Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * method_worker.c
4 : * AIO - perform AIO using worker processes
5 : *
6 : * IO workers consume IOs from a shared memory submission queue, run
7 : * traditional synchronous system calls, and perform the shared completion
8 : * handling immediately. Client code submits most requests by pushing IOs
9 : * into the submission queue, and waits (if necessary) using condition
10 : * variables. Some IOs cannot be performed in another process due to lack of
11 : * infrastructure for reopening the file, and must processed synchronously by
12 : * the client code when submitted.
13 : *
14 : * So that the submitter can make just one system call when submitting a batch
15 : * of IOs, wakeups "fan out"; each woken IO worker can wake two more. XXX This
16 : * could be improved by using futexes instead of latches to wake N waiters.
17 : *
18 : * This method of AIO is available in all builds on all operating systems, and
19 : * is the default.
20 : *
21 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
22 : * Portions Copyright (c) 1994, Regents of the University of California
23 : *
24 : * IDENTIFICATION
25 : * src/backend/storage/aio/method_worker.c
26 : *
27 : *-------------------------------------------------------------------------
28 : */
29 :
30 : #include "postgres.h"
31 :
32 : #include "libpq/pqsignal.h"
33 : #include "miscadmin.h"
34 : #include "port/pg_bitutils.h"
35 : #include "postmaster/auxprocess.h"
36 : #include "postmaster/interrupt.h"
37 : #include "storage/aio.h"
38 : #include "storage/aio_internal.h"
39 : #include "storage/aio_subsys.h"
40 : #include "storage/io_worker.h"
41 : #include "storage/ipc.h"
42 : #include "storage/latch.h"
43 : #include "storage/proc.h"
44 : #include "tcop/tcopprot.h"
45 : #include "utils/injection_point.h"
46 : #include "utils/memdebug.h"
47 : #include "utils/ps_status.h"
48 : #include "utils/wait_event.h"
49 :
50 :
51 : /* How many workers should each worker wake up if needed? */
52 : #define IO_WORKER_WAKEUP_FANOUT 2
53 :
54 :
55 : typedef struct PgAioWorkerSubmissionQueue
56 : {
57 : uint32 size;
58 : uint32 head;
59 : uint32 tail;
60 : int sqes[FLEXIBLE_ARRAY_MEMBER];
61 : } PgAioWorkerSubmissionQueue;
62 :
63 : typedef struct PgAioWorkerSlot
64 : {
65 : Latch *latch;
66 : bool in_use;
67 : } PgAioWorkerSlot;
68 :
69 : typedef struct PgAioWorkerControl
70 : {
71 : uint64 idle_worker_mask;
72 : PgAioWorkerSlot workers[FLEXIBLE_ARRAY_MEMBER];
73 : } PgAioWorkerControl;
74 :
75 :
76 : static size_t pgaio_worker_shmem_size(void);
77 : static void pgaio_worker_shmem_init(bool first_time);
78 :
79 : static bool pgaio_worker_needs_synchronous_execution(PgAioHandle *ioh);
80 : static int pgaio_worker_submit(uint16 num_staged_ios, PgAioHandle **staged_ios);
81 :
82 :
83 : const IoMethodOps pgaio_worker_ops = {
84 : .shmem_size = pgaio_worker_shmem_size,
85 : .shmem_init = pgaio_worker_shmem_init,
86 :
87 : .needs_synchronous_execution = pgaio_worker_needs_synchronous_execution,
88 : .submit = pgaio_worker_submit,
89 : };
90 :
91 :
92 : /* GUCs */
93 : int io_workers = 3;
94 :
95 :
96 : static int io_worker_queue_size = 64;
97 : static int MyIoWorkerId;
98 : static PgAioWorkerSubmissionQueue *io_worker_submission_queue;
99 : static PgAioWorkerControl *io_worker_control;
100 :
101 :
102 : static size_t
103 3375 : pgaio_worker_queue_shmem_size(int *queue_size)
104 : {
105 : /* Round size up to next power of two so we can make a mask. */
106 3375 : *queue_size = pg_nextpower2_32(io_worker_queue_size);
107 :
108 6750 : return offsetof(PgAioWorkerSubmissionQueue, sqes) +
109 3375 : sizeof(int) * *queue_size;
110 : }
111 :
112 : static size_t
113 3375 : pgaio_worker_control_shmem_size(void)
114 : {
115 3375 : return offsetof(PgAioWorkerControl, workers) +
116 : sizeof(PgAioWorkerSlot) * MAX_IO_WORKERS;
117 : }
118 :
119 : static size_t
120 2200 : pgaio_worker_shmem_size(void)
121 : {
122 : size_t sz;
123 : int queue_size;
124 :
125 2200 : sz = pgaio_worker_queue_shmem_size(&queue_size);
126 2200 : sz = add_size(sz, pgaio_worker_control_shmem_size());
127 :
128 2200 : return sz;
129 : }
130 :
131 : static void
132 1175 : pgaio_worker_shmem_init(bool first_time)
133 : {
134 : bool found;
135 : int queue_size;
136 :
137 1175 : io_worker_submission_queue =
138 1175 : ShmemInitStruct("AioWorkerSubmissionQueue",
139 : pgaio_worker_queue_shmem_size(&queue_size),
140 : &found);
141 1175 : if (!found)
142 : {
143 1175 : io_worker_submission_queue->size = queue_size;
144 1175 : io_worker_submission_queue->head = 0;
145 1175 : io_worker_submission_queue->tail = 0;
146 : }
147 :
148 1175 : io_worker_control =
149 1175 : ShmemInitStruct("AioWorkerControl",
150 : pgaio_worker_control_shmem_size(),
151 : &found);
152 1175 : if (!found)
153 : {
154 1175 : io_worker_control->idle_worker_mask = 0;
155 38775 : for (int i = 0; i < MAX_IO_WORKERS; ++i)
156 : {
157 37600 : io_worker_control->workers[i].latch = NULL;
158 37600 : io_worker_control->workers[i].in_use = false;
159 : }
160 : }
161 1175 : }
162 :
163 : static int
164 657031 : pgaio_worker_choose_idle(void)
165 : {
166 : int worker;
167 :
168 657031 : if (io_worker_control->idle_worker_mask == 0)
169 22135 : return -1;
170 :
171 : /* Find the lowest bit position, and clear it. */
172 634896 : worker = pg_rightmost_one_pos64(io_worker_control->idle_worker_mask);
173 634896 : io_worker_control->idle_worker_mask &= ~(UINT64_C(1) << worker);
174 : Assert(io_worker_control->workers[worker].in_use);
175 :
176 634896 : return worker;
177 : }
178 :
179 : static bool
180 641534 : pgaio_worker_submission_queue_insert(PgAioHandle *ioh)
181 : {
182 : PgAioWorkerSubmissionQueue *queue;
183 : uint32 new_head;
184 :
185 641534 : queue = io_worker_submission_queue;
186 641534 : new_head = (queue->head + 1) & (queue->size - 1);
187 641534 : if (new_head == queue->tail)
188 : {
189 0 : pgaio_debug(DEBUG3, "io queue is full, at %u elements",
190 : io_worker_submission_queue->size);
191 0 : return false; /* full */
192 : }
193 :
194 641534 : queue->sqes[queue->head] = pgaio_io_get_id(ioh);
195 641534 : queue->head = new_head;
196 :
197 641534 : return true;
198 : }
199 :
200 : static int
201 1009162 : pgaio_worker_submission_queue_consume(void)
202 : {
203 : PgAioWorkerSubmissionQueue *queue;
204 : int result;
205 :
206 1009162 : queue = io_worker_submission_queue;
207 1009162 : if (queue->tail == queue->head)
208 504340 : return -1; /* empty */
209 :
210 504822 : result = queue->sqes[queue->tail];
211 504822 : queue->tail = (queue->tail + 1) & (queue->size - 1);
212 :
213 504822 : return result;
214 : }
215 :
216 : static uint32
217 1008453 : pgaio_worker_submission_queue_depth(void)
218 : {
219 : uint32 head;
220 : uint32 tail;
221 :
222 1008453 : head = io_worker_submission_queue->head;
223 1008453 : tail = io_worker_submission_queue->tail;
224 :
225 1008453 : if (tail > head)
226 463 : head += io_worker_submission_queue->size;
227 :
228 : Assert(head >= tail);
229 :
230 1008453 : return head - tail;
231 : }
232 :
233 : static bool
234 648394 : pgaio_worker_needs_synchronous_execution(PgAioHandle *ioh)
235 : {
236 : return
237 648394 : !IsUnderPostmaster
238 644603 : || ioh->flags & PGAIO_HF_REFERENCES_LOCAL
239 1292997 : || !pgaio_io_can_reopen(ioh);
240 : }
241 :
242 : static void
243 642535 : pgaio_worker_submit_internal(int num_staged_ios, PgAioHandle **staged_ios)
244 : {
245 642535 : PgAioHandle **synchronous_ios = NULL;
246 642535 : int nsync = 0;
247 642535 : Latch *wakeup = NULL;
248 : int worker;
249 :
250 : Assert(num_staged_ios <= PGAIO_SUBMIT_BATCH_SIZE);
251 :
252 642535 : if (LWLockConditionalAcquire(AioWorkerSubmissionQueueLock, LW_EXCLUSIVE))
253 : {
254 1282530 : for (int i = 0; i < num_staged_ios; ++i)
255 : {
256 : Assert(!pgaio_worker_needs_synchronous_execution(staged_ios[i]));
257 641534 : if (!pgaio_worker_submission_queue_insert(staged_ios[i]))
258 : {
259 : /*
260 : * Do the rest synchronously. If the queue is full, give up
261 : * and do the rest synchronously. We're holding an exclusive
262 : * lock on the queue so nothing can consume entries.
263 : */
264 0 : synchronous_ios = &staged_ios[i];
265 0 : nsync = (num_staged_ios - i);
266 :
267 0 : break;
268 : }
269 :
270 641534 : if (wakeup == NULL)
271 : {
272 : /* Choose an idle worker to wake up if we haven't already. */
273 641024 : worker = pgaio_worker_choose_idle();
274 641024 : if (worker >= 0)
275 628421 : wakeup = io_worker_control->workers[worker].latch;
276 :
277 641024 : pgaio_debug_io(DEBUG4, staged_ios[i],
278 : "choosing worker %d",
279 : worker);
280 : }
281 : }
282 640996 : LWLockRelease(AioWorkerSubmissionQueueLock);
283 : }
284 : else
285 : {
286 : /* do everything synchronously, no wakeup needed */
287 1539 : synchronous_ios = staged_ios;
288 1539 : nsync = num_staged_ios;
289 : }
290 :
291 642535 : if (wakeup)
292 628421 : SetLatch(wakeup);
293 :
294 : /* Run whatever is left synchronously. */
295 642535 : if (nsync > 0)
296 : {
297 3078 : for (int i = 0; i < nsync; ++i)
298 : {
299 1539 : pgaio_io_perform_synchronously(synchronous_ios[i]);
300 : }
301 : }
302 642535 : }
303 :
304 : static int
305 642535 : pgaio_worker_submit(uint16 num_staged_ios, PgAioHandle **staged_ios)
306 : {
307 1285608 : for (int i = 0; i < num_staged_ios; i++)
308 : {
309 643073 : PgAioHandle *ioh = staged_ios[i];
310 :
311 643073 : pgaio_io_prepare_submit(ioh);
312 : }
313 :
314 642535 : pgaio_worker_submit_internal(num_staged_ios, staged_ios);
315 :
316 642535 : return num_staged_ios;
317 : }
318 :
319 : /*
320 : * on_shmem_exit() callback that releases the worker's slot in
321 : * io_worker_control.
322 : */
323 : static void
324 1867 : pgaio_worker_die(int code, Datum arg)
325 : {
326 1867 : LWLockAcquire(AioWorkerSubmissionQueueLock, LW_EXCLUSIVE);
327 : Assert(io_worker_control->workers[MyIoWorkerId].in_use);
328 : Assert(io_worker_control->workers[MyIoWorkerId].latch == MyLatch);
329 :
330 1867 : io_worker_control->idle_worker_mask &= ~(UINT64_C(1) << MyIoWorkerId);
331 1867 : io_worker_control->workers[MyIoWorkerId].in_use = false;
332 1867 : io_worker_control->workers[MyIoWorkerId].latch = NULL;
333 1867 : LWLockRelease(AioWorkerSubmissionQueueLock);
334 1867 : }
335 :
336 : /*
337 : * Register the worker in shared memory, assign MyIoWorkerId and register a
338 : * shutdown callback to release registration.
339 : */
340 : static void
341 1867 : pgaio_worker_register(void)
342 : {
343 1867 : MyIoWorkerId = -1;
344 :
345 : /*
346 : * XXX: This could do with more fine-grained locking. But it's also not
347 : * very common for the number of workers to change at the moment...
348 : */
349 1867 : LWLockAcquire(AioWorkerSubmissionQueueLock, LW_EXCLUSIVE);
350 :
351 4391 : for (int i = 0; i < MAX_IO_WORKERS; ++i)
352 : {
353 4391 : if (!io_worker_control->workers[i].in_use)
354 : {
355 : Assert(io_worker_control->workers[i].latch == NULL);
356 1867 : io_worker_control->workers[i].in_use = true;
357 1867 : MyIoWorkerId = i;
358 1867 : break;
359 : }
360 : else
361 : Assert(io_worker_control->workers[i].latch != NULL);
362 : }
363 :
364 1867 : if (MyIoWorkerId == -1)
365 0 : elog(ERROR, "couldn't find a free worker slot");
366 :
367 1867 : io_worker_control->idle_worker_mask |= (UINT64_C(1) << MyIoWorkerId);
368 1867 : io_worker_control->workers[MyIoWorkerId].latch = MyLatch;
369 1867 : LWLockRelease(AioWorkerSubmissionQueueLock);
370 :
371 1867 : on_shmem_exit(pgaio_worker_die, 0);
372 1867 : }
373 :
374 : static void
375 1169 : pgaio_worker_error_callback(void *arg)
376 : {
377 : ProcNumber owner;
378 : PGPROC *owner_proc;
379 : int32 owner_pid;
380 1169 : PgAioHandle *ioh = arg;
381 :
382 1169 : if (!ioh)
383 0 : return;
384 :
385 : Assert(ioh->owner_procno != MyProcNumber);
386 : Assert(MyBackendType == B_IO_WORKER);
387 :
388 1169 : owner = ioh->owner_procno;
389 1169 : owner_proc = GetPGProcByNumber(owner);
390 1169 : owner_pid = owner_proc->pid;
391 :
392 1169 : errcontext("I/O worker executing I/O on behalf of process %d", owner_pid);
393 : }
394 :
395 : void
396 1867 : IoWorkerMain(const void *startup_data, size_t startup_data_len)
397 : {
398 : sigjmp_buf local_sigjmp_buf;
399 1867 : PgAioHandle *volatile error_ioh = NULL;
400 1867 : ErrorContextCallback errcallback = {0};
401 1867 : volatile int error_errno = 0;
402 : char cmd[128];
403 :
404 1867 : AuxiliaryProcessMainCommon();
405 :
406 1867 : pqsignal(SIGHUP, SignalHandlerForConfigReload);
407 1867 : pqsignal(SIGINT, die); /* to allow manually triggering worker restart */
408 :
409 : /*
410 : * Ignore SIGTERM, will get explicit shutdown via SIGUSR2 later in the
411 : * shutdown sequence, similar to checkpointer.
412 : */
413 1867 : pqsignal(SIGTERM, SIG_IGN);
414 : /* SIGQUIT handler was already set up by InitPostmasterChild */
415 1867 : pqsignal(SIGALRM, SIG_IGN);
416 1867 : pqsignal(SIGPIPE, SIG_IGN);
417 1867 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
418 1867 : pqsignal(SIGUSR2, SignalHandlerForShutdownRequest);
419 :
420 : /* also registers a shutdown callback to unregister */
421 1867 : pgaio_worker_register();
422 :
423 1867 : sprintf(cmd, "%d", MyIoWorkerId);
424 1867 : set_ps_display(cmd);
425 :
426 1867 : errcallback.callback = pgaio_worker_error_callback;
427 1867 : errcallback.previous = error_context_stack;
428 1867 : error_context_stack = &errcallback;
429 :
430 : /* see PostgresMain() */
431 1867 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
432 : {
433 1 : error_context_stack = NULL;
434 1 : HOLD_INTERRUPTS();
435 :
436 1 : EmitErrorReport();
437 :
438 : /*
439 : * In the - very unlikely - case that the IO failed in a way that
440 : * raises an error we need to mark the IO as failed.
441 : *
442 : * Need to do just enough error recovery so that we can mark the IO as
443 : * failed and then exit (postmaster will start a new worker).
444 : */
445 1 : LWLockReleaseAll();
446 :
447 1 : if (error_ioh != NULL)
448 : {
449 : /* should never fail without setting error_errno */
450 : Assert(error_errno != 0);
451 :
452 1 : errno = error_errno;
453 :
454 1 : START_CRIT_SECTION();
455 1 : pgaio_io_process_completion(error_ioh, -error_errno);
456 1 : END_CRIT_SECTION();
457 : }
458 :
459 1 : proc_exit(1);
460 : }
461 :
462 : /* We can now handle ereport(ERROR) */
463 1867 : PG_exception_stack = &local_sigjmp_buf;
464 :
465 1867 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
466 :
467 1011003 : while (!ShutdownRequestPending)
468 : {
469 : uint32 io_index;
470 : Latch *latches[IO_WORKER_WAKEUP_FANOUT];
471 1009162 : int nlatches = 0;
472 1009162 : int nwakeups = 0;
473 : int worker;
474 :
475 : /*
476 : * Try to get a job to do.
477 : *
478 : * The lwlock acquisition also provides the necessary memory barrier
479 : * to ensure that we don't see an outdated data in the handle.
480 : */
481 1009162 : LWLockAcquire(AioWorkerSubmissionQueueLock, LW_EXCLUSIVE);
482 1009162 : if ((io_index = pgaio_worker_submission_queue_consume()) == -1)
483 : {
484 : /*
485 : * Nothing to do. Mark self idle.
486 : *
487 : * XXX: Invent some kind of back pressure to reduce useless
488 : * wakeups?
489 : */
490 504340 : io_worker_control->idle_worker_mask |= (UINT64_C(1) << MyIoWorkerId);
491 : }
492 : else
493 : {
494 : /* Got one. Clear idle flag. */
495 504822 : io_worker_control->idle_worker_mask &= ~(UINT64_C(1) << MyIoWorkerId);
496 :
497 : /* See if we can wake up some peers. */
498 504822 : nwakeups = Min(pgaio_worker_submission_queue_depth(),
499 : IO_WORKER_WAKEUP_FANOUT);
500 511297 : for (int i = 0; i < nwakeups; ++i)
501 : {
502 16007 : if ((worker = pgaio_worker_choose_idle()) < 0)
503 9532 : break;
504 6475 : latches[nlatches++] = io_worker_control->workers[worker].latch;
505 : }
506 : }
507 1009162 : LWLockRelease(AioWorkerSubmissionQueueLock);
508 :
509 1015637 : for (int i = 0; i < nlatches; ++i)
510 6475 : SetLatch(latches[i]);
511 :
512 1009162 : if (io_index != -1)
513 : {
514 504822 : PgAioHandle *ioh = NULL;
515 :
516 504822 : ioh = &pgaio_ctl->io_handles[io_index];
517 504822 : error_ioh = ioh;
518 504822 : errcallback.arg = ioh;
519 :
520 504822 : pgaio_debug_io(DEBUG4, ioh,
521 : "worker %d processing IO",
522 : MyIoWorkerId);
523 :
524 : /*
525 : * Prevent interrupts between pgaio_io_reopen() and
526 : * pgaio_io_perform_synchronously() that otherwise could lead to
527 : * the FD getting closed in that window.
528 : */
529 504822 : HOLD_INTERRUPTS();
530 :
531 : /*
532 : * It's very unlikely, but possible, that reopen fails. E.g. due
533 : * to memory allocations failing or file permissions changing or
534 : * such. In that case we need to fail the IO.
535 : *
536 : * There's not really a good errno we can report here.
537 : */
538 504822 : error_errno = ENOENT;
539 504822 : pgaio_io_reopen(ioh);
540 :
541 : /*
542 : * To be able to exercise the reopen-fails path, allow injection
543 : * points to trigger a failure at this point.
544 : */
545 504822 : INJECTION_POINT("aio-worker-after-reopen", ioh);
546 :
547 504821 : error_errno = 0;
548 504821 : error_ioh = NULL;
549 :
550 : /*
551 : * As part of IO completion the buffer will be marked as NOACCESS,
552 : * until the buffer is pinned again - which never happens in io
553 : * workers. Therefore the next time there is IO for the same
554 : * buffer, the memory will be considered inaccessible. To avoid
555 : * that, explicitly allow access to the memory before reading data
556 : * into it.
557 : */
558 : #ifdef USE_VALGRIND
559 : {
560 : struct iovec *iov;
561 : uint16 iov_length = pgaio_io_get_iovec_length(ioh, &iov);
562 :
563 : for (int i = 0; i < iov_length; i++)
564 : VALGRIND_MAKE_MEM_UNDEFINED(iov[i].iov_base, iov[i].iov_len);
565 : }
566 : #endif
567 :
568 : /*
569 : * We don't expect this to ever fail with ERROR or FATAL, no need
570 : * to keep error_ioh set to the IO.
571 : * pgaio_io_perform_synchronously() contains a critical section to
572 : * ensure we don't accidentally fail.
573 : */
574 504821 : pgaio_io_perform_synchronously(ioh);
575 :
576 504821 : RESUME_INTERRUPTS();
577 504821 : errcallback.arg = NULL;
578 : }
579 : else
580 : {
581 504340 : WaitLatch(MyLatch, WL_LATCH_SET | WL_EXIT_ON_PM_DEATH, -1,
582 : WAIT_EVENT_IO_WORKER_MAIN);
583 504319 : ResetLatch(MyLatch);
584 : }
585 :
586 1009140 : CHECK_FOR_INTERRUPTS();
587 :
588 1009136 : if (ConfigReloadPending)
589 : {
590 268 : ConfigReloadPending = false;
591 268 : ProcessConfigFile(PGC_SIGHUP);
592 : }
593 : }
594 :
595 1841 : error_context_stack = errcallback.previous;
596 1841 : proc_exit(0);
597 : }
598 :
599 : bool
600 174753 : pgaio_workers_enabled(void)
601 : {
602 174753 : return io_method == IOMETHOD_WORKER;
603 : }
|
