Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * execParallel.c
4 : : * Support routines for parallel execution.
5 : : *
6 : : * Portions Copyright (c) 1996-2026, 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/nodeAgg.h"
29 : : #include "executor/nodeAppend.h"
30 : : #include "executor/nodeBitmapHeapscan.h"
31 : : #include "executor/nodeBitmapIndexscan.h"
32 : : #include "executor/nodeCustom.h"
33 : : #include "executor/nodeForeignscan.h"
34 : : #include "executor/nodeHash.h"
35 : : #include "executor/nodeHashjoin.h"
36 : : #include "executor/nodeIncrementalSort.h"
37 : : #include "executor/nodeIndexonlyscan.h"
38 : : #include "executor/nodeIndexscan.h"
39 : : #include "executor/nodeMaterial.h"
40 : : #include "executor/nodeMemoize.h"
41 : : #include "executor/nodeSeqscan.h"
42 : : #include "executor/nodeSort.h"
43 : : #include "executor/nodeSubplan.h"
44 : : #include "executor/nodeTidrangescan.h"
45 : : #include "executor/tqueue.h"
46 : : #include "jit/jit.h"
47 : : #include "nodes/nodeFuncs.h"
48 : : #include "pgstat.h"
49 : : #include "storage/proc.h"
50 : : #include "tcop/tcopprot.h"
51 : : #include "utils/datum.h"
52 : : #include "utils/dsa.h"
53 : : #include "utils/lsyscache.h"
54 : : #include "utils/snapmgr.h"
55 : :
56 : : /*
57 : : * Magic numbers for parallel executor communication. We use constants
58 : : * greater than any 32-bit integer here so that values < 2^32 can be used
59 : : * by individual parallel nodes to store their own state.
60 : : */
61 : : #define PARALLEL_KEY_EXECUTOR_FIXED UINT64CONST(0xE000000000000001)
62 : : #define PARALLEL_KEY_PLANNEDSTMT UINT64CONST(0xE000000000000002)
63 : : #define PARALLEL_KEY_PARAMLISTINFO UINT64CONST(0xE000000000000003)
64 : : #define PARALLEL_KEY_BUFFER_USAGE UINT64CONST(0xE000000000000004)
65 : : #define PARALLEL_KEY_TUPLE_QUEUE UINT64CONST(0xE000000000000005)
66 : : #define PARALLEL_KEY_INSTRUMENTATION UINT64CONST(0xE000000000000006)
67 : : #define PARALLEL_KEY_DSA UINT64CONST(0xE000000000000007)
68 : : #define PARALLEL_KEY_QUERY_TEXT UINT64CONST(0xE000000000000008)
69 : : #define PARALLEL_KEY_JIT_INSTRUMENTATION UINT64CONST(0xE000000000000009)
70 : : #define PARALLEL_KEY_WAL_USAGE UINT64CONST(0xE00000000000000A)
71 : :
72 : : #define PARALLEL_TUPLE_QUEUE_SIZE 65536
73 : :
74 : : /*
75 : : * Fixed-size random stuff that we need to pass to parallel workers.
76 : : */
77 : : typedef struct FixedParallelExecutorState
78 : : {
79 : : int64 tuples_needed; /* tuple bound, see ExecSetTupleBound */
80 : : dsa_pointer param_exec;
81 : : int eflags;
82 : : int jit_flags;
83 : : } FixedParallelExecutorState;
84 : :
85 : : /*
86 : : * DSM structure for accumulating per-PlanState instrumentation.
87 : : *
88 : : * instrument_options: Same meaning here as in instrument.c.
89 : : *
90 : : * instrument_offset: Offset, relative to the start of this structure,
91 : : * of the first NodeInstrumentation object. This will depend on the length of
92 : : * the plan_node_id array.
93 : : *
94 : : * num_workers: Number of workers.
95 : : *
96 : : * num_plan_nodes: Number of plan nodes.
97 : : *
98 : : * plan_node_id: Array of plan nodes for which we are gathering instrumentation
99 : : * from parallel workers. The length of this array is given by num_plan_nodes.
100 : : */
101 : : struct SharedExecutorInstrumentation
102 : : {
103 : : int instrument_options;
104 : : int instrument_offset;
105 : : int num_workers;
106 : : int num_plan_nodes;
107 : : int plan_node_id[FLEXIBLE_ARRAY_MEMBER];
108 : :
109 : : /*
110 : : * Array of num_plan_nodes * num_workers NodeInstrumentation objects
111 : : * follows.
112 : : */
113 : : };
114 : : #define GetInstrumentationArray(sei) \
115 : : (StaticAssertVariableIsOfTypeMacro(sei, SharedExecutorInstrumentation *), \
116 : : (NodeInstrumentation *) (((char *) sei) + sei->instrument_offset))
117 : :
118 : : /* Context object for ExecParallelEstimate. */
119 : : typedef struct ExecParallelEstimateContext
120 : : {
121 : : ParallelContext *pcxt;
122 : : int nnodes;
123 : : } ExecParallelEstimateContext;
124 : :
125 : : /* Context object for ExecParallelInitializeDSM. */
126 : : typedef struct ExecParallelInitializeDSMContext
127 : : {
128 : : ParallelContext *pcxt;
129 : : SharedExecutorInstrumentation *instrumentation;
130 : : int nnodes;
131 : : } ExecParallelInitializeDSMContext;
132 : :
133 : : /* Helper functions that run in the parallel leader. */
134 : : static char *ExecSerializePlan(Plan *plan, EState *estate);
135 : : static bool ExecParallelEstimate(PlanState *planstate,
136 : : ExecParallelEstimateContext *e);
137 : : static bool ExecParallelInitializeDSM(PlanState *planstate,
138 : : ExecParallelInitializeDSMContext *d);
139 : : static shm_mq_handle **ExecParallelSetupTupleQueues(ParallelContext *pcxt,
140 : : bool reinitialize);
141 : : static bool ExecParallelReInitializeDSM(PlanState *planstate,
142 : : ParallelContext *pcxt);
143 : : static bool ExecParallelRetrieveInstrumentation(PlanState *planstate,
144 : : SharedExecutorInstrumentation *instrumentation);
145 : :
146 : : /* Helper function that runs in the parallel worker. */
147 : : static DestReceiver *ExecParallelGetReceiver(dsm_segment *seg, shm_toc *toc);
148 : :
149 : : /*
150 : : * Create a serialized representation of the plan to be sent to each worker.
151 : : */
152 : : static char *
153 : 512 : ExecSerializePlan(Plan *plan, EState *estate)
154 : : {
155 : : PlannedStmt *pstmt;
156 : : ListCell *lc;
157 : :
158 : : /* We can't scribble on the original plan, so make a copy. */
159 : 512 : plan = copyObject(plan);
160 : :
161 : : /*
162 : : * The worker will start its own copy of the executor, and that copy will
163 : : * insert a junk filter if the toplevel node has any resjunk entries. We
164 : : * don't want that to happen, because while resjunk columns shouldn't be
165 : : * sent back to the user, here the tuples are coming back to another
166 : : * backend which may very well need them. So mutate the target list
167 : : * accordingly. This is sort of a hack; there might be better ways to do
168 : : * this...
169 : : */
170 [ + + + + : 1398 : foreach(lc, plan->targetlist)
+ + ]
171 : : {
172 : 886 : TargetEntry *tle = lfirst_node(TargetEntry, lc);
173 : :
174 : 886 : tle->resjunk = false;
175 : : }
176 : :
177 : : /*
178 : : * Create a dummy PlannedStmt. Most of the fields don't need to be valid
179 : : * for our purposes, but the worker will need at least a minimal
180 : : * PlannedStmt to start the executor.
181 : : */
182 : 512 : pstmt = makeNode(PlannedStmt);
183 : 512 : pstmt->commandType = CMD_SELECT;
184 : 512 : pstmt->queryId = pgstat_get_my_query_id();
185 : 512 : pstmt->planId = pgstat_get_my_plan_id();
186 : 512 : pstmt->hasReturning = false;
187 : 512 : pstmt->hasModifyingCTE = false;
188 : 512 : pstmt->canSetTag = true;
189 : 512 : pstmt->transientPlan = false;
190 : 512 : pstmt->dependsOnRole = false;
191 : 512 : pstmt->parallelModeNeeded = false;
192 : 512 : pstmt->planTree = plan;
193 : 512 : pstmt->partPruneInfos = estate->es_part_prune_infos;
194 : 512 : pstmt->rtable = estate->es_range_table;
195 : 512 : pstmt->unprunableRelids = estate->es_unpruned_relids;
196 : 512 : pstmt->permInfos = estate->es_rteperminfos;
197 : 512 : pstmt->appendRelations = NIL;
198 : 512 : pstmt->planOrigin = PLAN_STMT_INTERNAL;
199 : :
200 : : /*
201 : : * Transfer only parallel-safe subplans, leaving a NULL "hole" in the list
202 : : * for unsafe ones (so that the list indexes of the safe ones are
203 : : * preserved). This positively ensures that the worker won't try to run,
204 : : * or even do ExecInitNode on, an unsafe subplan. That's important to
205 : : * protect, eg, non-parallel-aware FDWs from getting into trouble.
206 : : */
207 : 512 : pstmt->subplans = NIL;
208 [ + + + + : 548 : foreach(lc, estate->es_plannedstmt->subplans)
+ + ]
209 : : {
210 : 36 : Plan *subplan = (Plan *) lfirst(lc);
211 : :
212 [ + - + + ]: 36 : if (subplan && !subplan->parallel_safe)
213 : 8 : subplan = NULL;
214 : 36 : pstmt->subplans = lappend(pstmt->subplans, subplan);
215 : : }
216 : :
217 : 512 : pstmt->rewindPlanIDs = NULL;
218 : 512 : pstmt->rowMarks = NIL;
219 : :
220 : : /*
221 : : * Pass the row mark and result relation relids to parallel workers. They
222 : : * may need to check them to inform heuristics.
223 : : */
224 : 512 : pstmt->rowMarkRelids = estate->es_plannedstmt->rowMarkRelids;
225 : 512 : pstmt->resultRelationRelids = estate->es_plannedstmt->resultRelationRelids;
226 : 512 : pstmt->relationOids = NIL;
227 : 512 : pstmt->invalItems = NIL; /* workers can't replan anyway... */
228 : 512 : pstmt->paramExecTypes = estate->es_plannedstmt->paramExecTypes;
229 : 512 : pstmt->utilityStmt = NULL;
230 : 512 : pstmt->stmt_location = -1;
231 : 512 : pstmt->stmt_len = -1;
232 : :
233 : : /* Return serialized copy of our dummy PlannedStmt. */
234 : 512 : return nodeToString(pstmt);
235 : : }
236 : :
237 : : /*
238 : : * Parallel-aware plan nodes (and occasionally others) may need some state
239 : : * which is shared across all parallel workers. Before we size the DSM, give
240 : : * them a chance to call shm_toc_estimate_chunk or shm_toc_estimate_keys on
241 : : * &pcxt->estimator.
242 : : *
243 : : * While we're at it, count the number of PlanState nodes in the tree, so
244 : : * we know how many Instrumentation structures we need.
245 : : */
246 : : static bool
247 : 2621 : ExecParallelEstimate(PlanState *planstate, ExecParallelEstimateContext *e)
248 : : {
249 [ - + ]: 2621 : if (planstate == NULL)
250 : 0 : return false;
251 : :
252 : : /* Count this node. */
253 : 2621 : e->nnodes++;
254 : :
255 [ + + + + : 2621 : switch (nodeTag(planstate))
- + + + +
+ + + - +
+ - + ]
256 : : {
257 : 1332 : case T_SeqScanState:
258 [ + + ]: 1332 : if (planstate->plan->parallel_aware)
259 : 1174 : ExecSeqScanEstimate((SeqScanState *) planstate,
260 : : e->pcxt);
261 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
262 : 1332 : ExecSeqScanInstrumentEstimate((SeqScanState *) planstate,
263 : : e->pcxt);
264 : 1332 : break;
265 : 196 : case T_IndexScanState:
266 [ + + ]: 196 : if (planstate->plan->parallel_aware)
267 : 12 : ExecIndexScanEstimate((IndexScanState *) planstate,
268 : : e->pcxt);
269 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
270 : 196 : ExecIndexScanInstrumentEstimate((IndexScanState *) planstate,
271 : : e->pcxt);
272 : 196 : break;
273 : 38 : case T_IndexOnlyScanState:
274 [ + + ]: 38 : if (planstate->plan->parallel_aware)
275 : 30 : ExecIndexOnlyScanEstimate((IndexOnlyScanState *) planstate,
276 : : e->pcxt);
277 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
278 : 38 : ExecIndexOnlyScanInstrumentEstimate((IndexOnlyScanState *) planstate,
279 : : e->pcxt);
280 : 38 : break;
281 : 13 : case T_BitmapIndexScanState:
282 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
283 : 13 : ExecBitmapIndexScanEstimate((BitmapIndexScanState *) planstate,
284 : : e->pcxt);
285 : 13 : break;
286 : 0 : case T_ForeignScanState:
287 [ # # ]: 0 : if (planstate->plan->parallel_aware)
288 : 0 : ExecForeignScanEstimate((ForeignScanState *) planstate,
289 : : e->pcxt);
290 : 0 : break;
291 : 16 : case T_TidRangeScanState:
292 [ + - ]: 16 : if (planstate->plan->parallel_aware)
293 : 16 : ExecTidRangeScanEstimate((TidRangeScanState *) planstate,
294 : : e->pcxt);
295 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
296 : 16 : ExecTidRangeScanInstrumentEstimate((TidRangeScanState *) planstate,
297 : : e->pcxt);
298 : 16 : break;
299 : 128 : case T_AppendState:
300 [ + + ]: 128 : if (planstate->plan->parallel_aware)
301 : 96 : ExecAppendEstimate((AppendState *) planstate,
302 : : e->pcxt);
303 : 128 : break;
304 : 1 : case T_CustomScanState:
305 [ - + ]: 1 : if (planstate->plan->parallel_aware)
306 : 0 : ExecCustomScanEstimate((CustomScanState *) planstate,
307 : : e->pcxt);
308 : 1 : break;
309 : 13 : case T_BitmapHeapScanState:
310 [ + + ]: 13 : if (planstate->plan->parallel_aware)
311 : 12 : ExecBitmapHeapEstimate((BitmapHeapScanState *) planstate,
312 : : e->pcxt);
313 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
314 : 13 : ExecBitmapHeapInstrumentEstimate((BitmapHeapScanState *) planstate,
315 : : e->pcxt);
316 : 13 : break;
317 : 132 : case T_HashJoinState:
318 [ + + ]: 132 : if (planstate->plan->parallel_aware)
319 : 84 : ExecHashJoinEstimate((HashJoinState *) planstate,
320 : : e->pcxt);
321 : 132 : break;
322 : 132 : case T_HashState:
323 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
324 : 132 : ExecHashEstimate((HashState *) planstate, e->pcxt);
325 : 132 : break;
326 : 110 : case T_SortState:
327 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
328 : 110 : ExecSortEstimate((SortState *) planstate, e->pcxt);
329 : 110 : break;
330 : 0 : case T_IncrementalSortState:
331 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
332 : 0 : ExecIncrementalSortEstimate((IncrementalSortState *) planstate, e->pcxt);
333 : 0 : break;
334 : 396 : case T_AggState:
335 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
336 : 396 : ExecAggEstimate((AggState *) planstate, e->pcxt);
337 : 396 : break;
338 : 4 : case T_MemoizeState:
339 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
340 : 4 : ExecMemoizeEstimate((MemoizeState *) planstate, e->pcxt);
341 : 4 : break;
342 : 0 : case T_MaterialState:
343 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
344 : 0 : ExecMaterialEstimate((MaterialState *) planstate, e->pcxt);
345 : 0 : break;
346 : 110 : default:
347 : 110 : break;
348 : : }
349 : :
350 : 2621 : return planstate_tree_walker(planstate, ExecParallelEstimate, e);
351 : : }
352 : :
353 : : /*
354 : : * Estimate the amount of space required to serialize the indicated parameters.
355 : : */
356 : : static Size
357 : 16 : EstimateParamExecSpace(EState *estate, Bitmapset *params)
358 : : {
359 : : int paramid;
360 : 16 : Size sz = sizeof(int);
361 : :
362 : 16 : paramid = -1;
363 [ + + ]: 36 : while ((paramid = bms_next_member(params, paramid)) >= 0)
364 : : {
365 : : Oid typeOid;
366 : : int16 typLen;
367 : : bool typByVal;
368 : : ParamExecData *prm;
369 : :
370 : 20 : prm = &(estate->es_param_exec_vals[paramid]);
371 : 20 : typeOid = list_nth_oid(estate->es_plannedstmt->paramExecTypes,
372 : : paramid);
373 : :
374 : 20 : sz = add_size(sz, sizeof(int)); /* space for paramid */
375 : :
376 : : /* space for datum/isnull */
377 [ + - ]: 20 : if (OidIsValid(typeOid))
378 : 20 : get_typlenbyval(typeOid, &typLen, &typByVal);
379 : : else
380 : : {
381 : : /* If no type OID, assume by-value, like copyParamList does. */
382 : 0 : typLen = sizeof(Datum);
383 : 0 : typByVal = true;
384 : : }
385 : 20 : sz = add_size(sz,
386 : 20 : datumEstimateSpace(prm->value, prm->isnull,
387 : : typByVal, typLen));
388 : : }
389 : 16 : return sz;
390 : : }
391 : :
392 : : /*
393 : : * Serialize specified PARAM_EXEC parameters.
394 : : *
395 : : * We write the number of parameters first, as a 4-byte integer, and then
396 : : * write details for each parameter in turn. The details for each parameter
397 : : * consist of a 4-byte paramid (location of param in execution time internal
398 : : * parameter array) and then the datum as serialized by datumSerialize().
399 : : */
400 : : static dsa_pointer
401 : 16 : SerializeParamExecParams(EState *estate, Bitmapset *params, dsa_area *area)
402 : : {
403 : : Size size;
404 : : int nparams;
405 : : int paramid;
406 : : ParamExecData *prm;
407 : : dsa_pointer handle;
408 : : char *start_address;
409 : :
410 : : /* Allocate enough space for the current parameter values. */
411 : 16 : size = EstimateParamExecSpace(estate, params);
412 : 16 : handle = dsa_allocate(area, size);
413 : 16 : start_address = dsa_get_address(area, handle);
414 : :
415 : : /* First write the number of parameters as a 4-byte integer. */
416 : 16 : nparams = bms_num_members(params);
417 : 16 : memcpy(start_address, &nparams, sizeof(int));
418 : 16 : start_address += sizeof(int);
419 : :
420 : : /* Write details for each parameter in turn. */
421 : 16 : paramid = -1;
422 [ + + ]: 36 : while ((paramid = bms_next_member(params, paramid)) >= 0)
423 : : {
424 : : Oid typeOid;
425 : : int16 typLen;
426 : : bool typByVal;
427 : :
428 : 20 : prm = &(estate->es_param_exec_vals[paramid]);
429 : 20 : typeOid = list_nth_oid(estate->es_plannedstmt->paramExecTypes,
430 : : paramid);
431 : :
432 : : /* Write paramid. */
433 : 20 : memcpy(start_address, ¶mid, sizeof(int));
434 : 20 : start_address += sizeof(int);
435 : :
436 : : /* Write datum/isnull */
437 [ + - ]: 20 : if (OidIsValid(typeOid))
438 : 20 : get_typlenbyval(typeOid, &typLen, &typByVal);
439 : : else
440 : : {
441 : : /* If no type OID, assume by-value, like copyParamList does. */
442 : 0 : typLen = sizeof(Datum);
443 : 0 : typByVal = true;
444 : : }
445 : 20 : datumSerialize(prm->value, prm->isnull, typByVal, typLen,
446 : : &start_address);
447 : : }
448 : :
449 : 16 : return handle;
450 : : }
451 : :
452 : : /*
453 : : * Restore specified PARAM_EXEC parameters.
454 : : */
455 : : static void
456 : 48 : RestoreParamExecParams(char *start_address, EState *estate)
457 : : {
458 : : int nparams;
459 : : int i;
460 : : int paramid;
461 : :
462 : 48 : memcpy(&nparams, start_address, sizeof(int));
463 : 48 : start_address += sizeof(int);
464 : :
465 [ + + ]: 104 : for (i = 0; i < nparams; i++)
466 : : {
467 : : ParamExecData *prm;
468 : :
469 : : /* Read paramid */
470 : 56 : memcpy(¶mid, start_address, sizeof(int));
471 : 56 : start_address += sizeof(int);
472 : 56 : prm = &(estate->es_param_exec_vals[paramid]);
473 : :
474 : : /* Read datum/isnull. */
475 : 56 : prm->value = datumRestore(&start_address, &prm->isnull);
476 : 56 : prm->execPlan = NULL;
477 : : }
478 : 48 : }
479 : :
480 : : /*
481 : : * Initialize the dynamic shared memory segment that will be used to control
482 : : * parallel execution.
483 : : */
484 : : static bool
485 : 2621 : ExecParallelInitializeDSM(PlanState *planstate,
486 : : ExecParallelInitializeDSMContext *d)
487 : : {
488 [ - + ]: 2621 : if (planstate == NULL)
489 : 0 : return false;
490 : :
491 : : /* If instrumentation is enabled, initialize slot for this node. */
492 [ + + ]: 2621 : if (d->instrumentation != NULL)
493 : 684 : d->instrumentation->plan_node_id[d->nnodes] =
494 : 684 : planstate->plan->plan_node_id;
495 : :
496 : : /* Count this node. */
497 : 2621 : d->nnodes++;
498 : :
499 : : /*
500 : : * Call initializers for DSM-using plan nodes.
501 : : *
502 : : * Most plan nodes won't do anything here, but plan nodes that allocated
503 : : * DSM may need to initialize shared state in the DSM before parallel
504 : : * workers are launched. They can allocate the space they previously
505 : : * estimated using shm_toc_allocate, and add the keys they previously
506 : : * estimated using shm_toc_insert, in each case targeting pcxt->toc.
507 : : */
508 [ + + + + : 2621 : switch (nodeTag(planstate))
- + + + +
+ + + - +
+ - + ]
509 : : {
510 : 1332 : case T_SeqScanState:
511 [ + + ]: 1332 : if (planstate->plan->parallel_aware)
512 : 1174 : ExecSeqScanInitializeDSM((SeqScanState *) planstate,
513 : : d->pcxt);
514 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
515 : 1332 : ExecSeqScanInstrumentInitDSM((SeqScanState *) planstate,
516 : : d->pcxt);
517 : 1332 : break;
518 : 196 : case T_IndexScanState:
519 [ + + ]: 196 : if (planstate->plan->parallel_aware)
520 : 12 : ExecIndexScanInitializeDSM((IndexScanState *) planstate,
521 : : d->pcxt);
522 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
523 : 196 : ExecIndexScanInstrumentInitDSM((IndexScanState *) planstate,
524 : : d->pcxt);
525 : 196 : break;
526 : 38 : case T_IndexOnlyScanState:
527 [ + + ]: 38 : if (planstate->plan->parallel_aware)
528 : 30 : ExecIndexOnlyScanInitializeDSM((IndexOnlyScanState *) planstate,
529 : : d->pcxt);
530 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
531 : 38 : ExecIndexOnlyScanInstrumentInitDSM((IndexOnlyScanState *) planstate,
532 : : d->pcxt);
533 : 38 : break;
534 : 13 : case T_BitmapIndexScanState:
535 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
536 : 13 : ExecBitmapIndexScanInitializeDSM((BitmapIndexScanState *) planstate, d->pcxt);
537 : 13 : break;
538 : 0 : case T_ForeignScanState:
539 [ # # ]: 0 : if (planstate->plan->parallel_aware)
540 : 0 : ExecForeignScanInitializeDSM((ForeignScanState *) planstate,
541 : : d->pcxt);
542 : 0 : break;
543 : 16 : case T_TidRangeScanState:
544 [ + - ]: 16 : if (planstate->plan->parallel_aware)
545 : 16 : ExecTidRangeScanInitializeDSM((TidRangeScanState *) planstate,
546 : : d->pcxt);
547 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
548 : 16 : ExecTidRangeScanInstrumentInitDSM((TidRangeScanState *) planstate,
549 : : d->pcxt);
550 : 16 : break;
551 : 128 : case T_AppendState:
552 [ + + ]: 128 : if (planstate->plan->parallel_aware)
553 : 96 : ExecAppendInitializeDSM((AppendState *) planstate,
554 : : d->pcxt);
555 : 128 : break;
556 : 1 : case T_CustomScanState:
557 [ - + ]: 1 : if (planstate->plan->parallel_aware)
558 : 0 : ExecCustomScanInitializeDSM((CustomScanState *) planstate,
559 : : d->pcxt);
560 : 1 : break;
561 : 13 : case T_BitmapHeapScanState:
562 [ + + ]: 13 : if (planstate->plan->parallel_aware)
563 : 12 : ExecBitmapHeapInitializeDSM((BitmapHeapScanState *) planstate,
564 : : d->pcxt);
565 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
566 : 13 : ExecBitmapHeapInstrumentInitDSM((BitmapHeapScanState *) planstate,
567 : : d->pcxt);
568 : 13 : break;
569 : 132 : case T_HashJoinState:
570 [ + + ]: 132 : if (planstate->plan->parallel_aware)
571 : 84 : ExecHashJoinInitializeDSM((HashJoinState *) planstate,
572 : : d->pcxt);
573 : 132 : break;
574 : 132 : case T_HashState:
575 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
576 : 132 : ExecHashInitializeDSM((HashState *) planstate, d->pcxt);
577 : 132 : break;
578 : 110 : case T_SortState:
579 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
580 : 110 : ExecSortInitializeDSM((SortState *) planstate, d->pcxt);
581 : 110 : break;
582 : 0 : case T_IncrementalSortState:
583 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
584 : 0 : ExecIncrementalSortInitializeDSM((IncrementalSortState *) planstate, d->pcxt);
585 : 0 : break;
586 : 396 : case T_AggState:
587 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
588 : 396 : ExecAggInitializeDSM((AggState *) planstate, d->pcxt);
589 : 396 : break;
590 : 4 : case T_MemoizeState:
591 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
592 : 4 : ExecMemoizeInitializeDSM((MemoizeState *) planstate, d->pcxt);
593 : 4 : break;
594 : 0 : case T_MaterialState:
595 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
596 : 0 : ExecMaterialInitializeDSM((MaterialState *) planstate, d->pcxt);
597 : 0 : break;
598 : 110 : default:
599 : 110 : break;
600 : : }
601 : :
602 : 2621 : return planstate_tree_walker(planstate, ExecParallelInitializeDSM, d);
603 : : }
604 : :
605 : : /*
606 : : * It sets up the response queues for backend workers to return tuples
607 : : * to the main backend and start the workers.
608 : : */
609 : : static shm_mq_handle **
610 : 684 : ExecParallelSetupTupleQueues(ParallelContext *pcxt, bool reinitialize)
611 : : {
612 : : shm_mq_handle **responseq;
613 : : char *tqueuespace;
614 : : int i;
615 : :
616 : : /* Skip this if no workers. */
617 [ - + ]: 684 : if (pcxt->nworkers == 0)
618 : 0 : return NULL;
619 : :
620 : : /* Allocate memory for shared memory queue handles. */
621 : 684 : responseq = palloc_array(shm_mq_handle *, pcxt->nworkers);
622 : :
623 : : /*
624 : : * If not reinitializing, allocate space from the DSM for the queues;
625 : : * otherwise, find the already allocated space.
626 : : */
627 [ + + ]: 684 : if (!reinitialize)
628 : : tqueuespace =
629 : 512 : shm_toc_allocate(pcxt->toc,
630 : : mul_size(PARALLEL_TUPLE_QUEUE_SIZE,
631 : 512 : pcxt->nworkers));
632 : : else
633 : 172 : tqueuespace = shm_toc_lookup(pcxt->toc, PARALLEL_KEY_TUPLE_QUEUE, false);
634 : :
635 : : /* Create the queues, and become the receiver for each. */
636 [ + + ]: 2531 : for (i = 0; i < pcxt->nworkers; ++i)
637 : : {
638 : : shm_mq *mq;
639 : :
640 : 1847 : mq = shm_mq_create(tqueuespace +
641 : 1847 : ((Size) i) * PARALLEL_TUPLE_QUEUE_SIZE,
642 : : (Size) PARALLEL_TUPLE_QUEUE_SIZE);
643 : :
644 : 1847 : shm_mq_set_receiver(mq, MyProc);
645 : 1847 : responseq[i] = shm_mq_attach(mq, pcxt->seg, NULL);
646 : : }
647 : :
648 : : /* Add array of queues to shm_toc, so others can find it. */
649 [ + + ]: 684 : if (!reinitialize)
650 : 512 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_TUPLE_QUEUE, tqueuespace);
651 : :
652 : : /* Return array of handles. */
653 : 684 : return responseq;
654 : : }
655 : :
656 : : /*
657 : : * Sets up the required infrastructure for backend workers to perform
658 : : * execution and return results to the main backend.
659 : : */
660 : : ParallelExecutorInfo *
661 : 512 : ExecInitParallelPlan(PlanState *planstate, EState *estate,
662 : : Bitmapset *sendParams, int nworkers,
663 : : int64 tuples_needed)
664 : : {
665 : : ParallelExecutorInfo *pei;
666 : : ParallelContext *pcxt;
667 : : ExecParallelEstimateContext e;
668 : : ExecParallelInitializeDSMContext d;
669 : : FixedParallelExecutorState *fpes;
670 : : char *pstmt_data;
671 : : char *pstmt_space;
672 : : char *paramlistinfo_space;
673 : : BufferUsage *bufusage_space;
674 : : WalUsage *walusage_space;
675 : 512 : SharedExecutorInstrumentation *instrumentation = NULL;
676 : 512 : SharedJitInstrumentation *jit_instrumentation = NULL;
677 : : int pstmt_len;
678 : : int paramlistinfo_len;
679 : 512 : int instrumentation_len = 0;
680 : 512 : int jit_instrumentation_len = 0;
681 : 512 : int instrument_offset = 0;
682 : 512 : Size dsa_minsize = dsa_minimum_size();
683 : : char *query_string;
684 : : int query_len;
685 : :
686 : : /*
687 : : * Force any initplan outputs that we're going to pass to workers to be
688 : : * evaluated, if they weren't already.
689 : : *
690 : : * For simplicity, we use the EState's per-output-tuple ExprContext here.
691 : : * That risks intra-query memory leakage, since we might pass through here
692 : : * many times before that ExprContext gets reset; but ExecSetParamPlan
693 : : * doesn't normally leak any memory in the context (see its comments), so
694 : : * it doesn't seem worth complicating this function's API to pass it a
695 : : * shorter-lived ExprContext. This might need to change someday.
696 : : */
697 [ + + ]: 512 : ExecSetParamPlanMulti(sendParams, GetPerTupleExprContext(estate));
698 : :
699 : : /* Allocate object for return value. */
700 : 512 : pei = palloc0_object(ParallelExecutorInfo);
701 : 512 : pei->finished = false;
702 : 512 : pei->planstate = planstate;
703 : :
704 : : /* Fix up and serialize plan to be sent to workers. */
705 : 512 : pstmt_data = ExecSerializePlan(planstate->plan, estate);
706 : :
707 : : /* Create a parallel context. */
708 : 512 : pcxt = CreateParallelContext("postgres", "ParallelQueryMain", nworkers);
709 : 512 : pei->pcxt = pcxt;
710 : :
711 : : /*
712 : : * Before telling the parallel context to create a dynamic shared memory
713 : : * segment, we need to figure out how big it should be. Estimate space
714 : : * for the various things we need to store.
715 : : */
716 : :
717 : : /* Estimate space for fixed-size state. */
718 : 512 : shm_toc_estimate_chunk(&pcxt->estimator,
719 : : sizeof(FixedParallelExecutorState));
720 : 512 : shm_toc_estimate_keys(&pcxt->estimator, 1);
721 : :
722 : : /* Estimate space for query text. */
723 : 512 : query_len = strlen(estate->es_sourceText);
724 : 512 : shm_toc_estimate_chunk(&pcxt->estimator, query_len + 1);
725 : 512 : shm_toc_estimate_keys(&pcxt->estimator, 1);
726 : :
727 : : /* Estimate space for serialized PlannedStmt. */
728 : 512 : pstmt_len = strlen(pstmt_data) + 1;
729 : 512 : shm_toc_estimate_chunk(&pcxt->estimator, pstmt_len);
730 : 512 : shm_toc_estimate_keys(&pcxt->estimator, 1);
731 : :
732 : : /* Estimate space for serialized ParamListInfo. */
733 : 512 : paramlistinfo_len = EstimateParamListSpace(estate->es_param_list_info);
734 : 512 : shm_toc_estimate_chunk(&pcxt->estimator, paramlistinfo_len);
735 : 512 : shm_toc_estimate_keys(&pcxt->estimator, 1);
736 : :
737 : : /*
738 : : * Estimate space for BufferUsage.
739 : : *
740 : : * If EXPLAIN is not in use and there are no extensions loaded that care,
741 : : * we could skip this. But we have no way of knowing whether anyone's
742 : : * looking at pgBufferUsage, so do it unconditionally.
743 : : */
744 : 512 : shm_toc_estimate_chunk(&pcxt->estimator,
745 : : mul_size(sizeof(BufferUsage), pcxt->nworkers));
746 : 512 : shm_toc_estimate_keys(&pcxt->estimator, 1);
747 : :
748 : : /*
749 : : * Same thing for WalUsage.
750 : : */
751 : 512 : shm_toc_estimate_chunk(&pcxt->estimator,
752 : : mul_size(sizeof(WalUsage), pcxt->nworkers));
753 : 512 : shm_toc_estimate_keys(&pcxt->estimator, 1);
754 : :
755 : : /* Estimate space for tuple queues. */
756 : 512 : shm_toc_estimate_chunk(&pcxt->estimator,
757 : : mul_size(PARALLEL_TUPLE_QUEUE_SIZE, pcxt->nworkers));
758 : 512 : shm_toc_estimate_keys(&pcxt->estimator, 1);
759 : :
760 : : /*
761 : : * Give parallel-aware nodes a chance to add to the estimates, and get a
762 : : * count of how many PlanState nodes there are.
763 : : */
764 : 512 : e.pcxt = pcxt;
765 : 512 : e.nnodes = 0;
766 : 512 : ExecParallelEstimate(planstate, &e);
767 : :
768 : : /* Estimate space for instrumentation, if required. */
769 [ + + ]: 512 : if (estate->es_instrument)
770 : : {
771 : 120 : instrumentation_len =
772 : : offsetof(SharedExecutorInstrumentation, plan_node_id) +
773 : 120 : sizeof(int) * e.nnodes;
774 : 120 : instrumentation_len = MAXALIGN(instrumentation_len);
775 : 120 : instrument_offset = instrumentation_len;
776 : 120 : instrumentation_len +=
777 : 120 : mul_size(sizeof(NodeInstrumentation),
778 : 120 : mul_size(e.nnodes, nworkers));
779 : 120 : shm_toc_estimate_chunk(&pcxt->estimator, instrumentation_len);
780 : 120 : shm_toc_estimate_keys(&pcxt->estimator, 1);
781 : :
782 : : /* Estimate space for JIT instrumentation, if required. */
783 [ - + ]: 120 : if (estate->es_jit_flags != PGJIT_NONE)
784 : : {
785 : 0 : jit_instrumentation_len =
786 : 0 : offsetof(SharedJitInstrumentation, jit_instr) +
787 : : sizeof(JitInstrumentation) * nworkers;
788 : 0 : shm_toc_estimate_chunk(&pcxt->estimator, jit_instrumentation_len);
789 : 0 : shm_toc_estimate_keys(&pcxt->estimator, 1);
790 : : }
791 : : }
792 : :
793 : : /* Estimate space for DSA area. */
794 : 512 : shm_toc_estimate_chunk(&pcxt->estimator, dsa_minsize);
795 : 512 : shm_toc_estimate_keys(&pcxt->estimator, 1);
796 : :
797 : : /*
798 : : * InitializeParallelDSM() passes the active snapshot to the parallel
799 : : * worker, which uses it to set es_snapshot. Make sure we don't set
800 : : * es_snapshot differently in the child.
801 : : */
802 : : Assert(GetActiveSnapshot() == estate->es_snapshot);
803 : :
804 : : /* Everyone's had a chance to ask for space, so now create the DSM. */
805 : 512 : InitializeParallelDSM(pcxt);
806 : :
807 : : /*
808 : : * OK, now we have a dynamic shared memory segment, and it should be big
809 : : * enough to store all of the data we estimated we would want to put into
810 : : * it, plus whatever general stuff (not specifically executor-related) the
811 : : * ParallelContext itself needs to store there. None of the space we
812 : : * asked for has been allocated or initialized yet, though, so do that.
813 : : */
814 : :
815 : : /* Store fixed-size state. */
816 : 512 : fpes = shm_toc_allocate(pcxt->toc, sizeof(FixedParallelExecutorState));
817 : 512 : fpes->tuples_needed = tuples_needed;
818 : 512 : fpes->param_exec = InvalidDsaPointer;
819 : 512 : fpes->eflags = estate->es_top_eflags;
820 : 512 : fpes->jit_flags = estate->es_jit_flags;
821 : 512 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_EXECUTOR_FIXED, fpes);
822 : :
823 : : /* Store query string */
824 : 512 : query_string = shm_toc_allocate(pcxt->toc, query_len + 1);
825 : 512 : memcpy(query_string, estate->es_sourceText, query_len + 1);
826 : 512 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_QUERY_TEXT, query_string);
827 : :
828 : : /* Store serialized PlannedStmt. */
829 : 512 : pstmt_space = shm_toc_allocate(pcxt->toc, pstmt_len);
830 : 512 : memcpy(pstmt_space, pstmt_data, pstmt_len);
831 : 512 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_PLANNEDSTMT, pstmt_space);
832 : :
833 : : /* Store serialized ParamListInfo. */
834 : 512 : paramlistinfo_space = shm_toc_allocate(pcxt->toc, paramlistinfo_len);
835 : 512 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_PARAMLISTINFO, paramlistinfo_space);
836 : 512 : SerializeParamList(estate->es_param_list_info, ¶mlistinfo_space);
837 : :
838 : : /* Allocate space for each worker's BufferUsage; no need to initialize. */
839 : 512 : bufusage_space = shm_toc_allocate(pcxt->toc,
840 : 512 : mul_size(sizeof(BufferUsage), pcxt->nworkers));
841 : 512 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_BUFFER_USAGE, bufusage_space);
842 : 512 : pei->buffer_usage = bufusage_space;
843 : :
844 : : /* Same for WalUsage. */
845 : 512 : walusage_space = shm_toc_allocate(pcxt->toc,
846 : 512 : mul_size(sizeof(WalUsage), pcxt->nworkers));
847 : 512 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_WAL_USAGE, walusage_space);
848 : 512 : pei->wal_usage = walusage_space;
849 : :
850 : : /* Set up the tuple queues that the workers will write into. */
851 : 512 : pei->tqueue = ExecParallelSetupTupleQueues(pcxt, false);
852 : :
853 : : /* We don't need the TupleQueueReaders yet, though. */
854 : 512 : pei->reader = NULL;
855 : :
856 : : /*
857 : : * If instrumentation options were supplied, allocate space for the data.
858 : : * It only gets partially initialized here; the rest happens during
859 : : * ExecParallelInitializeDSM.
860 : : */
861 [ + + ]: 512 : if (estate->es_instrument)
862 : : {
863 : : NodeInstrumentation *instrument;
864 : : int i;
865 : :
866 : 120 : instrumentation = shm_toc_allocate(pcxt->toc, instrumentation_len);
867 : 120 : instrumentation->instrument_options = estate->es_instrument;
868 : 120 : instrumentation->instrument_offset = instrument_offset;
869 : 120 : instrumentation->num_workers = nworkers;
870 : 120 : instrumentation->num_plan_nodes = e.nnodes;
871 : 120 : instrument = GetInstrumentationArray(instrumentation);
872 [ + + ]: 1240 : for (i = 0; i < nworkers * e.nnodes; ++i)
873 : 1120 : InstrInitNode(&instrument[i], estate->es_instrument, false);
874 : 120 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_INSTRUMENTATION,
875 : : instrumentation);
876 : 120 : pei->instrumentation = instrumentation;
877 : :
878 [ - + ]: 120 : if (estate->es_jit_flags != PGJIT_NONE)
879 : : {
880 : 0 : jit_instrumentation = shm_toc_allocate(pcxt->toc,
881 : : jit_instrumentation_len);
882 : 0 : jit_instrumentation->num_workers = nworkers;
883 : 0 : memset(jit_instrumentation->jit_instr, 0,
884 : : sizeof(JitInstrumentation) * nworkers);
885 : 0 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_JIT_INSTRUMENTATION,
886 : : jit_instrumentation);
887 : 0 : pei->jit_instrumentation = jit_instrumentation;
888 : : }
889 : : }
890 : :
891 : : /*
892 : : * Create a DSA area that can be used by the leader and all workers.
893 : : * (However, if we failed to create a DSM and are using private memory
894 : : * instead, then skip this.)
895 : : */
896 [ + - ]: 512 : if (pcxt->seg != NULL)
897 : : {
898 : : char *area_space;
899 : :
900 : 512 : area_space = shm_toc_allocate(pcxt->toc, dsa_minsize);
901 : 512 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_DSA, area_space);
902 : 512 : pei->area = dsa_create_in_place(area_space, dsa_minsize,
903 : : LWTRANCHE_PARALLEL_QUERY_DSA,
904 : : pcxt->seg);
905 : :
906 : : /*
907 : : * Serialize parameters, if any, using DSA storage. We don't dare use
908 : : * the main parallel query DSM for this because we might relaunch
909 : : * workers after the values have changed (and thus the amount of
910 : : * storage required has changed).
911 : : */
912 [ + + ]: 512 : if (!bms_is_empty(sendParams))
913 : : {
914 : 16 : pei->param_exec = SerializeParamExecParams(estate, sendParams,
915 : : pei->area);
916 : 16 : fpes->param_exec = pei->param_exec;
917 : : }
918 : : }
919 : :
920 : : /*
921 : : * Give parallel-aware nodes a chance to initialize their shared data.
922 : : * This also initializes the elements of instrumentation->ps_instrument,
923 : : * if it exists.
924 : : */
925 : 512 : d.pcxt = pcxt;
926 : 512 : d.instrumentation = instrumentation;
927 : 512 : d.nnodes = 0;
928 : :
929 : : /* Install our DSA area while initializing the plan. */
930 : 512 : estate->es_query_dsa = pei->area;
931 : 512 : ExecParallelInitializeDSM(planstate, &d);
932 : 512 : estate->es_query_dsa = NULL;
933 : :
934 : : /*
935 : : * Make sure that the world hasn't shifted under our feet. This could
936 : : * probably just be an Assert(), but let's be conservative for now.
937 : : */
938 [ - + ]: 512 : if (e.nnodes != d.nnodes)
939 [ # # ]: 0 : elog(ERROR, "inconsistent count of PlanState nodes");
940 : :
941 : : /* OK, we're ready to rock and roll. */
942 : 512 : return pei;
943 : : }
944 : :
945 : : /*
946 : : * Set up tuple queue readers to read the results of a parallel subplan.
947 : : *
948 : : * This is separate from ExecInitParallelPlan() because we can launch the
949 : : * worker processes and let them start doing something before we do this.
950 : : */
951 : : void
952 : 672 : ExecParallelCreateReaders(ParallelExecutorInfo *pei)
953 : : {
954 : 672 : int nworkers = pei->pcxt->nworkers_launched;
955 : : int i;
956 : :
957 : : Assert(pei->reader == NULL);
958 : :
959 [ + - ]: 672 : if (nworkers > 0)
960 : : {
961 : 672 : pei->reader = palloc_array(TupleQueueReader *, nworkers);
962 : :
963 [ + + ]: 2464 : for (i = 0; i < nworkers; i++)
964 : : {
965 : 1792 : shm_mq_set_handle(pei->tqueue[i],
966 : 1792 : pei->pcxt->worker[i].bgwhandle);
967 : 1792 : pei->reader[i] = CreateTupleQueueReader(pei->tqueue[i]);
968 : : }
969 : : }
970 : 672 : }
971 : :
972 : : /*
973 : : * Re-initialize the parallel executor shared memory state before launching
974 : : * a fresh batch of workers.
975 : : */
976 : : void
977 : 172 : ExecParallelReinitialize(PlanState *planstate,
978 : : ParallelExecutorInfo *pei,
979 : : Bitmapset *sendParams)
980 : : {
981 : 172 : EState *estate = planstate->state;
982 : : FixedParallelExecutorState *fpes;
983 : :
984 : : /* Old workers must already be shut down */
985 : : Assert(pei->finished);
986 : :
987 : : /*
988 : : * Force any initplan outputs that we're going to pass to workers to be
989 : : * evaluated, if they weren't already (see comments in
990 : : * ExecInitParallelPlan).
991 : : */
992 [ + - ]: 172 : ExecSetParamPlanMulti(sendParams, GetPerTupleExprContext(estate));
993 : :
994 : 172 : ReinitializeParallelDSM(pei->pcxt);
995 : 172 : pei->tqueue = ExecParallelSetupTupleQueues(pei->pcxt, true);
996 : 172 : pei->reader = NULL;
997 : 172 : pei->finished = false;
998 : :
999 : 172 : fpes = shm_toc_lookup(pei->pcxt->toc, PARALLEL_KEY_EXECUTOR_FIXED, false);
1000 : :
1001 : : /* Free any serialized parameters from the last round. */
1002 [ - + ]: 172 : if (DsaPointerIsValid(fpes->param_exec))
1003 : : {
1004 : 0 : dsa_free(pei->area, fpes->param_exec);
1005 : 0 : fpes->param_exec = InvalidDsaPointer;
1006 : : }
1007 : :
1008 : : /* Serialize current parameter values if required. */
1009 [ - + ]: 172 : if (!bms_is_empty(sendParams))
1010 : : {
1011 : 0 : pei->param_exec = SerializeParamExecParams(estate, sendParams,
1012 : : pei->area);
1013 : 0 : fpes->param_exec = pei->param_exec;
1014 : : }
1015 : :
1016 : : /* Traverse plan tree and let each child node reset associated state. */
1017 : 172 : estate->es_query_dsa = pei->area;
1018 : 172 : ExecParallelReInitializeDSM(planstate, pei->pcxt);
1019 : 172 : estate->es_query_dsa = NULL;
1020 : 172 : }
1021 : :
1022 : : /*
1023 : : * Traverse plan tree to reinitialize per-node dynamic shared memory state
1024 : : */
1025 : : static bool
1026 : 444 : ExecParallelReInitializeDSM(PlanState *planstate,
1027 : : ParallelContext *pcxt)
1028 : : {
1029 [ - + ]: 444 : if (planstate == NULL)
1030 : 0 : return false;
1031 : :
1032 : : /*
1033 : : * Call reinitializers for DSM-using plan nodes.
1034 : : */
1035 [ + + + - : 444 : switch (nodeTag(planstate))
- - - + +
+ + ]
1036 : : {
1037 : 184 : case T_SeqScanState:
1038 [ + + ]: 184 : if (planstate->plan->parallel_aware)
1039 : 152 : ExecSeqScanReInitializeDSM((SeqScanState *) planstate,
1040 : : pcxt);
1041 : 184 : break;
1042 : 8 : case T_IndexScanState:
1043 [ + - ]: 8 : if (planstate->plan->parallel_aware)
1044 : 8 : ExecIndexScanReInitializeDSM((IndexScanState *) planstate,
1045 : : pcxt);
1046 : 8 : break;
1047 : 8 : case T_IndexOnlyScanState:
1048 [ + - ]: 8 : if (planstate->plan->parallel_aware)
1049 : 8 : ExecIndexOnlyScanReInitializeDSM((IndexOnlyScanState *) planstate,
1050 : : pcxt);
1051 : 8 : break;
1052 : 0 : case T_ForeignScanState:
1053 [ # # ]: 0 : if (planstate->plan->parallel_aware)
1054 : 0 : ExecForeignScanReInitializeDSM((ForeignScanState *) planstate,
1055 : : pcxt);
1056 : 0 : break;
1057 : 0 : case T_TidRangeScanState:
1058 [ # # ]: 0 : if (planstate->plan->parallel_aware)
1059 : 0 : ExecTidRangeScanReInitializeDSM((TidRangeScanState *) planstate,
1060 : : pcxt);
1061 : 0 : break;
1062 : 0 : case T_AppendState:
1063 [ # # ]: 0 : if (planstate->plan->parallel_aware)
1064 : 0 : ExecAppendReInitializeDSM((AppendState *) planstate, pcxt);
1065 : 0 : break;
1066 : 0 : case T_CustomScanState:
1067 [ # # ]: 0 : if (planstate->plan->parallel_aware)
1068 : 0 : ExecCustomScanReInitializeDSM((CustomScanState *) planstate,
1069 : : pcxt);
1070 : 0 : break;
1071 : 36 : case T_BitmapHeapScanState:
1072 [ + - ]: 36 : if (planstate->plan->parallel_aware)
1073 : 36 : ExecBitmapHeapReInitializeDSM((BitmapHeapScanState *) planstate,
1074 : : pcxt);
1075 : 36 : break;
1076 : 64 : case T_HashJoinState:
1077 [ + + ]: 64 : if (planstate->plan->parallel_aware)
1078 : 32 : ExecHashJoinReInitializeDSM((HashJoinState *) planstate,
1079 : : pcxt);
1080 : 64 : break;
1081 : 120 : case T_BitmapIndexScanState:
1082 : : case T_HashState:
1083 : : case T_SortState:
1084 : : case T_IncrementalSortState:
1085 : : case T_MemoizeState:
1086 : : case T_MaterialState:
1087 : : /* these nodes have DSM state, but no reinitialization is required */
1088 : 120 : break;
1089 : :
1090 : 24 : default:
1091 : 24 : break;
1092 : : }
1093 : :
1094 : 444 : return planstate_tree_walker(planstate, ExecParallelReInitializeDSM, pcxt);
1095 : : }
1096 : :
1097 : : /*
1098 : : * Copy instrumentation information about this node and its descendants from
1099 : : * dynamic shared memory.
1100 : : */
1101 : : static bool
1102 : 684 : ExecParallelRetrieveInstrumentation(PlanState *planstate,
1103 : : SharedExecutorInstrumentation *instrumentation)
1104 : : {
1105 : : NodeInstrumentation *instrument;
1106 : : int i;
1107 : : int n;
1108 : : int ibytes;
1109 : 684 : int plan_node_id = planstate->plan->plan_node_id;
1110 : : MemoryContext oldcontext;
1111 : :
1112 : : /* Find the instrumentation for this node. */
1113 [ + - ]: 3092 : for (i = 0; i < instrumentation->num_plan_nodes; ++i)
1114 [ + + ]: 3092 : if (instrumentation->plan_node_id[i] == plan_node_id)
1115 : 684 : break;
1116 [ - + ]: 684 : if (i >= instrumentation->num_plan_nodes)
1117 [ # # ]: 0 : elog(ERROR, "plan node %d not found", plan_node_id);
1118 : :
1119 : : /* Accumulate the statistics from all workers. */
1120 : 684 : instrument = GetInstrumentationArray(instrumentation);
1121 : 684 : instrument += i * instrumentation->num_workers;
1122 [ + + ]: 1804 : for (n = 0; n < instrumentation->num_workers; ++n)
1123 : 1120 : InstrAggNode(planstate->instrument, &instrument[n]);
1124 : :
1125 : : /*
1126 : : * Also store the per-worker detail.
1127 : : *
1128 : : * Worker instrumentation should be allocated in the same context as the
1129 : : * regular instrumentation information, which is the per-query context.
1130 : : * Switch into per-query memory context.
1131 : : */
1132 : 684 : oldcontext = MemoryContextSwitchTo(planstate->state->es_query_cxt);
1133 : 684 : ibytes = mul_size(instrumentation->num_workers, sizeof(NodeInstrumentation));
1134 : 684 : planstate->worker_instrument =
1135 : 684 : palloc(ibytes + offsetof(WorkerNodeInstrumentation, instrument));
1136 : 684 : MemoryContextSwitchTo(oldcontext);
1137 : :
1138 : 684 : planstate->worker_instrument->num_workers = instrumentation->num_workers;
1139 : 684 : memcpy(planstate->worker_instrument->instrument, instrument, ibytes);
1140 : :
1141 : : /* Perform any node-type-specific work that needs to be done. */
1142 [ + - - + : 684 : switch (nodeTag(planstate))
- + + - -
- + - + ]
1143 : : {
1144 : 180 : case T_IndexScanState:
1145 : 180 : ExecIndexScanRetrieveInstrumentation((IndexScanState *) planstate);
1146 : 180 : break;
1147 : 0 : case T_IndexOnlyScanState:
1148 : 0 : ExecIndexOnlyScanRetrieveInstrumentation((IndexOnlyScanState *) planstate);
1149 : 0 : break;
1150 : 0 : case T_BitmapIndexScanState:
1151 : 0 : ExecBitmapIndexScanRetrieveInstrumentation((BitmapIndexScanState *) planstate);
1152 : 0 : break;
1153 : 8 : case T_SortState:
1154 : 8 : ExecSortRetrieveInstrumentation((SortState *) planstate);
1155 : 8 : break;
1156 : 0 : case T_IncrementalSortState:
1157 : 0 : ExecIncrementalSortRetrieveInstrumentation((IncrementalSortState *) planstate);
1158 : 0 : break;
1159 : 56 : case T_HashState:
1160 : 56 : ExecHashRetrieveInstrumentation((HashState *) planstate);
1161 : 56 : break;
1162 : 68 : case T_AggState:
1163 : 68 : ExecAggRetrieveInstrumentation((AggState *) planstate);
1164 : 68 : break;
1165 : 0 : case T_MemoizeState:
1166 : 0 : ExecMemoizeRetrieveInstrumentation((MemoizeState *) planstate);
1167 : 0 : break;
1168 : 0 : case T_MaterialState:
1169 : 0 : ExecMaterialRetrieveInstrumentation((MaterialState *) planstate);
1170 : 0 : break;
1171 : 0 : case T_BitmapHeapScanState:
1172 : 0 : ExecBitmapHeapRetrieveInstrumentation((BitmapHeapScanState *) planstate);
1173 : 0 : break;
1174 : 232 : case T_SeqScanState:
1175 : 232 : ExecSeqScanRetrieveInstrumentation((SeqScanState *) planstate);
1176 : 232 : break;
1177 : 0 : case T_TidRangeScanState:
1178 : 0 : ExecTidRangeScanRetrieveInstrumentation((TidRangeScanState *) planstate);
1179 : 0 : break;
1180 : 140 : default:
1181 : 140 : break;
1182 : : }
1183 : :
1184 : 684 : return planstate_tree_walker(planstate, ExecParallelRetrieveInstrumentation,
1185 : : instrumentation);
1186 : : }
1187 : :
1188 : : /*
1189 : : * Add up the workers' JIT instrumentation from dynamic shared memory.
1190 : : */
1191 : : static void
1192 : 0 : ExecParallelRetrieveJitInstrumentation(PlanState *planstate,
1193 : : SharedJitInstrumentation *shared_jit)
1194 : : {
1195 : : JitInstrumentation *combined;
1196 : : int ibytes;
1197 : :
1198 : : int n;
1199 : :
1200 : : /*
1201 : : * Accumulate worker JIT instrumentation into the combined JIT
1202 : : * instrumentation, allocating it if required.
1203 : : */
1204 [ # # ]: 0 : if (!planstate->state->es_jit_worker_instr)
1205 : 0 : planstate->state->es_jit_worker_instr =
1206 : 0 : MemoryContextAllocZero(planstate->state->es_query_cxt, sizeof(JitInstrumentation));
1207 : 0 : combined = planstate->state->es_jit_worker_instr;
1208 : :
1209 : : /* Accumulate all the workers' instrumentations. */
1210 [ # # ]: 0 : for (n = 0; n < shared_jit->num_workers; ++n)
1211 : 0 : InstrJitAgg(combined, &shared_jit->jit_instr[n]);
1212 : :
1213 : : /*
1214 : : * Store the per-worker detail.
1215 : : *
1216 : : * Similar to ExecParallelRetrieveInstrumentation(), allocate the
1217 : : * instrumentation in per-query context.
1218 : : */
1219 : 0 : ibytes = offsetof(SharedJitInstrumentation, jit_instr)
1220 : 0 : + mul_size(shared_jit->num_workers, sizeof(JitInstrumentation));
1221 : 0 : planstate->worker_jit_instrument =
1222 : 0 : MemoryContextAlloc(planstate->state->es_query_cxt, ibytes);
1223 : :
1224 : 0 : memcpy(planstate->worker_jit_instrument, shared_jit, ibytes);
1225 : 0 : }
1226 : :
1227 : : /*
1228 : : * Finish parallel execution. We wait for parallel workers to finish, and
1229 : : * accumulate their buffer/WAL usage.
1230 : : */
1231 : : void
1232 : 1232 : ExecParallelFinish(ParallelExecutorInfo *pei)
1233 : : {
1234 : 1232 : int nworkers = pei->pcxt->nworkers_launched;
1235 : : int i;
1236 : :
1237 : : /* Make this be a no-op if called twice in a row. */
1238 [ + + ]: 1232 : if (pei->finished)
1239 : 556 : return;
1240 : :
1241 : : /*
1242 : : * Detach from tuple queues ASAP, so that any still-active workers will
1243 : : * notice that no further results are wanted.
1244 : : */
1245 [ + - ]: 676 : if (pei->tqueue != NULL)
1246 : : {
1247 [ + + ]: 2460 : for (i = 0; i < nworkers; i++)
1248 : 1784 : shm_mq_detach(pei->tqueue[i]);
1249 : 676 : pfree(pei->tqueue);
1250 : 676 : pei->tqueue = NULL;
1251 : : }
1252 : :
1253 : : /*
1254 : : * While we're waiting for the workers to finish, let's get rid of the
1255 : : * tuple queue readers. (Any other local cleanup could be done here too.)
1256 : : */
1257 [ + + ]: 676 : if (pei->reader != NULL)
1258 : : {
1259 [ + + ]: 2448 : for (i = 0; i < nworkers; i++)
1260 : 1784 : DestroyTupleQueueReader(pei->reader[i]);
1261 : 664 : pfree(pei->reader);
1262 : 664 : pei->reader = NULL;
1263 : : }
1264 : :
1265 : : /* Now wait for the workers to finish. */
1266 : 676 : WaitForParallelWorkersToFinish(pei->pcxt);
1267 : :
1268 : : /*
1269 : : * Next, accumulate buffer/WAL usage. (This must wait for the workers to
1270 : : * finish, or we might get incomplete data.)
1271 : : */
1272 [ + + ]: 2460 : for (i = 0; i < nworkers; i++)
1273 : 1784 : InstrAccumParallelQuery(&pei->buffer_usage[i], &pei->wal_usage[i]);
1274 : :
1275 : 676 : pei->finished = true;
1276 : : }
1277 : :
1278 : : /*
1279 : : * Accumulate instrumentation, and then clean up whatever ParallelExecutorInfo
1280 : : * resources still exist after ExecParallelFinish. We separate these
1281 : : * routines because someone might want to examine the contents of the DSM
1282 : : * after ExecParallelFinish and before calling this routine.
1283 : : */
1284 : : void
1285 : 504 : ExecParallelCleanup(ParallelExecutorInfo *pei)
1286 : : {
1287 : : /* Accumulate instrumentation, if any. */
1288 [ + + ]: 504 : if (pei->instrumentation)
1289 : 120 : ExecParallelRetrieveInstrumentation(pei->planstate,
1290 : : pei->instrumentation);
1291 : :
1292 : : /* Accumulate JIT instrumentation, if any. */
1293 [ - + ]: 504 : if (pei->jit_instrumentation)
1294 : 0 : ExecParallelRetrieveJitInstrumentation(pei->planstate,
1295 : 0 : pei->jit_instrumentation);
1296 : :
1297 : : /* Free any serialized parameters. */
1298 [ + + ]: 504 : if (DsaPointerIsValid(pei->param_exec))
1299 : : {
1300 : 16 : dsa_free(pei->area, pei->param_exec);
1301 : 16 : pei->param_exec = InvalidDsaPointer;
1302 : : }
1303 [ + - ]: 504 : if (pei->area != NULL)
1304 : : {
1305 : 504 : dsa_detach(pei->area);
1306 : 504 : pei->area = NULL;
1307 : : }
1308 [ + - ]: 504 : if (pei->pcxt != NULL)
1309 : : {
1310 : 504 : DestroyParallelContext(pei->pcxt);
1311 : 504 : pei->pcxt = NULL;
1312 : : }
1313 : 504 : pfree(pei);
1314 : 504 : }
1315 : :
1316 : : /*
1317 : : * Create a DestReceiver to write tuples we produce to the shm_mq designated
1318 : : * for that purpose.
1319 : : */
1320 : : static DestReceiver *
1321 : 1792 : ExecParallelGetReceiver(dsm_segment *seg, shm_toc *toc)
1322 : : {
1323 : : char *mqspace;
1324 : : shm_mq *mq;
1325 : :
1326 : 1792 : mqspace = shm_toc_lookup(toc, PARALLEL_KEY_TUPLE_QUEUE, false);
1327 : 1792 : mqspace += ParallelWorkerNumber * PARALLEL_TUPLE_QUEUE_SIZE;
1328 : 1792 : mq = (shm_mq *) mqspace;
1329 : 1792 : shm_mq_set_sender(mq, MyProc);
1330 : 1792 : return CreateTupleQueueDestReceiver(shm_mq_attach(mq, seg, NULL));
1331 : : }
1332 : :
1333 : : /*
1334 : : * Create a QueryDesc for the PlannedStmt we are to execute, and return it.
1335 : : */
1336 : : static QueryDesc *
1337 : 1792 : ExecParallelGetQueryDesc(shm_toc *toc, DestReceiver *receiver,
1338 : : int instrument_options)
1339 : : {
1340 : : char *pstmtspace;
1341 : : char *paramspace;
1342 : : PlannedStmt *pstmt;
1343 : : ParamListInfo paramLI;
1344 : : char *queryString;
1345 : :
1346 : : /* Get the query string from shared memory */
1347 : 1792 : queryString = shm_toc_lookup(toc, PARALLEL_KEY_QUERY_TEXT, false);
1348 : :
1349 : : /* Reconstruct leader-supplied PlannedStmt. */
1350 : 1792 : pstmtspace = shm_toc_lookup(toc, PARALLEL_KEY_PLANNEDSTMT, false);
1351 : 1792 : pstmt = (PlannedStmt *) stringToNode(pstmtspace);
1352 : :
1353 : : /* Reconstruct ParamListInfo. */
1354 : 1792 : paramspace = shm_toc_lookup(toc, PARALLEL_KEY_PARAMLISTINFO, false);
1355 : 1792 : paramLI = RestoreParamList(¶mspace);
1356 : :
1357 : : /* Create a QueryDesc for the query. */
1358 : 1792 : return CreateQueryDesc(pstmt,
1359 : : queryString,
1360 : : GetActiveSnapshot(), InvalidSnapshot,
1361 : : receiver, paramLI, NULL, instrument_options);
1362 : : }
1363 : :
1364 : : /*
1365 : : * Copy instrumentation information from this node and its descendants into
1366 : : * dynamic shared memory, so that the parallel leader can retrieve it.
1367 : : */
1368 : : static bool
1369 : 1582 : ExecParallelReportInstrumentation(PlanState *planstate,
1370 : : SharedExecutorInstrumentation *instrumentation)
1371 : : {
1372 : : int i;
1373 : 1582 : int plan_node_id = planstate->plan->plan_node_id;
1374 : : NodeInstrumentation *instrument;
1375 : :
1376 : 1582 : InstrEndLoop(planstate->instrument);
1377 : :
1378 : : /*
1379 : : * If we shuffled the plan_node_id values in ps_instrument into sorted
1380 : : * order, we could use binary search here. This might matter someday if
1381 : : * we're pushing down sufficiently large plan trees. For now, do it the
1382 : : * slow, dumb way.
1383 : : */
1384 [ + - ]: 5205 : for (i = 0; i < instrumentation->num_plan_nodes; ++i)
1385 [ + + ]: 5205 : if (instrumentation->plan_node_id[i] == plan_node_id)
1386 : 1582 : break;
1387 [ - + ]: 1582 : if (i >= instrumentation->num_plan_nodes)
1388 [ # # ]: 0 : elog(ERROR, "plan node %d not found", plan_node_id);
1389 : :
1390 : : /*
1391 : : * Add our statistics to the per-node, per-worker totals. It's possible
1392 : : * that this could happen more than once if we relaunched workers.
1393 : : */
1394 : 1582 : instrument = GetInstrumentationArray(instrumentation);
1395 : 1582 : instrument += i * instrumentation->num_workers;
1396 : : Assert(IsParallelWorker());
1397 : : Assert(ParallelWorkerNumber < instrumentation->num_workers);
1398 : 1582 : InstrAggNode(&instrument[ParallelWorkerNumber], planstate->instrument);
1399 : :
1400 : 1582 : return planstate_tree_walker(planstate, ExecParallelReportInstrumentation,
1401 : : instrumentation);
1402 : : }
1403 : :
1404 : : /*
1405 : : * Initialize the PlanState and its descendants with the information
1406 : : * retrieved from shared memory. This has to be done once the PlanState
1407 : : * is allocated and initialized by executor; that is, after ExecutorStart().
1408 : : */
1409 : : static bool
1410 : 6805 : ExecParallelInitializeWorker(PlanState *planstate, ParallelWorkerContext *pwcxt)
1411 : : {
1412 [ - + ]: 6805 : if (planstate == NULL)
1413 : 0 : return false;
1414 : :
1415 [ + + + + : 6805 : switch (nodeTag(planstate))
- + + + +
+ + + - +
+ - + ]
1416 : : {
1417 : 3359 : case T_SeqScanState:
1418 [ + + ]: 3359 : if (planstate->plan->parallel_aware)
1419 : 2941 : ExecSeqScanInitializeWorker((SeqScanState *) planstate, pwcxt);
1420 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1421 : 3359 : ExecSeqScanInstrumentInitWorker((SeqScanState *) planstate, pwcxt);
1422 : 3359 : break;
1423 : 264 : case T_IndexScanState:
1424 [ + + ]: 264 : if (planstate->plan->parallel_aware)
1425 : 80 : ExecIndexScanInitializeWorker((IndexScanState *) planstate,
1426 : : pwcxt);
1427 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1428 : 264 : ExecIndexScanInstrumentInitWorker((IndexScanState *) planstate,
1429 : : pwcxt);
1430 : 264 : break;
1431 : 160 : case T_IndexOnlyScanState:
1432 [ + + ]: 160 : if (planstate->plan->parallel_aware)
1433 : 136 : ExecIndexOnlyScanInitializeWorker((IndexOnlyScanState *) planstate,
1434 : : pwcxt);
1435 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1436 : 160 : ExecIndexOnlyScanInstrumentInitWorker((IndexOnlyScanState *) planstate,
1437 : : pwcxt);
1438 : 160 : break;
1439 : 181 : case T_BitmapIndexScanState:
1440 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1441 : 181 : ExecBitmapIndexScanInitializeWorker((BitmapIndexScanState *) planstate,
1442 : : pwcxt);
1443 : 181 : break;
1444 : 0 : case T_ForeignScanState:
1445 [ # # ]: 0 : if (planstate->plan->parallel_aware)
1446 : 0 : ExecForeignScanInitializeWorker((ForeignScanState *) planstate,
1447 : : pwcxt);
1448 : 0 : break;
1449 : 64 : case T_TidRangeScanState:
1450 [ + - ]: 64 : if (planstate->plan->parallel_aware)
1451 : 64 : ExecTidRangeScanInitializeWorker((TidRangeScanState *) planstate,
1452 : : pwcxt);
1453 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1454 : 64 : ExecTidRangeScanInstrumentInitWorker((TidRangeScanState *) planstate,
1455 : : pwcxt);
1456 : 64 : break;
1457 : 258 : case T_AppendState:
1458 [ + + ]: 258 : if (planstate->plan->parallel_aware)
1459 : 218 : ExecAppendInitializeWorker((AppendState *) planstate, pwcxt);
1460 : 258 : break;
1461 : 1 : case T_CustomScanState:
1462 [ - + ]: 1 : if (planstate->plan->parallel_aware)
1463 : 0 : ExecCustomScanInitializeWorker((CustomScanState *) planstate,
1464 : : pwcxt);
1465 : 1 : break;
1466 : 181 : case T_BitmapHeapScanState:
1467 [ + + ]: 181 : if (planstate->plan->parallel_aware)
1468 : 180 : ExecBitmapHeapInitializeWorker((BitmapHeapScanState *) planstate,
1469 : : pwcxt);
1470 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1471 : 181 : ExecBitmapHeapInstrumentInitWorker((BitmapHeapScanState *) planstate,
1472 : : pwcxt);
1473 : 181 : break;
1474 : 373 : case T_HashJoinState:
1475 [ + + ]: 373 : if (planstate->plan->parallel_aware)
1476 : 213 : ExecHashJoinInitializeWorker((HashJoinState *) planstate,
1477 : : pwcxt);
1478 : 373 : break;
1479 : 373 : case T_HashState:
1480 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1481 : 373 : ExecHashInitializeWorker((HashState *) planstate, pwcxt);
1482 : 373 : break;
1483 : 317 : case T_SortState:
1484 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1485 : 317 : ExecSortInitializeWorker((SortState *) planstate, pwcxt);
1486 : 317 : break;
1487 : 0 : case T_IncrementalSortState:
1488 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1489 : 0 : ExecIncrementalSortInitializeWorker((IncrementalSortState *) planstate,
1490 : : pwcxt);
1491 : 0 : break;
1492 : 1114 : case T_AggState:
1493 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1494 : 1114 : ExecAggInitializeWorker((AggState *) planstate, pwcxt);
1495 : 1114 : break;
1496 : 8 : case T_MemoizeState:
1497 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1498 : 8 : ExecMemoizeInitializeWorker((MemoizeState *) planstate, pwcxt);
1499 : 8 : break;
1500 : 0 : case T_MaterialState:
1501 : : /* even when not parallel-aware, for EXPLAIN ANALYZE */
1502 : 0 : ExecMaterialInitializeWorker((MaterialState *) planstate, pwcxt);
1503 : 0 : break;
1504 : 152 : default:
1505 : 152 : break;
1506 : : }
1507 : :
1508 : 6805 : return planstate_tree_walker(planstate, ExecParallelInitializeWorker,
1509 : : pwcxt);
1510 : : }
1511 : :
1512 : : /*
1513 : : * Main entrypoint for parallel query worker processes.
1514 : : *
1515 : : * We reach this function from ParallelWorkerMain, so the setup necessary to
1516 : : * create a sensible parallel environment has already been done;
1517 : : * ParallelWorkerMain worries about stuff like the transaction state, combo
1518 : : * CID mappings, and GUC values, so we don't need to deal with any of that
1519 : : * here.
1520 : : *
1521 : : * Our job is to deal with concerns specific to the executor. The parallel
1522 : : * group leader will have stored a serialized PlannedStmt, and it's our job
1523 : : * to execute that plan and write the resulting tuples to the appropriate
1524 : : * tuple queue. Various bits of supporting information that we need in order
1525 : : * to do this are also stored in the dsm_segment and can be accessed through
1526 : : * the shm_toc.
1527 : : */
1528 : : void
1529 : 1792 : ParallelQueryMain(dsm_segment *seg, shm_toc *toc)
1530 : : {
1531 : : FixedParallelExecutorState *fpes;
1532 : : BufferUsage *buffer_usage;
1533 : : WalUsage *wal_usage;
1534 : : DestReceiver *receiver;
1535 : : QueryDesc *queryDesc;
1536 : : SharedExecutorInstrumentation *instrumentation;
1537 : : SharedJitInstrumentation *jit_instrumentation;
1538 : 1792 : int instrument_options = 0;
1539 : : void *area_space;
1540 : : dsa_area *area;
1541 : : ParallelWorkerContext pwcxt;
1542 : :
1543 : : /* Get fixed-size state. */
1544 : 1792 : fpes = shm_toc_lookup(toc, PARALLEL_KEY_EXECUTOR_FIXED, false);
1545 : :
1546 : : /* Set up DestReceiver, SharedExecutorInstrumentation, and QueryDesc. */
1547 : 1792 : receiver = ExecParallelGetReceiver(seg, toc);
1548 : 1792 : instrumentation = shm_toc_lookup(toc, PARALLEL_KEY_INSTRUMENTATION, true);
1549 [ + + ]: 1792 : if (instrumentation != NULL)
1550 : 483 : instrument_options = instrumentation->instrument_options;
1551 : 1792 : jit_instrumentation = shm_toc_lookup(toc, PARALLEL_KEY_JIT_INSTRUMENTATION,
1552 : : true);
1553 : 1792 : queryDesc = ExecParallelGetQueryDesc(toc, receiver, instrument_options);
1554 : :
1555 : : /* Setting debug_query_string for individual workers */
1556 : 1792 : debug_query_string = queryDesc->sourceText;
1557 : :
1558 : : /* Report workers' query for monitoring purposes */
1559 : 1792 : pgstat_report_activity(STATE_RUNNING, debug_query_string);
1560 : :
1561 : : /* Attach to the dynamic shared memory area. */
1562 : 1792 : area_space = shm_toc_lookup(toc, PARALLEL_KEY_DSA, false);
1563 : 1792 : area = dsa_attach_in_place(area_space, seg);
1564 : :
1565 : : /* Start up the executor */
1566 : 1792 : queryDesc->plannedstmt->jitFlags = fpes->jit_flags;
1567 : 1792 : ExecutorStart(queryDesc, fpes->eflags);
1568 : :
1569 : : /* Special executor initialization steps for parallel workers */
1570 : 1792 : queryDesc->planstate->state->es_query_dsa = area;
1571 [ + + ]: 1792 : if (DsaPointerIsValid(fpes->param_exec))
1572 : : {
1573 : : char *paramexec_space;
1574 : :
1575 : 48 : paramexec_space = dsa_get_address(area, fpes->param_exec);
1576 : 48 : RestoreParamExecParams(paramexec_space, queryDesc->estate);
1577 : : }
1578 : 1792 : pwcxt.toc = toc;
1579 : 1792 : pwcxt.seg = seg;
1580 : 1792 : ExecParallelInitializeWorker(queryDesc->planstate, &pwcxt);
1581 : :
1582 : : /* Pass down any tuple bound */
1583 : 1792 : ExecSetTupleBound(fpes->tuples_needed, queryDesc->planstate);
1584 : :
1585 : : /*
1586 : : * Prepare to track buffer/WAL usage during query execution.
1587 : : *
1588 : : * We do this after starting up the executor to match what happens in the
1589 : : * leader, which also doesn't count buffer accesses and WAL activity that
1590 : : * occur during executor startup.
1591 : : */
1592 : 1792 : InstrStartParallelQuery();
1593 : :
1594 : : /*
1595 : : * Run the plan. If we specified a tuple bound, be careful not to demand
1596 : : * more tuples than that.
1597 : : */
1598 : 1792 : ExecutorRun(queryDesc,
1599 : : ForwardScanDirection,
1600 : 1792 : fpes->tuples_needed < 0 ? (int64) 0 : fpes->tuples_needed);
1601 : :
1602 : : /* Shut down the executor */
1603 : 1784 : ExecutorFinish(queryDesc);
1604 : :
1605 : : /* Report buffer/WAL usage during parallel execution. */
1606 : 1784 : buffer_usage = shm_toc_lookup(toc, PARALLEL_KEY_BUFFER_USAGE, false);
1607 : 1784 : wal_usage = shm_toc_lookup(toc, PARALLEL_KEY_WAL_USAGE, false);
1608 : 1784 : InstrEndParallelQuery(&buffer_usage[ParallelWorkerNumber],
1609 : 1784 : &wal_usage[ParallelWorkerNumber]);
1610 : :
1611 : : /* Report instrumentation data if any instrumentation options are set. */
1612 [ + + ]: 1784 : if (instrumentation != NULL)
1613 : 483 : ExecParallelReportInstrumentation(queryDesc->planstate,
1614 : : instrumentation);
1615 : :
1616 : : /* Report JIT instrumentation data if any */
1617 [ - + - - ]: 1784 : if (queryDesc->estate->es_jit && jit_instrumentation != NULL)
1618 : : {
1619 : : Assert(ParallelWorkerNumber < jit_instrumentation->num_workers);
1620 : 0 : jit_instrumentation->jit_instr[ParallelWorkerNumber] =
1621 : 0 : queryDesc->estate->es_jit->instr;
1622 : : }
1623 : :
1624 : : /* Must do this after capturing instrumentation. */
1625 : 1784 : ExecutorEnd(queryDesc);
1626 : :
1627 : : /* Cleanup. */
1628 : 1784 : dsa_detach(area);
1629 : 1784 : FreeQueryDesc(queryDesc);
1630 : 1784 : receiver->rDestroy(receiver);
1631 : 1784 : }
|
