Branch data Line data Source code
1 : : /*--------------------------------------------------------------------
2 : : * bgworker.c
3 : : * POSTGRES pluggable background workers implementation
4 : : *
5 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
6 : : *
7 : : * IDENTIFICATION
8 : : * src/backend/postmaster/bgworker.c
9 : : *
10 : : *-------------------------------------------------------------------------
11 : : */
12 : :
13 : : #include "postgres.h"
14 : :
15 : : #include "access/parallel.h"
16 : : #include "commands/repack.h"
17 : : #include "libpq/pqsignal.h"
18 : : #include "miscadmin.h"
19 : : #include "pgstat.h"
20 : : #include "port/atomics.h"
21 : : #include "postmaster/bgworker_internals.h"
22 : : #include "postmaster/datachecksum_state.h"
23 : : #include "postmaster/postmaster.h"
24 : : #include "replication/logicallauncher.h"
25 : : #include "replication/logicalworker.h"
26 : : #include "storage/ipc.h"
27 : : #include "storage/latch.h"
28 : : #include "storage/lwlock.h"
29 : : #include "storage/pmsignal.h"
30 : : #include "storage/proc.h"
31 : : #include "storage/procarray.h"
32 : : #include "storage/procsignal.h"
33 : : #include "storage/shmem.h"
34 : : #include "storage/subsystems.h"
35 : : #include "tcop/tcopprot.h"
36 : : #include "utils/ascii.h"
37 : : #include "utils/memutils.h"
38 : : #include "utils/ps_status.h"
39 : : #include "utils/timeout.h"
40 : : #include "utils/wait_event.h"
41 : :
42 : : /*
43 : : * The postmaster's list of registered background workers, in private memory.
44 : : */
45 : : dlist_head BackgroundWorkerList = DLIST_STATIC_INIT(BackgroundWorkerList);
46 : :
47 : : /*
48 : : * BackgroundWorkerSlots exist in shared memory and can be accessed (via
49 : : * the BackgroundWorkerArray) by both the postmaster and by regular backends.
50 : : * However, the postmaster cannot take locks, even spinlocks, because this
51 : : * might allow it to crash or become wedged if shared memory gets corrupted.
52 : : * Such an outcome is intolerable. Therefore, we need a lockless protocol
53 : : * for coordinating access to this data.
54 : : *
55 : : * The 'in_use' flag is used to hand off responsibility for the slot between
56 : : * the postmaster and the rest of the system. When 'in_use' is false,
57 : : * the postmaster will ignore the slot entirely, except for the 'in_use' flag
58 : : * itself, which it may read. In this state, regular backends may modify the
59 : : * slot. Once a backend sets 'in_use' to true, the slot becomes the
60 : : * responsibility of the postmaster. Regular backends may no longer modify it,
61 : : * but the postmaster may examine it. Thus, a backend initializing a slot
62 : : * must fully initialize the slot - and insert a write memory barrier - before
63 : : * marking it as in use.
64 : : *
65 : : * As an exception, however, even when the slot is in use, regular backends
66 : : * may set the 'terminate' flag for a slot, telling the postmaster not
67 : : * to restart it. Once the background worker is no longer running, the slot
68 : : * will be released for reuse.
69 : : *
70 : : * In addition to coordinating with the postmaster, backends modifying this
71 : : * data structure must coordinate with each other. Since they can take locks,
72 : : * this is straightforward: any backend wishing to manipulate a slot must
73 : : * take BackgroundWorkerLock in exclusive mode. Backends wishing to read
74 : : * data that might get concurrently modified by other backends should take
75 : : * this lock in shared mode. No matter what, backends reading this data
76 : : * structure must be able to tolerate concurrent modifications by the
77 : : * postmaster.
78 : : */
79 : : typedef struct BackgroundWorkerSlot
80 : : {
81 : : bool in_use;
82 : : bool terminate;
83 : : pid_t pid; /* InvalidPid = not started yet; 0 = dead */
84 : : uint64 generation; /* incremented when slot is recycled */
85 : : BackgroundWorker worker;
86 : : } BackgroundWorkerSlot;
87 : :
88 : : /*
89 : : * In order to limit the total number of parallel workers (according to
90 : : * max_parallel_workers GUC), we maintain the number of active parallel
91 : : * workers. Since the postmaster cannot take locks, two variables are used for
92 : : * this purpose: the number of registered parallel workers (modified by the
93 : : * backends, protected by BackgroundWorkerLock) and the number of terminated
94 : : * parallel workers (modified only by the postmaster, lockless). The active
95 : : * number of parallel workers is the number of registered workers minus the
96 : : * terminated ones. These counters can of course overflow, but it's not
97 : : * important here since the subtraction will still give the right number.
98 : : */
99 : : typedef struct BackgroundWorkerArray
100 : : {
101 : : int total_slots;
102 : : uint32 parallel_register_count;
103 : : uint32 parallel_terminate_count;
104 : : pg_attribute_counted_by(total_slots) BackgroundWorkerSlot slot[FLEXIBLE_ARRAY_MEMBER];
105 : : } BackgroundWorkerArray;
106 : :
107 : : struct BackgroundWorkerHandle
108 : : {
109 : : int slot;
110 : : uint64 generation;
111 : : };
112 : :
113 : : static BackgroundWorkerArray *BackgroundWorkerData;
114 : :
115 : : static void BackgroundWorkerShmemRequest(void *arg);
116 : : static void BackgroundWorkerShmemInit(void *arg);
117 : :
118 : : const ShmemCallbacks BackgroundWorkerShmemCallbacks = {
119 : : .request_fn = BackgroundWorkerShmemRequest,
120 : : .init_fn = BackgroundWorkerShmemInit,
121 : : };
122 : :
123 : : /*
124 : : * List of internal background worker entry points. We need this for
125 : : * reasons explained in LookupBackgroundWorkerFunction(), below.
126 : : */
127 : : static const struct
128 : : {
129 : : const char *fn_name;
130 : : bgworker_main_type fn_addr;
131 : : } InternalBGWorkers[] =
132 : :
133 : : {
134 : : {
135 : : .fn_name = "ApplyLauncherMain",
136 : : .fn_addr = ApplyLauncherMain
137 : : },
138 : : {
139 : : .fn_name = "ApplyWorkerMain",
140 : : .fn_addr = ApplyWorkerMain
141 : : },
142 : : {
143 : : .fn_name = "ParallelApplyWorkerMain",
144 : : .fn_addr = ParallelApplyWorkerMain
145 : : },
146 : : {
147 : : .fn_name = "ParallelWorkerMain",
148 : : .fn_addr = ParallelWorkerMain
149 : : },
150 : : {
151 : : .fn_name = "RepackWorkerMain",
152 : : .fn_addr = RepackWorkerMain
153 : : },
154 : : {
155 : : .fn_name = "SequenceSyncWorkerMain",
156 : : .fn_addr = SequenceSyncWorkerMain
157 : : },
158 : : {
159 : : .fn_name = "TableSyncWorkerMain",
160 : : .fn_addr = TableSyncWorkerMain
161 : : },
162 : : {
163 : : .fn_name = "DataChecksumsWorkerLauncherMain",
164 : : .fn_addr = DataChecksumsWorkerLauncherMain
165 : : },
166 : : {
167 : : .fn_name = "DataChecksumsWorkerMain",
168 : : .fn_addr = DataChecksumsWorkerMain
169 : : }
170 : : };
171 : :
172 : : /* Private functions. */
173 : : static bgworker_main_type LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname);
174 : :
175 : :
176 : : /*
177 : : * Register shared memory needed for background workers.
178 : : */
179 : : static void
180 : 1349 : BackgroundWorkerShmemRequest(void *arg)
181 : : {
182 : : Size size;
183 : :
184 : : /* Array of workers is variably sized. */
185 : 1349 : size = offsetof(BackgroundWorkerArray, slot);
186 : 1349 : size = add_size(size, mul_size(max_worker_processes,
187 : : sizeof(BackgroundWorkerSlot)));
188 : 1349 : ShmemRequestStruct(.name = "Background Worker Data",
189 : : .size = size,
190 : : .ptr = (void **) &BackgroundWorkerData,
191 : : );
192 : 1349 : }
193 : :
194 : : /*
195 : : * Initialize shared memory for background workers.
196 : : */
197 : : static void
198 : 1345 : BackgroundWorkerShmemInit(void *arg)
199 : : {
200 : : dlist_iter iter;
201 : 1345 : int slotno = 0;
202 : :
203 : 1345 : BackgroundWorkerData->total_slots = max_worker_processes;
204 : 1345 : BackgroundWorkerData->parallel_register_count = 0;
205 : 1345 : BackgroundWorkerData->parallel_terminate_count = 0;
206 : :
207 : : /*
208 : : * Copy contents of worker list into shared memory. Record the shared
209 : : * memory slot assigned to each worker. This ensures a 1-to-1
210 : : * correspondence between the postmaster's private list and the array in
211 : : * shared memory.
212 : : */
213 [ + - + + ]: 2370 : dlist_foreach(iter, &BackgroundWorkerList)
214 : : {
215 : 1025 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
216 : : RegisteredBgWorker *rw;
217 : :
218 : 1025 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
219 : : Assert(slotno < max_worker_processes);
220 : 1025 : slot->in_use = true;
221 : 1025 : slot->terminate = false;
222 : 1025 : slot->pid = InvalidPid;
223 : 1025 : slot->generation = 0;
224 : 1025 : rw->rw_shmem_slot = slotno;
225 : 1025 : rw->rw_worker.bgw_notify_pid = 0; /* might be reinit after crash */
226 : 1025 : memcpy(&slot->worker, &rw->rw_worker, sizeof(BackgroundWorker));
227 : 1025 : ++slotno;
228 : : }
229 : :
230 : : /*
231 : : * Mark any remaining slots as not in use.
232 : : */
233 [ + + ]: 11073 : while (slotno < max_worker_processes)
234 : : {
235 : 9728 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
236 : :
237 : 9728 : slot->in_use = false;
238 : 9728 : ++slotno;
239 : : }
240 : 1345 : }
241 : :
242 : : /*
243 : : * Search the postmaster's backend-private list of RegisteredBgWorker objects
244 : : * for the one that maps to the given slot number.
245 : : */
246 : : static RegisteredBgWorker *
247 : 6183 : FindRegisteredWorkerBySlotNumber(int slotno)
248 : : {
249 : : dlist_iter iter;
250 : :
251 [ + - + + ]: 14996 : dlist_foreach(iter, &BackgroundWorkerList)
252 : : {
253 : : RegisteredBgWorker *rw;
254 : :
255 : 12096 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
256 [ + + ]: 12096 : if (rw->rw_shmem_slot == slotno)
257 : 3283 : return rw;
258 : : }
259 : :
260 : 2900 : return NULL;
261 : : }
262 : :
263 : : /*
264 : : * Notice changes to shared memory made by other backends.
265 : : * Accept new worker requests only if allow_new_workers is true.
266 : : *
267 : : * This code runs in the postmaster, so we must be very careful not to assume
268 : : * that shared memory contents are sane. Otherwise, a rogue backend could
269 : : * take out the postmaster.
270 : : */
271 : : void
272 : 1891 : BackgroundWorkerStateChange(bool allow_new_workers)
273 : : {
274 : : int slotno;
275 : :
276 : : /*
277 : : * The total number of slots stored in shared memory should match our
278 : : * notion of max_worker_processes. If it does not, something is very
279 : : * wrong. Further down, we always refer to this value as
280 : : * max_worker_processes, in case shared memory gets corrupted while we're
281 : : * looping.
282 : : */
283 [ - + ]: 1891 : if (max_worker_processes != BackgroundWorkerData->total_slots)
284 : : {
285 [ # # ]: 0 : ereport(LOG,
286 : : (errmsg("inconsistent background worker state (\"max_worker_processes\"=%d, total slots=%d)",
287 : : max_worker_processes,
288 : : BackgroundWorkerData->total_slots)));
289 : 0 : return;
290 : : }
291 : :
292 : : /*
293 : : * Iterate through slots, looking for newly-registered workers or workers
294 : : * who must die.
295 : : */
296 [ + + ]: 17173 : for (slotno = 0; slotno < max_worker_processes; ++slotno)
297 : : {
298 : 15282 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
299 : : RegisteredBgWorker *rw;
300 : :
301 [ + + ]: 15282 : if (!slot->in_use)
302 : 9099 : continue;
303 : :
304 : : /*
305 : : * Make sure we don't see the in_use flag before the updated slot
306 : : * contents.
307 : : */
308 : 6183 : pg_read_barrier();
309 : :
310 : : /* See whether we already know about this worker. */
311 : 6183 : rw = FindRegisteredWorkerBySlotNumber(slotno);
312 [ + + ]: 6183 : if (rw != NULL)
313 : : {
314 : : /*
315 : : * In general, the worker data can't change after it's initially
316 : : * registered. However, someone can set the terminate flag.
317 : : */
318 [ + + + - ]: 3283 : if (slot->terminate && !rw->rw_terminate)
319 : : {
320 : 31 : rw->rw_terminate = true;
321 [ + - ]: 31 : if (rw->rw_pid != 0)
322 : 31 : kill(rw->rw_pid, SIGTERM);
323 : : else
324 : : {
325 : : /* Report never-started, now-terminated worker as dead. */
326 : 0 : ReportBackgroundWorkerPID(rw);
327 : : }
328 : : }
329 : 3283 : continue;
330 : : }
331 : :
332 : : /*
333 : : * If we aren't allowing new workers, then immediately mark it for
334 : : * termination; the next stanza will take care of cleaning it up.
335 : : * Doing this ensures that any process waiting for the worker will get
336 : : * awoken, even though the worker will never be allowed to run.
337 : : */
338 [ + + ]: 2900 : if (!allow_new_workers)
339 : 9 : slot->terminate = true;
340 : :
341 : : /*
342 : : * If the worker is marked for termination, we don't need to add it to
343 : : * the registered workers list; we can just free the slot. However, if
344 : : * bgw_notify_pid is set, the process that registered the worker may
345 : : * need to know that we've processed the terminate request, so be sure
346 : : * to signal it.
347 : : */
348 [ + + ]: 2900 : if (slot->terminate)
349 : 9 : {
350 : : int notify_pid;
351 : :
352 : : /*
353 : : * We need a memory barrier here to make sure that the load of
354 : : * bgw_notify_pid and the update of parallel_terminate_count
355 : : * complete before the store to in_use.
356 : : */
357 : 9 : notify_pid = slot->worker.bgw_notify_pid;
358 [ - + ]: 9 : if ((slot->worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
359 : 0 : BackgroundWorkerData->parallel_terminate_count++;
360 : 9 : slot->pid = 0;
361 : :
362 : 9 : pg_memory_barrier();
363 : 9 : slot->in_use = false;
364 : :
365 [ + - ]: 9 : if (notify_pid != 0)
366 : 9 : kill(notify_pid, SIGUSR1);
367 : :
368 : 9 : continue;
369 : : }
370 : :
371 : : /*
372 : : * Copy the registration data into the registered workers list.
373 : : */
374 : 2891 : rw = MemoryContextAllocExtended(PostmasterContext,
375 : : sizeof(RegisteredBgWorker),
376 : : MCXT_ALLOC_NO_OOM | MCXT_ALLOC_ZERO);
377 [ - + ]: 2891 : if (rw == NULL)
378 : : {
379 [ # # ]: 0 : ereport(LOG,
380 : : (errcode(ERRCODE_OUT_OF_MEMORY),
381 : : errmsg("out of memory")));
382 : 0 : return;
383 : : }
384 : :
385 : : /*
386 : : * Copy strings in a paranoid way. If shared memory is corrupted, the
387 : : * source data might not even be NUL-terminated.
388 : : */
389 : 2891 : ascii_safe_strlcpy(rw->rw_worker.bgw_name,
390 : 2891 : slot->worker.bgw_name, BGW_MAXLEN);
391 : 2891 : ascii_safe_strlcpy(rw->rw_worker.bgw_type,
392 : 2891 : slot->worker.bgw_type, BGW_MAXLEN);
393 : 2891 : ascii_safe_strlcpy(rw->rw_worker.bgw_library_name,
394 : 2891 : slot->worker.bgw_library_name, MAXPGPATH);
395 : 2891 : ascii_safe_strlcpy(rw->rw_worker.bgw_function_name,
396 : 2891 : slot->worker.bgw_function_name, BGW_MAXLEN);
397 : :
398 : : /*
399 : : * Copy various fixed-size fields.
400 : : *
401 : : * flags, start_time, and restart_time are examined by the postmaster,
402 : : * but nothing too bad will happen if they are corrupted. The
403 : : * remaining fields will only be examined by the child process. It
404 : : * might crash, but we won't.
405 : : */
406 : 2891 : rw->rw_worker.bgw_flags = slot->worker.bgw_flags;
407 : 2891 : rw->rw_worker.bgw_start_time = slot->worker.bgw_start_time;
408 : 2891 : rw->rw_worker.bgw_restart_time = slot->worker.bgw_restart_time;
409 : 2891 : rw->rw_worker.bgw_main_arg = slot->worker.bgw_main_arg;
410 : 2891 : memcpy(rw->rw_worker.bgw_extra, slot->worker.bgw_extra, BGW_EXTRALEN);
411 : :
412 : : /*
413 : : * Copy the PID to be notified about state changes, but only if the
414 : : * postmaster knows about a backend with that PID. It isn't an error
415 : : * if the postmaster doesn't know about the PID, because the backend
416 : : * that requested the worker could have died (or been killed) just
417 : : * after doing so. Nonetheless, at least until we get some experience
418 : : * with how this plays out in the wild, log a message at a relative
419 : : * high debug level.
420 : : */
421 : 2891 : rw->rw_worker.bgw_notify_pid = slot->worker.bgw_notify_pid;
422 [ - + ]: 2891 : if (!PostmasterMarkPIDForWorkerNotify(rw->rw_worker.bgw_notify_pid))
423 : : {
424 [ # # ]: 0 : elog(DEBUG1, "worker notification PID %d is not valid",
425 : : (int) rw->rw_worker.bgw_notify_pid);
426 : 0 : rw->rw_worker.bgw_notify_pid = 0;
427 : : }
428 : :
429 : : /* Initialize postmaster bookkeeping. */
430 : 2891 : rw->rw_pid = 0;
431 : 2891 : rw->rw_crashed_at = 0;
432 : 2891 : rw->rw_shmem_slot = slotno;
433 : 2891 : rw->rw_terminate = false;
434 : :
435 : : /* Log it! */
436 [ + + ]: 2891 : ereport(DEBUG1,
437 : : (errmsg_internal("registering background worker \"%s\"",
438 : : rw->rw_worker.bgw_name)));
439 : :
440 : 2891 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
441 : : }
442 : : }
443 : :
444 : : /*
445 : : * Forget about a background worker that's no longer needed.
446 : : *
447 : : * NOTE: The entry is unlinked from BackgroundWorkerList. If the caller is
448 : : * iterating through it, better use a mutable iterator!
449 : : *
450 : : * Caller is responsible for notifying bgw_notify_pid, if appropriate.
451 : : *
452 : : * This function must be invoked only in the postmaster.
453 : : */
454 : : void
455 : 2879 : ForgetBackgroundWorker(RegisteredBgWorker *rw)
456 : : {
457 : : BackgroundWorkerSlot *slot;
458 : :
459 : : Assert(rw->rw_shmem_slot < max_worker_processes);
460 : 2879 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
461 : : Assert(slot->in_use);
462 : :
463 : : /*
464 : : * We need a memory barrier here to make sure that the update of
465 : : * parallel_terminate_count completes before the store to in_use.
466 : : */
467 [ + + ]: 2879 : if ((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
468 : 2044 : BackgroundWorkerData->parallel_terminate_count++;
469 : :
470 : 2879 : pg_memory_barrier();
471 : 2879 : slot->in_use = false;
472 : :
473 [ + + ]: 2879 : ereport(DEBUG1,
474 : : (errmsg_internal("unregistering background worker \"%s\"",
475 : : rw->rw_worker.bgw_name)));
476 : :
477 : 2879 : dlist_delete(&rw->rw_lnode);
478 : 2879 : pfree(rw);
479 : 2879 : }
480 : :
481 : : /*
482 : : * Report the PID of a newly-launched background worker in shared memory.
483 : : *
484 : : * This function should only be called from the postmaster.
485 : : */
486 : : void
487 : 3784 : ReportBackgroundWorkerPID(RegisteredBgWorker *rw)
488 : : {
489 : : BackgroundWorkerSlot *slot;
490 : :
491 : : Assert(rw->rw_shmem_slot < max_worker_processes);
492 : 3784 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
493 : 3784 : slot->pid = rw->rw_pid;
494 : :
495 [ + + ]: 3784 : if (rw->rw_worker.bgw_notify_pid != 0)
496 : 2891 : kill(rw->rw_worker.bgw_notify_pid, SIGUSR1);
497 : 3784 : }
498 : :
499 : : /*
500 : : * Report that the PID of a background worker is now zero because a
501 : : * previously-running background worker has exited.
502 : : *
503 : : * NOTE: The entry may be unlinked from BackgroundWorkerList. If the caller
504 : : * is iterating through it, better use a mutable iterator!
505 : : *
506 : : * This function should only be called from the postmaster.
507 : : */
508 : : void
509 : 3440 : ReportBackgroundWorkerExit(RegisteredBgWorker *rw)
510 : : {
511 : : BackgroundWorkerSlot *slot;
512 : : int notify_pid;
513 : :
514 : : Assert(rw->rw_shmem_slot < max_worker_processes);
515 : 3440 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
516 : 3440 : slot->pid = rw->rw_pid;
517 : 3440 : notify_pid = rw->rw_worker.bgw_notify_pid;
518 : :
519 : : /*
520 : : * If this worker is slated for deregistration, do that before notifying
521 : : * the process which started it. Otherwise, if that process tries to
522 : : * reuse the slot immediately, it might not be available yet. In theory
523 : : * that could happen anyway if the process checks slot->pid at just the
524 : : * wrong moment, but this makes the window narrower.
525 : : */
526 [ + + ]: 3440 : if (rw->rw_terminate ||
527 [ + + ]: 962 : rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
528 : 2879 : ForgetBackgroundWorker(rw);
529 : :
530 [ + + ]: 3440 : if (notify_pid != 0)
531 : 2821 : kill(notify_pid, SIGUSR1);
532 : 3440 : }
533 : :
534 : : /*
535 : : * Cancel SIGUSR1 notifications for a PID belonging to an exiting backend.
536 : : *
537 : : * This function should only be called from the postmaster.
538 : : */
539 : : void
540 : 382 : BackgroundWorkerStopNotifications(pid_t pid)
541 : : {
542 : : dlist_iter iter;
543 : :
544 [ + - + + ]: 1200 : dlist_foreach(iter, &BackgroundWorkerList)
545 : : {
546 : : RegisteredBgWorker *rw;
547 : :
548 : 818 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
549 [ + + ]: 818 : if (rw->rw_worker.bgw_notify_pid == pid)
550 : 57 : rw->rw_worker.bgw_notify_pid = 0;
551 : : }
552 : 382 : }
553 : :
554 : : /*
555 : : * Cancel any not-yet-started worker requests that have waiting processes.
556 : : *
557 : : * This is called during a normal ("smart" or "fast") database shutdown.
558 : : * After this point, no new background workers will be started, so anything
559 : : * that might be waiting for them needs to be kicked off its wait. We do
560 : : * that by canceling the bgworker registration entirely, which is perhaps
561 : : * overkill, but since we're shutting down it does not matter whether the
562 : : * registration record sticks around.
563 : : *
564 : : * This function should only be called from the postmaster.
565 : : */
566 : : void
567 : 695 : ForgetUnstartedBackgroundWorkers(void)
568 : : {
569 : : dlist_mutable_iter iter;
570 : :
571 [ + - + + ]: 1378 : dlist_foreach_modify(iter, &BackgroundWorkerList)
572 : : {
573 : : RegisteredBgWorker *rw;
574 : : BackgroundWorkerSlot *slot;
575 : :
576 : 683 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
577 : : Assert(rw->rw_shmem_slot < max_worker_processes);
578 : 683 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
579 : :
580 : : /* If it's not yet started, and there's someone waiting ... */
581 [ + + ]: 683 : if (slot->pid == InvalidPid &&
582 [ - + ]: 68 : rw->rw_worker.bgw_notify_pid != 0)
583 : : {
584 : : /* ... then zap it, and notify the waiter */
585 : 0 : int notify_pid = rw->rw_worker.bgw_notify_pid;
586 : :
587 : 0 : ForgetBackgroundWorker(rw);
588 [ # # ]: 0 : if (notify_pid != 0)
589 : 0 : kill(notify_pid, SIGUSR1);
590 : : }
591 : : }
592 : 695 : }
593 : :
594 : : /*
595 : : * Reset background worker crash state.
596 : : *
597 : : * We assume that, after a crash-and-restart cycle, background workers without
598 : : * the never-restart flag should be restarted immediately, instead of waiting
599 : : * for bgw_restart_time to elapse. On the other hand, workers with that flag
600 : : * should be forgotten immediately, since we won't ever restart them.
601 : : *
602 : : * This function should only be called from the postmaster.
603 : : */
604 : : void
605 : 6 : ResetBackgroundWorkerCrashTimes(void)
606 : : {
607 : : dlist_mutable_iter iter;
608 : :
609 [ + - + + ]: 12 : dlist_foreach_modify(iter, &BackgroundWorkerList)
610 : : {
611 : : RegisteredBgWorker *rw;
612 : :
613 : 6 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
614 : :
615 [ - + ]: 6 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
616 : : {
617 : : /*
618 : : * Workers marked BGW_NEVER_RESTART shouldn't get relaunched after
619 : : * the crash, so forget about them. (If we wait until after the
620 : : * crash to forget about them, and they are parallel workers,
621 : : * parallel_terminate_count will get incremented after we've
622 : : * already zeroed parallel_register_count, which would be bad.)
623 : : */
624 : 0 : ForgetBackgroundWorker(rw);
625 : : }
626 : : else
627 : : {
628 : : /*
629 : : * The accounting which we do via parallel_register_count and
630 : : * parallel_terminate_count would get messed up if a worker marked
631 : : * parallel could survive a crash and restart cycle. All such
632 : : * workers should be marked BGW_NEVER_RESTART, and thus control
633 : : * should never reach this branch.
634 : : */
635 : : Assert((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) == 0);
636 : :
637 : : /*
638 : : * Allow this worker to be restarted immediately after we finish
639 : : * resetting.
640 : : */
641 : 6 : rw->rw_crashed_at = 0;
642 : 6 : rw->rw_pid = 0;
643 : :
644 : : /*
645 : : * If there was anyone waiting for it, they're history.
646 : : */
647 : 6 : rw->rw_worker.bgw_notify_pid = 0;
648 : : }
649 : : }
650 : 6 : }
651 : :
652 : : /*
653 : : * Complain about the BackgroundWorker definition using error level elevel.
654 : : * Return true if it looks ok, false if not (unless elevel >= ERROR, in
655 : : * which case we won't return at all in the not-OK case).
656 : : */
657 : : static bool
658 : 3765 : SanityCheckBackgroundWorker(BackgroundWorker *worker, int elevel)
659 : : {
660 : : /* sanity check for flags */
661 : :
662 : : /*
663 : : * We used to support workers not connected to shared memory, but don't
664 : : * anymore. Thus this is a required flag now. We're not removing the flag
665 : : * for compatibility reasons and because the flag still provides some
666 : : * signal when reading code.
667 : : */
668 [ - + ]: 3765 : if (!(worker->bgw_flags & BGWORKER_SHMEM_ACCESS))
669 : : {
670 [ # # ]: 0 : ereport(elevel,
671 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
672 : : errmsg("background worker \"%s\": background workers without shared memory access are not supported",
673 : : worker->bgw_name)));
674 : 0 : return false;
675 : : }
676 : :
677 [ + + ]: 3765 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
678 : : {
679 [ - + ]: 3752 : if (worker->bgw_start_time == BgWorkerStart_PostmasterStart)
680 : : {
681 [ # # ]: 0 : ereport(elevel,
682 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
683 : : errmsg("background worker \"%s\": cannot request database access if starting at postmaster start",
684 : : worker->bgw_name)));
685 : 0 : return false;
686 : : }
687 : :
688 : : /* XXX other checks? */
689 : : }
690 : :
691 : : /* Interruptible workers require a database connection */
692 [ + + ]: 3765 : if ((worker->bgw_flags & BGWORKER_INTERRUPTIBLE) &&
693 [ - + ]: 5 : !(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
694 : : {
695 [ # # ]: 0 : ereport(elevel,
696 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
697 : : errmsg("background worker \"%s\": cannot make background workers interruptible without database access",
698 : : worker->bgw_name)));
699 : 0 : return false;
700 : : }
701 : :
702 [ + + ]: 3765 : if ((worker->bgw_restart_time < 0 &&
703 [ + - ]: 2733 : worker->bgw_restart_time != BGW_NEVER_RESTART) ||
704 [ - + ]: 3765 : (worker->bgw_restart_time > USECS_PER_DAY / 1000))
705 : : {
706 [ # # ]: 0 : ereport(elevel,
707 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
708 : : errmsg("background worker \"%s\": invalid restart interval",
709 : : worker->bgw_name)));
710 : 0 : return false;
711 : : }
712 : :
713 : : /*
714 : : * Parallel workers may not be configured for restart, because the
715 : : * parallel_register_count/parallel_terminate_count accounting can't
716 : : * handle parallel workers lasting through a crash-and-restart cycle.
717 : : */
718 [ + + ]: 3765 : if (worker->bgw_restart_time != BGW_NEVER_RESTART &&
719 [ - + ]: 1032 : (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
720 : : {
721 [ # # ]: 0 : ereport(elevel,
722 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
723 : : errmsg("background worker \"%s\": parallel workers may not be configured for restart",
724 : : worker->bgw_name)));
725 : 0 : return false;
726 : : }
727 : :
728 : : /*
729 : : * If bgw_type is not filled in, use bgw_name.
730 : : */
731 [ - + ]: 3765 : if (strcmp(worker->bgw_type, "") == 0)
732 : 0 : strcpy(worker->bgw_type, worker->bgw_name);
733 : :
734 : 3765 : return true;
735 : : }
736 : :
737 : : /*
738 : : * Main entry point for background worker processes.
739 : : */
740 : : void
741 : 3447 : BackgroundWorkerMain(const void *startup_data, size_t startup_data_len)
742 : : {
743 : : sigjmp_buf local_sigjmp_buf;
744 : : BackgroundWorker *worker;
745 : : bgworker_main_type entrypt;
746 : :
747 [ - + ]: 3447 : if (startup_data == NULL)
748 [ # # ]: 0 : elog(FATAL, "unable to find bgworker entry");
749 : : Assert(startup_data_len == sizeof(BackgroundWorker));
750 : 3447 : worker = MemoryContextAlloc(TopMemoryContext, sizeof(BackgroundWorker));
751 : 3447 : memcpy(worker, startup_data, sizeof(BackgroundWorker));
752 : :
753 : : /*
754 : : * Now that we're done reading the startup data, release postmaster's
755 : : * working memory context.
756 : : */
757 [ + - ]: 3447 : if (PostmasterContext)
758 : : {
759 : 3447 : MemoryContextDelete(PostmasterContext);
760 : 3447 : PostmasterContext = NULL;
761 : : }
762 : :
763 : 3447 : MyBgworkerEntry = worker;
764 : 3447 : init_ps_display(worker->bgw_name);
765 : :
766 : : Assert(GetProcessingMode() == InitProcessing);
767 : :
768 : : /* Apply PostAuthDelay */
769 [ - + ]: 3447 : if (PostAuthDelay > 0)
770 : 0 : pg_usleep(PostAuthDelay * 1000000L);
771 : :
772 : : /*
773 : : * Set up signal handlers.
774 : : */
775 [ + + ]: 3447 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
776 : : {
777 : : /*
778 : : * SIGINT is used to signal canceling the current action
779 : : */
780 : 3434 : pqsignal(SIGINT, StatementCancelHandler);
781 : 3434 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
782 : 3434 : pqsignal(SIGFPE, FloatExceptionHandler);
783 : :
784 : : /* XXX Any other handlers needed here? */
785 : : }
786 : : else
787 : : {
788 : 13 : pqsignal(SIGINT, PG_SIG_IGN);
789 : 13 : pqsignal(SIGUSR1, PG_SIG_IGN);
790 : 13 : pqsignal(SIGFPE, PG_SIG_IGN);
791 : : }
792 : 3447 : pqsignal(SIGTERM, die);
793 : : /* SIGQUIT handler was already set up by InitPostmasterChild */
794 : 3447 : pqsignal(SIGHUP, PG_SIG_IGN);
795 : :
796 : 3447 : InitializeTimeouts(); /* establishes SIGALRM handler */
797 : :
798 : 3447 : pqsignal(SIGPIPE, PG_SIG_IGN);
799 : 3447 : pqsignal(SIGUSR2, PG_SIG_IGN);
800 : :
801 : : /*
802 : : * If an exception is encountered, processing resumes here.
803 : : *
804 : : * We just need to clean up, report the error, and go away.
805 : : */
806 [ + + ]: 3447 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
807 : : {
808 : : /* Since not using PG_TRY, must reset error stack by hand */
809 : 278 : error_context_stack = NULL;
810 : :
811 : : /* Prevent interrupts while cleaning up */
812 : 278 : HOLD_INTERRUPTS();
813 : :
814 : : /*
815 : : * sigsetjmp will have blocked all signals, but we may need to accept
816 : : * signals while communicating with our parallel leader. Once we've
817 : : * done HOLD_INTERRUPTS() it should be safe to unblock signals.
818 : : */
819 : 278 : BackgroundWorkerUnblockSignals();
820 : :
821 : : /* Report the error to the parallel leader and the server log */
822 : 278 : EmitErrorReport();
823 : :
824 : : /*
825 : : * Do we need more cleanup here? For shmem-connected bgworkers, we
826 : : * will call InitProcess below, which will install ProcKill as exit
827 : : * callback. That will take care of releasing locks, etc.
828 : : */
829 : :
830 : : /* and go away */
831 : 278 : proc_exit(1);
832 : : }
833 : :
834 : : /* We can now handle ereport(ERROR) */
835 : 3447 : PG_exception_stack = &local_sigjmp_buf;
836 : :
837 : : /*
838 : : * Create a per-backend PGPROC struct in shared memory. We must do this
839 : : * before we can use LWLocks or access any shared memory.
840 : : */
841 : 3447 : InitProcess();
842 : :
843 : : /*
844 : : * Early initialization.
845 : : */
846 : 3447 : BaseInit();
847 : :
848 : : /*
849 : : * Look up the entry point function, loading its library if necessary.
850 : : */
851 : 6894 : entrypt = LookupBackgroundWorkerFunction(worker->bgw_library_name,
852 : 3447 : worker->bgw_function_name);
853 : :
854 : : /*
855 : : * Note that in normal processes, we would call InitPostgres here. For a
856 : : * worker, however, we don't know what database to connect to, yet; so we
857 : : * need to wait until the user code does it via
858 : : * BackgroundWorkerInitializeConnection().
859 : : */
860 : :
861 : : /*
862 : : * Now invoke the user-defined worker code
863 : : */
864 : 3447 : entrypt(worker->bgw_main_arg);
865 : :
866 : : /* ... and if it returns, we're done */
867 : 2106 : proc_exit(0);
868 : : }
869 : :
870 : : /*
871 : : * Connect background worker to a database.
872 : : */
873 : : void
874 : 563 : BackgroundWorkerInitializeConnection(const char *dbname, const char *username, uint32 flags)
875 : : {
876 : 563 : BackgroundWorker *worker = MyBgworkerEntry;
877 : 563 : uint32 init_flags = 0; /* never honor session_preload_libraries */
878 : :
879 : : /* ignore datallowconn and ACL_CONNECT? */
880 [ - + ]: 563 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
881 : 0 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
882 : : /* ignore rolcanlogin? */
883 [ - + ]: 563 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
884 : 0 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
885 : :
886 : : /* XXX is this the right errcode? */
887 [ - + ]: 563 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
888 [ # # ]: 0 : ereport(FATAL,
889 : : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
890 : : errmsg("database connection requirement not indicated during registration")));
891 : :
892 : 563 : InitPostgres(dbname, InvalidOid, /* database to connect to */
893 : : username, InvalidOid, /* role to connect as */
894 : : init_flags,
895 : : NULL); /* no out_dbname */
896 : :
897 : : /* it had better not gotten out of "init" mode yet */
898 [ - + ]: 563 : if (!IsInitProcessingMode())
899 [ # # ]: 0 : ereport(ERROR,
900 : : (errmsg("invalid processing mode in background worker")));
901 : 563 : SetProcessingMode(NormalProcessing);
902 : 563 : }
903 : :
904 : : /*
905 : : * Connect background worker to a database using OIDs.
906 : : */
907 : : void
908 : 2870 : BackgroundWorkerInitializeConnectionByOid(Oid dboid, Oid useroid, uint32 flags)
909 : : {
910 : 2870 : BackgroundWorker *worker = MyBgworkerEntry;
911 : 2870 : uint32 init_flags = 0; /* never honor session_preload_libraries */
912 : :
913 : : /* ignore datallowconn and ACL_CONNECT? */
914 [ + + ]: 2870 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
915 : 2099 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
916 : : /* ignore rolcanlogin? */
917 [ + + ]: 2870 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
918 : 2056 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
919 : :
920 : : /* XXX is this the right errcode? */
921 [ - + ]: 2870 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
922 [ # # ]: 0 : ereport(FATAL,
923 : : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
924 : : errmsg("database connection requirement not indicated during registration")));
925 : :
926 : 2870 : InitPostgres(NULL, dboid, /* database to connect to */
927 : : NULL, useroid, /* role to connect as */
928 : : init_flags,
929 : : NULL); /* no out_dbname */
930 : :
931 : : /* it had better not gotten out of "init" mode yet */
932 [ - + ]: 2859 : if (!IsInitProcessingMode())
933 [ # # ]: 0 : ereport(ERROR,
934 : : (errmsg("invalid processing mode in background worker")));
935 : 2859 : SetProcessingMode(NormalProcessing);
936 : 2859 : }
937 : :
938 : : /*
939 : : * Block/unblock signals in a background worker
940 : : */
941 : : void
942 : 0 : BackgroundWorkerBlockSignals(void)
943 : : {
944 : 0 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
945 : 0 : }
946 : :
947 : : void
948 : 3725 : BackgroundWorkerUnblockSignals(void)
949 : : {
950 : 3725 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
951 : 3725 : }
952 : :
953 : : /*
954 : : * Register a new static background worker.
955 : : *
956 : : * This can only be called directly from postmaster or in the _PG_init
957 : : * function of a module library that's loaded by shared_preload_libraries;
958 : : * otherwise it will have no effect.
959 : : */
960 : : void
961 : 1034 : RegisterBackgroundWorker(BackgroundWorker *worker)
962 : : {
963 : : RegisteredBgWorker *rw;
964 : : static int numworkers = 0;
965 : :
966 : : /*
967 : : * Static background workers can only be registered in the postmaster
968 : : * process.
969 : : */
970 [ + - - + ]: 1034 : if (IsUnderPostmaster || !IsPostmasterEnvironment)
971 : : {
972 : : /*
973 : : * In EXEC_BACKEND or single-user mode, we process
974 : : * shared_preload_libraries in backend processes too. We cannot
975 : : * register static background workers at that stage, but many
976 : : * libraries' _PG_init() functions don't distinguish whether they're
977 : : * being loaded in the postmaster or in a backend, they just check
978 : : * process_shared_preload_libraries_in_progress. It's a bit sloppy,
979 : : * but for historical reasons we tolerate it. In EXEC_BACKEND mode,
980 : : * the background workers should already have been registered when the
981 : : * library was loaded in postmaster.
982 : : */
983 [ # # ]: 0 : if (process_shared_preload_libraries_in_progress)
984 : 0 : return;
985 [ # # ]: 0 : ereport(LOG,
986 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
987 : : errmsg("background worker \"%s\": must be registered in \"shared_preload_libraries\"",
988 : : worker->bgw_name)));
989 : 0 : return;
990 : : }
991 : :
992 : : /*
993 : : * Cannot register static background workers after calling
994 : : * BackgroundWorkerShmemInit().
995 : : */
996 [ - + ]: 1034 : if (BackgroundWorkerData != NULL)
997 [ # # ]: 0 : elog(ERROR, "cannot register background worker \"%s\" after shmem init",
998 : : worker->bgw_name);
999 : :
1000 [ + + ]: 1034 : ereport(DEBUG1,
1001 : : (errmsg_internal("registering background worker \"%s\"", worker->bgw_name)));
1002 : :
1003 [ - + ]: 1034 : if (!SanityCheckBackgroundWorker(worker, LOG))
1004 : 0 : return;
1005 : :
1006 [ - + ]: 1034 : if (worker->bgw_notify_pid != 0)
1007 : : {
1008 [ # # ]: 0 : ereport(LOG,
1009 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1010 : : errmsg("background worker \"%s\": only dynamic background workers can request notification",
1011 : : worker->bgw_name)));
1012 : 0 : return;
1013 : : }
1014 : :
1015 : : /*
1016 : : * Enforce maximum number of workers. Note this is overly restrictive: we
1017 : : * could allow more non-shmem-connected workers, because these don't count
1018 : : * towards the MAX_BACKENDS limit elsewhere. For now, it doesn't seem
1019 : : * important to relax this restriction.
1020 : : */
1021 [ - + ]: 1034 : if (++numworkers > max_worker_processes)
1022 : : {
1023 [ # # ]: 0 : ereport(LOG,
1024 : : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
1025 : : errmsg("too many background workers"),
1026 : : errdetail_plural("Up to %d background worker can be registered with the current settings.",
1027 : : "Up to %d background workers can be registered with the current settings.",
1028 : : max_worker_processes,
1029 : : max_worker_processes),
1030 : : errhint("Consider increasing the configuration parameter \"%s\".", "max_worker_processes")));
1031 : 0 : return;
1032 : : }
1033 : :
1034 : : /*
1035 : : * Copy the registration data into the registered workers list.
1036 : : */
1037 : 1034 : rw = MemoryContextAllocExtended(PostmasterContext,
1038 : : sizeof(RegisteredBgWorker),
1039 : : MCXT_ALLOC_NO_OOM);
1040 [ - + ]: 1034 : if (rw == NULL)
1041 : : {
1042 [ # # ]: 0 : ereport(LOG,
1043 : : (errcode(ERRCODE_OUT_OF_MEMORY),
1044 : : errmsg("out of memory")));
1045 : 0 : return;
1046 : : }
1047 : :
1048 : 1034 : rw->rw_worker = *worker;
1049 : 1034 : rw->rw_pid = 0;
1050 : 1034 : rw->rw_crashed_at = 0;
1051 : 1034 : rw->rw_terminate = false;
1052 : :
1053 : 1034 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
1054 : : }
1055 : :
1056 : : /*
1057 : : * Register a new background worker from a regular backend.
1058 : : *
1059 : : * Returns true on success and false on failure. Failure typically indicates
1060 : : * that no background worker slots are currently available.
1061 : : *
1062 : : * If handle != NULL, we'll set *handle to a pointer that can subsequently
1063 : : * be used as an argument to GetBackgroundWorkerPid(). The caller can
1064 : : * free this pointer using pfree(), if desired.
1065 : : */
1066 : : bool
1067 : 2731 : RegisterDynamicBackgroundWorker(BackgroundWorker *worker,
1068 : : BackgroundWorkerHandle **handle)
1069 : : {
1070 : : int slotno;
1071 : 2731 : bool success = false;
1072 : : bool parallel;
1073 : 2731 : uint64 generation = 0;
1074 : :
1075 : : /*
1076 : : * We can't register dynamic background workers from the postmaster. If
1077 : : * this is a standalone backend, we're the only process and can't start
1078 : : * any more. In a multi-process environment, it might be theoretically
1079 : : * possible, but we don't currently support it due to locking
1080 : : * considerations; see comments on the BackgroundWorkerSlot data
1081 : : * structure.
1082 : : */
1083 [ - + ]: 2731 : if (!IsUnderPostmaster)
1084 : 0 : return false;
1085 : :
1086 [ - + ]: 2731 : if (!SanityCheckBackgroundWorker(worker, ERROR))
1087 : 0 : return false;
1088 : :
1089 : 2731 : parallel = (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0;
1090 : :
1091 : 2731 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1092 : :
1093 : : /*
1094 : : * If this is a parallel worker, check whether there are already too many
1095 : : * parallel workers; if so, don't register another one. Our view of
1096 : : * parallel_terminate_count may be slightly stale, but that doesn't really
1097 : : * matter: we would have gotten the same result if we'd arrived here
1098 : : * slightly earlier anyway. There's no help for it, either, since the
1099 : : * postmaster must not take locks; a memory barrier wouldn't guarantee
1100 : : * anything useful.
1101 : : */
1102 [ + + ]: 2731 : if (parallel && (BackgroundWorkerData->parallel_register_count -
1103 [ + + ]: 2067 : BackgroundWorkerData->parallel_terminate_count) >=
1104 : : max_parallel_workers)
1105 : : {
1106 : : Assert(BackgroundWorkerData->parallel_register_count -
1107 : : BackgroundWorkerData->parallel_terminate_count <=
1108 : : MAX_PARALLEL_WORKER_LIMIT);
1109 : 15 : LWLockRelease(BackgroundWorkerLock);
1110 : 15 : return false;
1111 : : }
1112 : :
1113 : : /*
1114 : : * Look for an unused slot. If we find one, grab it.
1115 : : */
1116 [ + + ]: 8588 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; ++slotno)
1117 : : {
1118 : 8580 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1119 : :
1120 [ + + ]: 8580 : if (!slot->in_use)
1121 : : {
1122 : 2708 : memcpy(&slot->worker, worker, sizeof(BackgroundWorker));
1123 : 2708 : slot->pid = InvalidPid; /* indicates not started yet */
1124 : 2708 : slot->generation++;
1125 : 2708 : slot->terminate = false;
1126 : 2708 : generation = slot->generation;
1127 [ + + ]: 2708 : if (parallel)
1128 : 2044 : BackgroundWorkerData->parallel_register_count++;
1129 : :
1130 : : /*
1131 : : * Make sure postmaster doesn't see the slot as in use before it
1132 : : * sees the new contents.
1133 : : */
1134 : 2708 : pg_write_barrier();
1135 : :
1136 : 2708 : slot->in_use = true;
1137 : 2708 : success = true;
1138 : 2708 : break;
1139 : : }
1140 : : }
1141 : :
1142 : 2716 : LWLockRelease(BackgroundWorkerLock);
1143 : :
1144 : : /* If we found a slot, tell the postmaster to notice the change. */
1145 [ + + ]: 2716 : if (success)
1146 : 2708 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1147 : :
1148 : : /*
1149 : : * If we found a slot and the user has provided a handle, initialize it.
1150 : : */
1151 [ + + + - ]: 2716 : if (success && handle)
1152 : : {
1153 : 2708 : *handle = palloc_object(BackgroundWorkerHandle);
1154 : 2708 : (*handle)->slot = slotno;
1155 : 2708 : (*handle)->generation = generation;
1156 : : }
1157 : :
1158 : 2716 : return success;
1159 : : }
1160 : :
1161 : : /*
1162 : : * Get the PID of a dynamically-registered background worker.
1163 : : *
1164 : : * If the worker is determined to be running, the return value will be
1165 : : * BGWH_STARTED and *pidp will get the PID of the worker process. If the
1166 : : * postmaster has not yet attempted to start the worker, the return value will
1167 : : * be BGWH_NOT_YET_STARTED. Otherwise, the return value is BGWH_STOPPED.
1168 : : *
1169 : : * BGWH_STOPPED can indicate either that the worker is temporarily stopped
1170 : : * (because it is configured for automatic restart and exited non-zero),
1171 : : * or that the worker is permanently stopped (because it exited with exit
1172 : : * code 0, or was not configured for automatic restart), or even that the
1173 : : * worker was unregistered without ever starting (either because startup
1174 : : * failed and the worker is not configured for automatic restart, or because
1175 : : * TerminateBackgroundWorker was used before the worker was successfully
1176 : : * started).
1177 : : */
1178 : : BgwHandleStatus
1179 : 11314657 : GetBackgroundWorkerPid(BackgroundWorkerHandle *handle, pid_t *pidp)
1180 : : {
1181 : : BackgroundWorkerSlot *slot;
1182 : : pid_t pid;
1183 : :
1184 : : Assert(handle->slot < max_worker_processes);
1185 : 11314657 : slot = &BackgroundWorkerData->slot[handle->slot];
1186 : :
1187 : : /*
1188 : : * We could probably arrange to synchronize access to data using memory
1189 : : * barriers only, but for now, let's just keep it simple and grab the
1190 : : * lock. It seems unlikely that there will be enough traffic here to
1191 : : * result in meaningful contention.
1192 : : */
1193 : 11314657 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1194 : :
1195 : : /*
1196 : : * The generation number can't be concurrently changed while we hold the
1197 : : * lock. The pid, which is updated by the postmaster, can change at any
1198 : : * time, but we assume such changes are atomic. So the value we read
1199 : : * won't be garbage, but it might be out of date by the time the caller
1200 : : * examines it (but that's unavoidable anyway).
1201 : : *
1202 : : * The in_use flag could be in the process of changing from true to false,
1203 : : * but if it is already false then it can't change further.
1204 : : */
1205 [ + + + + ]: 11314657 : if (handle->generation != slot->generation || !slot->in_use)
1206 : 2100 : pid = 0;
1207 : : else
1208 : 11312557 : pid = slot->pid;
1209 : :
1210 : : /* All done. */
1211 : 11314657 : LWLockRelease(BackgroundWorkerLock);
1212 : :
1213 [ + + ]: 11314657 : if (pid == 0)
1214 : 2100 : return BGWH_STOPPED;
1215 [ + + ]: 11312557 : else if (pid == InvalidPid)
1216 : 1194032 : return BGWH_NOT_YET_STARTED;
1217 : 10118525 : *pidp = pid;
1218 : 10118525 : return BGWH_STARTED;
1219 : : }
1220 : :
1221 : : /*
1222 : : * Wait for a background worker to start up.
1223 : : *
1224 : : * This is like GetBackgroundWorkerPid(), except that if the worker has not
1225 : : * yet started, we wait for it to do so; thus, BGWH_NOT_YET_STARTED is never
1226 : : * returned. However, if the postmaster has died, we give up and return
1227 : : * BGWH_POSTMASTER_DIED, since it that case we know that startup will not
1228 : : * take place.
1229 : : *
1230 : : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1231 : : * else we will not be awoken promptly when the worker's state changes.
1232 : : */
1233 : : BgwHandleStatus
1234 : 12 : WaitForBackgroundWorkerStartup(BackgroundWorkerHandle *handle, pid_t *pidp)
1235 : : {
1236 : : BgwHandleStatus status;
1237 : : int rc;
1238 : :
1239 : : for (;;)
1240 : 16 : {
1241 : : pid_t pid;
1242 : :
1243 [ - + ]: 28 : CHECK_FOR_INTERRUPTS();
1244 : :
1245 : 28 : status = GetBackgroundWorkerPid(handle, &pid);
1246 [ + + ]: 28 : if (status == BGWH_STARTED)
1247 : 12 : *pidp = pid;
1248 [ + + ]: 28 : if (status != BGWH_NOT_YET_STARTED)
1249 : 12 : break;
1250 : :
1251 : 16 : rc = WaitLatch(MyLatch,
1252 : : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1253 : : WAIT_EVENT_BGWORKER_STARTUP);
1254 : :
1255 [ - + ]: 16 : if (rc & WL_POSTMASTER_DEATH)
1256 : : {
1257 : 0 : status = BGWH_POSTMASTER_DIED;
1258 : 0 : break;
1259 : : }
1260 : :
1261 : 16 : ResetLatch(MyLatch);
1262 : : }
1263 : :
1264 : 12 : return status;
1265 : : }
1266 : :
1267 : : /*
1268 : : * Wait for a background worker to stop.
1269 : : *
1270 : : * If the worker hasn't yet started, or is running, we wait for it to stop
1271 : : * and then return BGWH_STOPPED. However, if the postmaster has died, we give
1272 : : * up and return BGWH_POSTMASTER_DIED, because it's the postmaster that
1273 : : * notifies us when a worker's state changes.
1274 : : *
1275 : : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1276 : : * else we will not be awoken promptly when the worker's state changes.
1277 : : */
1278 : : BgwHandleStatus
1279 : 2058 : WaitForBackgroundWorkerShutdown(BackgroundWorkerHandle *handle)
1280 : : {
1281 : : BgwHandleStatus status;
1282 : : int rc;
1283 : :
1284 : : for (;;)
1285 : 2419 : {
1286 : : pid_t pid;
1287 : :
1288 [ + + ]: 4477 : CHECK_FOR_INTERRUPTS();
1289 : :
1290 : 4477 : status = GetBackgroundWorkerPid(handle, &pid);
1291 [ + + ]: 4477 : if (status == BGWH_STOPPED)
1292 : 2058 : break;
1293 : :
1294 : 2419 : rc = WaitLatch(MyLatch,
1295 : : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1296 : : WAIT_EVENT_BGWORKER_SHUTDOWN);
1297 : :
1298 [ - + ]: 2419 : if (rc & WL_POSTMASTER_DEATH)
1299 : : {
1300 : 0 : status = BGWH_POSTMASTER_DIED;
1301 : 0 : break;
1302 : : }
1303 : :
1304 : 2419 : ResetLatch(MyLatch);
1305 : : }
1306 : :
1307 : 2058 : return status;
1308 : : }
1309 : :
1310 : : /*
1311 : : * Instruct the postmaster to terminate a background worker.
1312 : : *
1313 : : * Note that it's safe to do this without regard to whether the worker is
1314 : : * still running, or even if the worker may already have exited and been
1315 : : * unregistered.
1316 : : */
1317 : : void
1318 : 31 : TerminateBackgroundWorker(BackgroundWorkerHandle *handle)
1319 : : {
1320 : : BackgroundWorkerSlot *slot;
1321 : 31 : bool signal_postmaster = false;
1322 : :
1323 : : Assert(handle->slot < max_worker_processes);
1324 : 31 : slot = &BackgroundWorkerData->slot[handle->slot];
1325 : :
1326 : : /* Set terminate flag in shared memory, unless slot has been reused. */
1327 : 31 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1328 [ + - ]: 31 : if (handle->generation == slot->generation)
1329 : : {
1330 : 31 : slot->terminate = true;
1331 : 31 : signal_postmaster = true;
1332 : : }
1333 : 31 : LWLockRelease(BackgroundWorkerLock);
1334 : :
1335 : : /* Make sure the postmaster notices the change to shared memory. */
1336 [ + - ]: 31 : if (signal_postmaster)
1337 : 31 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1338 : 31 : }
1339 : :
1340 : : /*
1341 : : * Look up (and possibly load) a bgworker entry point function.
1342 : : *
1343 : : * For functions contained in the core code, we use library name "postgres"
1344 : : * and consult the InternalBGWorkers array. External functions are
1345 : : * looked up, and loaded if necessary, using load_external_function().
1346 : : *
1347 : : * The point of this is to pass function names as strings across process
1348 : : * boundaries. We can't pass actual function addresses because of the
1349 : : * possibility that the function has been loaded at a different address
1350 : : * in a different process. This is obviously a hazard for functions in
1351 : : * loadable libraries, but it can happen even for functions in the core code
1352 : : * on platforms using EXEC_BACKEND (e.g., Windows).
1353 : : *
1354 : : * At some point it might be worthwhile to get rid of InternalBGWorkers[]
1355 : : * in favor of applying load_external_function() for core functions too;
1356 : : * but that raises portability issues that are not worth addressing now.
1357 : : */
1358 : : static bgworker_main_type
1359 : 3447 : LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname)
1360 : : {
1361 : : /*
1362 : : * If the function is to be loaded from postgres itself, search the
1363 : : * InternalBGWorkers array.
1364 : : */
1365 [ + + ]: 3447 : if (strcmp(libraryname, "postgres") == 0)
1366 : : {
1367 [ + - ]: 12033 : for (size_t i = 0; i < lengthof(InternalBGWorkers); i++)
1368 : : {
1369 [ + + ]: 12033 : if (strcmp(InternalBGWorkers[i].fn_name, funcname) == 0)
1370 : 3424 : return InternalBGWorkers[i].fn_addr;
1371 : : }
1372 : :
1373 : : /* We can only reach this by programming error. */
1374 [ # # ]: 0 : elog(ERROR, "internal function \"%s\" not found", funcname);
1375 : : }
1376 : :
1377 : : /* Otherwise load from external library. */
1378 : 23 : return (bgworker_main_type)
1379 : 23 : load_external_function(libraryname, funcname, true, NULL);
1380 : : }
1381 : :
1382 : : /*
1383 : : * Given a PID, get the bgw_type of the background worker. Returns NULL if
1384 : : * not a valid background worker.
1385 : : *
1386 : : * The return value is in static memory belonging to this function, so it has
1387 : : * to be used before calling this function again. This is so that the caller
1388 : : * doesn't have to worry about the background worker locking protocol.
1389 : : */
1390 : : const char *
1391 : 896 : GetBackgroundWorkerTypeByPid(pid_t pid)
1392 : : {
1393 : : int slotno;
1394 : 896 : bool found = false;
1395 : : static char result[BGW_MAXLEN];
1396 : :
1397 : 896 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1398 : :
1399 [ + - ]: 1073 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1400 : : {
1401 : 1073 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1402 : :
1403 [ + + + + ]: 1073 : if (slot->pid > 0 && slot->pid == pid)
1404 : : {
1405 : 896 : strcpy(result, slot->worker.bgw_type);
1406 : 896 : found = true;
1407 : 896 : break;
1408 : : }
1409 : : }
1410 : :
1411 : 896 : LWLockRelease(BackgroundWorkerLock);
1412 : :
1413 [ - + ]: 896 : if (!found)
1414 : 0 : return NULL;
1415 : :
1416 : 896 : return result;
1417 : : }
1418 : :
1419 : : /*
1420 : : * Terminate all background workers connected to the given database, if the
1421 : : * workers can be interrupted. The workers are signaled directly, so that
1422 : : * they can be restarted based on their bgw_restart_time.
1423 : : */
1424 : : void
1425 : 10 : TerminateBackgroundWorkersForDatabase(Oid databaseId)
1426 : : {
1427 : : pid_t *pids;
1428 : 10 : int npids = 0;
1429 : :
1430 [ + + ]: 10 : elog(DEBUG1, "attempting worker termination for database %u",
1431 : : databaseId);
1432 : :
1433 : 10 : pids = palloc_array(pid_t, BackgroundWorkerData->total_slots);
1434 : :
1435 : 10 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1436 : :
1437 : : /*
1438 : : * Iterate through slots, looking for workers connected to the given
1439 : : * database.
1440 : : */
1441 [ + + ]: 92 : for (int slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1442 : : {
1443 : 82 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1444 : :
1445 [ + + ]: 82 : if (slot->in_use &&
1446 [ + + ]: 21 : (slot->worker.bgw_flags & BGWORKER_INTERRUPTIBLE))
1447 : : {
1448 : : PGPROC *proc;
1449 : 5 : pid_t pid = slot->pid;
1450 : :
1451 : 5 : LWLockAcquire(ProcArrayLock, LW_SHARED);
1452 : 5 : proc = BackendPidGetProcWithLock(pid);
1453 [ + - + - ]: 5 : if (proc && proc->databaseId == databaseId)
1454 : 5 : pids[npids++] = pid;
1455 : 5 : LWLockRelease(ProcArrayLock);
1456 : : }
1457 : : }
1458 : :
1459 : 10 : LWLockRelease(BackgroundWorkerLock);
1460 : :
1461 : : /*
1462 : : * Signal the workers, avoiding kill() while holding the locks.
1463 : : */
1464 [ + + ]: 15 : for (int i = 0; i < npids; i++)
1465 : : {
1466 [ + - ]: 5 : elog(DEBUG1, "termination requested for worker (PID %d) on database %u",
1467 : : (int) pids[i], databaseId);
1468 : 5 : (void) kill(pids[i], SIGTERM); /* ignore any error */
1469 : : }
1470 : :
1471 : 10 : pfree(pids);
1472 : 10 : }
|
