->sortstate);
9210 inoue@tpf.co.jp 555 [ + + ]: 25231 : if (btspool2)
556 : : {
2401 alvherre@alvh.no-ip. 557 : 15 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
558 : : PROGRESS_BTREE_PHASE_PERFORMSORT_2);
9210 inoue@tpf.co.jp 559 : 15 : tuplesort_performsort(btspool2->sortstate);
560 : : }
561 : :
4655 tgl@sss.pgh.pa.us 562 : 25231 : wstate.heap = btspool->heap;
7818 563 : 25231 : wstate.index = btspool->index;
871 pg@bowt.ie 564 : 25231 : wstate.inskey = _bt_mkscankey(wstate.index, NULL);
565 : : /* _bt_mkscankey() won't set allequalimage without metapage */
2071 566 : 25231 : wstate.inskey->allequalimage = _bt_allequalimage(wstate.index, true);
567 : :
568 : : /* reserve the metapage */
7818 tgl@sss.pgh.pa.us 569 : 25231 : wstate.btws_pages_alloced = BTREE_METAPAGE + 1;
570 : :
2401 alvherre@alvh.no-ip. 571 : 25231 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
572 : : PROGRESS_BTREE_PHASE_LEAF_LOAD);
7818 tgl@sss.pgh.pa.us 573 : 25231 : _bt_load(&wstate, btspool, btspool2);
10703 scrappy@hub.org 574 : 25231 : }
575 : :
576 : : /*
577 : : * Per-tuple callback for table_index_build_scan
578 : : */
579 : : static void
2825 rhaas@postgresql.org 580 : 6382984 : _bt_build_callback(Relation index,
581 : : ItemPointer tid,
582 : : Datum *values,
583 : : bool *isnull,
584 : : bool tupleIsAlive,
585 : : void *state)
586 : : {
587 : 6382984 : BTBuildState *buildstate = (BTBuildState *) state;
588 : :
589 : : /*
590 : : * insert the index tuple into the appropriate spool file for subsequent
591 : : * processing
592 : : */
593 [ + + + + ]: 6382984 : if (tupleIsAlive || buildstate->spool2 == NULL)
2181 andres@anarazel.de 594 : 6382734 : _bt_spool(buildstate->spool, tid, values, isnull);
595 : : else
596 : : {
597 : : /* dead tuples are put into spool2 */
2825 rhaas@postgresql.org 598 : 250 : buildstate->havedead = true;
2181 andres@anarazel.de 599 : 250 : _bt_spool(buildstate->spool2, tid, values, isnull);
600 : : }
601 : :
2825 rhaas@postgresql.org 602 : 6382984 : buildstate->indtuples += 1;
603 : 6382984 : }
604 : :
605 : : /*
606 : : * allocate workspace for a new, clean btree page, not linked to any siblings.
607 : : */
608 : : static BulkWriteBuffer
613 heikki.linnakangas@i 609 : 26412 : _bt_blnewpage(BTWriteState *wstate, uint32 level)
610 : : {
611 : : BulkWriteBuffer buf;
612 : : Page page;
613 : : BTPageOpaque opaque;
614 : :
615 : 26412 : buf = smgr_bulk_get_buf(wstate->bulkstate);
616 : 26412 : page = (Page) buf;
617 : :
618 : : /* Zero the page and set up standard page header info */
7818 tgl@sss.pgh.pa.us 619 : 26412 : _bt_pageinit(page, BLCKSZ);
620 : :
621 : : /* Initialize BT opaque state */
1306 michael@paquier.xyz 622 : 26412 : opaque = BTPageGetOpaque(page);
10278 bruce@momjian.us 623 : 26412 : opaque->btpo_prev = opaque->btpo_next = P_NONE;
1707 pg@bowt.ie 624 : 26412 : opaque->btpo_level = level;
8285 tgl@sss.pgh.pa.us 625 : 26412 : opaque->btpo_flags = (level > 0) ? 0 : BTP_LEAF;
7113 626 : 26412 : opaque->btpo_cycleid = 0;
627 : :
628 : : /* Make the P_HIKEY line pointer appear allocated */
7818 629 : 26412 : ((PageHeader) page)->pd_lower += sizeof(ItemIdData);
630 : :
613 heikki.linnakangas@i 631 : 26412 : return buf;
632 : : }
633 : :
634 : : /*
635 : : * emit a completed btree page, and release the working storage.
636 : : */
637 : : static void
638 : 51643 : _bt_blwritepage(BTWriteState *wstate, BulkWriteBuffer buf, BlockNumber blkno)
639 : : {
640 : 51643 : smgr_bulk_write(wstate->bulkstate, blkno, buf, true);
641 : : /* smgr_bulk_write took ownership of 'buf' */
8285 tgl@sss.pgh.pa.us 642 : 51643 : }
643 : :
644 : : /*
645 : : * allocate and initialize a new BTPageState. the returned structure
646 : : * is suitable for immediate use by _bt_buildadd.
647 : : */
648 : : static BTPageState *
7818 649 : 6375 : _bt_pagestate(BTWriteState *wstate, uint32 level)
650 : : {
8385 bruce@momjian.us 651 : 6375 : BTPageState *state = (BTPageState *) palloc0(sizeof(BTPageState));
652 : :
653 : : /* create initial page for level */
613 heikki.linnakangas@i 654 : 6375 : state->btps_buf = _bt_blnewpage(wstate, level);
655 : :
656 : : /* and assign it a page position */
7818 tgl@sss.pgh.pa.us 657 : 6375 : state->btps_blkno = wstate->btws_pages_alloced++;
658 : :
2182 pg@bowt.ie 659 : 6375 : state->btps_lowkey = NULL;
660 : : /* initialize lastoff so first item goes into P_FIRSTKEY */
9230 tgl@sss.pgh.pa.us 661 : 6375 : state->btps_lastoff = P_HIKEY;
2071 pg@bowt.ie 662 : 6375 : state->btps_lastextra = 0;
9230 tgl@sss.pgh.pa.us 663 : 6375 : state->btps_level = level;
664 : : /* set "full" threshold based on level. See notes at head of file. */
7057 665 [ + + ]: 6375 : if (level > 0)
7049 666 : 1310 : state->btps_full = (BLCKSZ * (100 - BTREE_NONLEAF_FILLFACTOR) / 100);
667 : : else
2164 michael@paquier.xyz 668 [ + - - + : 5065 : state->btps_full = BTGetTargetPageFreeSpace(wstate->index);
+ + ]
669 : :
670 : : /* no parent level, yet */
7965 neilc@samurai.com 671 : 6375 : state->btps_next = NULL;
672 : :
9230 tgl@sss.pgh.pa.us 673 : 6375 : return state;
674 : : }
675 : :
676 : : /*
677 : : * Slide the array of ItemIds from the page back one slot (from P_FIRSTKEY to
678 : : * P_HIKEY, overwriting P_HIKEY).
679 : : *
680 : : * _bt_blnewpage() makes the P_HIKEY line pointer appear allocated, but the
681 : : * rightmost page on its level is not supposed to get a high key. Now that
682 : : * it's clear that this page is a rightmost page, remove the unneeded empty
683 : : * P_HIKEY line pointer space.
684 : : */
685 : : static void
1940 pg@bowt.ie 686 : 6375 : _bt_slideleft(Page rightmostpage)
687 : : {
688 : : OffsetNumber off;
689 : : OffsetNumber maxoff;
690 : : ItemId previi;
691 : :
692 : 6375 : maxoff = PageGetMaxOffsetNumber(rightmostpage);
693 [ - + ]: 6375 : Assert(maxoff >= P_FIRSTKEY);
694 : 6375 : previi = PageGetItemId(rightmostpage, P_HIKEY);
695 [ + + ]: 374034 : for (off = P_FIRSTKEY; off <= maxoff; off = OffsetNumberNext(off))
696 : : {
697 : 367659 : ItemId thisii = PageGetItemId(rightmostpage, off);
698 : :
699 : 367659 : *previi = *thisii;
700 : 367659 : previi = thisii;
701 : : }
702 : 6375 : ((PageHeader) rightmostpage)->pd_lower -= sizeof(ItemIdData);
10703 scrappy@hub.org 703 : 6375 : }
704 : :
705 : : /*
706 : : * Add an item to a page being built.
707 : : *
708 : : * This is very similar to nbtinsert.c's _bt_pgaddtup(), but this variant
709 : : * raises an error directly.
710 : : *
711 : : * Note that our nbtsort.c caller does not know yet if the page will be
712 : : * rightmost. Offset P_FIRSTKEY is always assumed to be the first data key by
713 : : * caller. Page that turns out to be the rightmost on its level is fixed by
714 : : * calling _bt_slideleft().
715 : : */
716 : : static void
9230 tgl@sss.pgh.pa.us 717 : 5740635 : _bt_sortaddtup(Page page,
718 : : Size itemsize,
719 : : const IndexTupleData *itup,
720 : : OffsetNumber itup_off,
721 : : bool newfirstdataitem)
722 : : {
723 : : IndexTupleData trunctuple;
724 : :
2024 pg@bowt.ie 725 [ + + ]: 5740635 : if (newfirstdataitem)
726 : : {
7216 tgl@sss.pgh.pa.us 727 : 1372 : trunctuple = *itup;
728 : 1372 : trunctuple.t_info = sizeof(IndexTupleData);
2030 pg@bowt.ie 729 : 1372 : BTreeTupleSetNAtts(&trunctuple, 0, false);
7216 tgl@sss.pgh.pa.us 730 : 1372 : itup = &trunctuple;
731 : 1372 : itemsize = sizeof(IndexTupleData);
732 : : }
733 : :
1 peter@eisentraut.org 734 [ - + ]:GNC 5740635 : if (PageAddItem(page, itup, itemsize, itup_off, false, false) == InvalidOffsetNumber)
8135 tgl@sss.pgh.pa.us 735 [ # # ]:UBC 0 : elog(ERROR, "failed to add item to the index page");
10485 scrappy@hub.org 736 :CBC 5740635 : }
737 : :
738 : : /*----------
739 : : * Add an item to a disk page from the sort output (or add a posting list
740 : : * item formed from the sort output).
741 : : *
742 : : * We must be careful to observe the page layout conventions of nbtsearch.c:
743 : : * - rightmost pages start data items at P_HIKEY instead of at P_FIRSTKEY.
744 : : * - on non-leaf pages, the key portion of the first item need not be
745 : : * stored, we should store only the link.
746 : : *
747 : : * A leaf page being built looks like:
748 : : *
749 : : * +----------------+---------------------------------+
750 : : * | PageHeaderData | linp0 linp1 linp2 ... |
751 : : * +-----------+----+---------------------------------+
752 : : * | ... linpN | |
753 : : * +-----------+--------------------------------------+
754 : : * | ^ last |
755 : : * | |
756 : : * +-------------+------------------------------------+
757 : : * | | itemN ... |
758 : : * +-------------+------------------+-----------------+
759 : : * | ... item3 item2 item1 | "special space" |
760 : : * +--------------------------------+-----------------+
761 : : *
762 : : * Contrast this with the diagram in bufpage.h; note the mismatch
763 : : * between linps and items. This is because we reserve linp0 as a
764 : : * placeholder for the pointer to the "high key" item; when we have
765 : : * filled up the page, we will set linp0 to point to itemN and clear
766 : : * linpN. On the other hand, if we find this is the last (rightmost)
767 : : * page, we leave the items alone and slide the linp array over. If
768 : : * the high key is to be truncated, offset 1 is deleted, and we insert
769 : : * the truncated high key at offset 1.
770 : : *
771 : : * 'last' pointer indicates the last offset added to the page.
772 : : *
773 : : * 'truncextra' is the size of the posting list in itup, if any. This
774 : : * information is stashed for the next call here, when we may benefit
775 : : * from considering the impact of truncating away the posting list on
776 : : * the page before deciding to finish the page off. Posting lists are
777 : : * often relatively large, so it is worth going to the trouble of
778 : : * accounting for the saving from truncating away the posting list of
779 : : * the tuple that becomes the high key (that may be the only way to
780 : : * get close to target free space on the page). Note that this is
781 : : * only used for the soft fillfactor-wise limit, not the critical hard
782 : : * limit.
783 : : *----------
784 : : */
785 : : static void
2071 pg@bowt.ie 786 : 5720598 : _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup,
787 : : Size truncextra)
788 : : {
789 : : BulkWriteBuffer nbuf;
790 : : Page npage;
791 : : BlockNumber nblkno;
792 : : OffsetNumber last_off;
793 : : Size last_truncextra;
794 : : Size pgspc;
795 : : Size itupsz;
796 : : bool isleaf;
797 : :
798 : : /*
799 : : * This is a handy place to check for cancel interrupts during the btree
800 : : * load phase of index creation.
801 : : */
7172 tgl@sss.pgh.pa.us 802 [ - + ]: 5720598 : CHECK_FOR_INTERRUPTS();
803 : :
613 heikki.linnakangas@i 804 : 5720598 : nbuf = state->btps_buf;
805 : 5720598 : npage = (Page) nbuf;
7818 tgl@sss.pgh.pa.us 806 : 5720598 : nblkno = state->btps_blkno;
10278 bruce@momjian.us 807 : 5720598 : last_off = state->btps_lastoff;
2071 pg@bowt.ie 808 : 5720598 : last_truncextra = state->btps_lastextra;
809 : 5720598 : state->btps_lastextra = truncextra;
810 : :
10278 bruce@momjian.us 811 : 5720598 : pgspc = PageGetFreeSpace(npage);
2799 tgl@sss.pgh.pa.us 812 : 5720598 : itupsz = IndexTupleSize(itup);
7216 813 : 5720598 : itupsz = MAXALIGN(itupsz);
814 : : /* Leaf case has slightly different rules due to suffix truncation */
2371 pg@bowt.ie 815 : 5720598 : isleaf = (state->btps_level == 0);
816 : :
817 : : /*
818 : : * Check whether the new item can fit on a btree page on current level at
819 : : * all.
820 : : *
821 : : * Every newly built index will treat heap TID as part of the keyspace,
822 : : * which imposes the requirement that new high keys must occasionally have
823 : : * a heap TID appended within _bt_truncate(). That may leave a new pivot
824 : : * tuple one or two MAXALIGN() quantums larger than the original
825 : : * firstright tuple it's derived from. v4 deals with the problem by
826 : : * decreasing the limit on the size of tuples inserted on the leaf level
827 : : * by the same small amount. Enforce the new v4+ limit on the leaf level,
828 : : * and the old limit on internal levels, since pivot tuples may need to
829 : : * make use of the reserved space. This should never fail on internal
830 : : * pages.
831 : : */
231 832 [ + + ]: 5720598 : if (unlikely(itupsz > BTMaxItemSize))
2371 833 : 132 : _bt_check_third_page(wstate->index, wstate->heap, isleaf, npage,
834 : : itup);
835 : :
836 : : /*
837 : : * Check to see if current page will fit new item, with space left over to
838 : : * append a heap TID during suffix truncation when page is a leaf page.
839 : : *
840 : : * It is guaranteed that we can fit at least 2 non-pivot tuples plus a
841 : : * high key with heap TID when finishing off a leaf page, since we rely on
842 : : * _bt_check_third_page() rejecting oversized non-pivot tuples. On
843 : : * internal pages we can always fit 3 pivot tuples with larger internal
844 : : * page tuple limit (includes page high key).
845 : : *
846 : : * Most of the time, a page is only "full" in the sense that the soft
847 : : * fillfactor-wise limit has been exceeded. However, we must always leave
848 : : * at least two items plus a high key on each page before starting a new
849 : : * page. Disregard fillfactor and insert on "full" current page if we
850 : : * don't have the minimum number of items yet. (Note that we deliberately
851 : : * assume that suffix truncation neither enlarges nor shrinks new high key
852 : : * when applying soft limit, except when last tuple has a posting list.)
853 : : */
2037 854 [ + + - + ]: 5720598 : Assert(last_truncextra == 0 || isleaf);
2371 855 [ + + + + ]: 5720598 : if (pgspc < itupsz + (isleaf ? MAXALIGN(sizeof(ItemPointerData)) : 0) ||
2071 856 [ + + + - ]: 5720020 : (pgspc + last_truncextra < state->btps_full && last_off > P_FIRSTKEY))
857 : : {
858 : : /*
859 : : * Finish off the page and write it out.
860 : : */
613 heikki.linnakangas@i 861 : 20037 : BulkWriteBuffer obuf = nbuf;
10277 bruce@momjian.us 862 : 20037 : Page opage = npage;
7730 863 : 20037 : BlockNumber oblkno = nblkno;
864 : : ItemId ii;
865 : : ItemId hii;
866 : : IndexTuple oitup;
867 : :
868 : : /* Create new page of same level */
613 heikki.linnakangas@i 869 : 20037 : nbuf = _bt_blnewpage(wstate, state->btps_level);
870 : 20037 : npage = (Page) nbuf;
871 : :
872 : : /* and assign it a page position */
7818 tgl@sss.pgh.pa.us 873 : 20037 : nblkno = wstate->btws_pages_alloced++;
874 : :
875 : : /*
876 : : * We copy the last item on the page into the new page, and then
877 : : * rearrange the old page so that the 'last item' becomes its high key
878 : : * rather than a true data item. There had better be at least two
879 : : * items on the page already, else the page would be empty of useful
880 : : * data.
881 : : */
9230 882 [ - + ]: 20037 : Assert(last_off > P_FIRSTKEY);
883 : 20037 : ii = PageGetItemId(opage, last_off);
7216 884 : 20037 : oitup = (IndexTuple) PageGetItem(opage, ii);
2024 pg@bowt.ie 885 : 20037 : _bt_sortaddtup(npage, ItemIdGetLength(ii), oitup, P_FIRSTKEY,
886 : 20037 : !isleaf);
887 : :
888 : : /*
889 : : * Move 'last' into the high key position on opage. _bt_blnewpage()
890 : : * allocated empty space for a line pointer when opage was first
891 : : * created, so this is a matter of rearranging already-allocated space
892 : : * on page, and initializing high key line pointer. (Actually, leaf
893 : : * pages must also swap oitup with a truncated version of oitup, which
894 : : * is sometimes larger than oitup, though never by more than the space
895 : : * needed to append a heap TID.)
896 : : */
10278 bruce@momjian.us 897 : 20037 : hii = PageGetItemId(opage, P_HIKEY);
898 : 20037 : *hii = *ii;
6621 tgl@sss.pgh.pa.us 899 : 20037 : ItemIdSetUnused(ii); /* redundant */
10278 bruce@momjian.us 900 : 20037 : ((PageHeader) opage)->pd_lower -= sizeof(ItemIdData);
901 : :
2371 pg@bowt.ie 902 [ + + ]: 20037 : if (isleaf)
903 : : {
904 : : IndexTuple lastleft;
905 : : IndexTuple truncated;
906 : :
907 : : /*
908 : : * Truncate away any unneeded attributes from high key on leaf
909 : : * level. This is only done at the leaf level because downlinks
910 : : * in internal pages are either negative infinity items, or get
911 : : * their contents from copying from one level down. See also:
912 : : * _bt_split().
913 : : *
914 : : * We don't try to bias our choice of split point to make it more
915 : : * likely that _bt_truncate() can truncate away more attributes,
916 : : * whereas the split point used within _bt_split() is chosen much
917 : : * more delicately. Even still, the lastleft and firstright
918 : : * tuples passed to _bt_truncate() here are at least not fully
919 : : * equal to each other when deduplication is used, unless there is
920 : : * a large group of duplicates (also, unique index builds usually
921 : : * have few or no spool2 duplicates). When the split point is
922 : : * between two unequal tuples, _bt_truncate() will avoid including
923 : : * a heap TID in the new high key, which is the most important
924 : : * benefit of suffix truncation.
925 : : *
926 : : * Overwrite the old item with new truncated high key directly.
927 : : * oitup is already located at the physical beginning of tuple
928 : : * space, so this should directly reuse the existing tuple space.
929 : : */
2414 930 : 19975 : ii = PageGetItemId(opage, OffsetNumberPrev(last_off));
931 : 19975 : lastleft = (IndexTuple) PageGetItem(opage, ii);
932 : :
2037 933 [ - + ]: 19975 : Assert(IndexTupleSize(oitup) > last_truncextra);
2414 934 : 19975 : truncated = _bt_truncate(wstate->index, lastleft, oitup,
935 : : wstate->inskey);
1 peter@eisentraut.org 936 [ - + ]:GNC 19975 : if (!PageIndexTupleOverwrite(opage, P_HIKEY, truncated, IndexTupleSize(truncated)))
2268 pg@bowt.ie 937 [ # # ]:UBC 0 : elog(ERROR, "failed to add high key to the index page");
2749 teodor@sigaev.ru 938 :CBC 19975 : pfree(truncated);
939 : :
940 : : /* oitup should continue to point to the page's high key */
941 : 19975 : hii = PageGetItemId(opage, P_HIKEY);
942 : 19975 : oitup = (IndexTuple) PageGetItem(opage, hii);
943 : : }
944 : :
945 : : /*
946 : : * Link the old page into its parent, using its low key. If we don't
947 : : * have a parent, we have to create one; this adds a new btree level.
948 : : */
7965 neilc@samurai.com 949 [ + + ]: 20037 : if (state->btps_next == NULL)
7818 tgl@sss.pgh.pa.us 950 : 1310 : state->btps_next = _bt_pagestate(wstate, state->btps_level + 1);
951 : :
2182 pg@bowt.ie 952 [ + - + - : 20037 : Assert((BTreeTupleGetNAtts(state->btps_lowkey, wstate->index) <=
+ - + + -
- - + ]
953 : : IndexRelationGetNumberOfKeyAttributes(wstate->index) &&
954 : : BTreeTupleGetNAtts(state->btps_lowkey, wstate->index) > 0) ||
955 : : P_LEFTMOST(BTPageGetOpaque(opage)));
956 [ + - + + : 20037 : Assert(BTreeTupleGetNAtts(state->btps_lowkey, wstate->index) == 0 ||
- + ]
957 : : !P_LEFTMOST(BTPageGetOpaque(opage)));
2143 958 : 20037 : BTreeTupleSetDownLink(state->btps_lowkey, oblkno);
2071 959 : 20037 : _bt_buildadd(wstate, state->btps_next, state->btps_lowkey, 0);
2182 960 : 20037 : pfree(state->btps_lowkey);
961 : :
962 : : /*
963 : : * Save a copy of the high key from the old page. It is also the low
964 : : * key for the new page.
965 : : */
966 : 20037 : state->btps_lowkey = CopyIndexTuple(oitup);
967 : :
968 : : /*
969 : : * Set the sibling links for both pages.
970 : : */
971 : : {
1306 michael@paquier.xyz 972 : 20037 : BTPageOpaque oopaque = BTPageGetOpaque(opage);
973 : 20037 : BTPageOpaque nopaque = BTPageGetOpaque(npage);
974 : :
7818 tgl@sss.pgh.pa.us 975 : 20037 : oopaque->btpo_next = nblkno;
976 : 20037 : nopaque->btpo_prev = oblkno;
3051 977 : 20037 : nopaque->btpo_next = P_NONE; /* redundant */
978 : : }
979 : :
980 : : /*
981 : : * Write out the old page. _bt_blwritepage takes ownership of the
982 : : * 'opage' buffer.
983 : : */
613 heikki.linnakangas@i 984 : 20037 : _bt_blwritepage(wstate, obuf, oblkno);
985 : :
986 : : /*
987 : : * Reset last_off to point to new page
988 : : */
9230 tgl@sss.pgh.pa.us 989 : 20037 : last_off = P_FIRSTKEY;
990 : : }
991 : :
992 : : /*
993 : : * By here, either original page is still the current page, or a new page
994 : : * was created that became the current page. Either way, the current page
995 : : * definitely has space for new item.
996 : : *
997 : : * If the new item is the first for its page, it must also be the first
998 : : * item on its entire level. On later same-level pages, a low key for a
999 : : * page will be copied from the prior page in the code above. Generate a
1000 : : * minus infinity low key here instead.
1001 : : */
1002 [ + + ]: 5720598 : if (last_off == P_HIKEY)
1003 : : {
2182 pg@bowt.ie 1004 [ - + ]: 6375 : Assert(state->btps_lowkey == NULL);
1005 : 6375 : state->btps_lowkey = palloc0(sizeof(IndexTupleData));
1006 : 6375 : state->btps_lowkey->t_info = sizeof(IndexTupleData);
2030 1007 : 6375 : BTreeTupleSetNAtts(state->btps_lowkey, 0, false);
1008 : : }
1009 : :
1010 : : /*
1011 : : * Add the new item into the current page.
1012 : : */
9230 tgl@sss.pgh.pa.us 1013 : 5720598 : last_off = OffsetNumberNext(last_off);
2024 pg@bowt.ie 1014 : 5720598 : _bt_sortaddtup(npage, itupsz, itup, last_off,
1015 [ + + + + ]: 5720598 : !isleaf && last_off == P_FIRSTKEY);
1016 : :
613 heikki.linnakangas@i 1017 : 5720598 : state->btps_buf = nbuf;
7818 tgl@sss.pgh.pa.us 1018 : 5720598 : state->btps_blkno = nblkno;
10278 bruce@momjian.us 1019 : 5720598 : state->btps_lastoff = last_off;
10485 scrappy@hub.org 1020 : 5720598 : }
1021 : :
1022 : : /*
1023 : : * Finalize pending posting list tuple, and add it to the index. Final tuple
1024 : : * is based on saved base tuple, and saved list of heap TIDs.
1025 : : *
1026 : : * This is almost like _bt_dedup_finish_pending(), but it adds a new tuple
1027 : : * using _bt_buildadd().
1028 : : */
1029 : : static void
2071 pg@bowt.ie 1030 : 2359303 : _bt_sort_dedup_finish_pending(BTWriteState *wstate, BTPageState *state,
1031 : : BTDedupState dstate)
1032 : : {
1033 [ - + ]: 2359303 : Assert(dstate->nitems > 0);
1034 : :
1035 [ + + ]: 2359303 : if (dstate->nitems == 1)
1036 : 2339205 : _bt_buildadd(wstate, state, dstate->base, 0);
1037 : : else
1038 : : {
1039 : : IndexTuple postingtuple;
1040 : : Size truncextra;
1041 : :
1042 : : /* form a tuple with a posting list */
1043 : 20098 : postingtuple = _bt_form_posting(dstate->base,
1044 : : dstate->htids,
1045 : : dstate->nhtids);
1046 : : /* Calculate posting list overhead */
1047 : 20098 : truncextra = IndexTupleSize(postingtuple) -
1048 : 20098 : BTreeTupleGetPostingOffset(postingtuple);
1049 : :
1050 : 20098 : _bt_buildadd(wstate, state, postingtuple, truncextra);
1051 : 20098 : pfree(postingtuple);
1052 : : }
1053 : :
1957 1054 : 2359303 : dstate->nmaxitems = 0;
2071 1055 : 2359303 : dstate->nhtids = 0;
1056 : 2359303 : dstate->nitems = 0;
1057 : 2359303 : dstate->phystupsize = 0;
1058 : 2359303 : }
1059 : :
1060 : : /*
1061 : : * Finish writing out the completed btree.
1062 : : */
1063 : : static void
7818 tgl@sss.pgh.pa.us 1064 : 25231 : _bt_uppershutdown(BTWriteState *wstate, BTPageState *state)
1065 : : {
1066 : : BTPageState *s;
7730 bruce@momjian.us 1067 : 25231 : BlockNumber rootblkno = P_NONE;
8065 tgl@sss.pgh.pa.us 1068 : 25231 : uint32 rootlevel = 0;
1069 : : BulkWriteBuffer metabuf;
1070 : :
1071 : : /*
1072 : : * Each iteration of this loop completes one more level of the tree.
1073 : : */
7965 neilc@samurai.com 1074 [ + + ]: 31606 : for (s = state; s != NULL; s = s->btps_next)
1075 : : {
1076 : : BlockNumber blkno;
1077 : : BTPageOpaque opaque;
1078 : :
7818 tgl@sss.pgh.pa.us 1079 : 6375 : blkno = s->btps_blkno;
613 heikki.linnakangas@i 1080 : 6375 : opaque = BTPageGetOpaque((Page) s->btps_buf);
1081 : :
1082 : : /*
1083 : : * We have to link the last page on this level to somewhere.
1084 : : *
1085 : : * If we're at the top, it's the root, so attach it to the metapage.
1086 : : * Otherwise, add an entry for it to its parent using its low key.
1087 : : * This may cause the last page of the parent level to split, but
1088 : : * that's not a problem -- we haven't gotten to it yet.
1089 : : */
7965 neilc@samurai.com 1090 [ + + ]: 6375 : if (s->btps_next == NULL)
1091 : : {
9230 tgl@sss.pgh.pa.us 1092 : 5065 : opaque->btpo_flags |= BTP_ROOT;
8065 1093 : 5065 : rootblkno = blkno;
1094 : 5065 : rootlevel = s->btps_level;
1095 : : }
1096 : : else
1097 : : {
2182 pg@bowt.ie 1098 [ + - + - : 1310 : Assert((BTreeTupleGetNAtts(s->btps_lowkey, wstate->index) <=
+ - - + -
- - - ]
1099 : : IndexRelationGetNumberOfKeyAttributes(wstate->index) &&
1100 : : BTreeTupleGetNAtts(s->btps_lowkey, wstate->index) > 0) ||
1101 : : P_LEFTMOST(opaque));
1102 [ + - + - : 1310 : Assert(BTreeTupleGetNAtts(s->btps_lowkey, wstate->index) == 0 ||
- + ]
1103 : : !P_LEFTMOST(opaque));
2143 1104 : 1310 : BTreeTupleSetDownLink(s->btps_lowkey, blkno);
2071 1105 : 1310 : _bt_buildadd(wstate, s->btps_next, s->btps_lowkey, 0);
2182 1106 : 1310 : pfree(s->btps_lowkey);
1107 : 1310 : s->btps_lowkey = NULL;
1108 : : }
1109 : :
1110 : : /*
1111 : : * This is the rightmost page, so the ItemId array needs to be slid
1112 : : * back one slot. Then we can dump out the page.
1113 : : */
613 heikki.linnakangas@i 1114 : 6375 : _bt_slideleft((Page) s->btps_buf);
1115 : 6375 : _bt_blwritepage(wstate, s->btps_buf, s->btps_blkno);
1116 : 6375 : s->btps_buf = NULL; /* writepage took ownership of the buffer */
1117 : : }
1118 : :
1119 : : /*
1120 : : * As the last step in the process, construct the metapage and make it
1121 : : * point to the new root (unless we had no data at all, in which case it's
1122 : : * set to point to "P_NONE"). This changes the index to the "valid" state
1123 : : * by filling in a valid magic number in the metapage.
1124 : : */
1125 : 25231 : metabuf = smgr_bulk_get_buf(wstate->bulkstate);
1126 : 25231 : _bt_initmetapage((Page) metabuf, rootblkno, rootlevel,
2071 pg@bowt.ie 1127 : 25231 : wstate->inskey->allequalimage);
613 heikki.linnakangas@i 1128 : 25231 : _bt_blwritepage(wstate, metabuf, BTREE_METAPAGE);
10703 scrappy@hub.org 1129 : 25231 : }
1130 : :
1131 : : /*
1132 : : * Read tuples in correct sort order from tuplesort, and load them into
1133 : : * btree leaves.
1134 : : */
1135 : : static void
7818 tgl@sss.pgh.pa.us 1136 : 25231 : _bt_load(BTWriteState *wstate, BTSpool *btspool, BTSpool *btspool2)
1137 : : {
9230 1138 : 25231 : BTPageState *state = NULL;
9210 inoue@tpf.co.jp 1139 : 25231 : bool merge = (btspool2 != NULL);
1140 : : IndexTuple itup,
7216 tgl@sss.pgh.pa.us 1141 : 25231 : itup2 = NULL;
1142 : : bool load1;
7818 1143 : 25231 : TupleDesc tupdes = RelationGetDescr(wstate->index);
1144 : : int i,
2761 teodor@sigaev.ru 1145 : 25231 : keysz = IndexRelationGetNumberOfKeyAttributes(wstate->index);
1146 : : SortSupport sortKeys;
2374 alvherre@alvh.no-ip. 1147 : 25231 : int64 tuples_done = 0;
1148 : : bool deduplicate;
1149 : :
613 heikki.linnakangas@i 1150 : 25231 : wstate->bulkstate = smgr_bulk_start_rel(wstate->index, MAIN_FORKNUM);
1151 : :
1929 pg@bowt.ie 1152 [ + + + + : 29736 : deduplicate = wstate->inskey->allequalimage && !btspool->isunique &&
+ - ]
2071 1153 [ + - - + : 4505 : BTGetDeduplicateItems(wstate->index);
+ + + - ]
1154 : :
9210 inoue@tpf.co.jp 1155 [ + + ]: 25231 : if (merge)
1156 : : {
1157 : : /*
1158 : : * Another BTSpool for dead tuples exists. Now we have to merge
1159 : : * btspool and btspool2.
1160 : : */
1161 : :
1162 : : /* the preparation of merge */
3242 rhaas@postgresql.org 1163 : 15 : itup = tuplesort_getindextuple(btspool->sortstate, true);
1164 : 15 : itup2 = tuplesort_getindextuple(btspool2->sortstate, true);
1165 : :
1166 : : /* Prepare SortSupport data for each column */
4008 1167 : 15 : sortKeys = (SortSupport) palloc0(keysz * sizeof(SortSupportData));
1168 : :
1169 [ + + ]: 31 : for (i = 0; i < keysz; i++)
1170 : : {
1171 : 16 : SortSupport sortKey = sortKeys + i;
2414 pg@bowt.ie 1172 : 16 : ScanKey scanKey = wstate->inskey->scankeys + i;
1173 : : bool reverse;
1174 : :
4008 rhaas@postgresql.org 1175 : 16 : sortKey->ssup_cxt = CurrentMemoryContext;
1176 : 16 : sortKey->ssup_collation = scanKey->sk_collation;
1177 : 16 : sortKey->ssup_nulls_first =
1178 : 16 : (scanKey->sk_flags & SK_BT_NULLS_FIRST) != 0;
1179 : 16 : sortKey->ssup_attno = scanKey->sk_attno;
1180 : : /* Abbreviation is not supported here */
3935 1181 : 16 : sortKey->abbreviate = false;
1182 : :
1096 peter@eisentraut.org 1183 [ - + ]: 16 : Assert(sortKey->ssup_attno != 0);
1184 : :
228 1185 : 16 : reverse = (scanKey->sk_flags & SK_BT_DESC) != 0;
1186 : :
1187 : 16 : PrepareSortSupportFromIndexRel(wstate->index, reverse, sortKey);
1188 : : }
1189 : :
1190 : : for (;;)
1191 : : {
8986 bruce@momjian.us 1192 : 1675 : load1 = true; /* load BTSpool next ? */
7216 tgl@sss.pgh.pa.us 1193 [ + + ]: 1675 : if (itup2 == NULL)
1194 : : {
1195 [ + + ]: 82 : if (itup == NULL)
9210 inoue@tpf.co.jp 1196 : 15 : break;
1197 : : }
7216 tgl@sss.pgh.pa.us 1198 [ + + ]: 1593 : else if (itup != NULL)
1199 : : {
2414 pg@bowt.ie 1200 : 1492 : int32 compare = 0;
1201 : :
9210 inoue@tpf.co.jp 1202 [ + + ]: 1613 : for (i = 1; i <= keysz; i++)
1203 : : {
1204 : : SortSupport entry;
1205 : : Datum attrDatum1,
1206 : : attrDatum2;
1207 : : bool isNull1,
1208 : : isNull2;
1209 : :
4008 rhaas@postgresql.org 1210 : 1518 : entry = sortKeys + i - 1;
6867 tgl@sss.pgh.pa.us 1211 : 1518 : attrDatum1 = index_getattr(itup, i, tupdes, &isNull1);
1212 : 1518 : attrDatum2 = index_getattr(itup2, i, tupdes, &isNull2);
1213 : :
4008 rhaas@postgresql.org 1214 : 1518 : compare = ApplySortComparator(attrDatum1, isNull1,
1215 : : attrDatum2, isNull2,
1216 : : entry);
6867 tgl@sss.pgh.pa.us 1217 [ + + ]: 1518 : if (compare > 0)
1218 : : {
1219 : 138 : load1 = false;
1220 : 1397 : break;
1221 : : }
1222 [ + + ]: 1380 : else if (compare < 0)
1223 : 1259 : 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 : : */
2414 pg@bowt.ie 1232 [ + + ]: 1492 : if (compare == 0)
1233 : : {
1234 : 95 : compare = ItemPointerCompare(&itup->t_tid, &itup2->t_tid);
1235 [ - + ]: 95 : Assert(compare != 0);
1236 [ + + ]: 95 : if (compare > 0)
1237 : 11 : load1 = false;
1238 : : }
1239 : : }
1240 : : else
9210 inoue@tpf.co.jp 1241 : 101 : load1 = false;
1242 : :
1243 : : /* When we see first tuple, create first index page */
1244 [ + + ]: 1660 : if (state == NULL)
7818 tgl@sss.pgh.pa.us 1245 : 15 : state = _bt_pagestate(wstate, 0);
1246 : :
9210 inoue@tpf.co.jp 1247 [ + + ]: 1660 : if (load1)
1248 : : {
2071 pg@bowt.ie 1249 : 1410 : _bt_buildadd(wstate, state, itup, 0);
3242 rhaas@postgresql.org 1250 : 1410 : itup = tuplesort_getindextuple(btspool->sortstate, true);
1251 : : }
1252 : : else
1253 : : {
2071 pg@bowt.ie 1254 : 250 : _bt_buildadd(wstate, state, itup2, 0);
3242 rhaas@postgresql.org 1255 : 250 : itup2 = tuplesort_getindextuple(btspool2->sortstate, true);
1256 : : }
1257 : :
1258 : : /* Report progress */
2401 alvherre@alvh.no-ip. 1259 : 1660 : pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_DONE,
1260 : : ++tuples_done);
1261 : : }
4008 rhaas@postgresql.org 1262 : 15 : pfree(sortKeys);
1263 : : }
2071 pg@bowt.ie 1264 [ + + ]: 25216 : else if (deduplicate)
1265 : : {
1266 : : /* merge is unnecessary, deduplicate into posting lists */
1267 : : BTDedupState dstate;
1268 : :
1269 : 4505 : dstate = (BTDedupState) palloc(sizeof(BTDedupStateData));
1270 : 4505 : dstate->deduplicate = true; /* unused */
1957 1271 : 4505 : dstate->nmaxitems = 0; /* unused */
2071 1272 : 4505 : dstate->maxpostingsize = 0; /* set later */
1273 : : /* Metadata about base tuple of current pending posting list */
1274 : 4505 : dstate->base = NULL;
1275 : 4505 : dstate->baseoff = InvalidOffsetNumber; /* unused */
1276 : 4505 : dstate->basetupsize = 0;
1277 : : /* Metadata about current pending posting list TIDs */
1278 : 4505 : dstate->htids = NULL;
1279 : 4505 : dstate->nhtids = 0;
1280 : 4505 : dstate->nitems = 0;
1281 : 4505 : dstate->phystupsize = 0; /* unused */
1282 : 4505 : dstate->nintervals = 0; /* unused */
1283 : :
1284 : 3047352 : while ((itup = tuplesort_getindextuple(btspool->sortstate,
1285 [ + + ]: 3047352 : true)) != NULL)
1286 : : {
1287 : : /* When we see first tuple, create first index page */
1288 [ + + ]: 3042847 : if (state == NULL)
1289 : : {
1290 : 1211 : 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 : 1211 : dstate->maxpostingsize = MAXALIGN_DOWN((BLCKSZ * 10 / 100)) -
1305 : : sizeof(ItemIdData);
231 1306 [ + - - + ]: 1211 : Assert(dstate->maxpostingsize <= BTMaxItemSize &&
1307 : : dstate->maxpostingsize <= INDEX_SIZE_MASK);
2071 1308 : 1211 : dstate->htids = palloc(dstate->maxpostingsize);
1309 : :
1310 : : /* start new pending posting list with itup copy */
1311 : 1211 : _bt_dedup_start_pending(dstate, CopyIndexTuple(itup),
1312 : : InvalidOffsetNumber);
1313 : : }
1314 [ + + ]: 3041636 : else if (_bt_keep_natts_fast(wstate->index, dstate->base,
1315 [ + + ]: 687514 : itup) > keysz &&
1316 : 687514 : _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 : 2358092 : _bt_sort_dedup_finish_pending(wstate, state, dstate);
1331 : 2358092 : pfree(dstate->base);
1332 : :
1333 : : /* start new pending posting list with itup copy */
1334 : 2358092 : _bt_dedup_start_pending(dstate, CopyIndexTuple(itup),
1335 : : InvalidOffsetNumber);
1336 : : }
1337 : :
1338 : : /* Report progress */
1339 : 3042847 : pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_DONE,
1340 : : ++tuples_done);
1341 : : }
1342 : :
1343 [ + + ]: 4505 : 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 : 1211 : _bt_sort_dedup_finish_pending(wstate, state, dstate);
1350 : 1211 : pfree(dstate->base);
1351 : 1211 : pfree(dstate->htids);
1352 : : }
1353 : :
1354 : 4505 : pfree(dstate);
1355 : : }
1356 : : else
1357 : : {
1358 : : /* merging and deduplication are both unnecessary */
7216 tgl@sss.pgh.pa.us 1359 : 3358999 : while ((itup = tuplesort_getindextuple(btspool->sortstate,
3242 rhaas@postgresql.org 1360 [ + + ]: 3358999 : true)) != NULL)
1361 : : {
1362 : : /* When we see first tuple, create first index page */
9210 inoue@tpf.co.jp 1363 [ + + ]: 3338288 : if (state == NULL)
7818 tgl@sss.pgh.pa.us 1364 : 3839 : state = _bt_pagestate(wstate, 0);
1365 : :
2071 pg@bowt.ie 1366 : 3338288 : _bt_buildadd(wstate, state, itup, 0);
1367 : :
1368 : : /* Report progress */
2401 alvherre@alvh.no-ip. 1369 : 3338288 : pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_DONE,
1370 : : ++tuples_done);
1371 : : }
1372 : : }
1373 : :
1374 : : /* Close down final pages and write the metapage */
7818 tgl@sss.pgh.pa.us 1375 : 25231 : _bt_uppershutdown(wstate, state);
613 heikki.linnakangas@i 1376 : 25231 : smgr_bulk_finish(wstate->bulkstate);
10703 scrappy@hub.org 1377 : 25231 : }
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
2825 rhaas@postgresql.org 1396 : 74 : _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 : 74 : BTSpool *btspool = buildstate->spool;
1407 : 74 : BTLeader *btleader = (BTLeader *) palloc0(sizeof(BTLeader));
1408 : : WalUsage *walusage;
1409 : : BufferUsage *bufferusage;
1410 : 74 : 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 : 74 : EnterParallelMode();
1422 [ - + ]: 74 : Assert(request > 0);
1423 : 74 : pcxt = CreateParallelContext("postgres", "_bt_parallel_build_main",
1424 : : request);
1425 : :
1426 [ + - ]: 74 : 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 [ + - ]: 74 : if (!isconcurrent)
1436 : 74 : snapshot = SnapshotAny;
1437 : : else
2825 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 : : */
2423 andres@anarazel.de 1444 :CBC 74 : estbtshared = _bt_parallel_estimate_shared(btspool->heap, snapshot);
2825 rhaas@postgresql.org 1445 : 74 : shm_toc_estimate_chunk(&pcxt->estimator, estbtshared);
1446 : 74 : estsort = tuplesort_estimate_shared(scantuplesortstates);
1447 : 74 : 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 [ + + ]: 74 : if (!btspool->isunique)
1454 : 42 : 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 : : */
2033 akapila@postgresql.o 1469 : 74 : shm_toc_estimate_chunk(&pcxt->estimator,
1470 : : mul_size(sizeof(WalUsage), pcxt->nworkers));
1471 : 74 : shm_toc_estimate_keys(&pcxt->estimator, 1);
2028 1472 : 74 : shm_toc_estimate_chunk(&pcxt->estimator,
1473 : : mul_size(sizeof(BufferUsage), pcxt->nworkers));
1474 : 74 : shm_toc_estimate_keys(&pcxt->estimator, 1);
1475 : :
1476 : : /* Finally, estimate PARALLEL_KEY_QUERY_TEXT space */
1823 noah@leadboat.com 1477 [ + - ]: 74 : if (debug_query_string)
1478 : : {
1479 : 74 : querylen = strlen(debug_query_string);
1480 : 74 : shm_toc_estimate_chunk(&pcxt->estimator, querylen + 1);
1481 : 74 : shm_toc_estimate_keys(&pcxt->estimator, 1);
1482 : : }
1483 : : else
1823 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 */
2825 rhaas@postgresql.org 1487 :CBC 74 : InitializeParallelDSM(pcxt);
1488 : :
1489 : : /* If no DSM segment was available, back out (do serial build) */
2092 tmunro@postgresql.or 1490 [ - + ]: 74 : if (pcxt->seg == NULL)
1491 : : {
2092 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 */
2825 rhaas@postgresql.org 1500 :CBC 74 : btshared = (BTShared *) shm_toc_allocate(pcxt->toc, estbtshared);
1501 : : /* Initialize immutable state */
1502 : 74 : btshared->heaprelid = RelationGetRelid(btspool->heap);
1503 : 74 : btshared->indexrelid = RelationGetRelid(btspool->index);
1504 : 74 : btshared->isunique = btspool->isunique;
1363 peter@eisentraut.org 1505 : 74 : btshared->nulls_not_distinct = btspool->nulls_not_distinct;
2825 rhaas@postgresql.org 1506 : 74 : btshared->isconcurrent = isconcurrent;
1507 : 74 : btshared->scantuplesortstates = scantuplesortstates;
393 michael@paquier.xyz 1508 : 74 : btshared->queryid = pgstat_get_my_query_id();
2825 rhaas@postgresql.org 1509 : 74 : ConditionVariableInit(&btshared->workersdonecv);
1510 : 74 : SpinLockInit(&btshared->mutex);
1511 : : /* Initialize mutable state */
1512 : 74 : btshared->nparticipantsdone = 0;
1513 : 74 : btshared->reltuples = 0.0;
1514 : 74 : btshared->havedead = false;
1515 : 74 : btshared->indtuples = 0.0;
1516 : 74 : btshared->brokenhotchain = false;
2423 andres@anarazel.de 1517 : 74 : table_parallelscan_initialize(btspool->heap,
1518 : : ParallelTableScanFromBTShared(btshared),
1519 : : snapshot);
1520 : :
1521 : : /*
1522 : : * Store shared tuplesort-private state, for which we reserved space.
1523 : : * Then, initialize opaque state using tuplesort routine.
1524 : : */
2825 rhaas@postgresql.org 1525 : 74 : sharedsort = (Sharedsort *) shm_toc_allocate(pcxt->toc, estsort);
1526 : 74 : tuplesort_initialize_shared(sharedsort, scantuplesortstates,
1527 : : pcxt->seg);
1528 : :
1529 : 74 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_BTREE_SHARED, btshared);
1530 : 74 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_TUPLESORT, sharedsort);
1531 : :
1532 : : /* Unique case requires a second spool, and associated shared state */
1533 [ + + ]: 74 : if (!btspool->isunique)
1534 : 42 : sharedsort2 = NULL;
1535 : : else
1536 : : {
1537 : : /*
1538 : : * Store additional shared tuplesort-private state, for which we
1539 : : * reserved space. Then, initialize opaque state using tuplesort
1540 : : * routine.
1541 : : */
1542 : 32 : sharedsort2 = (Sharedsort *) shm_toc_allocate(pcxt->toc, estsort);
1543 : 32 : tuplesort_initialize_shared(sharedsort2, scantuplesortstates,
1544 : : pcxt->seg);
1545 : :
1546 : 32 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_TUPLESORT_SPOOL2, sharedsort2);
1547 : : }
1548 : :
1549 : : /* Store query string for workers */
1823 noah@leadboat.com 1550 [ + - ]: 74 : if (debug_query_string)
1551 : : {
1552 : : char *sharedquery;
1553 : :
1554 : 74 : sharedquery = (char *) shm_toc_allocate(pcxt->toc, querylen + 1);
1555 : 74 : memcpy(sharedquery, debug_query_string, querylen + 1);
1556 : 74 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_QUERY_TEXT, sharedquery);
1557 : : }
1558 : :
1559 : : /*
1560 : : * Allocate space for each worker's WalUsage and BufferUsage; no need to
1561 : : * initialize.
1562 : : */
2033 akapila@postgresql.o 1563 : 74 : walusage = shm_toc_allocate(pcxt->toc,
1564 : 74 : mul_size(sizeof(WalUsage), pcxt->nworkers));
1565 : 74 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_WAL_USAGE, walusage);
2028 1566 : 74 : bufferusage = shm_toc_allocate(pcxt->toc,
1567 : 74 : mul_size(sizeof(BufferUsage), pcxt->nworkers));
1568 : 74 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_BUFFER_USAGE, bufferusage);
1569 : :
1570 : : /* Launch workers, saving status for leader/caller */
2825 rhaas@postgresql.org 1571 : 74 : LaunchParallelWorkers(pcxt);
1572 : 74 : btleader->pcxt = pcxt;
1573 : 74 : btleader->nparticipanttuplesorts = pcxt->nworkers_launched;
1574 [ + - ]: 74 : if (leaderparticipates)
1575 : 74 : btleader->nparticipanttuplesorts++;
1576 : 74 : btleader->btshared = btshared;
1577 : 74 : btleader->sharedsort = sharedsort;
1578 : 74 : btleader->sharedsort2 = sharedsort2;
1579 : 74 : btleader->snapshot = snapshot;
2033 akapila@postgresql.o 1580 : 74 : btleader->walusage = walusage;
2028 1581 : 74 : btleader->bufferusage = bufferusage;
1582 : :
1583 : : /* If no workers were successfully launched, back out (do serial build) */
2825 rhaas@postgresql.org 1584 [ - + ]: 74 : if (pcxt->nworkers_launched == 0)
1585 : : {
2825 rhaas@postgresql.org 1586 :UBC 0 : _bt_end_parallel(btleader);
1587 : 0 : return;
1588 : : }
1589 : :
1590 : : /* Save leader state now that it's clear build will be parallel */
2825 rhaas@postgresql.org 1591 :CBC 74 : buildstate->btleader = btleader;
1592 : :
1593 : : /* Join heap scan ourselves */
1594 [ + - ]: 74 : if (leaderparticipates)
1595 : 74 : _bt_leader_participate_as_worker(buildstate);
1596 : :
1597 : : /*
1598 : : * Caller needs to wait for all launched workers when we return. Make
1599 : : * sure that the failure-to-start case will not hang forever.
1600 : : */
1601 : 74 : WaitForParallelWorkersToAttach(pcxt);
1602 : : }
1603 : :
1604 : : /*
1605 : : * Shut down workers, destroy parallel context, and end parallel mode.
1606 : : */
1607 : : static void
1608 : 74 : _bt_end_parallel(BTLeader *btleader)
1609 : : {
1610 : : int i;
1611 : :
1612 : : /* Shutdown worker processes */
1613 : 74 : WaitForParallelWorkersToFinish(btleader->pcxt);
1614 : :
1615 : : /*
1616 : : * Next, accumulate WAL usage. (This must wait for the workers to finish,
1617 : : * or we might get incomplete data.)
1618 : : */
2033 akapila@postgresql.o 1619 [ + + ]: 148 : for (i = 0; i < btleader->pcxt->nworkers_launched; i++)
2028 1620 : 74 : InstrAccumParallelQuery(&btleader->bufferusage[i], &btleader->walusage[i]);
1621 : :
1622 : : /* Free last reference to MVCC snapshot, if one was used */
2825 rhaas@postgresql.org 1623 [ + - - + ]: 74 : if (IsMVCCSnapshot(btleader->snapshot))
2825 rhaas@postgresql.org 1624 :UBC 0 : UnregisterSnapshot(btleader->snapshot);
2825 rhaas@postgresql.org 1625 :CBC 74 : DestroyParallelContext(btleader->pcxt);
1626 : 74 : ExitParallelMode();
1627 : 74 : }
1628 : :
1629 : : /*
1630 : : * Returns size of shared memory required to store state for a parallel
1631 : : * btree index build based on the snapshot its parallel scan will use.
1632 : : */
1633 : : static Size
2423 andres@anarazel.de 1634 : 74 : _bt_parallel_estimate_shared(Relation heap, Snapshot snapshot)
1635 : : {
1636 : : /* c.f. shm_toc_allocate as to why BUFFERALIGN is used */
1637 : 74 : return add_size(BUFFERALIGN(sizeof(BTShared)),
1638 : : table_parallelscan_estimate(heap, snapshot));
1639 : : }
1640 : :
1641 : : /*
1642 : : * Within leader, wait for end of heap scan.
1643 : : *
1644 : : * When called, parallel heap scan started by _bt_begin_parallel() will
1645 : : * already be underway within worker processes (when leader participates
1646 : : * as a worker, we should end up here just as workers are finishing).
1647 : : *
1648 : : * Fills in fields needed for ambuild statistics, and lets caller set
1649 : : * field indicating that some worker encountered a broken HOT chain.
1650 : : *
1651 : : * Returns the total number of heap tuples scanned.
1652 : : */
1653 : : static double
2825 rhaas@postgresql.org 1654 : 74 : _bt_parallel_heapscan(BTBuildState *buildstate, bool *brokenhotchain)
1655 : : {
1656 : 74 : BTShared *btshared = buildstate->btleader->btshared;
1657 : : int nparticipanttuplesorts;
1658 : : double reltuples;
1659 : :
1660 : 74 : nparticipanttuplesorts = buildstate->btleader->nparticipanttuplesorts;
1661 : : for (;;)
1662 : : {
1663 [ - + ]: 198 : SpinLockAcquire(&btshared->mutex);
1664 [ + + ]: 198 : if (btshared->nparticipantsdone == nparticipanttuplesorts)
1665 : : {
1666 : 74 : buildstate->havedead = btshared->havedead;
1667 : 74 : buildstate->indtuples = btshared->indtuples;
1668 : 74 : *brokenhotchain = btshared->brokenhotchain;
1669 : 74 : reltuples = btshared->reltuples;
1670 : 74 : SpinLockRelease(&btshared->mutex);
1671 : 74 : break;
1672 : : }
1673 : 124 : SpinLockRelease(&btshared->mutex);
1674 : :
1675 : 124 : ConditionVariableSleep(&btshared->workersdonecv,
1676 : : WAIT_EVENT_PARALLEL_CREATE_INDEX_SCAN);
1677 : : }
1678 : :
1679 : 74 : ConditionVariableCancelSleep();
1680 : :
1681 : 74 : return reltuples;
1682 : : }
1683 : :
1684 : : /*
1685 : : * Within leader, participate as a parallel worker.
1686 : : */
1687 : : static void
1688 : 74 : _bt_leader_participate_as_worker(BTBuildState *buildstate)
1689 : : {
1690 : 74 : BTLeader *btleader = buildstate->btleader;
1691 : : BTSpool *leaderworker;
1692 : : BTSpool *leaderworker2;
1693 : : int sortmem;
1694 : :
1695 : : /* Allocate memory and initialize private spool */
1696 : 74 : leaderworker = (BTSpool *) palloc0(sizeof(BTSpool));
1697 : 74 : leaderworker->heap = buildstate->spool->heap;
1698 : 74 : leaderworker->index = buildstate->spool->index;
1699 : 74 : leaderworker->isunique = buildstate->spool->isunique;
1363 peter@eisentraut.org 1700 : 74 : leaderworker->nulls_not_distinct = buildstate->spool->nulls_not_distinct;
1701 : :
1702 : : /* Initialize second spool, if required */
2825 rhaas@postgresql.org 1703 [ + + ]: 74 : if (!btleader->btshared->isunique)
1704 : 42 : leaderworker2 = NULL;
1705 : : else
1706 : : {
1707 : : /* Allocate memory for worker's own private secondary spool */
1708 : 32 : leaderworker2 = (BTSpool *) palloc0(sizeof(BTSpool));
1709 : :
1710 : : /* Initialize worker's own secondary spool */
1711 : 32 : leaderworker2->heap = leaderworker->heap;
1712 : 32 : leaderworker2->index = leaderworker->index;
1713 : 32 : leaderworker2->isunique = false;
1714 : : }
1715 : :
1716 : : /*
1717 : : * Might as well use reliable figure when doling out maintenance_work_mem
1718 : : * (when requested number of workers were not launched, this will be
1719 : : * somewhat higher than it is for other workers).
1720 : : */
1721 : 74 : sortmem = maintenance_work_mem / btleader->nparticipanttuplesorts;
1722 : :
1723 : : /* Perform work common to all participants */
1724 : 74 : _bt_parallel_scan_and_sort(leaderworker, leaderworker2, btleader->btshared,
1725 : : btleader->sharedsort, btleader->sharedsort2,
1726 : : sortmem, true);
1727 : :
1728 : : #ifdef BTREE_BUILD_STATS
1729 : : if (log_btree_build_stats)
1730 : : {
1731 : : ShowUsage("BTREE BUILD (Leader Partial Spool) STATISTICS");
1732 : : ResetUsage();
1733 : : }
1734 : : #endif /* BTREE_BUILD_STATS */
1735 : 74 : }
1736 : :
1737 : : /*
1738 : : * Perform work within a launched parallel process.
1739 : : */
1740 : : void
1741 : 74 : _bt_parallel_build_main(dsm_segment *seg, shm_toc *toc)
1742 : : {
1743 : : char *sharedquery;
1744 : : BTSpool *btspool;
1745 : : BTSpool *btspool2;
1746 : : BTShared *btshared;
1747 : : Sharedsort *sharedsort;
1748 : : Sharedsort *sharedsort2;
1749 : : Relation heapRel;
1750 : : Relation indexRel;
1751 : : LOCKMODE heapLockmode;
1752 : : LOCKMODE indexLockmode;
1753 : : WalUsage *walusage;
1754 : : BufferUsage *bufferusage;
1755 : : int sortmem;
1756 : :
1757 : : #ifdef BTREE_BUILD_STATS
1758 : : if (log_btree_build_stats)
1759 : : ResetUsage();
1760 : : #endif /* BTREE_BUILD_STATS */
1761 : :
1762 : : /*
1763 : : * The only possible status flag that can be set to the parallel worker is
1764 : : * PROC_IN_SAFE_IC.
1765 : : */
1439 akapila@postgresql.o 1766 [ - + - - ]: 74 : Assert((MyProc->statusFlags == 0) ||
1767 : : (MyProc->statusFlags == PROC_IN_SAFE_IC));
1768 : :
1769 : : /* Set debug_query_string for individual workers first */
1823 noah@leadboat.com 1770 : 74 : sharedquery = shm_toc_lookup(toc, PARALLEL_KEY_QUERY_TEXT, true);
2777 rhaas@postgresql.org 1771 : 74 : debug_query_string = sharedquery;
1772 : :
1773 : : /* Report the query string from leader */
1774 : 74 : pgstat_report_activity(STATE_RUNNING, debug_query_string);
1775 : :
1776 : : /* Look up nbtree shared state */
2825 1777 : 74 : btshared = shm_toc_lookup(toc, PARALLEL_KEY_BTREE_SHARED, false);
1778 : :
1779 : : /* Open relations using lock modes known to be obtained by index.c */
1780 [ + - ]: 74 : if (!btshared->isconcurrent)
1781 : : {
1782 : 74 : heapLockmode = ShareLock;
1783 : 74 : indexLockmode = AccessExclusiveLock;
1784 : : }
1785 : : else
1786 : : {
2825 rhaas@postgresql.org 1787 :UBC 0 : heapLockmode = ShareUpdateExclusiveLock;
1788 : 0 : indexLockmode = RowExclusiveLock;
1789 : : }
1790 : :
1791 : : /* Track query ID */
393 michael@paquier.xyz 1792 :CBC 74 : pgstat_report_query_id(btshared->queryid, false);
1793 : :
1794 : : /* Open relations within worker */
2472 andres@anarazel.de 1795 : 74 : heapRel = table_open(btshared->heaprelid, heapLockmode);
2825 rhaas@postgresql.org 1796 : 74 : indexRel = index_open(btshared->indexrelid, indexLockmode);
1797 : :
1798 : : /* Initialize worker's own spool */
1799 : 74 : btspool = (BTSpool *) palloc0(sizeof(BTSpool));
1800 : 74 : btspool->heap = heapRel;
1801 : 74 : btspool->index = indexRel;
1802 : 74 : btspool->isunique = btshared->isunique;
1363 peter@eisentraut.org 1803 : 74 : btspool->nulls_not_distinct = btshared->nulls_not_distinct;
1804 : :
1805 : : /* Look up shared state private to tuplesort.c */
2825 rhaas@postgresql.org 1806 : 74 : sharedsort = shm_toc_lookup(toc, PARALLEL_KEY_TUPLESORT, false);
1807 : 74 : tuplesort_attach_shared(sharedsort, seg);
1808 [ + + ]: 74 : if (!btshared->isunique)
1809 : : {
1810 : 42 : btspool2 = NULL;
1811 : 42 : sharedsort2 = NULL;
1812 : : }
1813 : : else
1814 : : {
1815 : : /* Allocate memory for worker's own private secondary spool */
1816 : 32 : btspool2 = (BTSpool *) palloc0(sizeof(BTSpool));
1817 : :
1818 : : /* Initialize worker's own secondary spool */
1819 : 32 : btspool2->heap = btspool->heap;
1820 : 32 : btspool2->index = btspool->index;
1821 : 32 : btspool2->isunique = false;
1822 : : /* Look up shared state private to tuplesort.c */
1823 : 32 : sharedsort2 = shm_toc_lookup(toc, PARALLEL_KEY_TUPLESORT_SPOOL2, false);
1824 : 32 : tuplesort_attach_shared(sharedsort2, seg);
1825 : : }
1826 : :
1827 : : /* Prepare to track buffer usage during parallel execution */
2033 akapila@postgresql.o 1828 : 74 : InstrStartParallelQuery();
1829 : :
1830 : : /* Perform sorting of spool, and possibly a spool2 */
2825 rhaas@postgresql.org 1831 : 74 : sortmem = maintenance_work_mem / btshared->scantuplesortstates;
1832 : 74 : _bt_parallel_scan_and_sort(btspool, btspool2, btshared, sharedsort,
1833 : : sharedsort2, sortmem, false);
1834 : :
1835 : : /* Report WAL/buffer usage during parallel execution */
2028 akapila@postgresql.o 1836 : 74 : bufferusage = shm_toc_lookup(toc, PARALLEL_KEY_BUFFER_USAGE, false);
2033 1837 : 74 : walusage = shm_toc_lookup(toc, PARALLEL_KEY_WAL_USAGE, false);
2028 1838 : 74 : InstrEndParallelQuery(&bufferusage[ParallelWorkerNumber],
1839 : 74 : &walusage[ParallelWorkerNumber]);
1840 : :
1841 : : #ifdef BTREE_BUILD_STATS
1842 : : if (log_btree_build_stats)
1843 : : {
1844 : : ShowUsage("BTREE BUILD (Worker Partial Spool) STATISTICS");
1845 : : ResetUsage();
1846 : : }
1847 : : #endif /* BTREE_BUILD_STATS */
1848 : :
2825 rhaas@postgresql.org 1849 : 74 : index_close(indexRel, indexLockmode);
2472 andres@anarazel.de 1850 : 74 : table_close(heapRel, heapLockmode);
2825 rhaas@postgresql.org 1851 : 74 : }
1852 : :
1853 : : /*
1854 : : * Perform a worker's portion of a parallel sort.
1855 : : *
1856 : : * This generates a tuplesort for passed btspool, and a second tuplesort
1857 : : * state if a second btspool is need (i.e. for unique index builds). All
1858 : : * other spool fields should already be set when this is called.
1859 : : *
1860 : : * sortmem is the amount of working memory to use within each worker,
1861 : : * expressed in KBs.
1862 : : *
1863 : : * When this returns, workers are done, and need only release resources.
1864 : : */
1865 : : static void
1866 : 148 : _bt_parallel_scan_and_sort(BTSpool *btspool, BTSpool *btspool2,
1867 : : BTShared *btshared, Sharedsort *sharedsort,
1868 : : Sharedsort *sharedsort2, int sortmem, bool progress)
1869 : : {
1870 : : SortCoordinate coordinate;
1871 : : BTBuildState buildstate;
1872 : : TableScanDesc scan;
1873 : : double reltuples;
1874 : : IndexInfo *indexInfo;
1875 : :
1876 : : /* Initialize local tuplesort coordination state */
1877 : 148 : coordinate = palloc0(sizeof(SortCoordinateData));
1878 : 148 : coordinate->isWorker = true;
1879 : 148 : coordinate->nParticipants = -1;
1880 : 148 : coordinate->sharedsort = sharedsort;
1881 : :
1882 : : /* Begin "partial" tuplesort */
1883 : 296 : btspool->sortstate = tuplesort_begin_index_btree(btspool->heap,
1884 : : btspool->index,
1885 : 148 : btspool->isunique,
1363 peter@eisentraut.org 1886 : 148 : btspool->nulls_not_distinct,
1887 : : sortmem, coordinate,
1888 : : TUPLESORT_NONE);
1889 : :
1890 : : /*
1891 : : * Just as with serial case, there may be a second spool. If so, a
1892 : : * second, dedicated spool2 partial tuplesort is required.
1893 : : */
2825 rhaas@postgresql.org 1894 [ + + ]: 148 : if (btspool2)
1895 : : {
1896 : : SortCoordinate coordinate2;
1897 : :
1898 : : /*
1899 : : * We expect that the second one (for dead tuples) won't get very
1900 : : * full, so we give it only work_mem (unless sortmem is less for
1901 : : * worker). Worker processes are generally permitted to allocate
1902 : : * work_mem independently.
1903 : : */
1904 : 64 : coordinate2 = palloc0(sizeof(SortCoordinateData));
1905 : 64 : coordinate2->isWorker = true;
1906 : 64 : coordinate2->nParticipants = -1;
1907 : 64 : coordinate2->sharedsort = sharedsort2;
1908 : 64 : btspool2->sortstate =
1363 peter@eisentraut.org 1909 : 64 : tuplesort_begin_index_btree(btspool->heap, btspool->index, false, false,
1910 : : Min(sortmem, work_mem), coordinate2,
1911 : : false);
1912 : : }
1913 : :
1914 : : /* Fill in buildstate for _bt_build_callback() */
2825 rhaas@postgresql.org 1915 : 148 : buildstate.isunique = btshared->isunique;
1363 peter@eisentraut.org 1916 : 148 : buildstate.nulls_not_distinct = btshared->nulls_not_distinct;
2825 rhaas@postgresql.org 1917 : 148 : buildstate.havedead = false;
1918 : 148 : buildstate.heap = btspool->heap;
1919 : 148 : buildstate.spool = btspool;
1920 : 148 : buildstate.spool2 = btspool2;
1921 : 148 : buildstate.indtuples = 0;
1922 : 148 : buildstate.btleader = NULL;
1923 : :
1924 : : /* Join parallel scan */
1925 : 148 : indexInfo = BuildIndexInfo(btspool->index);
1926 : 148 : indexInfo->ii_Concurrent = btshared->isconcurrent;
2401 alvherre@alvh.no-ip. 1927 : 148 : scan = table_beginscan_parallel(btspool->heap,
1928 : : ParallelTableScanFromBTShared(btshared));
2407 andres@anarazel.de 1929 : 148 : reltuples = table_index_build_scan(btspool->heap, btspool->index, indexInfo,
1930 : : true, progress, _bt_build_callback,
1931 : : &buildstate, scan);
1932 : :
1933 : : /* Execute this worker's part of the sort */
1600 alvherre@alvh.no-ip. 1934 [ + + ]: 148 : if (progress)
1935 : 74 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
1936 : : PROGRESS_BTREE_PHASE_PERFORMSORT_1);
2825 rhaas@postgresql.org 1937 : 148 : tuplesort_performsort(btspool->sortstate);
1938 [ + + ]: 148 : if (btspool2)
1939 : : {
1600 alvherre@alvh.no-ip. 1940 [ + + ]: 64 : if (progress)
1941 : 32 : pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE,
1942 : : PROGRESS_BTREE_PHASE_PERFORMSORT_2);
2825 rhaas@postgresql.org 1943 : 64 : tuplesort_performsort(btspool2->sortstate);
1944 : : }
1945 : :
1946 : : /*
1947 : : * Done. Record ambuild statistics, and whether we encountered a broken
1948 : : * HOT chain.
1949 : : */
1950 [ - + ]: 148 : SpinLockAcquire(&btshared->mutex);
1951 : 148 : btshared->nparticipantsdone++;
1952 : 148 : btshared->reltuples += reltuples;
1953 [ - + ]: 148 : if (buildstate.havedead)
2825 rhaas@postgresql.org 1954 :UBC 0 : btshared->havedead = true;
2825 rhaas@postgresql.org 1955 :CBC 148 : btshared->indtuples += buildstate.indtuples;
1956 [ - + ]: 148 : if (indexInfo->ii_BrokenHotChain)
2825 rhaas@postgresql.org 1957 :UBC 0 : btshared->brokenhotchain = true;
2825 rhaas@postgresql.org 1958 :CBC 148 : SpinLockRelease(&btshared->mutex);
1959 : :
1960 : : /* Notify leader */
1961 : 148 : ConditionVariableSignal(&btshared->workersdonecv);
1962 : :
1963 : : /* We can end tuplesorts immediately */
1964 : 148 : tuplesort_end(btspool->sortstate);
1965 [ + + ]: 148 : if (btspool2)
1966 : 64 : tuplesort_end(btspool2->sortstate);
1967 : 148 : }
|
