Age Owner TLA Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * nodeGatherMerge.c
4 : * Scan a plan in multiple workers, and do order-preserving merge.
5 : *
6 : * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : * IDENTIFICATION
10 : * src/backend/executor/nodeGatherMerge.c
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 :
15 : #include "postgres.h"
16 :
17 : #include "access/relscan.h"
18 : #include "access/xact.h"
19 : #include "executor/execdebug.h"
20 : #include "executor/execParallel.h"
21 : #include "executor/nodeGatherMerge.h"
22 : #include "executor/nodeSubplan.h"
23 : #include "executor/tqueue.h"
24 : #include "lib/binaryheap.h"
25 : #include "miscadmin.h"
26 : #include "optimizer/optimizer.h"
27 : #include "utils/memutils.h"
28 : #include "utils/rel.h"
29 :
30 : /*
31 : * When we read tuples from workers, it's a good idea to read several at once
32 : * for efficiency when possible: this minimizes context-switching overhead.
33 : * But reading too many at a time wastes memory without improving performance.
34 : * We'll read up to MAX_TUPLE_STORE tuples (in addition to the first one).
35 : */
36 : #define MAX_TUPLE_STORE 10
37 :
38 : /*
39 : * Pending-tuple array for each worker. This holds additional tuples that
40 : * we were able to fetch from the worker, but can't process yet. In addition,
41 : * this struct holds the "done" flag indicating the worker is known to have
42 : * no more tuples. (We do not use this struct for the leader; we don't keep
43 : * any pending tuples for the leader, and the need_to_scan_locally flag serves
44 : * as its "done" indicator.)
45 : */
46 : typedef struct GMReaderTupleBuffer
47 : {
48 : MinimalTuple *tuple; /* array of length MAX_TUPLE_STORE */
49 : int nTuples; /* number of tuples currently stored */
50 : int readCounter; /* index of next tuple to extract */
51 : bool done; /* true if reader is known exhausted */
52 : } GMReaderTupleBuffer;
53 :
54 : static TupleTableSlot *ExecGatherMerge(PlanState *pstate);
55 : static int32 heap_compare_slots(Datum a, Datum b, void *arg);
56 : static TupleTableSlot *gather_merge_getnext(GatherMergeState *gm_state);
57 : static MinimalTuple gm_readnext_tuple(GatherMergeState *gm_state, int nreader,
58 : bool nowait, bool *done);
59 : static void ExecShutdownGatherMergeWorkers(GatherMergeState *node);
60 : static void gather_merge_setup(GatherMergeState *gm_state);
61 : static void gather_merge_init(GatherMergeState *gm_state);
62 : static void gather_merge_clear_tuples(GatherMergeState *gm_state);
63 : static bool gather_merge_readnext(GatherMergeState *gm_state, int reader,
64 : bool nowait);
65 : static void load_tuple_array(GatherMergeState *gm_state, int reader);
66 :
67 : /* ----------------------------------------------------------------
68 : * ExecInitGather
69 : * ----------------------------------------------------------------
70 : */
71 : GatherMergeState *
2222 rhaas 72 CBC 138 : ExecInitGatherMerge(GatherMerge *node, EState *estate, int eflags)
73 : {
74 : GatherMergeState *gm_state;
75 : Plan *outerNode;
76 : TupleDesc tupDesc;
77 :
78 : /* Gather merge node doesn't have innerPlan node. */
79 138 : Assert(innerPlan(node) == NULL);
80 :
81 : /*
82 : * create state structure
83 : */
84 138 : gm_state = makeNode(GatherMergeState);
85 138 : gm_state->ps.plan = (Plan *) node;
86 138 : gm_state->ps.state = estate;
2092 andres 87 138 : gm_state->ps.ExecProcNode = ExecGatherMerge;
88 :
2048 tgl 89 138 : gm_state->initialized = false;
90 138 : gm_state->gm_initialized = false;
2049 rhaas 91 138 : gm_state->tuples_needed = -1;
92 :
93 : /*
94 : * Miscellaneous initialization
95 : *
96 : * create expression context for node
97 : */
2222 98 138 : ExecAssignExprContext(estate, &gm_state->ps);
99 :
100 : /*
101 : * GatherMerge doesn't support checking a qual (it's always more efficient
102 : * to do it in the child node).
103 : */
2048 tgl 104 138 : Assert(!node->plan.qual);
105 :
106 : /*
107 : * now initialize outer plan
108 : */
2222 rhaas 109 138 : outerNode = outerPlan(node);
110 138 : outerPlanState(gm_state) = ExecInitNode(outerNode, estate, eflags);
111 :
112 : /*
113 : * Leader may access ExecProcNode result directly (if
114 : * need_to_scan_locally), or from workers via tuple queue. So we can't
115 : * trivially rely on the slot type being fixed for expressions evaluated
116 : * within this node.
117 : */
1606 andres 118 138 : gm_state->ps.outeropsset = true;
119 138 : gm_state->ps.outeropsfixed = false;
120 :
121 : /*
122 : * Store the tuple descriptor into gather merge state, so we can use it
123 : * while initializing the gather merge slots.
124 : */
1878 125 138 : tupDesc = ExecGetResultType(outerPlanState(gm_state));
1961 rhaas 126 138 : gm_state->tupDesc = tupDesc;
127 :
128 : /*
129 : * Initialize result type and projection.
130 : */
1612 andres 131 138 : ExecInitResultTypeTL(&gm_state->ps);
1961 rhaas 132 138 : ExecConditionalAssignProjectionInfo(&gm_state->ps, tupDesc, OUTER_VAR);
133 :
134 : /*
135 : * Without projections result slot type is not trivially known, see
136 : * comment above.
137 : */
1606 andres 138 138 : if (gm_state->ps.ps_ProjInfo == NULL)
139 : {
140 132 : gm_state->ps.resultopsset = true;
141 132 : gm_state->ps.resultopsfixed = false;
142 : }
143 :
144 : /*
145 : * initialize sort-key information
146 : */
2222 rhaas 147 138 : if (node->numCols)
148 : {
149 : int i;
150 :
151 138 : gm_state->gm_nkeys = node->numCols;
152 138 : gm_state->gm_sortkeys =
153 138 : palloc0(sizeof(SortSupportData) * node->numCols);
154 :
155 306 : for (i = 0; i < node->numCols; i++)
156 : {
157 168 : SortSupport sortKey = gm_state->gm_sortkeys + i;
158 :
159 168 : sortKey->ssup_cxt = CurrentMemoryContext;
160 168 : sortKey->ssup_collation = node->collations[i];
161 168 : sortKey->ssup_nulls_first = node->nullsFirst[i];
162 168 : sortKey->ssup_attno = node->sortColIdx[i];
163 :
164 : /*
165 : * We don't perform abbreviated key conversion here, for the same
166 : * reasons that it isn't used in MergeAppend
167 : */
168 168 : sortKey->abbreviate = false;
169 :
170 168 : PrepareSortSupportFromOrderingOp(node->sortOperators[i], sortKey);
171 : }
172 : }
173 :
174 : /* Now allocate the workspace for gather merge */
2047 tgl 175 138 : gather_merge_setup(gm_state);
176 :
2222 rhaas 177 138 : return gm_state;
178 : }
179 :
180 : /* ----------------------------------------------------------------
181 : * ExecGatherMerge(node)
182 : *
183 : * Scans the relation via multiple workers and returns
184 : * the next qualifying tuple.
185 : * ----------------------------------------------------------------
186 : */
187 : static TupleTableSlot *
2092 andres 188 127379 : ExecGatherMerge(PlanState *pstate)
189 : {
190 127379 : GatherMergeState *node = castNode(GatherMergeState, pstate);
191 : TupleTableSlot *slot;
192 : ExprContext *econtext;
193 :
2084 194 127379 : CHECK_FOR_INTERRUPTS();
195 :
196 : /*
197 : * As with Gather, we don't launch workers until this node is actually
198 : * executed.
199 : */
2222 rhaas 200 127379 : if (!node->initialized)
201 : {
202 72 : EState *estate = node->ps.state;
2048 tgl 203 72 : GatherMerge *gm = castNode(GatherMerge, node->ps.plan);
204 :
205 : /*
206 : * Sometimes we might have to run without parallelism; but if parallel
207 : * mode is active then we can try to fire up some workers.
208 : */
1990 rhaas 209 72 : if (gm->num_workers > 0 && estate->es_use_parallel_mode)
210 : {
211 : ParallelContext *pcxt;
212 :
213 : /* Initialize, or re-initialize, shared state needed by workers. */
2222 214 72 : if (!node->pei)
276 tgl 215 GNC 57 : node->pei = ExecInitParallelPlan(outerPlanState(node),
216 : estate,
217 : gm->initParam,
218 : gm->num_workers,
219 : node->tuples_needed);
220 : else
221 15 : ExecParallelReinitialize(outerPlanState(node),
1970 rhaas 222 CBC 15 : node->pei,
223 : gm->initParam);
224 :
225 : /* Try to launch workers. */
2222 226 72 : pcxt = node->pei->pcxt;
227 72 : LaunchParallelWorkers(pcxt);
228 : /* We save # workers launched for the benefit of EXPLAIN */
229 72 : node->nworkers_launched = pcxt->nworkers_launched;
230 :
231 : /* Set up tuple queue readers to read the results. */
232 72 : if (pcxt->nworkers_launched > 0)
233 : {
2033 andres 234 66 : ExecParallelCreateReaders(node->pei);
235 : /* Make a working array showing the active readers */
2046 tgl 236 66 : node->nreaders = pcxt->nworkers_launched;
237 66 : node->reader = (TupleQueueReader **)
238 66 : palloc(node->nreaders * sizeof(TupleQueueReader *));
239 66 : memcpy(node->reader, node->pei->reader,
240 66 : node->nreaders * sizeof(TupleQueueReader *));
241 : }
242 : else
243 : {
244 : /* No workers? Then never mind. */
245 6 : node->nreaders = 0;
246 6 : node->reader = NULL;
247 : }
248 : }
249 :
250 : /* allow leader to participate if enabled or no choice */
1971 rhaas 251 72 : if (parallel_leader_participation || node->nreaders == 0)
252 69 : node->need_to_scan_locally = true;
2222 253 72 : node->initialized = true;
254 : }
255 :
256 : /*
257 : * Reset per-tuple memory context to free any expression evaluation
258 : * storage allocated in the previous tuple cycle.
259 : */
260 127379 : econtext = node->ps.ps_ExprContext;
261 127379 : ResetExprContext(econtext);
262 :
263 : /*
264 : * Get next tuple, either from one of our workers, or by running the plan
265 : * ourselves.
266 : */
267 127379 : slot = gather_merge_getnext(node);
268 127379 : if (TupIsNull(slot))
269 57 : return NULL;
270 :
271 : /* If no projection is required, we're done. */
1961 272 127322 : if (node->ps.ps_ProjInfo == NULL)
273 127322 : return slot;
274 :
275 : /*
276 : * Form the result tuple using ExecProject(), and return it.
277 : */
2222 rhaas 278 UBC 0 : econtext->ecxt_outertuple = slot;
279 0 : return ExecProject(node->ps.ps_ProjInfo);
280 : }
281 :
282 : /* ----------------------------------------------------------------
283 : * ExecEndGatherMerge
284 : *
285 : * frees any storage allocated through C routines.
286 : * ----------------------------------------------------------------
287 : */
288 : void
2222 rhaas 289 CBC 138 : ExecEndGatherMerge(GatherMergeState *node)
290 : {
2153 bruce 291 138 : ExecEndNode(outerPlanState(node)); /* let children clean up first */
2222 rhaas 292 138 : ExecShutdownGatherMerge(node);
293 138 : ExecFreeExprContext(&node->ps);
1612 andres 294 138 : if (node->ps.ps_ResultTupleSlot)
295 6 : ExecClearTuple(node->ps.ps_ResultTupleSlot);
2222 rhaas 296 138 : }
297 :
298 : /* ----------------------------------------------------------------
299 : * ExecShutdownGatherMerge
300 : *
301 : * Destroy the setup for parallel workers including parallel context.
302 : * ----------------------------------------------------------------
303 : */
304 : void
305 198 : ExecShutdownGatherMerge(GatherMergeState *node)
306 : {
307 198 : ExecShutdownGatherMergeWorkers(node);
308 :
309 : /* Now destroy the parallel context. */
310 198 : if (node->pei != NULL)
311 : {
312 57 : ExecParallelCleanup(node->pei);
313 57 : node->pei = NULL;
314 : }
315 198 : }
316 :
317 : /* ----------------------------------------------------------------
318 : * ExecShutdownGatherMergeWorkers
319 : *
320 : * Stop all the parallel workers.
321 : * ----------------------------------------------------------------
322 : */
323 : static void
324 222 : ExecShutdownGatherMergeWorkers(GatherMergeState *node)
325 : {
326 222 : if (node->pei != NULL)
327 72 : ExecParallelFinish(node->pei);
328 :
329 : /* Flush local copy of reader array */
2046 tgl 330 222 : if (node->reader)
331 66 : pfree(node->reader);
332 222 : node->reader = NULL;
2222 rhaas 333 222 : }
334 :
335 : /* ----------------------------------------------------------------
336 : * ExecReScanGatherMerge
337 : *
338 : * Prepare to re-scan the result of a GatherMerge.
339 : * ----------------------------------------------------------------
340 : */
341 : void
342 24 : ExecReScanGatherMerge(GatherMergeState *node)
343 : {
2048 tgl 344 24 : GatherMerge *gm = (GatherMerge *) node->ps.plan;
345 24 : PlanState *outerPlan = outerPlanState(node);
346 :
347 : /* Make sure any existing workers are gracefully shut down */
2222 rhaas 348 24 : ExecShutdownGatherMergeWorkers(node);
349 :
350 : /* Free any unused tuples, so we don't leak memory across rescans */
2047 tgl 351 24 : gather_merge_clear_tuples(node);
352 :
353 : /* Mark node so that shared state will be rebuilt at next call */
2222 rhaas 354 24 : node->initialized = false;
2061 tgl 355 24 : node->gm_initialized = false;
356 :
357 : /*
358 : * Set child node's chgParam to tell it that the next scan might deliver a
359 : * different set of rows within the leader process. (The overall rowset
360 : * shouldn't change, but the leader process's subset might; hence nodes
361 : * between here and the parallel table scan node mustn't optimize on the
362 : * assumption of an unchanging rowset.)
363 : */
2048 364 24 : if (gm->rescan_param >= 0)
365 24 : outerPlan->chgParam = bms_add_member(outerPlan->chgParam,
366 : gm->rescan_param);
367 :
368 : /*
369 : * If chgParam of subnode is not null then plan will be re-scanned by
370 : * first ExecProcNode. Note: because this does nothing if we have a
371 : * rescan_param, it's currently guaranteed that parallel-aware child nodes
372 : * will not see a ReScan call until after they get a ReInitializeDSM call.
373 : * That ordering might not be something to rely on, though. A good rule
374 : * of thumb is that ReInitializeDSM should reset only shared state, ReScan
375 : * should reset only local state, and anything that depends on both of
376 : * those steps being finished must wait until the first ExecProcNode call.
377 : */
378 24 : if (outerPlan->chgParam == NULL)
2048 tgl 379 UBC 0 : ExecReScan(outerPlan);
2222 rhaas 380 CBC 24 : }
381 :
382 : /*
383 : * Set up the data structures that we'll need for Gather Merge.
384 : *
385 : * We allocate these once on the basis of gm->num_workers, which is an
386 : * upper bound for the number of workers we'll actually have. During
387 : * a rescan, we reset the structures to empty. This approach simplifies
388 : * not leaking memory across rescans.
389 : *
390 : * In the gm_slots[] array, index 0 is for the leader, and indexes 1 to n
391 : * are for workers. The values placed into gm_heap correspond to indexes
392 : * in gm_slots[]. The gm_tuple_buffers[] array, however, is indexed from
393 : * 0 to n-1; it has no entry for the leader.
394 : */
395 : static void
2047 tgl 396 138 : gather_merge_setup(GatherMergeState *gm_state)
397 : {
398 138 : GatherMerge *gm = castNode(GatherMerge, gm_state->ps.plan);
399 138 : int nreaders = gm->num_workers;
400 : int i;
401 :
402 : /*
403 : * Allocate gm_slots for the number of workers + one more slot for leader.
404 : * Slot 0 is always for the leader. Leader always calls ExecProcNode() to
405 : * read the tuple, and then stores it directly into its gm_slots entry.
406 : * For other slots, code below will call ExecInitExtraTupleSlot() to
407 : * create a slot for the worker's results. Note that during any single
408 : * scan, we might have fewer than num_workers available workers, in which
409 : * case the extra array entries go unused.
410 : */
411 138 : gm_state->gm_slots = (TupleTableSlot **)
412 138 : palloc0((nreaders + 1) * sizeof(TupleTableSlot *));
413 :
414 : /* Allocate the tuple slot and tuple array for each worker */
415 138 : gm_state->gm_tuple_buffers = (GMReaderTupleBuffer *)
416 138 : palloc0(nreaders * sizeof(GMReaderTupleBuffer));
417 :
418 516 : for (i = 0; i < nreaders; i++)
419 : {
420 : /* Allocate the tuple array with length MAX_TUPLE_STORE */
2222 rhaas 421 756 : gm_state->gm_tuple_buffers[i].tuple =
996 tmunro 422 378 : (MinimalTuple *) palloc0(sizeof(MinimalTuple) * MAX_TUPLE_STORE);
423 :
424 : /* Initialize tuple slot for worker */
1878 andres 425 378 : gm_state->gm_slots[i + 1] =
1606 426 378 : ExecInitExtraTupleSlot(gm_state->ps.state, gm_state->tupDesc,
427 : &TTSOpsMinimalTuple);
428 : }
429 :
430 : /* Allocate the resources for the merge */
2047 tgl 431 138 : gm_state->gm_heap = binaryheap_allocate(nreaders + 1,
432 : heap_compare_slots,
433 : gm_state);
434 138 : }
435 :
436 : /*
437 : * Initialize the Gather Merge.
438 : *
439 : * Reset data structures to ensure they're empty. Then pull at least one
440 : * tuple from leader + each worker (or set its "done" indicator), and set up
441 : * the heap.
442 : */
443 : static void
444 72 : gather_merge_init(GatherMergeState *gm_state)
445 : {
446 72 : int nreaders = gm_state->nreaders;
447 72 : bool nowait = true;
448 : int i;
449 :
450 : /* Assert that gather_merge_setup made enough space */
451 72 : Assert(nreaders <= castNode(GatherMerge, gm_state->ps.plan)->num_workers);
452 :
453 : /* Reset leader's tuple slot to empty */
454 72 : gm_state->gm_slots[0] = NULL;
455 :
456 : /* Reset the tuple slot and tuple array for each worker */
457 264 : for (i = 0; i < nreaders; i++)
458 : {
459 : /* Reset tuple array to empty */
460 192 : gm_state->gm_tuple_buffers[i].nTuples = 0;
461 192 : gm_state->gm_tuple_buffers[i].readCounter = 0;
462 : /* Reset done flag to not-done */
463 192 : gm_state->gm_tuple_buffers[i].done = false;
464 : /* Ensure output slot is empty */
465 192 : ExecClearTuple(gm_state->gm_slots[i + 1]);
466 : }
467 :
468 : /* Reset binary heap to empty */
469 72 : binaryheap_reset(gm_state->gm_heap);
470 :
471 : /*
472 : * First, try to read a tuple from each worker (including leader) in
473 : * nowait mode. After this, if not all workers were able to produce a
474 : * tuple (or a "done" indication), then re-read from remaining workers,
475 : * this time using wait mode. Add all live readers (those producing at
476 : * least one tuple) to the heap.
477 : */
2222 rhaas 478 129 : reread:
2047 tgl 479 624 : for (i = 0; i <= nreaders; i++)
480 : {
2084 andres 481 495 : CHECK_FOR_INTERRUPTS();
482 :
483 : /* skip this source if already known done */
2047 tgl 484 861 : if ((i == 0) ? gm_state->need_to_scan_locally :
485 366 : !gm_state->gm_tuple_buffers[i - 1].done)
486 : {
2048 487 485 : if (TupIsNull(gm_state->gm_slots[i]))
488 : {
489 : /* Don't have a tuple yet, try to get one */
490 429 : if (gather_merge_readnext(gm_state, i, nowait))
491 101 : binaryheap_add_unordered(gm_state->gm_heap,
492 : Int32GetDatum(i));
493 : }
494 : else
495 : {
496 : /*
497 : * We already got at least one tuple from this worker, but
498 : * might as well see if it has any more ready by now.
499 : */
500 56 : load_tuple_array(gm_state, i);
501 : }
502 : }
503 : }
504 :
505 : /* need not recheck leader, since nowait doesn't matter for it */
2047 506 324 : for (i = 1; i <= nreaders; i++)
507 : {
508 252 : if (!gm_state->gm_tuple_buffers[i - 1].done &&
2048 509 81 : TupIsNull(gm_state->gm_slots[i]))
510 : {
511 57 : nowait = false;
2222 rhaas 512 57 : goto reread;
513 : }
514 : }
515 :
516 : /* Now heapify the heap. */
517 72 : binaryheap_build(gm_state->gm_heap);
518 :
519 72 : gm_state->gm_initialized = true;
520 72 : }
521 :
522 : /*
523 : * Clear out the tuple table slot, and any unused pending tuples,
524 : * for each gather merge input.
525 : */
526 : static void
2047 tgl 527 81 : gather_merge_clear_tuples(GatherMergeState *gm_state)
528 : {
529 : int i;
530 :
2222 rhaas 531 306 : for (i = 0; i < gm_state->nreaders; i++)
532 : {
2047 tgl 533 225 : GMReaderTupleBuffer *tuple_buffer = &gm_state->gm_tuple_buffers[i];
534 :
535 225 : while (tuple_buffer->readCounter < tuple_buffer->nTuples)
996 tmunro 536 UBC 0 : pfree(tuple_buffer->tuple[tuple_buffer->readCounter++]);
537 :
2047 tgl 538 CBC 225 : ExecClearTuple(gm_state->gm_slots[i + 1]);
539 : }
2222 rhaas 540 81 : }
541 :
542 : /*
543 : * Read the next tuple for gather merge.
544 : *
545 : * Fetch the sorted tuple out of the heap.
546 : */
547 : static TupleTableSlot *
548 127379 : gather_merge_getnext(GatherMergeState *gm_state)
549 : {
550 : int i;
551 :
2219 tgl 552 127379 : if (!gm_state->gm_initialized)
553 : {
554 : /*
555 : * First time through: pull the first tuple from each participant, and
556 : * set up the heap.
557 : */
2222 rhaas 558 72 : gather_merge_init(gm_state);
559 : }
560 : else
561 : {
562 : /*
563 : * Otherwise, pull the next tuple from whichever participant we
564 : * returned from last time, and reinsert that participant's index into
565 : * the heap, because it might now compare differently against the
566 : * other elements of the heap.
567 : */
568 127307 : i = DatumGetInt32(binaryheap_first(gm_state->gm_heap));
569 :
570 127307 : if (gather_merge_readnext(gm_state, i, false))
571 127221 : binaryheap_replace_first(gm_state->gm_heap, Int32GetDatum(i));
572 : else
573 : {
574 : /* reader exhausted, remove it from heap */
575 86 : (void) binaryheap_remove_first(gm_state->gm_heap);
576 : }
577 : }
578 :
579 127379 : if (binaryheap_empty(gm_state->gm_heap))
580 : {
581 : /* All the queues are exhausted, and so is the heap */
2047 tgl 582 57 : gather_merge_clear_tuples(gm_state);
2200 rhaas 583 57 : return NULL;
584 : }
585 : else
586 : {
587 : /* Return next tuple from whichever participant has the leading one */
2222 588 127322 : i = DatumGetInt32(binaryheap_first(gm_state->gm_heap));
589 127322 : return gm_state->gm_slots[i];
590 : }
591 : }
592 :
593 : /*
594 : * Read tuple(s) for given reader in nowait mode, and load into its tuple
595 : * array, until we have MAX_TUPLE_STORE of them or would have to block.
596 : */
597 : static void
2048 tgl 598 113 : load_tuple_array(GatherMergeState *gm_state, int reader)
599 : {
600 : GMReaderTupleBuffer *tuple_buffer;
601 : int i;
602 :
603 : /* Don't do anything if this is the leader. */
2047 604 113 : if (reader == 0)
2222 rhaas 605 54 : return;
606 :
2047 tgl 607 59 : tuple_buffer = &gm_state->gm_tuple_buffers[reader - 1];
608 :
609 : /* If there's nothing in the array, reset the counters to zero. */
2222 rhaas 610 59 : if (tuple_buffer->nTuples == tuple_buffer->readCounter)
611 57 : tuple_buffer->nTuples = tuple_buffer->readCounter = 0;
612 :
613 : /* Try to fill additional slots in the array. */
614 486 : for (i = tuple_buffer->nTuples; i < MAX_TUPLE_STORE; i++)
615 : {
616 : MinimalTuple tuple;
617 :
2048 tgl 618 459 : tuple = gm_readnext_tuple(gm_state,
619 : reader,
620 : true,
621 : &tuple_buffer->done);
996 tmunro 622 459 : if (!tuple)
2222 rhaas 623 32 : break;
1952 624 427 : tuple_buffer->tuple[i] = tuple;
2222 625 427 : tuple_buffer->nTuples++;
626 : }
627 : }
628 :
629 : /*
630 : * Store the next tuple for a given reader into the appropriate slot.
631 : *
632 : * Returns true if successful, false if not (either reader is exhausted,
633 : * or we didn't want to wait for a tuple). Sets done flag if reader
634 : * is found to be exhausted.
635 : */
636 : static bool
637 127736 : gather_merge_readnext(GatherMergeState *gm_state, int reader, bool nowait)
638 : {
639 : GMReaderTupleBuffer *tuple_buffer;
640 : MinimalTuple tup;
641 :
642 : /*
643 : * If we're being asked to generate a tuple from the leader, then we just
644 : * call ExecProcNode as normal to produce one.
645 : */
2047 tgl 646 127736 : if (reader == 0)
647 : {
2222 rhaas 648 126892 : if (gm_state->need_to_scan_locally)
649 : {
650 126892 : PlanState *outerPlan = outerPlanState(gm_state);
651 : TupleTableSlot *outerTupleSlot;
1809 tgl 652 126892 : EState *estate = gm_state->ps.state;
653 :
654 : /* Install our DSA area while executing the plan. */
1938 rhaas 655 126892 : estate->es_query_dsa = gm_state->pei ? gm_state->pei->area : NULL;
2222 656 126892 : outerTupleSlot = ExecProcNode(outerPlan);
1938 657 126892 : estate->es_query_dsa = NULL;
658 :
2222 659 126892 : if (!TupIsNull(outerTupleSlot))
660 : {
2047 tgl 661 126838 : gm_state->gm_slots[0] = outerTupleSlot;
2222 rhaas 662 126838 : return true;
663 : }
664 : /* need_to_scan_locally serves as "done" flag for leader */
665 54 : gm_state->need_to_scan_locally = false;
666 : }
667 54 : return false;
668 : }
669 :
670 : /* Otherwise, check the state of the relevant tuple buffer. */
2047 tgl 671 844 : tuple_buffer = &gm_state->gm_tuple_buffers[reader - 1];
672 :
2222 rhaas 673 844 : if (tuple_buffer->nTuples > tuple_buffer->readCounter)
674 : {
675 : /* Return any tuple previously read that is still buffered. */
676 427 : tup = tuple_buffer->tuple[tuple_buffer->readCounter++];
677 : }
678 417 : else if (tuple_buffer->done)
679 : {
680 : /* Reader is known to be exhausted. */
681 32 : return false;
682 : }
683 : else
684 : {
685 : /* Read and buffer next tuple. */
2048 tgl 686 385 : tup = gm_readnext_tuple(gm_state,
687 : reader,
688 : nowait,
689 : &tuple_buffer->done);
996 tmunro 690 385 : if (!tup)
2048 tgl 691 328 : return false;
692 :
693 : /*
694 : * Attempt to read more tuples in nowait mode and store them in the
695 : * pending-tuple array for the reader.
696 : */
697 57 : load_tuple_array(gm_state, reader);
698 : }
699 :
996 tmunro 700 484 : Assert(tup);
701 :
702 : /* Build the TupleTableSlot for the given tuple */
697 tgl 703 484 : ExecStoreMinimalTuple(tup, /* tuple to store */
704 484 : gm_state->gm_slots[reader], /* slot in which to
705 : * store the tuple */
706 : true); /* pfree tuple when done with it */
707 :
2222 rhaas 708 484 : return true;
709 : }
710 :
711 : /*
712 : * Attempt to read a tuple from given worker.
713 : */
714 : static MinimalTuple
715 844 : gm_readnext_tuple(GatherMergeState *gm_state, int nreader, bool nowait,
716 : bool *done)
717 : {
718 : TupleQueueReader *reader;
719 : MinimalTuple tup;
720 :
721 : /* Check for async events, particularly messages from workers. */
722 844 : CHECK_FOR_INTERRUPTS();
723 :
724 : /*
725 : * Attempt to read a tuple.
726 : *
727 : * Note that TupleQueueReaderNext will just return NULL for a worker which
728 : * fails to initialize. We'll treat that worker as having produced no
729 : * tuples; WaitForParallelWorkersToFinish will error out when we get
730 : * there.
731 : */
2047 tgl 732 844 : reader = gm_state->reader[nreader - 1];
2222 rhaas 733 844 : tup = TupleQueueReaderNext(reader, nowait, done);
734 :
735 : /*
736 : * Since we'll be buffering these across multiple calls, we need to make a
737 : * copy.
738 : */
996 tmunro 739 844 : return tup ? heap_copy_minimal_tuple(tup) : NULL;
740 : }
741 :
742 : /*
743 : * We have one slot for each item in the heap array. We use SlotNumber
744 : * to store slot indexes. This doesn't actually provide any formal
745 : * type-safety, but it makes the code more self-documenting.
746 : */
747 : typedef int32 SlotNumber;
748 :
749 : /*
750 : * Compare the tuples in the two given slots.
751 : */
752 : static int32
2222 rhaas 753 1572 : heap_compare_slots(Datum a, Datum b, void *arg)
754 : {
755 1572 : GatherMergeState *node = (GatherMergeState *) arg;
756 1572 : SlotNumber slot1 = DatumGetInt32(a);
757 1572 : SlotNumber slot2 = DatumGetInt32(b);
758 :
759 1572 : TupleTableSlot *s1 = node->gm_slots[slot1];
760 1572 : TupleTableSlot *s2 = node->gm_slots[slot2];
761 : int nkey;
762 :
763 1572 : Assert(!TupIsNull(s1));
764 1572 : Assert(!TupIsNull(s2));
765 :
766 2410 : for (nkey = 0; nkey < node->gm_nkeys; nkey++)
767 : {
768 1572 : SortSupport sortKey = node->gm_sortkeys + nkey;
769 1572 : AttrNumber attno = sortKey->ssup_attno;
770 : Datum datum1,
771 : datum2;
772 : bool isNull1,
773 : isNull2;
774 : int compare;
775 :
776 1572 : datum1 = slot_getattr(s1, attno, &isNull1);
777 1572 : datum2 = slot_getattr(s2, attno, &isNull2);
778 :
779 1572 : compare = ApplySortComparator(datum1, isNull1,
780 : datum2, isNull2,
781 : sortKey);
782 1572 : if (compare != 0)
783 : {
1647 tgl 784 734 : INVERT_COMPARE_RESULT(compare);
785 734 : return compare;
786 : }
787 : }
2222 rhaas 788 838 : return 0;
789 : }
|
