Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * execParallel.c
4 : * Support routines for parallel execution.
5 : *
6 : * Portions Copyright (c) 1996-2020, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : * This file contains routines that are intended to support setting up,
10 : * using, and tearing down a ParallelContext from within the PostgreSQL
11 : * executor. The ParallelContext machinery will handle starting the
12 : * workers and ensuring that their state generally matches that of the
13 : * leader; see src/backend/access/transam/README.parallel for details.
14 : * However, we must save and restore relevant executor state, such as
15 : * any ParamListInfo associated with the query, buffer/WAL usage info, and
16 : * the actual plan to be passed down to the worker.
17 : *
18 : * IDENTIFICATION
19 : * src/backend/executor/execParallel.c
20 : *
21 : *-------------------------------------------------------------------------
22 : */
23 :
24 : #include "postgres.h"
25 :
26 : #include "executor/execParallel.h"
27 : #include "executor/executor.h"
28 : #include "executor/nodeAppend.h"
29 : #include "executor/nodeBitmapHeapscan.h"
30 : #include "executor/nodeCustom.h"
31 : #include "executor/nodeForeignscan.h"
32 : #include "executor/nodeHash.h"
33 : #include "executor/nodeHashjoin.h"
34 : #include "executor/nodeIncrementalSort.h"
35 : #include "executor/nodeIndexonlyscan.h"
36 : #include "executor/nodeIndexscan.h"
37 : #include "executor/nodeSeqscan.h"
38 : #include "executor/nodeSort.h"
39 : #include "executor/nodeSubplan.h"
40 : #include "executor/tqueue.h"
41 : #include "jit/jit.h"
42 : #include "nodes/nodeFuncs.h"
43 : #include "pgstat.h"
44 : #include "storage/spin.h"
45 : #include "tcop/tcopprot.h"
46 : #include "utils/datum.h"
47 : #include "utils/dsa.h"
48 : #include "utils/lsyscache.h"
49 : #include "utils/memutils.h"
50 : #include "utils/snapmgr.h"
51 :
52 : /*
53 : * Magic numbers for parallel executor communication. We use constants
54 : * greater than any 32-bit integer here so that values < 2^32 can be used
55 : * by individual parallel nodes to store their own state.
56 : */
57 : #define PARALLEL_KEY_EXECUTOR_FIXED UINT64CONST(0xE000000000000001)
58 : #define PARALLEL_KEY_PLANNEDSTMT UINT64CONST(0xE000000000000002)
59 : #define PARALLEL_KEY_PARAMLISTINFO UINT64CONST(0xE000000000000003)
60 : #define PARALLEL_KEY_BUFFER_USAGE UINT64CONST(0xE000000000000004)
61 : #define PARALLEL_KEY_TUPLE_QUEUE UINT64CONST(0xE000000000000005)
62 : #define PARALLEL_KEY_INSTRUMENTATION UINT64CONST(0xE000000000000006)
63 : #define PARALLEL_KEY_DSA UINT64CONST(0xE000000000000007)
64 : #define PARALLEL_KEY_QUERY_TEXT UINT64CONST(0xE000000000000008)
65 : #define PARALLEL_KEY_JIT_INSTRUMENTATION UINT64CONST(0xE000000000000009)
66 : #define PARALLEL_KEY_WAL_USAGE UINT64CONST(0xE00000000000000A)
67 :
68 : #define PARALLEL_TUPLE_QUEUE_SIZE 65536
69 :
70 : /*
71 : * Fixed-size random stuff that we need to pass to parallel workers.
72 : */
73 : typedef struct FixedParallelExecutorState
74 : {
75 : int64 tuples_needed; /* tuple bound, see ExecSetTupleBound */
76 : dsa_pointer param_exec;
77 : int eflags;
78 : int jit_flags;
79 : } FixedParallelExecutorState;
80 :
81 : /*
82 : * DSM structure for accumulating per-PlanState instrumentation.
83 : *
84 : * instrument_options: Same meaning here as in instrument.c.
85 : *
86 : * instrument_offset: Offset, relative to the start of this structure,
87 : * of the first Instrumentation object. This will depend on the length of
88 : * the plan_node_id array.
89 : *
90 : * num_workers: Number of workers.
91 : *
92 : * num_plan_nodes: Number of plan nodes.
93 : *
94 : * plan_node_id: Array of plan nodes for which we are gathering instrumentation
95 : * from parallel workers. The length of this array is given by num_plan_nodes.
96 : */
97 : struct SharedExecutorInstrumentation
98 : {
99 : int instrument_options;
100 : int instrument_offset;
101 : int num_workers;
102 : int num_plan_nodes;
103 : int plan_node_id[FLEXIBLE_ARRAY_MEMBER];
104 : /* array of num_plan_nodes * num_workers Instrumentation objects follows */
105 : };
106 : #define GetInstrumentationArray(sei) \
107 : (AssertVariableIsOfTypeMacro(sei, SharedExecutorInstrumentation *), \
108 : (Instrumentation *) (((char *) sei) + sei->instrument_offset))
109 :
110 : /* Context object for ExecParallelEstimate. */
111 : typedef struct ExecParallelEstimateContext
112 : {
113 : ParallelContext *pcxt;
114 : int nnodes;
115 : } ExecParallelEstimateContext;
116 :
117 : /* Context object for ExecParallelInitializeDSM. */
118 : typedef struct ExecParallelInitializeDSMContext
119 : {
120 : ParallelContext *pcxt;
121 : SharedExecutorInstrumentation *instrumentation;
122 : int nnodes;
123 : } ExecParallelInitializeDSMContext;
124 :
125 : /* Helper functions that run in the parallel leader. */
126 : static char *ExecSerializePlan(Plan *plan, EState *estate);
127 : static bool ExecParallelEstimate(PlanState *node,
128 : ExecParallelEstimateContext *e);
129 : static bool ExecParallelInitializeDSM(PlanState *node,
130 : ExecParallelInitializeDSMContext *d);
131 : static shm_mq_handle **ExecParallelSetupTupleQueues(ParallelContext *pcxt,
132 : bool reinitialize);
133 : static bool ExecParallelReInitializeDSM(PlanState *planstate,
134 : ParallelContext *pcxt);
135 : static bool ExecParallelRetrieveInstrumentation(PlanState *planstate,
136 : SharedExecutorInstrumentation *instrumentation);
137 :
138 : /* Helper function that runs in the parallel worker. */
139 : static DestReceiver *ExecParallelGetReceiver(dsm_segment *seg, shm_toc *toc);
140 :
141 : /*
142 : * Create a serialized representation of the plan to be sent to each worker.
143 : */
144 : static char *
145 370 : ExecSerializePlan(Plan *plan, EState *estate)
146 : {
147 : PlannedStmt *pstmt;
148 : ListCell *lc;
149 :
150 : /* We can't scribble on the original plan, so make a copy. */
151 370 : plan = copyObject(plan);
152 :
153 : /*
154 : * The worker will start its own copy of the executor, and that copy will
155 : * insert a junk filter if the toplevel node has any resjunk entries. We
156 : * don't want that to happen, because while resjunk columns shouldn't be
157 : * sent back to the user, here the tuples are coming back to another
158 : * backend which may very well need them. So mutate the target list
159 : * accordingly. This is sort of a hack; there might be better ways to do
160 : * this...
161 : */
162 1026 : foreach(lc, plan->targetlist)
163 : {
164 656 : TargetEntry *tle = lfirst_node(TargetEntry, lc);
165 :
166 656 : tle->resjunk = false;
167 : }
168 :
169 : /*
170 : * Create a dummy PlannedStmt. Most of the fields don't need to be valid
171 : * for our purposes, but the worker will need at least a minimal
172 : * PlannedStmt to start the executor.
173 : */
174 370 : pstmt = makeNode(PlannedStmt);
175 370 : pstmt->commandType = CMD_SELECT;
176 370 : pstmt->queryId = UINT64CONST(0);
177 370 : pstmt->hasReturning = false;
178 370 : pstmt->hasModifyingCTE = false;
179 370 : pstmt->canSetTag = true;
180 370 : pstmt->transientPlan = false;
181 370 : pstmt->dependsOnRole = false;
182 370 : pstmt->parallelModeNeeded = false;
183 370 : pstmt->planTree = plan;
184 370 : pstmt->rtable = estate->es_range_table;
185 370 : pstmt->resultRelations = NIL;
186 370 : pstmt->rootResultRelations = NIL;
187 370 : pstmt->appendRelations = NIL;
188 :
189 : /*
190 : * Transfer only parallel-safe subplans, leaving a NULL "hole" in the list
191 : * for unsafe ones (so that the list indexes of the safe ones are
192 : * preserved). This positively ensures that the worker won't try to run,
193 : * or even do ExecInitNode on, an unsafe subplan. That's important to
194 : * protect, eg, non-parallel-aware FDWs from getting into trouble.
195 : */
196 370 : pstmt->subplans = NIL;
197 390 : foreach(lc, estate->es_plannedstmt->subplans)
198 : {
199 20 : Plan *subplan = (Plan *) lfirst(lc);
200 :
201 20 : if (subplan && !subplan->parallel_safe)
202 16 : subplan = NULL;
203 20 : pstmt->subplans = lappend(pstmt->subplans, subplan);
204 : }
205 :
206 370 : pstmt->rewindPlanIDs = NULL;
207 370 : pstmt->rowMarks = NIL;
208 370 : pstmt->relationOids = NIL;
209 370 : pstmt->invalItems = NIL; /* workers can't replan anyway... */
210 370 : pstmt->paramExecTypes = estate->es_plannedstmt->paramExecTypes;
211 370 : pstmt->utilityStmt = NULL;
212 370 : pstmt->stmt_location = -1;
213 370 : pstmt->stmt_len = -1;
214 :
215 : /* Return serialized copy of our dummy PlannedStmt. */
216 370 : return nodeToString(pstmt);
217 : }
218 :
219 : /*
220 : * Parallel-aware plan nodes (and occasionally others) may need some state
221 : * which is shared across all parallel workers. Before we size the DSM, give
222 : * them a chance to call shm_toc_estimate_chunk or shm_toc_estimate_keys on
223 : * &pcxt->estimator.
224 : *
225 : * While we're at it, count the number of PlanState nodes in the tree, so
226 : * we know how many Instrumentation structures we need.
227 : */
228 : static bool
229 1674 : ExecParallelEstimate(PlanState *planstate, ExecParallelEstimateContext *e)
230 : {
231 1674 : if (planstate == NULL)
232 0 : return false;
233 :
234 : /* Count this node. */
235 1674 : e->nnodes++;
236 :
237 1674 : switch (nodeTag(planstate))
238 : {
239 644 : case T_SeqScanState:
240 644 : if (planstate->plan->parallel_aware)
241 496 : ExecSeqScanEstimate((SeqScanState *) planstate,
242 : e->pcxt);
243 644 : break;
244 192 : case T_IndexScanState:
245 192 : if (planstate->plan->parallel_aware)
246 8 : ExecIndexScanEstimate((IndexScanState *) planstate,
247 : e->pcxt);
248 192 : break;
249 28 : case T_IndexOnlyScanState:
250 28 : if (planstate->plan->parallel_aware)
251 24 : ExecIndexOnlyScanEstimate((IndexOnlyScanState *) planstate,
252 : e->pcxt);
253 28 : break;
254 0 : case T_ForeignScanState:
255 0 : if (planstate->plan->parallel_aware)
256 0 : ExecForeignScanEstimate((ForeignScanState *) planstate,
257 : e->pcxt);
258 0 : break;
259 100 : case T_AppendState:
260 100 : if (planstate->plan->parallel_aware)
261 68 : ExecAppendEstimate((AppendState *) planstate,
262 : e->pcxt);
263 100 : break;
264 0 : case T_CustomScanState:
265 0 : if (planstate->plan->parallel_aware)
266 0 : ExecCustomScanEstimate((CustomScanState *) planstate,
267 : e->pcxt);
268 0 : break;
269 10 : case T_BitmapHeapScanState:
270 10 : if (planstate->plan->parallel_aware)
271 8 : ExecBitmapHeapEstimate((BitmapHeapScanState *) planstate,
272 : e->pcxt);
273 10 : break;
274 112 : case T_HashJoinState:
275 112 : if (planstate->plan->parallel_aware)
276 64 : ExecHashJoinEstimate((HashJoinState *) planstate,
277 : e->pcxt);
278 112 : break;
279 112 : case T_HashState:
280 : /* even when not parallel-aware, for EXPLAIN ANALYZE */
281 112 : ExecHashEstimate((HashState *) planstate, e->pcxt);
282 112 : break;
283 94 : case T_SortState:
284 : /* even when not parallel-aware, for EXPLAIN ANALYZE */
285 94 : ExecSortEstimate((SortState *) planstate, e->pcxt);
286 94 : break;
287 0 : case T_IncrementalSortState:
288 : /* even when not parallel-aware, for EXPLAIN ANALYZE */
289 0 : ExecIncrementalSortEstimate((IncrementalSortState *) planstate, e->pcxt);
290 0 : break;
291 :
292 382 : default:
293 382 : break;
294 : }
295 :
296 1674 : return planstate_tree_walker(planstate, ExecParallelEstimate, e);
297 : }
298 :
299 : /*
300 : * Estimate the amount of space required to serialize the indicated parameters.
301 : */
302 : static Size
303 8 : EstimateParamExecSpace(EState *estate, Bitmapset *params)
304 : {
305 : int paramid;
306 8 : Size sz = sizeof(int);
307 :
308 8 : paramid = -1;
309 20 : while ((paramid = bms_next_member(params, paramid)) >= 0)
310 : {
311 : Oid typeOid;
312 : int16 typLen;
313 : bool typByVal;
314 : ParamExecData *prm;
315 :
316 12 : prm = &(estate->es_param_exec_vals[paramid]);
317 12 : typeOid = list_nth_oid(estate->es_plannedstmt->paramExecTypes,
318 : paramid);
319 :
320 12 : sz = add_size(sz, sizeof(int)); /* space for paramid */
321 :
322 : /* space for datum/isnull */
323 12 : if (OidIsValid(typeOid))
324 12 : get_typlenbyval(typeOid, &typLen, &typByVal);
325 : else
326 : {
327 : /* If no type OID, assume by-value, like copyParamList does. */
328 0 : typLen = sizeof(Datum);
329 0 : typByVal = true;
330 : }
331 12 : sz = add_size(sz,
332 12 : datumEstimateSpace(prm->value, prm->isnull,
333 : typByVal, typLen));
334 : }
335 8 : return sz;
336 : }
337 :
338 : /*
339 : * Serialize specified PARAM_EXEC parameters.
340 : *
341 : * We write the number of parameters first, as a 4-byte integer, and then
342 : * write details for each parameter in turn. The details for each parameter
343 : * consist of a 4-byte paramid (location of param in execution time internal
344 : * parameter array) and then the datum as serialized by datumSerialize().
345 : */
346 : static dsa_pointer
347 8 : SerializeParamExecParams(EState *estate, Bitmapset *params, dsa_area *area)
348 : {
349 : Size size;
350 : int nparams;
351 : int paramid;
352 : ParamExecData *prm;
353 : dsa_pointer handle;
354 : char *start_address;
355 :
356 : /* Allocate enough space for the current parameter values. */
357 8 : size = EstimateParamExecSpace(estate, params);
358 8 : handle = dsa_allocate(area, size);
359 8 : start_address = dsa_get_address(area, handle);
360 :
361 : /* First write the number of parameters as a 4-byte integer. */
362 8 : nparams = bms_num_members(params);
363 8 : memcpy(start_address, &nparams, sizeof(int));
364 8 : start_address += sizeof(int);
365 :
366 : /* Write details for each parameter in turn. */
367 8 : paramid = -1;
368 20 : while ((paramid = bms_next_member(params, paramid)) >= 0)
369 : {
370 : Oid typeOid;
371 : int16 typLen;
372 : bool typByVal;
373 :
374 12 : prm = &(estate->es_param_exec_vals[paramid]);
375 12 : typeOid = list_nth_oid(estate->es_plannedstmt->paramExecTypes,
376 : paramid);
377 :
378 : /* Write paramid. */
379 12 : memcpy(start_address, ¶mid, sizeof(int));
380 12 : start_address += sizeof(int);
381 :
382 : /* Write datum/isnull */
383 12 : if (OidIsValid(typeOid))
384 12 : get_typlenbyval(typeOid, &typLen, &typByVal);
385 : else
386 : {
387 : /* If no type OID, assume by-value, like copyParamList does. */
388 0 : typLen = sizeof(Datum);
389 0 : typByVal = true;
390 : }
391 12 : datumSerialize(prm->value, prm->isnull, typByVal, typLen,
392 : &start_address);
393 : }
394 :
395 8 : return handle;
396 : }
397 :
398 : /*
399 : * Restore specified PARAM_EXEC parameters.
400 : */
401 : static void
402 24 : RestoreParamExecParams(char *start_address, EState *estate)
403 : {
404 : int nparams;
405 : int i;
406 : int paramid;
407 :
408 24 : memcpy(&nparams, start_address, sizeof(int));
409 24 : start_address += sizeof(int);
410 :
411 56 : for (i = 0; i < nparams; i++)
412 : {
413 : ParamExecData *prm;
414 :
415 : /* Read paramid */
416 32 : memcpy(¶mid, start_address, sizeof(int));
417 32 : start_address += sizeof(int);
418 32 : prm = &(estate->es_param_exec_vals[paramid]);
419 :
420 : /* Read datum/isnull. */
421 32 : prm->value = datumRestore(&start_address, &prm->isnull);
422 32 : prm->execPlan = NULL;
423 : }
424 24 : }
425 :
426 : /*
427 : * Initialize the dynamic shared memory segment that will be used to control
428 : * parallel execution.
429 : */
430 : static bool
431 1674 : ExecParallelInitializeDSM(PlanState *planstate,
432 : ExecParallelInitializeDSMContext *d)
433 : {
434 1674 : if (planstate == NULL)
435 0 : return false;
436 :
437 : /* If instrumentation is enabled, initialize slot for this node. */
438 1674 : if (d->instrumentation != NULL)
439 1280 : d->instrumentation->plan_node_id[d->nnodes] =
440 640 : planstate->plan->plan_node_id;
441 :
442 : /* Count this node. */
443 1674 : d->nnodes++;
444 :
445 : /*
446 : * Call initializers for DSM-using plan nodes.
447 : *
448 : * Most plan nodes won't do anything here, but plan nodes that allocated
449 : * DSM may need to initialize shared state in the DSM before parallel
450 : * workers are launched. They can allocate the space they previously
451 : * estimated using shm_toc_allocate, and add the keys they previously
452 : * estimated using shm_toc_insert, in each case targeting pcxt->toc.
453 : */
454 1674 : switch (nodeTag(planstate))
455 : {
456 644 : case T_SeqScanState:
457 644 : if (planstate->plan->parallel_aware)
458 496 : ExecSeqScanInitializeDSM((SeqScanState *) planstate,
459 : d->pcxt);
460 644 : break;
461 192 : case T_IndexScanState:
462 192 : if (planstate->plan->parallel_aware)
463 8 : ExecIndexScanInitializeDSM((IndexScanState *) planstate,
464 : d->pcxt);
465 192 : break;
466 28 : case T_IndexOnlyScanState:
467 28 : if (planstate->plan->parallel_aware)
468 24 : ExecIndexOnlyScanInitializeDSM((IndexOnlyScanState *) planstate,
469 : d->pcxt);
470 28 : break;
471 0 : case T_ForeignScanState:
472 0 : if (planstate->plan->parallel_aware)
473 0 : ExecForeignScanInitializeDSM((ForeignScanState *) planstate,
474 : d->pcxt);
475 0 : break;
476 100 : case T_AppendState:
477 100 : if (planstate->plan->parallel_aware)
478 68 : ExecAppendInitializeDSM((AppendState *) planstate,
479 : d->pcxt);
480 100 : break;
481 0 : case T_CustomScanState:
482 0 : if (planstate->plan->parallel_aware)
483 0 : ExecCustomScanInitializeDSM((CustomScanState *) planstate,
484 : d->pcxt);
485 0 : break;
486 10 : case T_BitmapHeapScanState:
487 10 : if (planstate->plan->parallel_aware)
488 8 : ExecBitmapHeapInitializeDSM((BitmapHeapScanState *) planstate,
489 : d->pcxt);
490 10 : break;
491 112 : case T_HashJoinState:
492 112 : if (planstate->plan->parallel_aware)
493 64 : ExecHashJoinInitializeDSM((HashJoinState *) planstate,
494 : d->pcxt);
495 112 : break;
496 112 : case T_HashState:
497 : /* even when not parallel-aware, for EXPLAIN ANALYZE */
498 112 : ExecHashInitializeDSM((HashState *) planstate, d->pcxt);
499 112 : break;
500 94 : case T_SortState:
501 : /* even when not parallel-aware, for EXPLAIN ANALYZE */
502 94 : ExecSortInitializeDSM((SortState *) planstate, d->pcxt);
503 94 : break;
504 0 : case T_IncrementalSortState:
505 : /* even when not parallel-aware, for EXPLAIN ANALYZE */
506 0 : ExecIncrementalSortInitializeDSM((IncrementalSortState *) planstate, d->pcxt);
507 0 : break;
508 :
509 382 : default:
510 382 : break;
511 : }
512 :
513 1674 : return planstate_tree_walker(planstate, ExecParallelInitializeDSM, d);
514 : }
515 :
516 : /*
517 : * It sets up the response queues for backend workers to return tuples
518 : * to the main backend and start the workers.
519 : */
520 : static shm_mq_handle **
521 542 : ExecParallelSetupTupleQueues(ParallelContext *pcxt, bool reinitialize)
522 : {
523 : shm_mq_handle **responseq;
524 : char *tqueuespace;
525 : int i;
526 :
527 : /* Skip this if no workers. */
528 542 : if (pcxt->nworkers == 0)
529 0 : return NULL;
530 :
531 : /* Allocate memory for shared memory queue handles. */
532 : responseq = (shm_mq_handle **)
533 542 : palloc(pcxt->nworkers * sizeof(shm_mq_handle *));
534 :
535 : /*
536 : * If not reinitializing, allocate space from the DSM for the queues;
537 : * otherwise, find the already allocated space.
538 : */
539 542 : if (!reinitialize)
540 : tqueuespace =
541 370 : shm_toc_allocate(pcxt->toc,
542 : mul_size(PARALLEL_TUPLE_QUEUE_SIZE,
543 370 : pcxt->nworkers));
544 : else
545 172 : tqueuespace = shm_toc_lookup(pcxt->toc, PARALLEL_KEY_TUPLE_QUEUE, false);
546 :
547 : /* Create the queues, and become the receiver for each. */
548 2052 : for (i = 0; i < pcxt->nworkers; ++i)
549 : {
550 : shm_mq *mq;
551 :
552 1510 : mq = shm_mq_create(tqueuespace +
553 1510 : ((Size) i) * PARALLEL_TUPLE_QUEUE_SIZE,
554 : (Size) PARALLEL_TUPLE_QUEUE_SIZE);
555 :
556 1510 : shm_mq_set_receiver(mq, MyProc);
557 1510 : responseq[i] = shm_mq_attach(mq, pcxt->seg, NULL);
558 : }
559 :
560 : /* Add array of queues to shm_toc, so others can find it. */
561 542 : if (!reinitialize)
562 370 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_TUPLE_QUEUE, tqueuespace);
563 :
564 : /* Return array of handles. */
565 542 : return responseq;
566 : }
567 :
568 : /*
569 : * Sets up the required infrastructure for backend workers to perform
570 : * execution and return results to the main backend.
571 : */
572 : ParallelExecutorInfo *
573 370 : ExecInitParallelPlan(PlanState *planstate, EState *estate,
574 : Bitmapset *sendParams, int nworkers,
575 : int64 tuples_needed)
576 : {
577 : ParallelExecutorInfo *pei;
578 : ParallelContext *pcxt;
579 : ExecParallelEstimateContext e;
580 : ExecParallelInitializeDSMContext d;
581 : FixedParallelExecutorState *fpes;
582 : char *pstmt_data;
583 : char *pstmt_space;
584 : char *paramlistinfo_space;
585 : BufferUsage *bufusage_space;
586 : WalUsage *walusage_space;
587 370 : SharedExecutorInstrumentation *instrumentation = NULL;
588 370 : SharedJitInstrumentation *jit_instrumentation = NULL;
589 : int pstmt_len;
590 : int paramlistinfo_len;
591 370 : int instrumentation_len = 0;
592 370 : int jit_instrumentation_len = 0;
593 370 : int instrument_offset = 0;
594 370 : Size dsa_minsize = dsa_minimum_size();
595 : char *query_string;
596 : int query_len;
597 :
598 : /*
599 : * Force any initplan outputs that we're going to pass to workers to be
600 : * evaluated, if they weren't already.
601 : *
602 : * For simplicity, we use the EState's per-output-tuple ExprContext here.
603 : * That risks intra-query memory leakage, since we might pass through here
604 : * many times before that ExprContext gets reset; but ExecSetParamPlan
605 : * doesn't normally leak any memory in the context (see its comments), so
606 : * it doesn't seem worth complicating this function's API to pass it a
607 : * shorter-lived ExprContext. This might need to change someday.
608 : */
609 370 : ExecSetParamPlanMulti(sendParams, GetPerTupleExprContext(estate));
610 :
611 : /* Allocate object for return value. */
612 370 : pei = palloc0(sizeof(ParallelExecutorInfo));
613 370 : pei->finished = false;
614 370 : pei->planstate = planstate;
615 :
616 : /* Fix up and serialize plan to be sent to workers. */
617 370 : pstmt_data = ExecSerializePlan(planstate->plan, estate);
618 :
619 : /* Create a parallel context. */
620 370 : pcxt = CreateParallelContext("postgres", "ParallelQueryMain", nworkers);
621 370 : pei->pcxt = pcxt;
622 :
623 : /*
624 : * Before telling the parallel context to create a dynamic shared memory
625 : * segment, we need to figure out how big it should be. Estimate space
626 : * for the various things we need to store.
627 : */
628 :
629 : /* Estimate space for fixed-size state. */
630 370 : shm_toc_estimate_chunk(&pcxt->estimator,
631 : sizeof(FixedParallelExecutorState));
632 370 : shm_toc_estimate_keys(&pcxt->estimator, 1);
633 :
634 : /* Estimate space for query text. */
635 370 : query_len = strlen(estate->es_sourceText);
636 370 : shm_toc_estimate_chunk(&pcxt->estimator, query_len + 1);
637 370 : shm_toc_estimate_keys(&pcxt->estimator, 1);
638 :
639 : /* Estimate space for serialized PlannedStmt. */
640 370 : pstmt_len = strlen(pstmt_data) + 1;
641 370 : shm_toc_estimate_chunk(&pcxt->estimator, pstmt_len);
642 370 : shm_toc_estimate_keys(&pcxt->estimator, 1);
643 :
644 : /* Estimate space for serialized ParamListInfo. */
645 370 : paramlistinfo_len = EstimateParamListSpace(estate->es_param_list_info);
646 370 : shm_toc_estimate_chunk(&pcxt->estimator, paramlistinfo_len);
647 370 : shm_toc_estimate_keys(&pcxt->estimator, 1);
648 :
649 : /*
650 : * Estimate space for BufferUsage.
651 : *
652 : * If EXPLAIN is not in use and there are no extensions loaded that care,
653 : * we could skip this. But we have no way of knowing whether anyone's
654 : * looking at pgBufferUsage, so do it unconditionally.
655 : */
656 370 : shm_toc_estimate_chunk(&pcxt->estimator,
657 : mul_size(sizeof(BufferUsage), pcxt->nworkers));
658 370 : shm_toc_estimate_keys(&pcxt->estimator, 1);
659 :
660 : /*
661 : * Same thing for WalUsage.
662 : */
663 370 : shm_toc_estimate_chunk(&pcxt->estimator,
664 : mul_size(sizeof(WalUsage), pcxt->nworkers));
665 370 : shm_toc_estimate_keys(&pcxt->estimator, 1);
666 :
667 : /* Estimate space for tuple queues. */
668 370 : shm_toc_estimate_chunk(&pcxt->estimator,
669 : mul_size(PARALLEL_TUPLE_QUEUE_SIZE, pcxt->nworkers));
670 370 : shm_toc_estimate_keys(&pcxt->estimator, 1);
671 :
672 : /*
673 : * Give parallel-aware nodes a chance to add to the estimates, and get a
674 : * count of how many PlanState nodes there are.
675 : */
676 370 : e.pcxt = pcxt;
677 370 : e.nnodes = 0;
678 370 : ExecParallelEstimate(planstate, &e);
679 :
680 : /* Estimate space for instrumentation, if required. */
681 370 : if (estate->es_instrument)
682 : {
683 112 : instrumentation_len =
684 : offsetof(SharedExecutorInstrumentation, plan_node_id) +
685 112 : sizeof(int) * e.nnodes;
686 112 : instrumentation_len = MAXALIGN(instrumentation_len);
687 112 : instrument_offset = instrumentation_len;
688 112 : instrumentation_len +=
689 112 : mul_size(sizeof(Instrumentation),
690 112 : mul_size(e.nnodes, nworkers));
691 112 : shm_toc_estimate_chunk(&pcxt->estimator, instrumentation_len);
692 112 : shm_toc_estimate_keys(&pcxt->estimator, 1);
693 :
694 : /* Estimate space for JIT instrumentation, if required. */
695 112 : if (estate->es_jit_flags != PGJIT_NONE)
696 : {
697 16 : jit_instrumentation_len =
698 16 : offsetof(SharedJitInstrumentation, jit_instr) +
699 : sizeof(JitInstrumentation) * nworkers;
700 16 : shm_toc_estimate_chunk(&pcxt->estimator, jit_instrumentation_len);
701 16 : shm_toc_estimate_keys(&pcxt->estimator, 1);
702 : }
703 : }
704 :
705 : /* Estimate space for DSA area. */
706 370 : shm_toc_estimate_chunk(&pcxt->estimator, dsa_minsize);
707 370 : shm_toc_estimate_keys(&pcxt->estimator, 1);
708 :
709 : /* Everyone's had a chance to ask for space, so now create the DSM. */
710 370 : InitializeParallelDSM(pcxt);
711 :
712 : /*
713 : * OK, now we have a dynamic shared memory segment, and it should be big
714 : * enough to store all of the data we estimated we would want to put into
715 : * it, plus whatever general stuff (not specifically executor-related) the
716 : * ParallelContext itself needs to store there. None of the space we
717 : * asked for has been allocated or initialized yet, though, so do that.
718 : */
719 :
720 : /* Store fixed-size state. */
721 370 : fpes = shm_toc_allocate(pcxt->toc, sizeof(FixedParallelExecutorState));
722 370 : fpes->tuples_needed = tuples_needed;
723 370 : fpes->param_exec = InvalidDsaPointer;
724 370 : fpes->eflags = estate->es_top_eflags;
725 370 : fpes->jit_flags = estate->es_jit_flags;
726 370 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_EXECUTOR_FIXED, fpes);
727 :
728 : /* Store query string */
729 370 : query_string = shm_toc_allocate(pcxt->toc, query_len + 1);
730 370 : memcpy(query_string, estate->es_sourceText, query_len + 1);
731 370 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_QUERY_TEXT, query_string);
732 :
733 : /* Store serialized PlannedStmt. */
734 370 : pstmt_space = shm_toc_allocate(pcxt->toc, pstmt_len);
735 370 : memcpy(pstmt_space, pstmt_data, pstmt_len);
736 370 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_PLANNEDSTMT, pstmt_space);
737 :
738 : /* Store serialized ParamListInfo. */
739 370 : paramlistinfo_space = shm_toc_allocate(pcxt->toc, paramlistinfo_len);
740 370 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_PARAMLISTINFO, paramlistinfo_space);
741 370 : SerializeParamList(estate->es_param_list_info, ¶mlistinfo_space);
742 :
743 : /* Allocate space for each worker's BufferUsage; no need to initialize. */
744 370 : bufusage_space = shm_toc_allocate(pcxt->toc,
745 370 : mul_size(sizeof(BufferUsage), pcxt->nworkers));
746 370 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_BUFFER_USAGE, bufusage_space);
747 370 : pei->buffer_usage = bufusage_space;
748 :
749 : /* Same for WalUsage. */
750 370 : walusage_space = shm_toc_allocate(pcxt->toc,
751 370 : mul_size(sizeof(WalUsage), pcxt->nworkers));
752 370 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_WAL_USAGE, walusage_space);
753 370 : pei->wal_usage = walusage_space;
754 :
755 : /* Set up the tuple queues that the workers will write into. */
756 370 : pei->tqueue = ExecParallelSetupTupleQueues(pcxt, false);
757 :
758 : /* We don't need the TupleQueueReaders yet, though. */
759 370 : pei->reader = NULL;
760 :
761 : /*
762 : * If instrumentation options were supplied, allocate space for the data.
763 : * It only gets partially initialized here; the rest happens during
764 : * ExecParallelInitializeDSM.
765 : */
766 370 : if (estate->es_instrument)
767 : {
768 : Instrumentation *instrument;
769 : int i;
770 :
771 112 : instrumentation = shm_toc_allocate(pcxt->toc, instrumentation_len);
772 112 : instrumentation->instrument_options = estate->es_instrument;
773 112 : instrumentation->instrument_offset = instrument_offset;
774 112 : instrumentation->num_workers = nworkers;
775 112 : instrumentation->num_plan_nodes = e.nnodes;
776 112 : instrument = GetInstrumentationArray(instrumentation);
777 1140 : for (i = 0; i < nworkers * e.nnodes; ++i)
778 1028 : InstrInit(&instrument[i], estate->es_instrument);
779 112 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_INSTRUMENTATION,
780 : instrumentation);
781 112 : pei->instrumentation = instrumentation;
782 :
783 112 : if (estate->es_jit_flags != PGJIT_NONE)
784 : {
785 16 : jit_instrumentation = shm_toc_allocate(pcxt->toc,
786 : jit_instrumentation_len);
787 16 : jit_instrumentation->num_workers = nworkers;
788 16 : memset(jit_instrumentation->jit_instr, 0,
789 : sizeof(JitInstrumentation) * nworkers);
790 16 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_JIT_INSTRUMENTATION,
791 : jit_instrumentation);
792 16 : pei->jit_instrumentation = jit_instrumentation;
793 : }
794 : }
795 :
796 : /*
797 : * Create a DSA area that can be used by the leader and all workers.
798 : * (However, if we failed to create a DSM and are using private memory
799 : * instead, then skip this.)
800 : */
801 370 : if (pcxt->seg != NULL)
802 : {
803 : char *area_space;
804 :
805 370 : area_space = shm_toc_allocate(pcxt->toc, dsa_minsize);
806 370 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_DSA, area_space);
807 370 : pei->area = dsa_create_in_place(area_space, dsa_minsize,
808 : LWTRANCHE_PARALLEL_QUERY_DSA,
809 : pcxt->seg);
810 :
811 : /*
812 : * Serialize parameters, if any, using DSA storage. We don't dare use
813 : * the main parallel query DSM for this because we might relaunch
814 : * workers after the values have changed (and thus the amount of
815 : * storage required has changed).
816 : */
817 370 : if (!bms_is_empty(sendParams))
818 : {
819 8 : pei->param_exec = SerializeParamExecParams(estate, sendParams,
820 : pei->area);
821 8 : fpes->param_exec = pei->param_exec;
822 : }
823 : }
824 :
825 : /*
826 : * Give parallel-aware nodes a chance to initialize their shared data.
827 : * This also initializes the elements of instrumentation->ps_instrument,
828 : * if it exists.
829 : */
830 370 : d.pcxt = pcxt;
831 370 : d.instrumentation = instrumentation;
832 370 : d.nnodes = 0;
833 :
834 : /* Install our DSA area while initializing the plan. */
835 370 : estate->es_query_dsa = pei->area;
836 370 : ExecParallelInitializeDSM(planstate, &d);
837 370 : estate->es_query_dsa = NULL;
838 :
839 : /*
840 : * Make sure that the world hasn't shifted under our feet. This could
841 : * probably just be an Assert(), but let's be conservative for now.
842 : */
843 370 : if (e.nnodes != d.nnodes)
844 0 : elog(ERROR, "inconsistent count of PlanState nodes");
845 :
846 : /* OK, we're ready to rock and roll. */
847 370 : return pei;
848 : }
849 :
850 : /*
851 : * Set up tuple queue readers to read the results of a parallel subplan.
852 : *
853 : * This is separate from ExecInitParallelPlan() because we can launch the
854 : * worker processes and let them start doing something before we do this.
855 : */
856 : void
857 530 : ExecParallelCreateReaders(ParallelExecutorInfo *pei)
858 : {
859 530 : int nworkers = pei->pcxt->nworkers_launched;
860 : int i;
861 :
862 : Assert(pei->reader == NULL);
863 :
864 530 : if (nworkers > 0)
865 : {
866 530 : pei->reader = (TupleQueueReader **)
867 530 : palloc(nworkers * sizeof(TupleQueueReader *));
868 :
869 1990 : for (i = 0; i < nworkers; i++)
870 : {
871 1460 : shm_mq_set_handle(pei->tqueue[i],
872 1460 : pei->pcxt->worker[i].bgwhandle);
873 1460 : pei->reader[i] = CreateTupleQueueReader(pei->tqueue[i]);
874 : }
875 : }
876 530 : }
877 :
878 : /*
879 : * Re-initialize the parallel executor shared memory state before launching
880 : * a fresh batch of workers.
881 : */
882 : void
883 172 : ExecParallelReinitialize(PlanState *planstate,
884 : ParallelExecutorInfo *pei,
885 : Bitmapset *sendParams)
886 : {
887 172 : EState *estate = planstate->state;
888 : FixedParallelExecutorState *fpes;
889 :
890 : /* Old workers must already be shut down */
891 : Assert(pei->finished);
892 :
893 : /*
894 : * Force any initplan outputs that we're going to pass to workers to be
895 : * evaluated, if they weren't already (see comments in
896 : * ExecInitParallelPlan).
897 : */
898 172 : ExecSetParamPlanMulti(sendParams, GetPerTupleExprContext(estate));
899 :
900 172 : ReinitializeParallelDSM(pei->pcxt);
901 172 : pei->tqueue = ExecParallelSetupTupleQueues(pei->pcxt, true);
902 172 : pei->reader = NULL;
903 172 : pei->finished = false;
904 :
905 172 : fpes = shm_toc_lookup(pei->pcxt->toc, PARALLEL_KEY_EXECUTOR_FIXED, false);
906 :
907 : /* Free any serialized parameters from the last round. */
908 172 : if (DsaPointerIsValid(fpes->param_exec))
909 : {
910 0 : dsa_free(pei->area, fpes->param_exec);
911 0 : fpes->param_exec = InvalidDsaPointer;
912 : }
913 :
914 : /* Serialize current parameter values if required. */
915 172 : if (!bms_is_empty(sendParams))
916 : {
917 0 : pei->param_exec = SerializeParamExecParams(estate, sendParams,
918 : pei->area);
919 0 : fpes->param_exec = pei->param_exec;
920 : }
921 :
922 : /* Traverse plan tree and let each child node reset associated state. */
923 172 : estate->es_query_dsa = pei->area;
924 172 : ExecParallelReInitializeDSM(planstate, pei->pcxt);
925 172 : estate->es_query_dsa = NULL;
926 172 : }
927 :
928 : /*
929 : * Traverse plan tree to reinitialize per-node dynamic shared memory state
930 : */
931 : static bool
932 444 : ExecParallelReInitializeDSM(PlanState *planstate,
933 : ParallelContext *pcxt)
934 : {
935 444 : if (planstate == NULL)
936 0 : return false;
937 :
938 : /*
939 : * Call reinitializers for DSM-using plan nodes.
940 : */
941 444 : switch (nodeTag(planstate))
942 : {
943 184 : case T_SeqScanState:
944 184 : if (planstate->plan->parallel_aware)
945 152 : ExecSeqScanReInitializeDSM((SeqScanState *) planstate,
946 : pcxt);
947 184 : break;
948 8 : case T_IndexScanState:
949 8 : if (planstate->plan->parallel_aware)
950 8 : ExecIndexScanReInitializeDSM((IndexScanState *) planstate,
951 : pcxt);
952 8 : break;
953 8 : case T_IndexOnlyScanState:
954 8 : if (planstate->plan->parallel_aware)
955 8 : ExecIndexOnlyScanReInitializeDSM((IndexOnlyScanState *) planstate,
956 : pcxt);
957 8 : break;
958 0 : case T_ForeignScanState:
959 0 : if (planstate->plan->parallel_aware)
960 0 : ExecForeignScanReInitializeDSM((ForeignScanState *) planstate,
961 : pcxt);
962 0 : break;
963 0 : case T_AppendState:
964 0 : if (planstate->plan->parallel_aware)
965 0 : ExecAppendReInitializeDSM((AppendState *) planstate, pcxt);
966 0 : break;
967 0 : case T_CustomScanState:
968 0 : if (planstate->plan->parallel_aware)
969 0 : ExecCustomScanReInitializeDSM((CustomScanState *) planstate,
970 : pcxt);
971 0 : break;
972 36 : case T_BitmapHeapScanState:
973 36 : if (planstate->plan->parallel_aware)
974 36 : ExecBitmapHeapReInitializeDSM((BitmapHeapScanState *) planstate,
975 : pcxt);
976 36 : break;
977 64 : case T_HashJoinState:
978 64 : if (planstate->plan->parallel_aware)
979 32 : ExecHashJoinReInitializeDSM((HashJoinState *) planstate,
980 : pcxt);
981 64 : break;
982 84 : case T_HashState:
983 : case T_SortState:
984 : case T_IncrementalSortState:
985 : /* these nodes have DSM state, but no reinitialization is required */
986 84 : break;
987 :
988 60 : default:
989 60 : break;
990 : }
991 :
992 444 : return planstate_tree_walker(planstate, ExecParallelReInitializeDSM, pcxt);
993 : }
994 :
995 : /*
996 : * Copy instrumentation information about this node and its descendants from
997 : * dynamic shared memory.
998 : */
999 : static bool
1000 640 : ExecParallelRetrieveInstrumentation(PlanState *planstate,
1001 : SharedExecutorInstrumentation *instrumentation)
1002 : {
1003 : Instrumentation *instrument;
1004 : int i;
1005 : int n;
1006 : int ibytes;
1007 640 : int plan_node_id = planstate->plan->plan_node_id;
1008 : MemoryContext oldcontext;
1009 :
1010 : /* Find the instrumentation for this node. */
1011 2908 : for (i = 0; i < instrumentation->num_plan_nodes; ++i)
1012 2908 : if (instrumentation->plan_node_id[i] == plan_node_id)
1013 640 : break;
1014 640 : if (i >= instrumentation->num_plan_nodes)
1015 0 : elog(ERROR, "plan node %d not found", plan_node_id);
1016 :
1017 : /* Accumulate the statistics from all workers. */
1018 640 : instrument = GetInstrumentationArray(instrumentation);
1019 640 : instrument += i * instrumentation->num_workers;
1020 1668 : for (n = 0; n < instrumentation->num_workers; ++n)
1021 1028 : InstrAggNode(planstate->instrument, &instrument[n]);
1022 :
1023 : /*
1024 : * Also store the per-worker detail.
1025 : *
1026 : * Worker instrumentation should be allocated in the same context as the
1027 : * regular instrumentation information, which is the per-query context.
1028 : * Switch into per-query memory context.
1029 : */
1030 640 : oldcontext = MemoryContextSwitchTo(planstate->state->es_query_cxt);
1031 640 : ibytes = mul_size(instrumentation->num_workers, sizeof(Instrumentation));
1032 640 : planstate->worker_instrument =
1033 640 : palloc(ibytes + offsetof(WorkerInstrumentation, instrument));
1034 640 : MemoryContextSwitchTo(oldcontext);
1035 :
1036 640 : planstate->worker_instrument->num_workers = instrumentation->num_workers;
1037 640 : memcpy(&planstate->worker_instrument->instrument, instrument, ibytes);
1038 :
1039 : /* Perform any node-type-specific work that needs to be done. */
1040 640 : switch (nodeTag(planstate))
1041 : {
1042 8 : case T_SortState:
1043 8 : ExecSortRetrieveInstrumentation((SortState *) planstate);
1044 8 : break;
1045 0 : case T_IncrementalSortState:
1046 0 : ExecIncrementalSortRetrieveInstrumentation((IncrementalSortState *) planstate);
1047 0 : break;
1048 56 : case T_HashState:
1049 56 : ExecHashRetrieveInstrumentation((HashState *) planstate);
1050 56 : break;
1051 576 : default:
1052 576 : break;
1053 : }
1054 :
1055 640 : return planstate_tree_walker(planstate, ExecParallelRetrieveInstrumentation,
1056 : instrumentation);
1057 : }
1058 :
1059 : /*
1060 : * Add up the workers' JIT instrumentation from dynamic shared memory.
1061 : */
1062 : static void
1063 16 : ExecParallelRetrieveJitInstrumentation(PlanState *planstate,
1064 : SharedJitInstrumentation *shared_jit)
1065 : {
1066 : JitInstrumentation *combined;
1067 : int ibytes;
1068 :
1069 : int n;
1070 :
1071 : /*
1072 : * Accumulate worker JIT instrumentation into the combined JIT
1073 : * instrumentation, allocating it if required.
1074 : */
1075 16 : if (!planstate->state->es_jit_worker_instr)
1076 16 : planstate->state->es_jit_worker_instr =
1077 16 : MemoryContextAllocZero(planstate->state->es_query_cxt, sizeof(JitInstrumentation));
1078 16 : combined = planstate->state->es_jit_worker_instr;
1079 :
1080 : /* Accumulate all the workers' instrumentations. */
1081 48 : for (n = 0; n < shared_jit->num_workers; ++n)
1082 32 : InstrJitAgg(combined, &shared_jit->jit_instr[n]);
1083 :
1084 : /*
1085 : * Store the per-worker detail.
1086 : *
1087 : * Similar to ExecParallelRetrieveInstrumentation(), allocate the
1088 : * instrumentation in per-query context.
1089 : */
1090 16 : ibytes = offsetof(SharedJitInstrumentation, jit_instr)
1091 16 : + mul_size(shared_jit->num_workers, sizeof(JitInstrumentation));
1092 16 : planstate->worker_jit_instrument =
1093 16 : MemoryContextAlloc(planstate->state->es_query_cxt, ibytes);
1094 :
1095 16 : memcpy(planstate->worker_jit_instrument, shared_jit, ibytes);
1096 16 : }
1097 :
1098 : /*
1099 : * Finish parallel execution. We wait for parallel workers to finish, and
1100 : * accumulate their buffer/WAL usage.
1101 : */
1102 : void
1103 976 : ExecParallelFinish(ParallelExecutorInfo *pei)
1104 : {
1105 976 : int nworkers = pei->pcxt->nworkers_launched;
1106 : int i;
1107 :
1108 : /* Make this be a no-op if called twice in a row. */
1109 976 : if (pei->finished)
1110 438 : return;
1111 :
1112 : /*
1113 : * Detach from tuple queues ASAP, so that any still-active workers will
1114 : * notice that no further results are wanted.
1115 : */
1116 538 : if (pei->tqueue != NULL)
1117 : {
1118 1994 : for (i = 0; i < nworkers; i++)
1119 1456 : shm_mq_detach(pei->tqueue[i]);
1120 538 : pfree(pei->tqueue);
1121 538 : pei->tqueue = NULL;
1122 : }
1123 :
1124 : /*
1125 : * While we're waiting for the workers to finish, let's get rid of the
1126 : * tuple queue readers. (Any other local cleanup could be done here too.)
1127 : */
1128 538 : if (pei->reader != NULL)
1129 : {
1130 1982 : for (i = 0; i < nworkers; i++)
1131 1456 : DestroyTupleQueueReader(pei->reader[i]);
1132 526 : pfree(pei->reader);
1133 526 : pei->reader = NULL;
1134 : }
1135 :
1136 : /* Now wait for the workers to finish. */
1137 538 : WaitForParallelWorkersToFinish(pei->pcxt);
1138 :
1139 : /*
1140 : * Next, accumulate buffer/WAL usage. (This must wait for the workers to
1141 : * finish, or we might get incomplete data.)
1142 : */
1143 1994 : for (i = 0; i < nworkers; i++)
1144 1456 : InstrAccumParallelQuery(&pei->buffer_usage[i], &pei->wal_usage[i]);
1145 :
1146 538 : pei->finished = true;
1147 : }
1148 :
1149 : /*
1150 : * Accumulate instrumentation, and then clean up whatever ParallelExecutorInfo
1151 : * resources still exist after ExecParallelFinish. We separate these
1152 : * routines because someone might want to examine the contents of the DSM
1153 : * after ExecParallelFinish and before calling this routine.
1154 : */
1155 : void
1156 366 : ExecParallelCleanup(ParallelExecutorInfo *pei)
1157 : {
1158 : /* Accumulate instrumentation, if any. */
1159 366 : if (pei->instrumentation)
1160 112 : ExecParallelRetrieveInstrumentation(pei->planstate,
1161 : pei->instrumentation);
1162 :
1163 : /* Accumulate JIT instrumentation, if any. */
1164 366 : if (pei->jit_instrumentation)
1165 16 : ExecParallelRetrieveJitInstrumentation(pei->planstate,
1166 16 : pei->jit_instrumentation);
1167 :
1168 : /* Free any serialized parameters. */
1169 366 : if (DsaPointerIsValid(pei->param_exec))
1170 : {
1171 8 : dsa_free(pei->area, pei->param_exec);
1172 8 : pei->param_exec = InvalidDsaPointer;
1173 : }
1174 366 : if (pei->area != NULL)
1175 : {
1176 366 : dsa_detach(pei->area);
1177 366 : pei->area = NULL;
1178 : }
1179 366 : if (pei->pcxt != NULL)
1180 : {
1181 366 : DestroyParallelContext(pei->pcxt);
1182 366 : pei->pcxt = NULL;
1183 : }
1184 366 : pfree(pei);
1185 366 : }
1186 :
1187 : /*
1188 : * Create a DestReceiver to write tuples we produce to the shm_mq designated
1189 : * for that purpose.
1190 : */
1191 : static DestReceiver *
1192 1460 : ExecParallelGetReceiver(dsm_segment *seg, shm_toc *toc)
1193 : {
1194 : char *mqspace;
1195 : shm_mq *mq;
1196 :
1197 1460 : mqspace = shm_toc_lookup(toc, PARALLEL_KEY_TUPLE_QUEUE, false);
1198 1460 : mqspace += ParallelWorkerNumber * PARALLEL_TUPLE_QUEUE_SIZE;
1199 1460 : mq = (shm_mq *) mqspace;
1200 1460 : shm_mq_set_sender(mq, MyProc);
1201 1460 : return CreateTupleQueueDestReceiver(shm_mq_attach(mq, seg, NULL));
1202 : }
1203 :
1204 : /*
1205 : * Create a QueryDesc for the PlannedStmt we are to execute, and return it.
1206 : */
1207 : static QueryDesc *
1208 1460 : ExecParallelGetQueryDesc(shm_toc *toc, DestReceiver *receiver,
1209 : int instrument_options)
1210 : {
1211 : char *pstmtspace;
1212 : char *paramspace;
1213 : PlannedStmt *pstmt;
1214 : ParamListInfo paramLI;
1215 : char *queryString;
1216 :
1217 : /* Get the query string from shared memory */
1218 1460 : queryString = shm_toc_lookup(toc, PARALLEL_KEY_QUERY_TEXT, false);
1219 :
1220 : /* Reconstruct leader-supplied PlannedStmt. */
1221 1460 : pstmtspace = shm_toc_lookup(toc, PARALLEL_KEY_PLANNEDSTMT, false);
1222 1460 : pstmt = (PlannedStmt *) stringToNode(pstmtspace);
1223 :
1224 : /* Reconstruct ParamListInfo. */
1225 1460 : paramspace = shm_toc_lookup(toc, PARALLEL_KEY_PARAMLISTINFO, false);
1226 1460 : paramLI = RestoreParamList(¶mspace);
1227 :
1228 : /* Create a QueryDesc for the query. */
1229 1460 : return CreateQueryDesc(pstmt,
1230 : queryString,
1231 : GetActiveSnapshot(), InvalidSnapshot,
1232 : receiver, paramLI, NULL, instrument_options);
1233 : }
1234 :
1235 : /*
1236 : * Copy instrumentation information from this node and its descendants into
1237 : * dynamic shared memory, so that the parallel leader can retrieve it.
1238 : */
1239 : static bool
1240 1492 : ExecParallelReportInstrumentation(PlanState *planstate,
1241 : SharedExecutorInstrumentation *instrumentation)
1242 : {
1243 : int i;
1244 1492 : int plan_node_id = planstate->plan->plan_node_id;
1245 : Instrumentation *instrument;
1246 :
1247 1492 : InstrEndLoop(planstate->instrument);
1248 :
1249 : /*
1250 : * If we shuffled the plan_node_id values in ps_instrument into sorted
1251 : * order, we could use binary search here. This might matter someday if
1252 : * we're pushing down sufficiently large plan trees. For now, do it the
1253 : * slow, dumb way.
1254 : */
1255 4820 : for (i = 0; i < instrumentation->num_plan_nodes; ++i)
1256 4820 : if (instrumentation->plan_node_id[i] == plan_node_id)
1257 1492 : break;
1258 1492 : if (i >= instrumentation->num_plan_nodes)
1259 0 : elog(ERROR, "plan node %d not found", plan_node_id);
1260 :
1261 : /*
1262 : * Add our statistics to the per-node, per-worker totals. It's possible
1263 : * that this could happen more than once if we relaunched workers.
1264 : */
1265 1492 : instrument = GetInstrumentationArray(instrumentation);
1266 1492 : instrument += i * instrumentation->num_workers;
1267 : Assert(IsParallelWorker());
1268 : Assert(ParallelWorkerNumber < instrumentation->num_workers);
1269 1492 : InstrAggNode(&instrument[ParallelWorkerNumber], planstate->instrument);
1270 :
1271 1492 : return planstate_tree_walker(planstate, ExecParallelReportInstrumentation,
1272 : instrumentation);
1273 : }
1274 :
1275 : /*
1276 : * Initialize the PlanState and its descendants with the information
1277 : * retrieved from shared memory. This has to be done once the PlanState
1278 : * is allocated and initialized by executor; that is, after ExecutorStart().
1279 : */
1280 : static bool
1281 4758 : ExecParallelInitializeWorker(PlanState *planstate, ParallelWorkerContext *pwcxt)
1282 : {
1283 4758 : if (planstate == NULL)
1284 0 : return false;
1285 :
1286 4758 : switch (nodeTag(planstate))
1287 : {
1288 1894 : case T_SeqScanState:
1289 1894 : if (planstate->plan->parallel_aware)
1290 1498 : ExecSeqScanInitializeWorker((SeqScanState *) planstate, pwcxt);
1291 1894 : break;
1292 248 : case T_IndexScanState:
1293 248 : if (planstate->plan->parallel_aware)
1294 64 : ExecIndexScanInitializeWorker((IndexScanState *) planstate,
1295 : pwcxt);
1296 248 : break;
1297 144 : case T_IndexOnlyScanState:
1298 144 : if (planstate->plan->parallel_aware)
1299 128 : ExecIndexOnlyScanInitializeWorker((IndexOnlyScanState *) planstate,
1300 : pwcxt);
1301 144 : break;
1302 0 : case T_ForeignScanState:
1303 0 : if (planstate->plan->parallel_aware)
1304 0 : ExecForeignScanInitializeWorker((ForeignScanState *) planstate,
1305 : pwcxt);
1306 0 : break;
1307 188 : case T_AppendState:
1308 188 : if (planstate->plan->parallel_aware)
1309 148 : ExecAppendInitializeWorker((AppendState *) planstate, pwcxt);
1310 188 : break;
1311 0 : case T_CustomScanState:
1312 0 : if (planstate->plan->parallel_aware)
1313 0 : ExecCustomScanInitializeWorker((CustomScanState *) planstate,
1314 : pwcxt);
1315 0 : break;
1316 178 : case T_BitmapHeapScanState:
1317 178 : if (planstate->plan->parallel_aware)
1318 176 : ExecBitmapHeapInitializeWorker((BitmapHeapScanState *) planstate,
1319 : pwcxt);
1320 178 : break;
1321 340 : case T_HashJoinState:
1322 340 : if (planstate->plan->parallel_aware)
1323 180 : ExecHashJoinInitializeWorker((HashJoinState *) planstate,
1324 : pwcxt);
1325 340 : break;
1326 340 : case T_HashState:
1327 : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1328 340 : ExecHashInitializeWorker((HashState *) planstate, pwcxt);
1329 340 : break;
1330 286 : case T_SortState:
1331 : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1332 286 : ExecSortInitializeWorker((SortState *) planstate, pwcxt);
1333 286 : break;
1334 0 : case T_IncrementalSortState:
1335 : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1336 0 : ExecIncrementalSortInitializeWorker((IncrementalSortState *) planstate,
1337 : pwcxt);
1338 0 : break;
1339 :
1340 1140 : default:
1341 1140 : break;
1342 : }
1343 :
1344 4758 : return planstate_tree_walker(planstate, ExecParallelInitializeWorker,
1345 : pwcxt);
1346 : }
1347 :
1348 : /*
1349 : * Main entrypoint for parallel query worker processes.
1350 : *
1351 : * We reach this function from ParallelWorkerMain, so the setup necessary to
1352 : * create a sensible parallel environment has already been done;
1353 : * ParallelWorkerMain worries about stuff like the transaction state, combo
1354 : * CID mappings, and GUC values, so we don't need to deal with any of that
1355 : * here.
1356 : *
1357 : * Our job is to deal with concerns specific to the executor. The parallel
1358 : * group leader will have stored a serialized PlannedStmt, and it's our job
1359 : * to execute that plan and write the resulting tuples to the appropriate
1360 : * tuple queue. Various bits of supporting information that we need in order
1361 : * to do this are also stored in the dsm_segment and can be accessed through
1362 : * the shm_toc.
1363 : */
1364 : void
1365 1460 : ParallelQueryMain(dsm_segment *seg, shm_toc *toc)
1366 : {
1367 : FixedParallelExecutorState *fpes;
1368 : BufferUsage *buffer_usage;
1369 : WalUsage *wal_usage;
1370 : DestReceiver *receiver;
1371 : QueryDesc *queryDesc;
1372 : SharedExecutorInstrumentation *instrumentation;
1373 : SharedJitInstrumentation *jit_instrumentation;
1374 1460 : int instrument_options = 0;
1375 : void *area_space;
1376 : dsa_area *area;
1377 : ParallelWorkerContext pwcxt;
1378 :
1379 : /* Get fixed-size state. */
1380 1460 : fpes = shm_toc_lookup(toc, PARALLEL_KEY_EXECUTOR_FIXED, false);
1381 :
1382 : /* Set up DestReceiver, SharedExecutorInstrumentation, and QueryDesc. */
1383 1460 : receiver = ExecParallelGetReceiver(seg, toc);
1384 1460 : instrumentation = shm_toc_lookup(toc, PARALLEL_KEY_INSTRUMENTATION, true);
1385 1460 : if (instrumentation != NULL)
1386 468 : instrument_options = instrumentation->instrument_options;
1387 1460 : jit_instrumentation = shm_toc_lookup(toc, PARALLEL_KEY_JIT_INSTRUMENTATION,
1388 : true);
1389 1460 : queryDesc = ExecParallelGetQueryDesc(toc, receiver, instrument_options);
1390 :
1391 : /* Setting debug_query_string for individual workers */
1392 1460 : debug_query_string = queryDesc->sourceText;
1393 :
1394 : /* Report workers' query for monitoring purposes */
1395 1460 : pgstat_report_activity(STATE_RUNNING, debug_query_string);
1396 :
1397 : /* Attach to the dynamic shared memory area. */
1398 1460 : area_space = shm_toc_lookup(toc, PARALLEL_KEY_DSA, false);
1399 1460 : area = dsa_attach_in_place(area_space, seg);
1400 :
1401 : /* Start up the executor */
1402 1460 : queryDesc->plannedstmt->jitFlags = fpes->jit_flags;
1403 1460 : ExecutorStart(queryDesc, fpes->eflags);
1404 :
1405 : /* Special executor initialization steps for parallel workers */
1406 1460 : queryDesc->planstate->state->es_query_dsa = area;
1407 1460 : if (DsaPointerIsValid(fpes->param_exec))
1408 : {
1409 : char *paramexec_space;
1410 :
1411 24 : paramexec_space = dsa_get_address(area, fpes->param_exec);
1412 24 : RestoreParamExecParams(paramexec_space, queryDesc->estate);
1413 :
1414 : }
1415 1460 : pwcxt.toc = toc;
1416 1460 : pwcxt.seg = seg;
1417 1460 : ExecParallelInitializeWorker(queryDesc->planstate, &pwcxt);
1418 :
1419 : /* Pass down any tuple bound */
1420 1460 : ExecSetTupleBound(fpes->tuples_needed, queryDesc->planstate);
1421 :
1422 : /*
1423 : * Prepare to track buffer/WAL usage during query execution.
1424 : *
1425 : * We do this after starting up the executor to match what happens in the
1426 : * leader, which also doesn't count buffer accesses and WAL activity that
1427 : * occur during executor startup.
1428 : */
1429 1460 : InstrStartParallelQuery();
1430 :
1431 : /*
1432 : * Run the plan. If we specified a tuple bound, be careful not to demand
1433 : * more tuples than that.
1434 : */
1435 1460 : ExecutorRun(queryDesc,
1436 : ForwardScanDirection,
1437 1460 : fpes->tuples_needed < 0 ? (int64) 0 : fpes->tuples_needed,
1438 : true);
1439 :
1440 : /* Shut down the executor */
1441 1456 : ExecutorFinish(queryDesc);
1442 :
1443 : /* Report buffer/WAL usage during parallel execution. */
1444 1456 : buffer_usage = shm_toc_lookup(toc, PARALLEL_KEY_BUFFER_USAGE, false);
1445 1456 : wal_usage = shm_toc_lookup(toc, PARALLEL_KEY_WAL_USAGE, false);
1446 1456 : InstrEndParallelQuery(&buffer_usage[ParallelWorkerNumber],
1447 1456 : &wal_usage[ParallelWorkerNumber]);
1448 :
1449 : /* Report instrumentation data if any instrumentation options are set. */
1450 1456 : if (instrumentation != NULL)
1451 468 : ExecParallelReportInstrumentation(queryDesc->planstate,
1452 : instrumentation);
1453 :
1454 : /* Report JIT instrumentation data if any */
1455 1456 : if (queryDesc->estate->es_jit && jit_instrumentation != NULL)
1456 : {
1457 : Assert(ParallelWorkerNumber < jit_instrumentation->num_workers);
1458 96 : jit_instrumentation->jit_instr[ParallelWorkerNumber] =
1459 96 : queryDesc->estate->es_jit->instr;
1460 : }
1461 :
1462 : /* Must do this after capturing instrumentation. */
1463 1456 : ExecutorEnd(queryDesc);
1464 :
1465 : /* Cleanup. */
1466 1456 : dsa_detach(area);
1467 1456 : FreeQueryDesc(queryDesc);
1468 1456 : receiver->rDestroy(receiver);
1469 1456 : }
|
