Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * parallel.c
4 : * Infrastructure for launching parallel workers
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : * IDENTIFICATION
10 : * src/backend/access/transam/parallel.c
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 :
15 : #include "postgres.h"
16 :
17 : #include "access/brin.h"
18 : #include "access/gin.h"
19 : #include "access/nbtree.h"
20 : #include "access/parallel.h"
21 : #include "access/session.h"
22 : #include "access/xact.h"
23 : #include "access/xlog.h"
24 : #include "catalog/index.h"
25 : #include "catalog/namespace.h"
26 : #include "catalog/pg_enum.h"
27 : #include "catalog/storage.h"
28 : #include "commands/async.h"
29 : #include "commands/vacuum.h"
30 : #include "executor/execParallel.h"
31 : #include "libpq/libpq.h"
32 : #include "libpq/pqformat.h"
33 : #include "libpq/pqmq.h"
34 : #include "miscadmin.h"
35 : #include "optimizer/optimizer.h"
36 : #include "pgstat.h"
37 : #include "storage/ipc.h"
38 : #include "storage/predicate.h"
39 : #include "storage/spin.h"
40 : #include "tcop/tcopprot.h"
41 : #include "utils/combocid.h"
42 : #include "utils/guc.h"
43 : #include "utils/inval.h"
44 : #include "utils/memutils.h"
45 : #include "utils/relmapper.h"
46 : #include "utils/snapmgr.h"
47 :
48 : /*
49 : * We don't want to waste a lot of memory on an error queue which, most of
50 : * the time, will process only a handful of small messages. However, it is
51 : * desirable to make it large enough that a typical ErrorResponse can be sent
52 : * without blocking. That way, a worker that errors out can write the whole
53 : * message into the queue and terminate without waiting for the user backend.
54 : */
55 : #define PARALLEL_ERROR_QUEUE_SIZE 16384
56 :
57 : /* Magic number for parallel context TOC. */
58 : #define PARALLEL_MAGIC 0x50477c7c
59 :
60 : /*
61 : * Magic numbers for per-context parallel state sharing. Higher-level code
62 : * should use smaller values, leaving these very large ones for use by this
63 : * module.
64 : */
65 : #define PARALLEL_KEY_FIXED UINT64CONST(0xFFFFFFFFFFFF0001)
66 : #define PARALLEL_KEY_ERROR_QUEUE UINT64CONST(0xFFFFFFFFFFFF0002)
67 : #define PARALLEL_KEY_LIBRARY UINT64CONST(0xFFFFFFFFFFFF0003)
68 : #define PARALLEL_KEY_GUC UINT64CONST(0xFFFFFFFFFFFF0004)
69 : #define PARALLEL_KEY_COMBO_CID UINT64CONST(0xFFFFFFFFFFFF0005)
70 : #define PARALLEL_KEY_TRANSACTION_SNAPSHOT UINT64CONST(0xFFFFFFFFFFFF0006)
71 : #define PARALLEL_KEY_ACTIVE_SNAPSHOT UINT64CONST(0xFFFFFFFFFFFF0007)
72 : #define PARALLEL_KEY_TRANSACTION_STATE UINT64CONST(0xFFFFFFFFFFFF0008)
73 : #define PARALLEL_KEY_ENTRYPOINT UINT64CONST(0xFFFFFFFFFFFF0009)
74 : #define PARALLEL_KEY_SESSION_DSM UINT64CONST(0xFFFFFFFFFFFF000A)
75 : #define PARALLEL_KEY_PENDING_SYNCS UINT64CONST(0xFFFFFFFFFFFF000B)
76 : #define PARALLEL_KEY_REINDEX_STATE UINT64CONST(0xFFFFFFFFFFFF000C)
77 : #define PARALLEL_KEY_RELMAPPER_STATE UINT64CONST(0xFFFFFFFFFFFF000D)
78 : #define PARALLEL_KEY_UNCOMMITTEDENUMS UINT64CONST(0xFFFFFFFFFFFF000E)
79 : #define PARALLEL_KEY_CLIENTCONNINFO UINT64CONST(0xFFFFFFFFFFFF000F)
80 :
81 : /* Fixed-size parallel state. */
82 : typedef struct FixedParallelState
83 : {
84 : /* Fixed-size state that workers must restore. */
85 : Oid database_id;
86 : Oid authenticated_user_id;
87 : Oid session_user_id;
88 : Oid outer_user_id;
89 : Oid current_user_id;
90 : Oid temp_namespace_id;
91 : Oid temp_toast_namespace_id;
92 : int sec_context;
93 : bool session_user_is_superuser;
94 : bool role_is_superuser;
95 : PGPROC *parallel_leader_pgproc;
96 : pid_t parallel_leader_pid;
97 : ProcNumber parallel_leader_proc_number;
98 : TimestampTz xact_ts;
99 : TimestampTz stmt_ts;
100 : SerializableXactHandle serializable_xact_handle;
101 :
102 : /* Mutex protects remaining fields. */
103 : slock_t mutex;
104 :
105 : /* Maximum XactLastRecEnd of any worker. */
106 : XLogRecPtr last_xlog_end;
107 : } FixedParallelState;
108 :
109 : /*
110 : * Our parallel worker number. We initialize this to -1, meaning that we are
111 : * not a parallel worker. In parallel workers, it will be set to a value >= 0
112 : * and < the number of workers before any user code is invoked; each parallel
113 : * worker will get a different parallel worker number.
114 : */
115 : int ParallelWorkerNumber = -1;
116 :
117 : /* Is there a parallel message pending which we need to receive? */
118 : volatile sig_atomic_t ParallelMessagePending = false;
119 :
120 : /* Are we initializing a parallel worker? */
121 : bool InitializingParallelWorker = false;
122 :
123 : /* Pointer to our fixed parallel state. */
124 : static FixedParallelState *MyFixedParallelState;
125 :
126 : /* List of active parallel contexts. */
127 : static dlist_head pcxt_list = DLIST_STATIC_INIT(pcxt_list);
128 :
129 : /* Backend-local copy of data from FixedParallelState. */
130 : static pid_t ParallelLeaderPid;
131 :
132 : /*
133 : * List of internal parallel worker entry points. We need this for
134 : * reasons explained in LookupParallelWorkerFunction(), below.
135 : */
136 : static const struct
137 : {
138 : const char *fn_name;
139 : parallel_worker_main_type fn_addr;
140 : } InternalParallelWorkers[] =
141 :
142 : {
143 : {
144 : "ParallelQueryMain", ParallelQueryMain
145 : },
146 : {
147 : "_bt_parallel_build_main", _bt_parallel_build_main
148 : },
149 : {
150 : "_brin_parallel_build_main", _brin_parallel_build_main
151 : },
152 : {
153 : "_gin_parallel_build_main", _gin_parallel_build_main
154 : },
155 : {
156 : "parallel_vacuum_main", parallel_vacuum_main
157 : }
158 : };
159 :
160 : /* Private functions. */
161 : static void ProcessParallelMessage(ParallelContext *pcxt, int i, StringInfo msg);
162 : static void WaitForParallelWorkersToExit(ParallelContext *pcxt);
163 : static parallel_worker_main_type LookupParallelWorkerFunction(const char *libraryname, const char *funcname);
164 : static void ParallelWorkerShutdown(int code, Datum arg);
165 :
166 :
167 : /*
168 : * Establish a new parallel context. This should be done after entering
169 : * parallel mode, and (unless there is an error) the context should be
170 : * destroyed before exiting the current subtransaction.
171 : */
172 : ParallelContext *
173 910 : CreateParallelContext(const char *library_name, const char *function_name,
174 : int nworkers)
175 : {
176 : MemoryContext oldcontext;
177 : ParallelContext *pcxt;
178 :
179 : /* It is unsafe to create a parallel context if not in parallel mode. */
180 : Assert(IsInParallelMode());
181 :
182 : /* Number of workers should be non-negative. */
183 : Assert(nworkers >= 0);
184 :
185 : /* We might be running in a short-lived memory context. */
186 910 : oldcontext = MemoryContextSwitchTo(TopTransactionContext);
187 :
188 : /* Initialize a new ParallelContext. */
189 910 : pcxt = palloc0(sizeof(ParallelContext));
190 910 : pcxt->subid = GetCurrentSubTransactionId();
191 910 : pcxt->nworkers = nworkers;
192 910 : pcxt->nworkers_to_launch = nworkers;
193 910 : pcxt->library_name = pstrdup(library_name);
194 910 : pcxt->function_name = pstrdup(function_name);
195 910 : pcxt->error_context_stack = error_context_stack;
196 910 : shm_toc_initialize_estimator(&pcxt->estimator);
197 910 : dlist_push_head(&pcxt_list, &pcxt->node);
198 :
199 : /* Restore previous memory context. */
200 910 : MemoryContextSwitchTo(oldcontext);
201 :
202 910 : return pcxt;
203 : }
204 :
205 : /*
206 : * Establish the dynamic shared memory segment for a parallel context and
207 : * copy state and other bookkeeping information that will be needed by
208 : * parallel workers into it.
209 : */
210 : void
211 910 : InitializeParallelDSM(ParallelContext *pcxt)
212 : {
213 : MemoryContext oldcontext;
214 910 : Size library_len = 0;
215 910 : Size guc_len = 0;
216 910 : Size combocidlen = 0;
217 910 : Size tsnaplen = 0;
218 910 : Size asnaplen = 0;
219 910 : Size tstatelen = 0;
220 910 : Size pendingsyncslen = 0;
221 910 : Size reindexlen = 0;
222 910 : Size relmapperlen = 0;
223 910 : Size uncommittedenumslen = 0;
224 910 : Size clientconninfolen = 0;
225 910 : Size segsize = 0;
226 : int i;
227 : FixedParallelState *fps;
228 910 : dsm_handle session_dsm_handle = DSM_HANDLE_INVALID;
229 910 : Snapshot transaction_snapshot = GetTransactionSnapshot();
230 910 : Snapshot active_snapshot = GetActiveSnapshot();
231 :
232 : /* We might be running in a very short-lived memory context. */
233 910 : oldcontext = MemoryContextSwitchTo(TopTransactionContext);
234 :
235 : /* Allow space to store the fixed-size parallel state. */
236 910 : shm_toc_estimate_chunk(&pcxt->estimator, sizeof(FixedParallelState));
237 910 : shm_toc_estimate_keys(&pcxt->estimator, 1);
238 :
239 : /*
240 : * If we manage to reach here while non-interruptible, it's unsafe to
241 : * launch any workers: we would fail to process interrupts sent by them.
242 : * We can deal with that edge case by pretending no workers were
243 : * requested.
244 : */
245 910 : if (!INTERRUPTS_CAN_BE_PROCESSED())
246 0 : pcxt->nworkers = 0;
247 :
248 : /*
249 : * Normally, the user will have requested at least one worker process, but
250 : * if by chance they have not, we can skip a bunch of things here.
251 : */
252 910 : if (pcxt->nworkers > 0)
253 : {
254 : /* Get (or create) the per-session DSM segment's handle. */
255 910 : session_dsm_handle = GetSessionDsmHandle();
256 :
257 : /*
258 : * If we weren't able to create a per-session DSM segment, then we can
259 : * continue but we can't safely launch any workers because their
260 : * record typmods would be incompatible so they couldn't exchange
261 : * tuples.
262 : */
263 910 : if (session_dsm_handle == DSM_HANDLE_INVALID)
264 0 : pcxt->nworkers = 0;
265 : }
266 :
267 910 : if (pcxt->nworkers > 0)
268 : {
269 : /* Estimate space for various kinds of state sharing. */
270 910 : library_len = EstimateLibraryStateSpace();
271 910 : shm_toc_estimate_chunk(&pcxt->estimator, library_len);
272 910 : guc_len = EstimateGUCStateSpace();
273 910 : shm_toc_estimate_chunk(&pcxt->estimator, guc_len);
274 910 : combocidlen = EstimateComboCIDStateSpace();
275 910 : shm_toc_estimate_chunk(&pcxt->estimator, combocidlen);
276 910 : if (IsolationUsesXactSnapshot())
277 : {
278 22 : tsnaplen = EstimateSnapshotSpace(transaction_snapshot);
279 22 : shm_toc_estimate_chunk(&pcxt->estimator, tsnaplen);
280 : }
281 910 : asnaplen = EstimateSnapshotSpace(active_snapshot);
282 910 : shm_toc_estimate_chunk(&pcxt->estimator, asnaplen);
283 910 : tstatelen = EstimateTransactionStateSpace();
284 910 : shm_toc_estimate_chunk(&pcxt->estimator, tstatelen);
285 910 : shm_toc_estimate_chunk(&pcxt->estimator, sizeof(dsm_handle));
286 910 : pendingsyncslen = EstimatePendingSyncsSpace();
287 910 : shm_toc_estimate_chunk(&pcxt->estimator, pendingsyncslen);
288 910 : reindexlen = EstimateReindexStateSpace();
289 910 : shm_toc_estimate_chunk(&pcxt->estimator, reindexlen);
290 910 : relmapperlen = EstimateRelationMapSpace();
291 910 : shm_toc_estimate_chunk(&pcxt->estimator, relmapperlen);
292 910 : uncommittedenumslen = EstimateUncommittedEnumsSpace();
293 910 : shm_toc_estimate_chunk(&pcxt->estimator, uncommittedenumslen);
294 910 : clientconninfolen = EstimateClientConnectionInfoSpace();
295 910 : shm_toc_estimate_chunk(&pcxt->estimator, clientconninfolen);
296 : /* If you add more chunks here, you probably need to add keys. */
297 910 : shm_toc_estimate_keys(&pcxt->estimator, 12);
298 :
299 : /* Estimate space need for error queues. */
300 : StaticAssertStmt(BUFFERALIGN(PARALLEL_ERROR_QUEUE_SIZE) ==
301 : PARALLEL_ERROR_QUEUE_SIZE,
302 : "parallel error queue size not buffer-aligned");
303 910 : shm_toc_estimate_chunk(&pcxt->estimator,
304 : mul_size(PARALLEL_ERROR_QUEUE_SIZE,
305 : pcxt->nworkers));
306 910 : shm_toc_estimate_keys(&pcxt->estimator, 1);
307 :
308 : /* Estimate how much we'll need for the entrypoint info. */
309 910 : shm_toc_estimate_chunk(&pcxt->estimator, strlen(pcxt->library_name) +
310 : strlen(pcxt->function_name) + 2);
311 910 : shm_toc_estimate_keys(&pcxt->estimator, 1);
312 : }
313 :
314 : /*
315 : * Create DSM and initialize with new table of contents. But if the user
316 : * didn't request any workers, then don't bother creating a dynamic shared
317 : * memory segment; instead, just use backend-private memory.
318 : *
319 : * Also, if we can't create a dynamic shared memory segment because the
320 : * maximum number of segments have already been created, then fall back to
321 : * backend-private memory, and plan not to use any workers. We hope this
322 : * won't happen very often, but it's better to abandon the use of
323 : * parallelism than to fail outright.
324 : */
325 910 : segsize = shm_toc_estimate(&pcxt->estimator);
326 910 : if (pcxt->nworkers > 0)
327 910 : pcxt->seg = dsm_create(segsize, DSM_CREATE_NULL_IF_MAXSEGMENTS);
328 910 : if (pcxt->seg != NULL)
329 910 : pcxt->toc = shm_toc_create(PARALLEL_MAGIC,
330 : dsm_segment_address(pcxt->seg),
331 : segsize);
332 : else
333 : {
334 0 : pcxt->nworkers = 0;
335 0 : pcxt->private_memory = MemoryContextAlloc(TopMemoryContext, segsize);
336 0 : pcxt->toc = shm_toc_create(PARALLEL_MAGIC, pcxt->private_memory,
337 : segsize);
338 : }
339 :
340 : /* Initialize fixed-size state in shared memory. */
341 : fps = (FixedParallelState *)
342 910 : shm_toc_allocate(pcxt->toc, sizeof(FixedParallelState));
343 910 : fps->database_id = MyDatabaseId;
344 910 : fps->authenticated_user_id = GetAuthenticatedUserId();
345 910 : fps->session_user_id = GetSessionUserId();
346 910 : fps->outer_user_id = GetCurrentRoleId();
347 910 : GetUserIdAndSecContext(&fps->current_user_id, &fps->sec_context);
348 910 : fps->session_user_is_superuser = GetSessionUserIsSuperuser();
349 910 : fps->role_is_superuser = current_role_is_superuser;
350 910 : GetTempNamespaceState(&fps->temp_namespace_id,
351 : &fps->temp_toast_namespace_id);
352 910 : fps->parallel_leader_pgproc = MyProc;
353 910 : fps->parallel_leader_pid = MyProcPid;
354 910 : fps->parallel_leader_proc_number = MyProcNumber;
355 910 : fps->xact_ts = GetCurrentTransactionStartTimestamp();
356 910 : fps->stmt_ts = GetCurrentStatementStartTimestamp();
357 910 : fps->serializable_xact_handle = ShareSerializableXact();
358 910 : SpinLockInit(&fps->mutex);
359 910 : fps->last_xlog_end = 0;
360 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_FIXED, fps);
361 :
362 : /* We can skip the rest of this if we're not budgeting for any workers. */
363 910 : if (pcxt->nworkers > 0)
364 : {
365 : char *libraryspace;
366 : char *gucspace;
367 : char *combocidspace;
368 : char *tsnapspace;
369 : char *asnapspace;
370 : char *tstatespace;
371 : char *pendingsyncsspace;
372 : char *reindexspace;
373 : char *relmapperspace;
374 : char *error_queue_space;
375 : char *session_dsm_handle_space;
376 : char *entrypointstate;
377 : char *uncommittedenumsspace;
378 : char *clientconninfospace;
379 : Size lnamelen;
380 :
381 : /* Serialize shared libraries we have loaded. */
382 910 : libraryspace = shm_toc_allocate(pcxt->toc, library_len);
383 910 : SerializeLibraryState(library_len, libraryspace);
384 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_LIBRARY, libraryspace);
385 :
386 : /* Serialize GUC settings. */
387 910 : gucspace = shm_toc_allocate(pcxt->toc, guc_len);
388 910 : SerializeGUCState(guc_len, gucspace);
389 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_GUC, gucspace);
390 :
391 : /* Serialize combo CID state. */
392 910 : combocidspace = shm_toc_allocate(pcxt->toc, combocidlen);
393 910 : SerializeComboCIDState(combocidlen, combocidspace);
394 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_COMBO_CID, combocidspace);
395 :
396 : /*
397 : * Serialize the transaction snapshot if the transaction isolation
398 : * level uses a transaction snapshot.
399 : */
400 910 : if (IsolationUsesXactSnapshot())
401 : {
402 22 : tsnapspace = shm_toc_allocate(pcxt->toc, tsnaplen);
403 22 : SerializeSnapshot(transaction_snapshot, tsnapspace);
404 22 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_TRANSACTION_SNAPSHOT,
405 : tsnapspace);
406 : }
407 :
408 : /* Serialize the active snapshot. */
409 910 : asnapspace = shm_toc_allocate(pcxt->toc, asnaplen);
410 910 : SerializeSnapshot(active_snapshot, asnapspace);
411 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_ACTIVE_SNAPSHOT, asnapspace);
412 :
413 : /* Provide the handle for per-session segment. */
414 910 : session_dsm_handle_space = shm_toc_allocate(pcxt->toc,
415 : sizeof(dsm_handle));
416 910 : *(dsm_handle *) session_dsm_handle_space = session_dsm_handle;
417 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_SESSION_DSM,
418 : session_dsm_handle_space);
419 :
420 : /* Serialize transaction state. */
421 910 : tstatespace = shm_toc_allocate(pcxt->toc, tstatelen);
422 910 : SerializeTransactionState(tstatelen, tstatespace);
423 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_TRANSACTION_STATE, tstatespace);
424 :
425 : /* Serialize pending syncs. */
426 910 : pendingsyncsspace = shm_toc_allocate(pcxt->toc, pendingsyncslen);
427 910 : SerializePendingSyncs(pendingsyncslen, pendingsyncsspace);
428 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_PENDING_SYNCS,
429 : pendingsyncsspace);
430 :
431 : /* Serialize reindex state. */
432 910 : reindexspace = shm_toc_allocate(pcxt->toc, reindexlen);
433 910 : SerializeReindexState(reindexlen, reindexspace);
434 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_REINDEX_STATE, reindexspace);
435 :
436 : /* Serialize relmapper state. */
437 910 : relmapperspace = shm_toc_allocate(pcxt->toc, relmapperlen);
438 910 : SerializeRelationMap(relmapperlen, relmapperspace);
439 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_RELMAPPER_STATE,
440 : relmapperspace);
441 :
442 : /* Serialize uncommitted enum state. */
443 910 : uncommittedenumsspace = shm_toc_allocate(pcxt->toc,
444 : uncommittedenumslen);
445 910 : SerializeUncommittedEnums(uncommittedenumsspace, uncommittedenumslen);
446 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_UNCOMMITTEDENUMS,
447 : uncommittedenumsspace);
448 :
449 : /* Serialize our ClientConnectionInfo. */
450 910 : clientconninfospace = shm_toc_allocate(pcxt->toc, clientconninfolen);
451 910 : SerializeClientConnectionInfo(clientconninfolen, clientconninfospace);
452 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_CLIENTCONNINFO,
453 : clientconninfospace);
454 :
455 : /* Allocate space for worker information. */
456 910 : pcxt->worker = palloc0(sizeof(ParallelWorkerInfo) * pcxt->nworkers);
457 :
458 : /*
459 : * Establish error queues in dynamic shared memory.
460 : *
461 : * These queues should be used only for transmitting ErrorResponse,
462 : * NoticeResponse, and NotifyResponse protocol messages. Tuple data
463 : * should be transmitted via separate (possibly larger?) queues.
464 : */
465 : error_queue_space =
466 910 : shm_toc_allocate(pcxt->toc,
467 : mul_size(PARALLEL_ERROR_QUEUE_SIZE,
468 910 : pcxt->nworkers));
469 2910 : for (i = 0; i < pcxt->nworkers; ++i)
470 : {
471 : char *start;
472 : shm_mq *mq;
473 :
474 2000 : start = error_queue_space + i * PARALLEL_ERROR_QUEUE_SIZE;
475 2000 : mq = shm_mq_create(start, PARALLEL_ERROR_QUEUE_SIZE);
476 2000 : shm_mq_set_receiver(mq, MyProc);
477 2000 : pcxt->worker[i].error_mqh = shm_mq_attach(mq, pcxt->seg, NULL);
478 : }
479 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_ERROR_QUEUE, error_queue_space);
480 :
481 : /*
482 : * Serialize entrypoint information. It's unsafe to pass function
483 : * pointers across processes, as the function pointer may be different
484 : * in each process in EXEC_BACKEND builds, so we always pass library
485 : * and function name. (We use library name "postgres" for functions
486 : * in the core backend.)
487 : */
488 910 : lnamelen = strlen(pcxt->library_name);
489 910 : entrypointstate = shm_toc_allocate(pcxt->toc, lnamelen +
490 910 : strlen(pcxt->function_name) + 2);
491 910 : strcpy(entrypointstate, pcxt->library_name);
492 910 : strcpy(entrypointstate + lnamelen + 1, pcxt->function_name);
493 910 : shm_toc_insert(pcxt->toc, PARALLEL_KEY_ENTRYPOINT, entrypointstate);
494 : }
495 :
496 : /* Update nworkers_to_launch, in case we changed nworkers above. */
497 910 : pcxt->nworkers_to_launch = pcxt->nworkers;
498 :
499 : /* Restore previous memory context. */
500 910 : MemoryContextSwitchTo(oldcontext);
501 910 : }
502 :
503 : /*
504 : * Reinitialize the dynamic shared memory segment for a parallel context such
505 : * that we could launch workers for it again.
506 : */
507 : void
508 270 : ReinitializeParallelDSM(ParallelContext *pcxt)
509 : {
510 : FixedParallelState *fps;
511 :
512 : /* Wait for any old workers to exit. */
513 270 : if (pcxt->nworkers_launched > 0)
514 : {
515 270 : WaitForParallelWorkersToFinish(pcxt);
516 270 : WaitForParallelWorkersToExit(pcxt);
517 270 : pcxt->nworkers_launched = 0;
518 270 : if (pcxt->known_attached_workers)
519 : {
520 270 : pfree(pcxt->known_attached_workers);
521 270 : pcxt->known_attached_workers = NULL;
522 270 : pcxt->nknown_attached_workers = 0;
523 : }
524 : }
525 :
526 : /* Reset a few bits of fixed parallel state to a clean state. */
527 270 : fps = shm_toc_lookup(pcxt->toc, PARALLEL_KEY_FIXED, false);
528 270 : fps->last_xlog_end = 0;
529 :
530 : /* Recreate error queues (if they exist). */
531 270 : if (pcxt->nworkers > 0)
532 : {
533 : char *error_queue_space;
534 : int i;
535 :
536 : error_queue_space =
537 270 : shm_toc_lookup(pcxt->toc, PARALLEL_KEY_ERROR_QUEUE, false);
538 1108 : for (i = 0; i < pcxt->nworkers; ++i)
539 : {
540 : char *start;
541 : shm_mq *mq;
542 :
543 838 : start = error_queue_space + i * PARALLEL_ERROR_QUEUE_SIZE;
544 838 : mq = shm_mq_create(start, PARALLEL_ERROR_QUEUE_SIZE);
545 838 : shm_mq_set_receiver(mq, MyProc);
546 838 : pcxt->worker[i].error_mqh = shm_mq_attach(mq, pcxt->seg, NULL);
547 : }
548 : }
549 270 : }
550 :
551 : /*
552 : * Reinitialize parallel workers for a parallel context such that we could
553 : * launch a different number of workers. This is required for cases where
554 : * we need to reuse the same DSM segment, but the number of workers can
555 : * vary from run-to-run.
556 : */
557 : void
558 46 : ReinitializeParallelWorkers(ParallelContext *pcxt, int nworkers_to_launch)
559 : {
560 : /*
561 : * The number of workers that need to be launched must be less than the
562 : * number of workers with which the parallel context is initialized. But
563 : * the caller might not know that InitializeParallelDSM reduced nworkers,
564 : * so just silently trim the request.
565 : */
566 46 : pcxt->nworkers_to_launch = Min(pcxt->nworkers, nworkers_to_launch);
567 46 : }
568 :
569 : /*
570 : * Launch parallel workers.
571 : */
572 : void
573 1180 : LaunchParallelWorkers(ParallelContext *pcxt)
574 : {
575 : MemoryContext oldcontext;
576 : BackgroundWorker worker;
577 : int i;
578 1180 : bool any_registrations_failed = false;
579 :
580 : /* Skip this if we have no workers. */
581 1180 : if (pcxt->nworkers == 0 || pcxt->nworkers_to_launch == 0)
582 0 : return;
583 :
584 : /* We need to be a lock group leader. */
585 1180 : BecomeLockGroupLeader();
586 :
587 : /* If we do have workers, we'd better have a DSM segment. */
588 : Assert(pcxt->seg != NULL);
589 :
590 : /* We might be running in a short-lived memory context. */
591 1180 : oldcontext = MemoryContextSwitchTo(TopTransactionContext);
592 :
593 : /* Configure a worker. */
594 1180 : memset(&worker, 0, sizeof(worker));
595 1180 : snprintf(worker.bgw_name, BGW_MAXLEN, "parallel worker for PID %d",
596 : MyProcPid);
597 1180 : snprintf(worker.bgw_type, BGW_MAXLEN, "parallel worker");
598 1180 : worker.bgw_flags =
599 : BGWORKER_SHMEM_ACCESS | BGWORKER_BACKEND_DATABASE_CONNECTION
600 : | BGWORKER_CLASS_PARALLEL;
601 1180 : worker.bgw_start_time = BgWorkerStart_ConsistentState;
602 1180 : worker.bgw_restart_time = BGW_NEVER_RESTART;
603 1180 : sprintf(worker.bgw_library_name, "postgres");
604 1180 : sprintf(worker.bgw_function_name, "ParallelWorkerMain");
605 1180 : worker.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(pcxt->seg));
606 1180 : worker.bgw_notify_pid = MyProcPid;
607 :
608 : /*
609 : * Start workers.
610 : *
611 : * The caller must be able to tolerate ending up with fewer workers than
612 : * expected, so there is no need to throw an error here if registration
613 : * fails. It wouldn't help much anyway, because registering the worker in
614 : * no way guarantees that it will start up and initialize successfully.
615 : */
616 4014 : for (i = 0; i < pcxt->nworkers_to_launch; ++i)
617 : {
618 2834 : memcpy(worker.bgw_extra, &i, sizeof(int));
619 5610 : if (!any_registrations_failed &&
620 2776 : RegisterDynamicBackgroundWorker(&worker,
621 2776 : &pcxt->worker[i].bgwhandle))
622 : {
623 2746 : shm_mq_set_handle(pcxt->worker[i].error_mqh,
624 2746 : pcxt->worker[i].bgwhandle);
625 2746 : pcxt->nworkers_launched++;
626 : }
627 : else
628 : {
629 : /*
630 : * If we weren't able to register the worker, then we've bumped up
631 : * against the max_worker_processes limit, and future
632 : * registrations will probably fail too, so arrange to skip them.
633 : * But we still have to execute this code for the remaining slots
634 : * to make sure that we forget about the error queues we budgeted
635 : * for those workers. Otherwise, we'll wait for them to start,
636 : * but they never will.
637 : */
638 88 : any_registrations_failed = true;
639 88 : pcxt->worker[i].bgwhandle = NULL;
640 88 : shm_mq_detach(pcxt->worker[i].error_mqh);
641 88 : pcxt->worker[i].error_mqh = NULL;
642 : }
643 : }
644 :
645 : /*
646 : * Now that nworkers_launched has taken its final value, we can initialize
647 : * known_attached_workers.
648 : */
649 1180 : if (pcxt->nworkers_launched > 0)
650 : {
651 1160 : pcxt->known_attached_workers =
652 1160 : palloc0(sizeof(bool) * pcxt->nworkers_launched);
653 1160 : pcxt->nknown_attached_workers = 0;
654 : }
655 :
656 : /* Restore previous memory context. */
657 1180 : MemoryContextSwitchTo(oldcontext);
658 : }
659 :
660 : /*
661 : * Wait for all workers to attach to their error queues, and throw an error if
662 : * any worker fails to do this.
663 : *
664 : * Callers can assume that if this function returns successfully, then the
665 : * number of workers given by pcxt->nworkers_launched have initialized and
666 : * attached to their error queues. Whether or not these workers are guaranteed
667 : * to still be running depends on what code the caller asked them to run;
668 : * this function does not guarantee that they have not exited. However, it
669 : * does guarantee that any workers which exited must have done so cleanly and
670 : * after successfully performing the work with which they were tasked.
671 : *
672 : * If this function is not called, then some of the workers that were launched
673 : * may not have been started due to a fork() failure, or may have exited during
674 : * early startup prior to attaching to the error queue, so nworkers_launched
675 : * cannot be viewed as completely reliable. It will never be less than the
676 : * number of workers which actually started, but it might be more. Any workers
677 : * that failed to start will still be discovered by
678 : * WaitForParallelWorkersToFinish and an error will be thrown at that time,
679 : * provided that function is eventually reached.
680 : *
681 : * In general, the leader process should do as much work as possible before
682 : * calling this function. fork() failures and other early-startup failures
683 : * are very uncommon, and having the leader sit idle when it could be doing
684 : * useful work is undesirable. However, if the leader needs to wait for
685 : * all of its workers or for a specific worker, it may want to call this
686 : * function before doing so. If not, it must make some other provision for
687 : * the failure-to-start case, lest it wait forever. On the other hand, a
688 : * leader which never waits for a worker that might not be started yet, or
689 : * at least never does so prior to WaitForParallelWorkersToFinish(), need not
690 : * call this function at all.
691 : */
692 : void
693 156 : WaitForParallelWorkersToAttach(ParallelContext *pcxt)
694 : {
695 : int i;
696 :
697 : /* Skip this if we have no launched workers. */
698 156 : if (pcxt->nworkers_launched == 0)
699 0 : return;
700 :
701 : for (;;)
702 : {
703 : /*
704 : * This will process any parallel messages that are pending and it may
705 : * also throw an error propagated from a worker.
706 : */
707 12641354 : CHECK_FOR_INTERRUPTS();
708 :
709 25624304 : for (i = 0; i < pcxt->nworkers_launched; ++i)
710 : {
711 : BgwHandleStatus status;
712 : shm_mq *mq;
713 : int rc;
714 : pid_t pid;
715 :
716 12982950 : if (pcxt->known_attached_workers[i])
717 341610 : continue;
718 :
719 : /*
720 : * If error_mqh is NULL, then the worker has already exited
721 : * cleanly.
722 : */
723 12641340 : if (pcxt->worker[i].error_mqh == NULL)
724 : {
725 0 : pcxt->known_attached_workers[i] = true;
726 0 : ++pcxt->nknown_attached_workers;
727 0 : continue;
728 : }
729 :
730 12641340 : status = GetBackgroundWorkerPid(pcxt->worker[i].bgwhandle, &pid);
731 12641340 : if (status == BGWH_STARTED)
732 : {
733 : /* Has the worker attached to the error queue? */
734 12641192 : mq = shm_mq_get_queue(pcxt->worker[i].error_mqh);
735 12641192 : if (shm_mq_get_sender(mq) != NULL)
736 : {
737 : /* Yes, so it is known to be attached. */
738 136 : pcxt->known_attached_workers[i] = true;
739 136 : ++pcxt->nknown_attached_workers;
740 : }
741 : }
742 148 : else if (status == BGWH_STOPPED)
743 : {
744 : /*
745 : * If the worker stopped without attaching to the error queue,
746 : * throw an error.
747 : */
748 0 : mq = shm_mq_get_queue(pcxt->worker[i].error_mqh);
749 0 : if (shm_mq_get_sender(mq) == NULL)
750 0 : ereport(ERROR,
751 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
752 : errmsg("parallel worker failed to initialize"),
753 : errhint("More details may be available in the server log.")));
754 :
755 0 : pcxt->known_attached_workers[i] = true;
756 0 : ++pcxt->nknown_attached_workers;
757 : }
758 : else
759 : {
760 : /*
761 : * Worker not yet started, so we must wait. The postmaster
762 : * will notify us if the worker's state changes. Our latch
763 : * might also get set for some other reason, but if so we'll
764 : * just end up waiting for the same worker again.
765 : */
766 148 : rc = WaitLatch(MyLatch,
767 : WL_LATCH_SET | WL_EXIT_ON_PM_DEATH,
768 : -1, WAIT_EVENT_BGWORKER_STARTUP);
769 :
770 148 : if (rc & WL_LATCH_SET)
771 148 : ResetLatch(MyLatch);
772 : }
773 : }
774 :
775 : /* If all workers are known to have started, we're done. */
776 12641354 : if (pcxt->nknown_attached_workers >= pcxt->nworkers_launched)
777 : {
778 : Assert(pcxt->nknown_attached_workers == pcxt->nworkers_launched);
779 156 : break;
780 : }
781 : }
782 : }
783 :
784 : /*
785 : * Wait for all workers to finish computing.
786 : *
787 : * Even if the parallel operation seems to have completed successfully, it's
788 : * important to call this function afterwards. We must not miss any errors
789 : * the workers may have thrown during the parallel operation, or any that they
790 : * may yet throw while shutting down.
791 : *
792 : * Also, we want to update our notion of XactLastRecEnd based on worker
793 : * feedback.
794 : */
795 : void
796 2838 : WaitForParallelWorkersToFinish(ParallelContext *pcxt)
797 : {
798 : for (;;)
799 1400 : {
800 2838 : bool anyone_alive = false;
801 2838 : int nfinished = 0;
802 : int i;
803 :
804 : /*
805 : * This will process any parallel messages that are pending, which may
806 : * change the outcome of the loop that follows. It may also throw an
807 : * error propagated from a worker.
808 : */
809 2838 : CHECK_FOR_INTERRUPTS();
810 :
811 9800 : for (i = 0; i < pcxt->nworkers_launched; ++i)
812 : {
813 : /*
814 : * If error_mqh is NULL, then the worker has already exited
815 : * cleanly. If we have received a message through error_mqh from
816 : * the worker, we know it started up cleanly, and therefore we're
817 : * certain to be notified when it exits.
818 : */
819 7010 : if (pcxt->worker[i].error_mqh == NULL)
820 5540 : ++nfinished;
821 1470 : else if (pcxt->known_attached_workers[i])
822 : {
823 48 : anyone_alive = true;
824 48 : break;
825 : }
826 : }
827 :
828 2838 : if (!anyone_alive)
829 : {
830 : /* If all workers are known to have finished, we're done. */
831 2790 : if (nfinished >= pcxt->nworkers_launched)
832 : {
833 : Assert(nfinished == pcxt->nworkers_launched);
834 1438 : break;
835 : }
836 :
837 : /*
838 : * We didn't detect any living workers, but not all workers are
839 : * known to have exited cleanly. Either not all workers have
840 : * launched yet, or maybe some of them failed to start or
841 : * terminated abnormally.
842 : */
843 4748 : for (i = 0; i < pcxt->nworkers_launched; ++i)
844 : {
845 : pid_t pid;
846 : shm_mq *mq;
847 :
848 : /*
849 : * If the worker is BGWH_NOT_YET_STARTED or BGWH_STARTED, we
850 : * should just keep waiting. If it is BGWH_STOPPED, then
851 : * further investigation is needed.
852 : */
853 3396 : if (pcxt->worker[i].error_mqh == NULL ||
854 2844 : pcxt->worker[i].bgwhandle == NULL ||
855 1422 : GetBackgroundWorkerPid(pcxt->worker[i].bgwhandle,
856 : &pid) != BGWH_STOPPED)
857 3396 : continue;
858 :
859 : /*
860 : * Check whether the worker ended up stopped without ever
861 : * attaching to the error queue. If so, the postmaster was
862 : * unable to fork the worker or it exited without initializing
863 : * properly. We must throw an error, since the caller may
864 : * have been expecting the worker to do some work before
865 : * exiting.
866 : */
867 0 : mq = shm_mq_get_queue(pcxt->worker[i].error_mqh);
868 0 : if (shm_mq_get_sender(mq) == NULL)
869 0 : ereport(ERROR,
870 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
871 : errmsg("parallel worker failed to initialize"),
872 : errhint("More details may be available in the server log.")));
873 :
874 : /*
875 : * The worker is stopped, but is attached to the error queue.
876 : * Unless there's a bug somewhere, this will only happen when
877 : * the worker writes messages and terminates after the
878 : * CHECK_FOR_INTERRUPTS() near the top of this function and
879 : * before the call to GetBackgroundWorkerPid(). In that case,
880 : * or latch should have been set as well and the right things
881 : * will happen on the next pass through the loop.
882 : */
883 : }
884 : }
885 :
886 1400 : (void) WaitLatch(MyLatch, WL_LATCH_SET | WL_EXIT_ON_PM_DEATH, -1,
887 : WAIT_EVENT_PARALLEL_FINISH);
888 1400 : ResetLatch(MyLatch);
889 : }
890 :
891 1438 : if (pcxt->toc != NULL)
892 : {
893 : FixedParallelState *fps;
894 :
895 1438 : fps = shm_toc_lookup(pcxt->toc, PARALLEL_KEY_FIXED, false);
896 1438 : if (fps->last_xlog_end > XactLastRecEnd)
897 32 : XactLastRecEnd = fps->last_xlog_end;
898 : }
899 1438 : }
900 :
901 : /*
902 : * Wait for all workers to exit.
903 : *
904 : * This function ensures that workers have been completely shutdown. The
905 : * difference between WaitForParallelWorkersToFinish and this function is
906 : * that the former just ensures that last message sent by a worker backend is
907 : * received by the leader backend whereas this ensures the complete shutdown.
908 : */
909 : static void
910 1180 : WaitForParallelWorkersToExit(ParallelContext *pcxt)
911 : {
912 : int i;
913 :
914 : /* Wait until the workers actually die. */
915 3926 : for (i = 0; i < pcxt->nworkers_launched; ++i)
916 : {
917 : BgwHandleStatus status;
918 :
919 2746 : if (pcxt->worker == NULL || pcxt->worker[i].bgwhandle == NULL)
920 0 : continue;
921 :
922 2746 : status = WaitForBackgroundWorkerShutdown(pcxt->worker[i].bgwhandle);
923 :
924 : /*
925 : * If the postmaster kicked the bucket, we have no chance of cleaning
926 : * up safely -- we won't be able to tell when our workers are actually
927 : * dead. This doesn't necessitate a PANIC since they will all abort
928 : * eventually, but we can't safely continue this session.
929 : */
930 2746 : if (status == BGWH_POSTMASTER_DIED)
931 0 : ereport(FATAL,
932 : (errcode(ERRCODE_ADMIN_SHUTDOWN),
933 : errmsg("postmaster exited during a parallel transaction")));
934 :
935 : /* Release memory. */
936 2746 : pfree(pcxt->worker[i].bgwhandle);
937 2746 : pcxt->worker[i].bgwhandle = NULL;
938 : }
939 1180 : }
940 :
941 : /*
942 : * Destroy a parallel context.
943 : *
944 : * If expecting a clean exit, you should use WaitForParallelWorkersToFinish()
945 : * first, before calling this function. When this function is invoked, any
946 : * remaining workers are forcibly killed; the dynamic shared memory segment
947 : * is unmapped; and we then wait (uninterruptibly) for the workers to exit.
948 : */
949 : void
950 910 : DestroyParallelContext(ParallelContext *pcxt)
951 : {
952 : int i;
953 :
954 : /*
955 : * Be careful about order of operations here! We remove the parallel
956 : * context from the list before we do anything else; otherwise, if an
957 : * error occurs during a subsequent step, we might try to nuke it again
958 : * from AtEOXact_Parallel or AtEOSubXact_Parallel.
959 : */
960 910 : dlist_delete(&pcxt->node);
961 :
962 : /* Kill each worker in turn, and forget their error queues. */
963 910 : if (pcxt->worker != NULL)
964 : {
965 2824 : for (i = 0; i < pcxt->nworkers_launched; ++i)
966 : {
967 1914 : if (pcxt->worker[i].error_mqh != NULL)
968 : {
969 12 : TerminateBackgroundWorker(pcxt->worker[i].bgwhandle);
970 :
971 12 : shm_mq_detach(pcxt->worker[i].error_mqh);
972 12 : pcxt->worker[i].error_mqh = NULL;
973 : }
974 : }
975 : }
976 :
977 : /*
978 : * If we have allocated a shared memory segment, detach it. This will
979 : * implicitly detach the error queues, and any other shared memory queues,
980 : * stored there.
981 : */
982 910 : if (pcxt->seg != NULL)
983 : {
984 910 : dsm_detach(pcxt->seg);
985 910 : pcxt->seg = NULL;
986 : }
987 :
988 : /*
989 : * If this parallel context is actually in backend-private memory rather
990 : * than shared memory, free that memory instead.
991 : */
992 910 : if (pcxt->private_memory != NULL)
993 : {
994 0 : pfree(pcxt->private_memory);
995 0 : pcxt->private_memory = NULL;
996 : }
997 :
998 : /*
999 : * We can't finish transaction commit or abort until all of the workers
1000 : * have exited. This means, in particular, that we can't respond to
1001 : * interrupts at this stage.
1002 : */
1003 910 : HOLD_INTERRUPTS();
1004 910 : WaitForParallelWorkersToExit(pcxt);
1005 910 : RESUME_INTERRUPTS();
1006 :
1007 : /* Free the worker array itself. */
1008 910 : if (pcxt->worker != NULL)
1009 : {
1010 910 : pfree(pcxt->worker);
1011 910 : pcxt->worker = NULL;
1012 : }
1013 :
1014 : /* Free memory. */
1015 910 : pfree(pcxt->library_name);
1016 910 : pfree(pcxt->function_name);
1017 910 : pfree(pcxt);
1018 910 : }
1019 :
1020 : /*
1021 : * Are there any parallel contexts currently active?
1022 : */
1023 : bool
1024 0 : ParallelContextActive(void)
1025 : {
1026 0 : return !dlist_is_empty(&pcxt_list);
1027 : }
1028 :
1029 : /*
1030 : * Handle receipt of an interrupt indicating a parallel worker message.
1031 : *
1032 : * Note: this is called within a signal handler! All we can do is set
1033 : * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
1034 : * ProcessParallelMessages().
1035 : */
1036 : void
1037 3784 : HandleParallelMessageInterrupt(void)
1038 : {
1039 3784 : InterruptPending = true;
1040 3784 : ParallelMessagePending = true;
1041 3784 : SetLatch(MyLatch);
1042 3784 : }
1043 :
1044 : /*
1045 : * Process any queued protocol messages received from parallel workers.
1046 : */
1047 : void
1048 3716 : ProcessParallelMessages(void)
1049 : {
1050 : dlist_iter iter;
1051 : MemoryContext oldcontext;
1052 :
1053 : static MemoryContext hpm_context = NULL;
1054 :
1055 : /*
1056 : * This is invoked from ProcessInterrupts(), and since some of the
1057 : * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
1058 : * for recursive calls if more signals are received while this runs. It's
1059 : * unclear that recursive entry would be safe, and it doesn't seem useful
1060 : * even if it is safe, so let's block interrupts until done.
1061 : */
1062 3716 : HOLD_INTERRUPTS();
1063 :
1064 : /*
1065 : * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
1066 : * don't want to risk leaking data into long-lived contexts, so let's do
1067 : * our work here in a private context that we can reset on each use.
1068 : */
1069 3716 : if (hpm_context == NULL) /* first time through? */
1070 138 : hpm_context = AllocSetContextCreate(TopMemoryContext,
1071 : "ProcessParallelMessages",
1072 : ALLOCSET_DEFAULT_SIZES);
1073 : else
1074 3578 : MemoryContextReset(hpm_context);
1075 :
1076 3716 : oldcontext = MemoryContextSwitchTo(hpm_context);
1077 :
1078 : /* OK to process messages. Reset the flag saying there are more to do. */
1079 3716 : ParallelMessagePending = false;
1080 :
1081 7524 : dlist_foreach(iter, &pcxt_list)
1082 : {
1083 : ParallelContext *pcxt;
1084 : int i;
1085 :
1086 3820 : pcxt = dlist_container(ParallelContext, node, iter.cur);
1087 3820 : if (pcxt->worker == NULL)
1088 0 : continue;
1089 :
1090 15776 : for (i = 0; i < pcxt->nworkers_launched; ++i)
1091 : {
1092 : /*
1093 : * Read as many messages as we can from each worker, but stop when
1094 : * either (1) the worker's error queue goes away, which can happen
1095 : * if we receive a Terminate message from the worker; or (2) no
1096 : * more messages can be read from the worker without blocking.
1097 : */
1098 14702 : while (pcxt->worker[i].error_mqh != NULL)
1099 : {
1100 : shm_mq_result res;
1101 : Size nbytes;
1102 : void *data;
1103 :
1104 6546 : res = shm_mq_receive(pcxt->worker[i].error_mqh, &nbytes,
1105 : &data, true);
1106 6546 : if (res == SHM_MQ_WOULD_BLOCK)
1107 3800 : break;
1108 2746 : else if (res == SHM_MQ_SUCCESS)
1109 : {
1110 : StringInfoData msg;
1111 :
1112 2746 : initStringInfo(&msg);
1113 2746 : appendBinaryStringInfo(&msg, data, nbytes);
1114 2746 : ProcessParallelMessage(pcxt, i, &msg);
1115 2734 : pfree(msg.data);
1116 : }
1117 : else
1118 0 : ereport(ERROR,
1119 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1120 : errmsg("lost connection to parallel worker")));
1121 : }
1122 : }
1123 : }
1124 :
1125 3704 : MemoryContextSwitchTo(oldcontext);
1126 :
1127 : /* Might as well clear the context on our way out */
1128 3704 : MemoryContextReset(hpm_context);
1129 :
1130 3704 : RESUME_INTERRUPTS();
1131 3704 : }
1132 :
1133 : /*
1134 : * Process a single protocol message received from a single parallel worker.
1135 : */
1136 : static void
1137 2746 : ProcessParallelMessage(ParallelContext *pcxt, int i, StringInfo msg)
1138 : {
1139 : char msgtype;
1140 :
1141 2746 : if (pcxt->known_attached_workers != NULL &&
1142 2746 : !pcxt->known_attached_workers[i])
1143 : {
1144 2610 : pcxt->known_attached_workers[i] = true;
1145 2610 : pcxt->nknown_attached_workers++;
1146 : }
1147 :
1148 2746 : msgtype = pq_getmsgbyte(msg);
1149 :
1150 2746 : switch (msgtype)
1151 : {
1152 12 : case PqMsg_ErrorResponse:
1153 : case PqMsg_NoticeResponse:
1154 : {
1155 : ErrorData edata;
1156 : ErrorContextCallback *save_error_context_stack;
1157 :
1158 : /* Parse ErrorResponse or NoticeResponse. */
1159 12 : pq_parse_errornotice(msg, &edata);
1160 :
1161 : /* Death of a worker isn't enough justification for suicide. */
1162 12 : edata.elevel = Min(edata.elevel, ERROR);
1163 :
1164 : /*
1165 : * If desired, add a context line to show that this is a
1166 : * message propagated from a parallel worker. Otherwise, it
1167 : * can sometimes be confusing to understand what actually
1168 : * happened. (We don't do this in DEBUG_PARALLEL_REGRESS mode
1169 : * because it causes test-result instability depending on
1170 : * whether a parallel worker is actually used or not.)
1171 : */
1172 12 : if (debug_parallel_query != DEBUG_PARALLEL_REGRESS)
1173 : {
1174 12 : if (edata.context)
1175 6 : edata.context = psprintf("%s\n%s", edata.context,
1176 : _("parallel worker"));
1177 : else
1178 6 : edata.context = pstrdup(_("parallel worker"));
1179 : }
1180 :
1181 : /*
1182 : * Context beyond that should use the error context callbacks
1183 : * that were in effect when the ParallelContext was created,
1184 : * not the current ones.
1185 : */
1186 12 : save_error_context_stack = error_context_stack;
1187 12 : error_context_stack = pcxt->error_context_stack;
1188 :
1189 : /* Rethrow error or print notice. */
1190 12 : ThrowErrorData(&edata);
1191 :
1192 : /* Not an error, so restore previous context stack. */
1193 0 : error_context_stack = save_error_context_stack;
1194 :
1195 0 : break;
1196 : }
1197 :
1198 0 : case PqMsg_NotificationResponse:
1199 : {
1200 : /* Propagate NotifyResponse. */
1201 : int32 pid;
1202 : const char *channel;
1203 : const char *payload;
1204 :
1205 0 : pid = pq_getmsgint(msg, 4);
1206 0 : channel = pq_getmsgrawstring(msg);
1207 0 : payload = pq_getmsgrawstring(msg);
1208 0 : pq_endmessage(msg);
1209 :
1210 0 : NotifyMyFrontEnd(channel, payload, pid);
1211 :
1212 0 : break;
1213 : }
1214 :
1215 0 : case PqMsg_Progress:
1216 : {
1217 : /*
1218 : * Only incremental progress reporting is currently supported.
1219 : * However, it's possible to add more fields to the message to
1220 : * allow for handling of other backend progress APIs.
1221 : */
1222 0 : int index = pq_getmsgint(msg, 4);
1223 0 : int64 incr = pq_getmsgint64(msg);
1224 :
1225 0 : pq_getmsgend(msg);
1226 :
1227 0 : pgstat_progress_incr_param(index, incr);
1228 :
1229 0 : break;
1230 : }
1231 :
1232 2734 : case PqMsg_Terminate:
1233 : {
1234 2734 : shm_mq_detach(pcxt->worker[i].error_mqh);
1235 2734 : pcxt->worker[i].error_mqh = NULL;
1236 2734 : break;
1237 : }
1238 :
1239 0 : default:
1240 : {
1241 0 : elog(ERROR, "unrecognized message type received from parallel worker: %c (message length %d bytes)",
1242 : msgtype, msg->len);
1243 : }
1244 : }
1245 2734 : }
1246 :
1247 : /*
1248 : * End-of-subtransaction cleanup for parallel contexts.
1249 : *
1250 : * Here we remove only parallel contexts initiated within the current
1251 : * subtransaction.
1252 : */
1253 : void
1254 19942 : AtEOSubXact_Parallel(bool isCommit, SubTransactionId mySubId)
1255 : {
1256 19948 : while (!dlist_is_empty(&pcxt_list))
1257 : {
1258 : ParallelContext *pcxt;
1259 :
1260 6 : pcxt = dlist_head_element(ParallelContext, node, &pcxt_list);
1261 6 : if (pcxt->subid != mySubId)
1262 0 : break;
1263 6 : if (isCommit)
1264 0 : elog(WARNING, "leaked parallel context");
1265 6 : DestroyParallelContext(pcxt);
1266 : }
1267 19942 : }
1268 :
1269 : /*
1270 : * End-of-transaction cleanup for parallel contexts.
1271 : *
1272 : * We nuke all remaining parallel contexts.
1273 : */
1274 : void
1275 818130 : AtEOXact_Parallel(bool isCommit)
1276 : {
1277 818136 : while (!dlist_is_empty(&pcxt_list))
1278 : {
1279 : ParallelContext *pcxt;
1280 :
1281 6 : pcxt = dlist_head_element(ParallelContext, node, &pcxt_list);
1282 6 : if (isCommit)
1283 0 : elog(WARNING, "leaked parallel context");
1284 6 : DestroyParallelContext(pcxt);
1285 : }
1286 818130 : }
1287 :
1288 : /*
1289 : * Main entrypoint for parallel workers.
1290 : */
1291 : void
1292 2746 : ParallelWorkerMain(Datum main_arg)
1293 : {
1294 : dsm_segment *seg;
1295 : shm_toc *toc;
1296 : FixedParallelState *fps;
1297 : char *error_queue_space;
1298 : shm_mq *mq;
1299 : shm_mq_handle *mqh;
1300 : char *libraryspace;
1301 : char *entrypointstate;
1302 : char *library_name;
1303 : char *function_name;
1304 : parallel_worker_main_type entrypt;
1305 : char *gucspace;
1306 : char *combocidspace;
1307 : char *tsnapspace;
1308 : char *asnapspace;
1309 : char *tstatespace;
1310 : char *pendingsyncsspace;
1311 : char *reindexspace;
1312 : char *relmapperspace;
1313 : char *uncommittedenumsspace;
1314 : char *clientconninfospace;
1315 : char *session_dsm_handle_space;
1316 : Snapshot tsnapshot;
1317 : Snapshot asnapshot;
1318 :
1319 : /* Set flag to indicate that we're initializing a parallel worker. */
1320 2746 : InitializingParallelWorker = true;
1321 :
1322 : /* Establish signal handlers. */
1323 2746 : pqsignal(SIGTERM, die);
1324 2746 : BackgroundWorkerUnblockSignals();
1325 :
1326 : /* Determine and set our parallel worker number. */
1327 : Assert(ParallelWorkerNumber == -1);
1328 2746 : memcpy(&ParallelWorkerNumber, MyBgworkerEntry->bgw_extra, sizeof(int));
1329 :
1330 : /* Set up a memory context to work in, just for cleanliness. */
1331 2746 : CurrentMemoryContext = AllocSetContextCreate(TopMemoryContext,
1332 : "Parallel worker",
1333 : ALLOCSET_DEFAULT_SIZES);
1334 :
1335 : /*
1336 : * Attach to the dynamic shared memory segment for the parallel query, and
1337 : * find its table of contents.
1338 : *
1339 : * Note: at this point, we have not created any ResourceOwner in this
1340 : * process. This will result in our DSM mapping surviving until process
1341 : * exit, which is fine. If there were a ResourceOwner, it would acquire
1342 : * ownership of the mapping, but we have no need for that.
1343 : */
1344 2746 : seg = dsm_attach(DatumGetUInt32(main_arg));
1345 2746 : if (seg == NULL)
1346 0 : ereport(ERROR,
1347 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1348 : errmsg("could not map dynamic shared memory segment")));
1349 2746 : toc = shm_toc_attach(PARALLEL_MAGIC, dsm_segment_address(seg));
1350 2746 : if (toc == NULL)
1351 0 : ereport(ERROR,
1352 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1353 : errmsg("invalid magic number in dynamic shared memory segment")));
1354 :
1355 : /* Look up fixed parallel state. */
1356 2746 : fps = shm_toc_lookup(toc, PARALLEL_KEY_FIXED, false);
1357 2746 : MyFixedParallelState = fps;
1358 :
1359 : /* Arrange to signal the leader if we exit. */
1360 2746 : ParallelLeaderPid = fps->parallel_leader_pid;
1361 2746 : ParallelLeaderProcNumber = fps->parallel_leader_proc_number;
1362 2746 : before_shmem_exit(ParallelWorkerShutdown, PointerGetDatum(seg));
1363 :
1364 : /*
1365 : * Now we can find and attach to the error queue provided for us. That's
1366 : * good, because until we do that, any errors that happen here will not be
1367 : * reported back to the process that requested that this worker be
1368 : * launched.
1369 : */
1370 2746 : error_queue_space = shm_toc_lookup(toc, PARALLEL_KEY_ERROR_QUEUE, false);
1371 2746 : mq = (shm_mq *) (error_queue_space +
1372 2746 : ParallelWorkerNumber * PARALLEL_ERROR_QUEUE_SIZE);
1373 2746 : shm_mq_set_sender(mq, MyProc);
1374 2746 : mqh = shm_mq_attach(mq, seg, NULL);
1375 2746 : pq_redirect_to_shm_mq(seg, mqh);
1376 2746 : pq_set_parallel_leader(fps->parallel_leader_pid,
1377 : fps->parallel_leader_proc_number);
1378 :
1379 : /*
1380 : * Hooray! Primary initialization is complete. Now, we need to set up our
1381 : * backend-local state to match the original backend.
1382 : */
1383 :
1384 : /*
1385 : * Join locking group. We must do this before anything that could try to
1386 : * acquire a heavyweight lock, because any heavyweight locks acquired to
1387 : * this point could block either directly against the parallel group
1388 : * leader or against some process which in turn waits for a lock that
1389 : * conflicts with the parallel group leader, causing an undetected
1390 : * deadlock. (If we can't join the lock group, the leader has gone away,
1391 : * so just exit quietly.)
1392 : */
1393 2746 : if (!BecomeLockGroupMember(fps->parallel_leader_pgproc,
1394 : fps->parallel_leader_pid))
1395 0 : return;
1396 :
1397 : /*
1398 : * Restore transaction and statement start-time timestamps. This must
1399 : * happen before anything that would start a transaction, else asserts in
1400 : * xact.c will fire.
1401 : */
1402 2746 : SetParallelStartTimestamps(fps->xact_ts, fps->stmt_ts);
1403 :
1404 : /*
1405 : * Identify the entry point to be called. In theory this could result in
1406 : * loading an additional library, though most likely the entry point is in
1407 : * the core backend or in a library we just loaded.
1408 : */
1409 2746 : entrypointstate = shm_toc_lookup(toc, PARALLEL_KEY_ENTRYPOINT, false);
1410 2746 : library_name = entrypointstate;
1411 2746 : function_name = entrypointstate + strlen(library_name) + 1;
1412 :
1413 2746 : entrypt = LookupParallelWorkerFunction(library_name, function_name);
1414 :
1415 : /*
1416 : * Restore current session authorization and role id. No verification
1417 : * happens here, we just blindly adopt the leader's state. Note that this
1418 : * has to happen before InitPostgres, since InitializeSessionUserId will
1419 : * not set these variables.
1420 : */
1421 2746 : SetAuthenticatedUserId(fps->authenticated_user_id);
1422 2746 : SetSessionAuthorization(fps->session_user_id,
1423 2746 : fps->session_user_is_superuser);
1424 2746 : SetCurrentRoleId(fps->outer_user_id, fps->role_is_superuser);
1425 :
1426 : /*
1427 : * Restore database connection. We skip connection authorization checks,
1428 : * reasoning that (a) the leader checked these things when it started, and
1429 : * (b) we do not want parallel mode to cause these failures, because that
1430 : * would make use of parallel query plans not transparent to applications.
1431 : */
1432 2746 : BackgroundWorkerInitializeConnectionByOid(fps->database_id,
1433 : fps->authenticated_user_id,
1434 : BGWORKER_BYPASS_ALLOWCONN |
1435 : BGWORKER_BYPASS_ROLELOGINCHECK);
1436 :
1437 : /*
1438 : * Set the client encoding to the database encoding, since that is what
1439 : * the leader will expect. (We're cheating a bit by not calling
1440 : * PrepareClientEncoding first. It's okay because this call will always
1441 : * result in installing a no-op conversion. No error should be possible,
1442 : * but check anyway.)
1443 : */
1444 2746 : if (SetClientEncoding(GetDatabaseEncoding()) < 0)
1445 0 : elog(ERROR, "SetClientEncoding(%d) failed", GetDatabaseEncoding());
1446 :
1447 : /*
1448 : * Load libraries that were loaded by original backend. We want to do
1449 : * this before restoring GUCs, because the libraries might define custom
1450 : * variables.
1451 : */
1452 2746 : libraryspace = shm_toc_lookup(toc, PARALLEL_KEY_LIBRARY, false);
1453 2746 : StartTransactionCommand();
1454 2746 : RestoreLibraryState(libraryspace);
1455 2746 : CommitTransactionCommand();
1456 :
1457 : /* Crank up a transaction state appropriate to a parallel worker. */
1458 2746 : tstatespace = shm_toc_lookup(toc, PARALLEL_KEY_TRANSACTION_STATE, false);
1459 2746 : StartParallelWorkerTransaction(tstatespace);
1460 :
1461 : /*
1462 : * Restore state that affects catalog access. Ideally we'd do this even
1463 : * before calling InitPostgres, but that has order-of-initialization
1464 : * problems, and also the relmapper would get confused during the
1465 : * CommitTransactionCommand call above.
1466 : */
1467 2746 : pendingsyncsspace = shm_toc_lookup(toc, PARALLEL_KEY_PENDING_SYNCS,
1468 : false);
1469 2746 : RestorePendingSyncs(pendingsyncsspace);
1470 2746 : relmapperspace = shm_toc_lookup(toc, PARALLEL_KEY_RELMAPPER_STATE, false);
1471 2746 : RestoreRelationMap(relmapperspace);
1472 2746 : reindexspace = shm_toc_lookup(toc, PARALLEL_KEY_REINDEX_STATE, false);
1473 2746 : RestoreReindexState(reindexspace);
1474 2746 : combocidspace = shm_toc_lookup(toc, PARALLEL_KEY_COMBO_CID, false);
1475 2746 : RestoreComboCIDState(combocidspace);
1476 :
1477 : /* Attach to the per-session DSM segment and contained objects. */
1478 : session_dsm_handle_space =
1479 2746 : shm_toc_lookup(toc, PARALLEL_KEY_SESSION_DSM, false);
1480 2746 : AttachSession(*(dsm_handle *) session_dsm_handle_space);
1481 :
1482 : /*
1483 : * If the transaction isolation level is REPEATABLE READ or SERIALIZABLE,
1484 : * the leader has serialized the transaction snapshot and we must restore
1485 : * it. At lower isolation levels, there is no transaction-lifetime
1486 : * snapshot, but we need TransactionXmin to get set to a value which is
1487 : * less than or equal to the xmin of every snapshot that will be used by
1488 : * this worker. The easiest way to accomplish that is to install the
1489 : * active snapshot as the transaction snapshot. Code running in this
1490 : * parallel worker might take new snapshots via GetTransactionSnapshot()
1491 : * or GetLatestSnapshot(), but it shouldn't have any way of acquiring a
1492 : * snapshot older than the active snapshot.
1493 : */
1494 2746 : asnapspace = shm_toc_lookup(toc, PARALLEL_KEY_ACTIVE_SNAPSHOT, false);
1495 2746 : tsnapspace = shm_toc_lookup(toc, PARALLEL_KEY_TRANSACTION_SNAPSHOT, true);
1496 2746 : asnapshot = RestoreSnapshot(asnapspace);
1497 2746 : tsnapshot = tsnapspace ? RestoreSnapshot(tsnapspace) : asnapshot;
1498 2746 : RestoreTransactionSnapshot(tsnapshot,
1499 2746 : fps->parallel_leader_pgproc);
1500 2746 : PushActiveSnapshot(asnapshot);
1501 :
1502 : /*
1503 : * We've changed which tuples we can see, and must therefore invalidate
1504 : * system caches.
1505 : */
1506 2746 : InvalidateSystemCaches();
1507 :
1508 : /*
1509 : * Restore GUC values from launching backend. We can't do this earlier,
1510 : * because GUC check hooks that do catalog lookups need to see the same
1511 : * database state as the leader. Also, the check hooks for
1512 : * session_authorization and role assume we already set the correct role
1513 : * OIDs.
1514 : */
1515 2746 : gucspace = shm_toc_lookup(toc, PARALLEL_KEY_GUC, false);
1516 2746 : RestoreGUCState(gucspace);
1517 :
1518 : /*
1519 : * Restore current user ID and security context. No verification happens
1520 : * here, we just blindly adopt the leader's state. We can't do this till
1521 : * after restoring GUCs, else we'll get complaints about restoring
1522 : * session_authorization and role. (In effect, we're assuming that all
1523 : * the restored values are okay to set, even if we are now inside a
1524 : * restricted context.)
1525 : */
1526 2746 : SetUserIdAndSecContext(fps->current_user_id, fps->sec_context);
1527 :
1528 : /* Restore temp-namespace state to ensure search path matches leader's. */
1529 2746 : SetTempNamespaceState(fps->temp_namespace_id,
1530 : fps->temp_toast_namespace_id);
1531 :
1532 : /* Restore uncommitted enums. */
1533 2746 : uncommittedenumsspace = shm_toc_lookup(toc, PARALLEL_KEY_UNCOMMITTEDENUMS,
1534 : false);
1535 2746 : RestoreUncommittedEnums(uncommittedenumsspace);
1536 :
1537 : /* Restore the ClientConnectionInfo. */
1538 2746 : clientconninfospace = shm_toc_lookup(toc, PARALLEL_KEY_CLIENTCONNINFO,
1539 : false);
1540 2746 : RestoreClientConnectionInfo(clientconninfospace);
1541 :
1542 : /*
1543 : * Initialize SystemUser now that MyClientConnectionInfo is restored. Also
1544 : * ensure that auth_method is actually valid, aka authn_id is not NULL.
1545 : */
1546 2746 : if (MyClientConnectionInfo.authn_id)
1547 4 : InitializeSystemUser(MyClientConnectionInfo.authn_id,
1548 : hba_authname(MyClientConnectionInfo.auth_method));
1549 :
1550 : /* Attach to the leader's serializable transaction, if SERIALIZABLE. */
1551 2746 : AttachSerializableXact(fps->serializable_xact_handle);
1552 :
1553 : /*
1554 : * We've initialized all of our state now; nothing should change
1555 : * hereafter.
1556 : */
1557 2746 : InitializingParallelWorker = false;
1558 2746 : EnterParallelMode();
1559 :
1560 : /*
1561 : * Time to do the real work: invoke the caller-supplied code.
1562 : */
1563 2746 : entrypt(seg, toc);
1564 :
1565 : /* Must exit parallel mode to pop active snapshot. */
1566 2734 : ExitParallelMode();
1567 :
1568 : /* Must pop active snapshot so snapmgr.c doesn't complain. */
1569 2734 : PopActiveSnapshot();
1570 :
1571 : /* Shut down the parallel-worker transaction. */
1572 2734 : EndParallelWorkerTransaction();
1573 :
1574 : /* Detach from the per-session DSM segment. */
1575 2734 : DetachSession();
1576 :
1577 : /* Report success. */
1578 2734 : pq_putmessage(PqMsg_Terminate, NULL, 0);
1579 : }
1580 :
1581 : /*
1582 : * Update shared memory with the ending location of the last WAL record we
1583 : * wrote, if it's greater than the value already stored there.
1584 : */
1585 : void
1586 2734 : ParallelWorkerReportLastRecEnd(XLogRecPtr last_xlog_end)
1587 : {
1588 2734 : FixedParallelState *fps = MyFixedParallelState;
1589 :
1590 : Assert(fps != NULL);
1591 2734 : SpinLockAcquire(&fps->mutex);
1592 2734 : if (fps->last_xlog_end < last_xlog_end)
1593 140 : fps->last_xlog_end = last_xlog_end;
1594 2734 : SpinLockRelease(&fps->mutex);
1595 2734 : }
1596 :
1597 : /*
1598 : * Make sure the leader tries to read from our error queue one more time.
1599 : * This guards against the case where we exit uncleanly without sending an
1600 : * ErrorResponse to the leader, for example because some code calls proc_exit
1601 : * directly.
1602 : *
1603 : * Also explicitly detach from dsm segment so that subsystems using
1604 : * on_dsm_detach() have a chance to send stats before the stats subsystem is
1605 : * shut down as part of a before_shmem_exit() hook.
1606 : *
1607 : * One might think this could instead be solved by carefully ordering the
1608 : * attaching to dsm segments, so that the pgstats segments get detached from
1609 : * later than the parallel query one. That turns out to not work because the
1610 : * stats hash might need to grow which can cause new segments to be allocated,
1611 : * which then will be detached from earlier.
1612 : */
1613 : static void
1614 2746 : ParallelWorkerShutdown(int code, Datum arg)
1615 : {
1616 2746 : SendProcSignal(ParallelLeaderPid,
1617 : PROCSIG_PARALLEL_MESSAGE,
1618 : ParallelLeaderProcNumber);
1619 :
1620 2746 : dsm_detach((dsm_segment *) DatumGetPointer(arg));
1621 2746 : }
1622 :
1623 : /*
1624 : * Look up (and possibly load) a parallel worker entry point function.
1625 : *
1626 : * For functions contained in the core code, we use library name "postgres"
1627 : * and consult the InternalParallelWorkers array. External functions are
1628 : * looked up, and loaded if necessary, using load_external_function().
1629 : *
1630 : * The point of this is to pass function names as strings across process
1631 : * boundaries. We can't pass actual function addresses because of the
1632 : * possibility that the function has been loaded at a different address
1633 : * in a different process. This is obviously a hazard for functions in
1634 : * loadable libraries, but it can happen even for functions in the core code
1635 : * on platforms using EXEC_BACKEND (e.g., Windows).
1636 : *
1637 : * At some point it might be worthwhile to get rid of InternalParallelWorkers[]
1638 : * in favor of applying load_external_function() for core functions too;
1639 : * but that raises portability issues that are not worth addressing now.
1640 : */
1641 : static parallel_worker_main_type
1642 2746 : LookupParallelWorkerFunction(const char *libraryname, const char *funcname)
1643 : {
1644 : /*
1645 : * If the function is to be loaded from postgres itself, search the
1646 : * InternalParallelWorkers array.
1647 : */
1648 2746 : if (strcmp(libraryname, "postgres") == 0)
1649 : {
1650 : int i;
1651 :
1652 3150 : for (i = 0; i < lengthof(InternalParallelWorkers); i++)
1653 : {
1654 3150 : if (strcmp(InternalParallelWorkers[i].fn_name, funcname) == 0)
1655 2746 : return InternalParallelWorkers[i].fn_addr;
1656 : }
1657 :
1658 : /* We can only reach this by programming error. */
1659 0 : elog(ERROR, "internal function \"%s\" not found", funcname);
1660 : }
1661 :
1662 : /* Otherwise load from external library. */
1663 0 : return (parallel_worker_main_type)
1664 0 : load_external_function(libraryname, funcname, true, NULL);
1665 : }
|
