Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * async.c
4 : * Asynchronous notification: NOTIFY, LISTEN, UNLISTEN
5 : *
6 : * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : * IDENTIFICATION
10 : * src/backend/commands/async.c
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 :
15 : /*-------------------------------------------------------------------------
16 : * Async Notification Model as of 9.0:
17 : *
18 : * 1. Multiple backends on same machine. Multiple backends listening on
19 : * several channels. (Channels are also called "conditions" in other
20 : * parts of the code.)
21 : *
22 : * 2. There is one central queue in disk-based storage (directory pg_notify/),
23 : * with actively-used pages mapped into shared memory by the slru.c module.
24 : * All notification messages are placed in the queue and later read out
25 : * by listening backends.
26 : *
27 : * There is no central knowledge of which backend listens on which channel;
28 : * every backend has its own list of interesting channels.
29 : *
30 : * Although there is only one queue, notifications are treated as being
31 : * database-local; this is done by including the sender's database OID
32 : * in each notification message. Listening backends ignore messages
33 : * that don't match their database OID. This is important because it
34 : * ensures senders and receivers have the same database encoding and won't
35 : * misinterpret non-ASCII text in the channel name or payload string.
36 : *
37 : * Since notifications are not expected to survive database crashes,
38 : * we can simply clean out the pg_notify data at any reboot, and there
39 : * is no need for WAL support or fsync'ing.
40 : *
41 : * 3. Every backend that is listening on at least one channel registers by
42 : * entering its PID into the array in AsyncQueueControl. It then scans all
43 : * incoming notifications in the central queue and first compares the
44 : * database OID of the notification with its own database OID and then
45 : * compares the notified channel with the list of channels that it listens
46 : * to. In case there is a match it delivers the notification event to its
47 : * frontend. Non-matching events are simply skipped.
48 : *
49 : * 4. The NOTIFY statement (routine Async_Notify) stores the notification in
50 : * a backend-local list which will not be processed until transaction end.
51 : *
52 : * Duplicate notifications from the same transaction are sent out as one
53 : * notification only. This is done to save work when for example a trigger
54 : * on a 2 million row table fires a notification for each row that has been
55 : * changed. If the application needs to receive every single notification
56 : * that has been sent, it can easily add some unique string into the extra
57 : * payload parameter.
58 : *
59 : * When the transaction is ready to commit, PreCommit_Notify() adds the
60 : * pending notifications to the head of the queue. The head pointer of the
61 : * queue always points to the next free position and a position is just a
62 : * page number and the offset in that page. This is done before marking the
63 : * transaction as committed in clog. If we run into problems writing the
64 : * notifications, we can still call elog(ERROR, ...) and the transaction
65 : * will roll back.
66 : *
67 : * Once we have put all of the notifications into the queue, we return to
68 : * CommitTransaction() which will then do the actual transaction commit.
69 : *
70 : * After commit we are called another time (AtCommit_Notify()). Here we
71 : * make the actual updates to the effective listen state (listenChannels).
72 : *
73 : * Finally, after we are out of the transaction altogether, we check if
74 : * we need to signal listening backends. In SignalBackends() we scan the
75 : * list of listening backends and send a PROCSIG_NOTIFY_INTERRUPT signal
76 : * to every listening backend (we don't know which backend is listening on
77 : * which channel so we must signal them all). We can exclude backends that
78 : * are already up to date, though. We don't bother with a self-signal
79 : * either, but just process the queue directly.
80 : *
81 : * 5. Upon receipt of a PROCSIG_NOTIFY_INTERRUPT signal, the signal handler
82 : * sets the process's latch, which triggers the event to be processed
83 : * immediately if this backend is idle (i.e., it is waiting for a frontend
84 : * command and is not within a transaction block. C.f.
85 : * ProcessClientReadInterrupt()). Otherwise the handler may only set a
86 : * flag, which will cause the processing to occur just before we next go
87 : * idle.
88 : *
89 : * Inbound-notify processing consists of reading all of the notifications
90 : * that have arrived since scanning last time. We read every notification
91 : * until we reach either a notification from an uncommitted transaction or
92 : * the head pointer's position. Then we check if we were the laziest
93 : * backend: if our pointer is set to the same position as the global tail
94 : * pointer is set, then we move the global tail pointer ahead to where the
95 : * second-laziest backend is (in general, we take the MIN of the current
96 : * head position and all active backends' new tail pointers). Whenever we
97 : * move the global tail pointer we also truncate now-unused pages (i.e.,
98 : * delete files in pg_notify/ that are no longer used).
99 : *
100 : * An application that listens on the same channel it notifies will get
101 : * NOTIFY messages for its own NOTIFYs. These can be ignored, if not useful,
102 : * by comparing be_pid in the NOTIFY message to the application's own backend's
103 : * PID. (As of FE/BE protocol 2.0, the backend's PID is provided to the
104 : * frontend during startup.) The above design guarantees that notifies from
105 : * other backends will never be missed by ignoring self-notifies.
106 : *
107 : * The amount of shared memory used for notify management (NUM_ASYNC_BUFFERS)
108 : * can be varied without affecting anything but performance. The maximum
109 : * amount of notification data that can be queued at one time is determined
110 : * by slru.c's wraparound limit; see QUEUE_MAX_PAGE below.
111 : *-------------------------------------------------------------------------
112 : */
113 :
114 : #include "postgres.h"
115 :
116 : #include <limits.h>
117 : #include <unistd.h>
118 : #include <signal.h>
119 :
120 : #include "access/parallel.h"
121 : #include "access/slru.h"
122 : #include "access/transam.h"
123 : #include "access/xact.h"
124 : #include "catalog/pg_database.h"
125 : #include "commands/async.h"
126 : #include "funcapi.h"
127 : #include "libpq/libpq.h"
128 : #include "libpq/pqformat.h"
129 : #include "miscadmin.h"
130 : #include "storage/ipc.h"
131 : #include "storage/lmgr.h"
132 : #include "storage/proc.h"
133 : #include "storage/procarray.h"
134 : #include "storage/procsignal.h"
135 : #include "storage/sinval.h"
136 : #include "tcop/tcopprot.h"
137 : #include "utils/builtins.h"
138 : #include "utils/memutils.h"
139 : #include "utils/ps_status.h"
140 : #include "utils/timestamp.h"
141 :
142 :
143 : /*
144 : * Maximum size of a NOTIFY payload, including terminating NULL. This
145 : * must be kept small enough so that a notification message fits on one
146 : * SLRU page. The magic fudge factor here is noncritical as long as it's
147 : * more than AsyncQueueEntryEmptySize --- we make it significantly bigger
148 : * than that, so changes in that data structure won't affect user-visible
149 : * restrictions.
150 : */
151 : #define NOTIFY_PAYLOAD_MAX_LENGTH (BLCKSZ - NAMEDATALEN - 128)
152 :
153 : /*
154 : * Struct representing an entry in the global notify queue
155 : *
156 : * This struct declaration has the maximal length, but in a real queue entry
157 : * the data area is only big enough for the actual channel and payload strings
158 : * (each null-terminated). AsyncQueueEntryEmptySize is the minimum possible
159 : * entry size, if both channel and payload strings are empty (but note it
160 : * doesn't include alignment padding).
161 : *
162 : * The "length" field should always be rounded up to the next QUEUEALIGN
163 : * multiple so that all fields are properly aligned.
164 : */
165 : typedef struct AsyncQueueEntry
166 : {
167 : int length; /* total allocated length of entry */
168 : Oid dboid; /* sender's database OID */
169 : TransactionId xid; /* sender's XID */
170 : int32 srcPid; /* sender's PID */
171 : char data[NAMEDATALEN + NOTIFY_PAYLOAD_MAX_LENGTH];
172 : } AsyncQueueEntry;
173 :
174 : /* Currently, no field of AsyncQueueEntry requires more than int alignment */
175 : #define QUEUEALIGN(len) INTALIGN(len)
176 :
177 : #define AsyncQueueEntryEmptySize (offsetof(AsyncQueueEntry, data) + 2)
178 :
179 : /*
180 : * Struct describing a queue position, and assorted macros for working with it
181 : */
182 : typedef struct QueuePosition
183 : {
184 : int page; /* SLRU page number */
185 : int offset; /* byte offset within page */
186 : } QueuePosition;
187 :
188 : #define QUEUE_POS_PAGE(x) ((x).page)
189 : #define QUEUE_POS_OFFSET(x) ((x).offset)
190 :
191 : #define SET_QUEUE_POS(x,y,z) \
192 : do { \
193 : (x).page = (y); \
194 : (x).offset = (z); \
195 : } while (0)
196 :
197 : #define QUEUE_POS_EQUAL(x,y) \
198 : ((x).page == (y).page && (x).offset == (y).offset)
199 :
200 : /* choose logically smaller QueuePosition */
201 : #define QUEUE_POS_MIN(x,y) \
202 : (asyncQueuePagePrecedes((x).page, (y).page) ? (x) : \
203 : (x).page != (y).page ? (y) : \
204 : (x).offset < (y).offset ? (x) : (y))
205 :
206 : /* choose logically larger QueuePosition */
207 : #define QUEUE_POS_MAX(x,y) \
208 : (asyncQueuePagePrecedes((x).page, (y).page) ? (y) : \
209 : (x).page != (y).page ? (x) : \
210 : (x).offset > (y).offset ? (x) : (y))
211 :
212 : /*
213 : * Struct describing a listening backend's status
214 : */
215 : typedef struct QueueBackendStatus
216 : {
217 : int32 pid; /* either a PID or InvalidPid */
218 : Oid dboid; /* backend's database OID, or InvalidOid */
219 : QueuePosition pos; /* backend has read queue up to here */
220 : } QueueBackendStatus;
221 :
222 : /*
223 : * Shared memory state for LISTEN/NOTIFY (excluding its SLRU stuff)
224 : *
225 : * The AsyncQueueControl structure is protected by the AsyncQueueLock.
226 : *
227 : * When holding the lock in SHARED mode, backends may only inspect their own
228 : * entries as well as the head and tail pointers. Consequently we can allow a
229 : * backend to update its own record while holding only SHARED lock (since no
230 : * other backend will inspect it).
231 : *
232 : * When holding the lock in EXCLUSIVE mode, backends can inspect the entries
233 : * of other backends and also change the head and tail pointers.
234 : *
235 : * AsyncCtlLock is used as the control lock for the pg_notify SLRU buffers.
236 : * In order to avoid deadlocks, whenever we need both locks, we always first
237 : * get AsyncQueueLock and then AsyncCtlLock.
238 : *
239 : * Each backend uses the backend[] array entry with index equal to its
240 : * BackendId (which can range from 1 to MaxBackends). We rely on this to make
241 : * SendProcSignal fast.
242 : */
243 : typedef struct AsyncQueueControl
244 : {
245 : QueuePosition head; /* head points to the next free location */
246 : QueuePosition tail; /* the global tail is equivalent to the pos of
247 : * the "slowest" backend */
248 : TimestampTz lastQueueFillWarn; /* time of last queue-full msg */
249 : QueueBackendStatus backend[FLEXIBLE_ARRAY_MEMBER];
250 : /* backend[0] is not used; used entries are from [1] to [MaxBackends] */
251 : } AsyncQueueControl;
252 :
253 : static AsyncQueueControl *asyncQueueControl;
254 :
255 : #define QUEUE_HEAD (asyncQueueControl->head)
256 : #define QUEUE_TAIL (asyncQueueControl->tail)
257 : #define QUEUE_BACKEND_PID(i) (asyncQueueControl->backend[i].pid)
258 : #define QUEUE_BACKEND_DBOID(i) (asyncQueueControl->backend[i].dboid)
259 : #define QUEUE_BACKEND_POS(i) (asyncQueueControl->backend[i].pos)
260 :
261 : /*
262 : * The SLRU buffer area through which we access the notification queue
263 : */
264 : static SlruCtlData AsyncCtlData;
265 :
266 : #define AsyncCtl (&AsyncCtlData)
267 : #define QUEUE_PAGESIZE BLCKSZ
268 : #define QUEUE_FULL_WARN_INTERVAL 5000 /* warn at most once every 5s */
269 :
270 : /*
271 : * slru.c currently assumes that all filenames are four characters of hex
272 : * digits. That means that we can use segments 0000 through FFFF.
273 : * Each segment contains SLRU_PAGES_PER_SEGMENT pages which gives us
274 : * the pages from 0 to SLRU_PAGES_PER_SEGMENT * 0x10000 - 1.
275 : *
276 : * It's of course possible to enhance slru.c, but this gives us so much
277 : * space already that it doesn't seem worth the trouble.
278 : *
279 : * The most data we can have in the queue at a time is QUEUE_MAX_PAGE/2
280 : * pages, because more than that would confuse slru.c into thinking there
281 : * was a wraparound condition. With the default BLCKSZ this means there
282 : * can be up to 8GB of queued-and-not-read data.
283 : *
284 : * Note: it's possible to redefine QUEUE_MAX_PAGE with a smaller multiple of
285 : * SLRU_PAGES_PER_SEGMENT, for easier testing of queue-full behaviour.
286 : */
287 : #define QUEUE_MAX_PAGE (SLRU_PAGES_PER_SEGMENT * 0x10000 - 1)
288 :
289 : /*
290 : * listenChannels identifies the channels we are actually listening to
291 : * (ie, have committed a LISTEN on). It is a simple list of channel names,
292 : * allocated in TopMemoryContext.
293 : */
294 : static List *listenChannels = NIL; /* list of C strings */
295 :
296 : /*
297 : * State for pending LISTEN/UNLISTEN actions consists of an ordered list of
298 : * all actions requested in the current transaction. As explained above,
299 : * we don't actually change listenChannels until we reach transaction commit.
300 : *
301 : * The list is kept in CurTransactionContext. In subtransactions, each
302 : * subtransaction has its own list in its own CurTransactionContext, but
303 : * successful subtransactions attach their lists to their parent's list.
304 : * Failed subtransactions simply discard their lists.
305 : */
306 : typedef enum
307 : {
308 : LISTEN_LISTEN,
309 : LISTEN_UNLISTEN,
310 : LISTEN_UNLISTEN_ALL
311 : } ListenActionKind;
312 :
313 : typedef struct
314 : {
315 : ListenActionKind action;
316 : char channel[FLEXIBLE_ARRAY_MEMBER]; /* nul-terminated string */
317 : } ListenAction;
318 :
319 : static List *pendingActions = NIL; /* list of ListenAction */
320 :
321 : static List *upperPendingActions = NIL; /* list of upper-xact lists */
322 :
323 : /*
324 : * State for outbound notifies consists of a list of all channels+payloads
325 : * NOTIFYed in the current transaction. We do not actually perform a NOTIFY
326 : * until and unless the transaction commits. pendingNotifies is NIL if no
327 : * NOTIFYs have been done in the current transaction.
328 : *
329 : * The list is kept in CurTransactionContext. In subtransactions, each
330 : * subtransaction has its own list in its own CurTransactionContext, but
331 : * successful subtransactions attach their lists to their parent's list.
332 : * Failed subtransactions simply discard their lists.
333 : *
334 : * Note: the action and notify lists do not interact within a transaction.
335 : * In particular, if a transaction does NOTIFY and then LISTEN on the same
336 : * condition name, it will get a self-notify at commit. This is a bit odd
337 : * but is consistent with our historical behavior.
338 : */
339 : typedef struct Notification
340 : {
341 : char *channel; /* channel name */
342 : char *payload; /* payload string (can be empty) */
343 : } Notification;
344 :
345 : static List *pendingNotifies = NIL; /* list of Notifications */
346 :
347 : static List *upperPendingNotifies = NIL; /* list of upper-xact lists */
348 :
349 : /*
350 : * Inbound notifications are initially processed by HandleNotifyInterrupt(),
351 : * called from inside a signal handler. That just sets the
352 : * notifyInterruptPending flag and sets the process
353 : * latch. ProcessNotifyInterrupt() will then be called whenever it's safe to
354 : * actually deal with the interrupt.
355 : */
356 : volatile sig_atomic_t notifyInterruptPending = false;
357 :
358 : /* True if we've registered an on_shmem_exit cleanup */
359 : static bool unlistenExitRegistered = false;
360 :
361 : /* True if we're currently registered as a listener in asyncQueueControl */
362 : static bool amRegisteredListener = false;
363 :
364 : /* has this backend sent notifications in the current transaction? */
365 : static bool backendHasSentNotifications = false;
366 :
367 : /* GUC parameter */
368 : bool Trace_notify = false;
369 :
370 : /* local function prototypes */
371 : static bool asyncQueuePagePrecedes(int p, int q);
372 : static void queue_listen(ListenActionKind action, const char *channel);
373 : static void Async_UnlistenOnExit(int code, Datum arg);
374 : static void Exec_ListenPreCommit(void);
375 : static void Exec_ListenCommit(const char *channel);
376 : static void Exec_UnlistenCommit(const char *channel);
377 : static void Exec_UnlistenAllCommit(void);
378 : static bool IsListeningOn(const char *channel);
379 : static void asyncQueueUnregister(void);
380 : static bool asyncQueueIsFull(void);
381 : static bool asyncQueueAdvance(volatile QueuePosition *position, int entryLength);
382 : static void asyncQueueNotificationToEntry(Notification *n, AsyncQueueEntry *qe);
383 : static ListCell *asyncQueueAddEntries(ListCell *nextNotify);
384 : static double asyncQueueUsage(void);
385 : static void asyncQueueFillWarning(void);
386 : static bool SignalBackends(void);
387 : static void asyncQueueReadAllNotifications(void);
388 : static bool asyncQueueProcessPageEntries(volatile QueuePosition *current,
389 : QueuePosition stop,
390 : char *page_buffer);
391 : static void asyncQueueAdvanceTail(void);
392 : static void ProcessIncomingNotify(void);
393 : static bool AsyncExistsPendingNotify(const char *channel, const char *payload);
394 : static void ClearPendingActionsAndNotifies(void);
395 :
396 : /*
397 : * We will work on the page range of 0..QUEUE_MAX_PAGE.
398 : */
399 : static bool
400 10 : asyncQueuePagePrecedes(int p, int q)
401 : {
402 : int diff;
403 :
404 : /*
405 : * We have to compare modulo (QUEUE_MAX_PAGE+1)/2. Both inputs should be
406 : * in the range 0..QUEUE_MAX_PAGE.
407 : */
408 10 : Assert(p >= 0 && p <= QUEUE_MAX_PAGE);
409 10 : Assert(q >= 0 && q <= QUEUE_MAX_PAGE);
410 :
411 10 : diff = p - q;
412 10 : if (diff >= ((QUEUE_MAX_PAGE + 1) / 2))
413 0 : diff -= QUEUE_MAX_PAGE + 1;
414 10 : else if (diff < -((QUEUE_MAX_PAGE + 1) / 2))
415 0 : diff += QUEUE_MAX_PAGE + 1;
416 10 : return diff < 0;
417 : }
418 :
419 : /*
420 : * Report space needed for our shared memory area
421 : */
422 : Size
423 5 : AsyncShmemSize(void)
424 : {
425 : Size size;
426 :
427 : /* This had better match AsyncShmemInit */
428 5 : size = mul_size(MaxBackends + 1, sizeof(QueueBackendStatus));
429 5 : size = add_size(size, offsetof(AsyncQueueControl, backend));
430 :
431 5 : size = add_size(size, SimpleLruShmemSize(NUM_ASYNC_BUFFERS, 0));
432 :
433 5 : return size;
434 : }
435 :
436 : /*
437 : * Initialize our shared memory area
438 : */
439 : void
440 5 : AsyncShmemInit(void)
441 : {
442 : bool found;
443 : int slotno;
444 : Size size;
445 :
446 : /*
447 : * Create or attach to the AsyncQueueControl structure.
448 : *
449 : * The used entries in the backend[] array run from 1 to MaxBackends; the
450 : * zero'th entry is unused but must be allocated.
451 : */
452 5 : size = mul_size(MaxBackends + 1, sizeof(QueueBackendStatus));
453 5 : size = add_size(size, offsetof(AsyncQueueControl, backend));
454 :
455 5 : asyncQueueControl = (AsyncQueueControl *)
456 5 : ShmemInitStruct("Async Queue Control", size, &found);
457 :
458 5 : if (!found)
459 : {
460 : /* First time through, so initialize it */
461 : int i;
462 :
463 5 : SET_QUEUE_POS(QUEUE_HEAD, 0, 0);
464 5 : SET_QUEUE_POS(QUEUE_TAIL, 0, 0);
465 5 : asyncQueueControl->lastQueueFillWarn = 0;
466 : /* zero'th entry won't be used, but let's initialize it anyway */
467 570 : for (i = 0; i <= MaxBackends; i++)
468 : {
469 565 : QUEUE_BACKEND_PID(i) = InvalidPid;
470 565 : QUEUE_BACKEND_DBOID(i) = InvalidOid;
471 565 : SET_QUEUE_POS(QUEUE_BACKEND_POS(i), 0, 0);
472 : }
473 : }
474 :
475 : /*
476 : * Set up SLRU management of the pg_notify data.
477 : */
478 5 : AsyncCtl->PagePrecedes = asyncQueuePagePrecedes;
479 5 : SimpleLruInit(AsyncCtl, "async", NUM_ASYNC_BUFFERS, 0,
480 5 : AsyncCtlLock, "pg_notify", LWTRANCHE_ASYNC_BUFFERS);
481 : /* Override default assumption that writes should be fsync'd */
482 5 : AsyncCtl->do_fsync = false;
483 :
484 5 : if (!found)
485 : {
486 : /*
487 : * During start or reboot, clean out the pg_notify directory.
488 : */
489 5 : (void) SlruScanDirectory(AsyncCtl, SlruScanDirCbDeleteAll, NULL);
490 :
491 : /* Now initialize page zero to empty */
492 5 : LWLockAcquire(AsyncCtlLock, LW_EXCLUSIVE);
493 5 : slotno = SimpleLruZeroPage(AsyncCtl, QUEUE_POS_PAGE(QUEUE_HEAD));
494 : /* This write is just to verify that pg_notify/ is writable */
495 5 : SimpleLruWritePage(AsyncCtl, slotno);
496 5 : LWLockRelease(AsyncCtlLock);
497 : }
498 5 : }
499 :
500 :
501 : /*
502 : * pg_notify -
503 : * SQL function to send a notification event
504 : */
505 : Datum
506 6 : pg_notify(PG_FUNCTION_ARGS)
507 : {
508 : const char *channel;
509 : const char *payload;
510 :
511 6 : if (PG_ARGISNULL(0))
512 1 : channel = "";
513 : else
514 5 : channel = text_to_cstring(PG_GETARG_TEXT_PP(0));
515 :
516 6 : if (PG_ARGISNULL(1))
517 1 : payload = "";
518 : else
519 5 : payload = text_to_cstring(PG_GETARG_TEXT_PP(1));
520 :
521 : /* For NOTIFY as a statement, this is checked in ProcessUtility */
522 6 : PreventCommandDuringRecovery("NOTIFY");
523 :
524 6 : Async_Notify(channel, payload);
525 :
526 3 : PG_RETURN_VOID();
527 : }
528 :
529 :
530 : /*
531 : * Async_Notify
532 : *
533 : * This is executed by the SQL notify command.
534 : *
535 : * Adds the message to the list of pending notifies.
536 : * Actual notification happens during transaction commit.
537 : * ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
538 : */
539 : void
540 7 : Async_Notify(const char *channel, const char *payload)
541 : {
542 : Notification *n;
543 : MemoryContext oldcontext;
544 :
545 7 : if (IsParallelWorker())
546 0 : elog(ERROR, "cannot send notifications from a parallel worker");
547 :
548 7 : if (Trace_notify)
549 0 : elog(DEBUG1, "Async_Notify(%s)", channel);
550 :
551 : /* a channel name must be specified */
552 7 : if (!channel || !strlen(channel))
553 2 : ereport(ERROR,
554 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
555 : errmsg("channel name cannot be empty")));
556 :
557 5 : if (strlen(channel) >= NAMEDATALEN)
558 1 : ereport(ERROR,
559 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
560 : errmsg("channel name too long")));
561 :
562 4 : if (payload)
563 : {
564 3 : if (strlen(payload) >= NOTIFY_PAYLOAD_MAX_LENGTH)
565 0 : ereport(ERROR,
566 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
567 : errmsg("payload string too long")));
568 : }
569 :
570 : /* no point in making duplicate entries in the list ... */
571 4 : if (AsyncExistsPendingNotify(channel, payload))
572 4 : return;
573 :
574 : /*
575 : * The notification list needs to live until end of transaction, so store
576 : * it in the transaction context.
577 : */
578 4 : oldcontext = MemoryContextSwitchTo(CurTransactionContext);
579 :
580 4 : n = (Notification *) palloc(sizeof(Notification));
581 4 : n->channel = pstrdup(channel);
582 4 : if (payload)
583 3 : n->payload = pstrdup(payload);
584 : else
585 1 : n->payload = "";
586 :
587 : /*
588 : * We want to preserve the order so we need to append every notification.
589 : * See comments at AsyncExistsPendingNotify().
590 : */
591 4 : pendingNotifies = lappend(pendingNotifies, n);
592 :
593 4 : MemoryContextSwitchTo(oldcontext);
594 : }
595 :
596 : /*
597 : * queue_listen
598 : * Common code for listen, unlisten, unlisten all commands.
599 : *
600 : * Adds the request to the list of pending actions.
601 : * Actual update of the listenChannels list happens during transaction
602 : * commit.
603 : */
604 : static void
605 5 : queue_listen(ListenActionKind action, const char *channel)
606 : {
607 : MemoryContext oldcontext;
608 : ListenAction *actrec;
609 :
610 : /*
611 : * Unlike Async_Notify, we don't try to collapse out duplicates. It would
612 : * be too complicated to ensure we get the right interactions of
613 : * conflicting LISTEN/UNLISTEN/UNLISTEN_ALL, and it's unlikely that there
614 : * would be any performance benefit anyway in sane applications.
615 : */
616 5 : oldcontext = MemoryContextSwitchTo(CurTransactionContext);
617 :
618 : /* space for terminating null is included in sizeof(ListenAction) */
619 5 : actrec = (ListenAction *) palloc(offsetof(ListenAction, channel) +
620 5 : strlen(channel) + 1);
621 5 : actrec->action = action;
622 5 : strcpy(actrec->channel, channel);
623 :
624 5 : pendingActions = lappend(pendingActions, actrec);
625 :
626 5 : MemoryContextSwitchTo(oldcontext);
627 5 : }
628 :
629 : /*
630 : * Async_Listen
631 : *
632 : * This is executed by the SQL listen command.
633 : */
634 : void
635 2 : Async_Listen(const char *channel)
636 : {
637 2 : if (Trace_notify)
638 0 : elog(DEBUG1, "Async_Listen(%s,%d)", channel, MyProcPid);
639 :
640 2 : queue_listen(LISTEN_LISTEN, channel);
641 2 : }
642 :
643 : /*
644 : * Async_Unlisten
645 : *
646 : * This is executed by the SQL unlisten command.
647 : */
648 : void
649 1 : Async_Unlisten(const char *channel)
650 : {
651 1 : if (Trace_notify)
652 0 : elog(DEBUG1, "Async_Unlisten(%s,%d)", channel, MyProcPid);
653 :
654 : /* If we couldn't possibly be listening, no need to queue anything */
655 1 : if (pendingActions == NIL && !unlistenExitRegistered)
656 1 : return;
657 :
658 1 : queue_listen(LISTEN_UNLISTEN, channel);
659 : }
660 :
661 : /*
662 : * Async_UnlistenAll
663 : *
664 : * This is invoked by UNLISTEN * command, and also at backend exit.
665 : */
666 : void
667 2 : Async_UnlistenAll(void)
668 : {
669 2 : if (Trace_notify)
670 0 : elog(DEBUG1, "Async_UnlistenAll(%d)", MyProcPid);
671 :
672 : /* If we couldn't possibly be listening, no need to queue anything */
673 2 : if (pendingActions == NIL && !unlistenExitRegistered)
674 2 : return;
675 :
676 2 : queue_listen(LISTEN_UNLISTEN_ALL, "");
677 : }
678 :
679 : /*
680 : * SQL function: return a set of the channel names this backend is actively
681 : * listening to.
682 : *
683 : * Note: this coding relies on the fact that the listenChannels list cannot
684 : * change within a transaction.
685 : */
686 : Datum
687 3 : pg_listening_channels(PG_FUNCTION_ARGS)
688 : {
689 : FuncCallContext *funcctx;
690 : ListCell **lcp;
691 :
692 : /* stuff done only on the first call of the function */
693 3 : if (SRF_IS_FIRSTCALL())
694 : {
695 : MemoryContext oldcontext;
696 :
697 : /* create a function context for cross-call persistence */
698 2 : funcctx = SRF_FIRSTCALL_INIT();
699 :
700 : /* switch to memory context appropriate for multiple function calls */
701 2 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
702 :
703 : /* allocate memory for user context */
704 2 : lcp = (ListCell **) palloc(sizeof(ListCell *));
705 2 : *lcp = list_head(listenChannels);
706 2 : funcctx->user_fctx = (void *) lcp;
707 :
708 2 : MemoryContextSwitchTo(oldcontext);
709 : }
710 :
711 : /* stuff done on every call of the function */
712 3 : funcctx = SRF_PERCALL_SETUP();
713 3 : lcp = (ListCell **) funcctx->user_fctx;
714 :
715 6 : while (*lcp != NULL)
716 : {
717 1 : char *channel = (char *) lfirst(*lcp);
718 :
719 1 : *lcp = lnext(*lcp);
720 1 : SRF_RETURN_NEXT(funcctx, CStringGetTextDatum(channel));
721 : }
722 :
723 2 : SRF_RETURN_DONE(funcctx);
724 : }
725 :
726 : /*
727 : * Async_UnlistenOnExit
728 : *
729 : * This is executed at backend exit if we have done any LISTENs in this
730 : * backend. It might not be necessary anymore, if the user UNLISTENed
731 : * everything, but we don't try to detect that case.
732 : */
733 : static void
734 2 : Async_UnlistenOnExit(int code, Datum arg)
735 : {
736 2 : Exec_UnlistenAllCommit();
737 2 : asyncQueueUnregister();
738 2 : }
739 :
740 : /*
741 : * AtPrepare_Notify
742 : *
743 : * This is called at the prepare phase of a two-phase
744 : * transaction. Save the state for possible commit later.
745 : */
746 : void
747 6 : AtPrepare_Notify(void)
748 : {
749 : /* It's not allowed to have any pending LISTEN/UNLISTEN/NOTIFY actions */
750 6 : if (pendingActions || pendingNotifies)
751 0 : ereport(ERROR,
752 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
753 : errmsg("cannot PREPARE a transaction that has executed LISTEN, UNLISTEN, or NOTIFY")));
754 6 : }
755 :
756 : /*
757 : * PreCommit_Notify
758 : *
759 : * This is called at transaction commit, before actually committing to
760 : * clog.
761 : *
762 : * If there are pending LISTEN actions, make sure we are listed in the
763 : * shared-memory listener array. This must happen before commit to
764 : * ensure we don't miss any notifies from transactions that commit
765 : * just after ours.
766 : *
767 : * If there are outbound notify requests in the pendingNotifies list,
768 : * add them to the global queue. We do that before commit so that
769 : * we can still throw error if we run out of queue space.
770 : */
771 : void
772 22906 : PreCommit_Notify(void)
773 : {
774 : ListCell *p;
775 :
776 22906 : if (pendingActions == NIL && pendingNotifies == NIL)
777 45803 : return; /* no relevant statements in this xact */
778 :
779 9 : if (Trace_notify)
780 0 : elog(DEBUG1, "PreCommit_Notify");
781 :
782 : /* Preflight for any pending listen/unlisten actions */
783 14 : foreach(p, pendingActions)
784 : {
785 5 : ListenAction *actrec = (ListenAction *) lfirst(p);
786 :
787 5 : switch (actrec->action)
788 : {
789 : case LISTEN_LISTEN:
790 2 : Exec_ListenPreCommit();
791 2 : break;
792 : case LISTEN_UNLISTEN:
793 : /* there is no Exec_UnlistenPreCommit() */
794 1 : break;
795 : case LISTEN_UNLISTEN_ALL:
796 : /* there is no Exec_UnlistenAllPreCommit() */
797 2 : break;
798 : }
799 : }
800 :
801 : /* Queue any pending notifies */
802 9 : if (pendingNotifies)
803 : {
804 : ListCell *nextNotify;
805 :
806 : /*
807 : * Make sure that we have an XID assigned to the current transaction.
808 : * GetCurrentTransactionId is cheap if we already have an XID, but not
809 : * so cheap if we don't, and we'd prefer not to do that work while
810 : * holding AsyncQueueLock.
811 : */
812 4 : (void) GetCurrentTransactionId();
813 :
814 : /*
815 : * Serialize writers by acquiring a special lock that we hold till
816 : * after commit. This ensures that queue entries appear in commit
817 : * order, and in particular that there are never uncommitted queue
818 : * entries ahead of committed ones, so an uncommitted transaction
819 : * can't block delivery of deliverable notifications.
820 : *
821 : * We use a heavyweight lock so that it'll automatically be released
822 : * after either commit or abort. This also allows deadlocks to be
823 : * detected, though really a deadlock shouldn't be possible here.
824 : *
825 : * The lock is on "database 0", which is pretty ugly but it doesn't
826 : * seem worth inventing a special locktag category just for this.
827 : * (Historical note: before PG 9.0, a similar lock on "database 0" was
828 : * used by the flatfiles mechanism.)
829 : */
830 4 : LockSharedObject(DatabaseRelationId, InvalidOid, 0,
831 : AccessExclusiveLock);
832 :
833 : /* Now push the notifications into the queue */
834 4 : backendHasSentNotifications = true;
835 :
836 4 : nextNotify = list_head(pendingNotifies);
837 12 : while (nextNotify != NULL)
838 : {
839 : /*
840 : * Add the pending notifications to the queue. We acquire and
841 : * release AsyncQueueLock once per page, which might be overkill
842 : * but it does allow readers to get in while we're doing this.
843 : *
844 : * A full queue is very uncommon and should really not happen,
845 : * given that we have so much space available in the SLRU pages.
846 : * Nevertheless we need to deal with this possibility. Note that
847 : * when we get here we are in the process of committing our
848 : * transaction, but we have not yet committed to clog, so at this
849 : * point in time we can still roll the transaction back.
850 : */
851 4 : LWLockAcquire(AsyncQueueLock, LW_EXCLUSIVE);
852 4 : asyncQueueFillWarning();
853 4 : if (asyncQueueIsFull())
854 0 : ereport(ERROR,
855 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
856 : errmsg("too many notifications in the NOTIFY queue")));
857 4 : nextNotify = asyncQueueAddEntries(nextNotify);
858 4 : LWLockRelease(AsyncQueueLock);
859 : }
860 : }
861 : }
862 :
863 : /*
864 : * AtCommit_Notify
865 : *
866 : * This is called at transaction commit, after committing to clog.
867 : *
868 : * Update listenChannels and clear transaction-local state.
869 : */
870 : void
871 22906 : AtCommit_Notify(void)
872 : {
873 : ListCell *p;
874 :
875 : /*
876 : * Allow transactions that have not executed LISTEN/UNLISTEN/NOTIFY to
877 : * return as soon as possible
878 : */
879 22906 : if (!pendingActions && !pendingNotifies)
880 45803 : return;
881 :
882 9 : if (Trace_notify)
883 0 : elog(DEBUG1, "AtCommit_Notify");
884 :
885 : /* Perform any pending listen/unlisten actions */
886 14 : foreach(p, pendingActions)
887 : {
888 5 : ListenAction *actrec = (ListenAction *) lfirst(p);
889 :
890 5 : switch (actrec->action)
891 : {
892 : case LISTEN_LISTEN:
893 2 : Exec_ListenCommit(actrec->channel);
894 2 : break;
895 : case LISTEN_UNLISTEN:
896 1 : Exec_UnlistenCommit(actrec->channel);
897 1 : break;
898 : case LISTEN_UNLISTEN_ALL:
899 2 : Exec_UnlistenAllCommit();
900 2 : break;
901 : }
902 : }
903 :
904 : /* If no longer listening to anything, get out of listener array */
905 9 : if (amRegisteredListener && listenChannels == NIL)
906 2 : asyncQueueUnregister();
907 :
908 : /* And clean up */
909 9 : ClearPendingActionsAndNotifies();
910 : }
911 :
912 : /*
913 : * Exec_ListenPreCommit --- subroutine for PreCommit_Notify
914 : *
915 : * This function must make sure we are ready to catch any incoming messages.
916 : */
917 : static void
918 2 : Exec_ListenPreCommit(void)
919 : {
920 : QueuePosition head;
921 : QueuePosition max;
922 : int i;
923 :
924 : /*
925 : * Nothing to do if we are already listening to something, nor if we
926 : * already ran this routine in this transaction.
927 : */
928 2 : if (amRegisteredListener)
929 2 : return;
930 :
931 2 : if (Trace_notify)
932 0 : elog(DEBUG1, "Exec_ListenPreCommit(%d)", MyProcPid);
933 :
934 : /*
935 : * Before registering, make sure we will unlisten before dying. (Note:
936 : * this action does not get undone if we abort later.)
937 : */
938 2 : if (!unlistenExitRegistered)
939 : {
940 2 : before_shmem_exit(Async_UnlistenOnExit, 0);
941 2 : unlistenExitRegistered = true;
942 : }
943 :
944 : /*
945 : * This is our first LISTEN, so establish our pointer.
946 : *
947 : * We set our pointer to the global tail pointer and then move it forward
948 : * over already-committed notifications. This ensures we cannot miss any
949 : * not-yet-committed notifications. We might get a few more but that
950 : * doesn't hurt.
951 : *
952 : * In some scenarios there might be a lot of committed notifications that
953 : * have not yet been pruned away (because some backend is being lazy about
954 : * reading them). To reduce our startup time, we can look at other
955 : * backends and adopt the maximum "pos" pointer of any backend that's in
956 : * our database; any notifications it's already advanced over are surely
957 : * committed and need not be re-examined by us. (We must consider only
958 : * backends connected to our DB, because others will not have bothered to
959 : * check committed-ness of notifications in our DB.) But we only bother
960 : * with that if there's more than a page worth of notifications
961 : * outstanding, otherwise scanning all the other backends isn't worth it.
962 : *
963 : * We need exclusive lock here so we can look at other backends' entries.
964 : */
965 2 : LWLockAcquire(AsyncQueueLock, LW_EXCLUSIVE);
966 2 : head = QUEUE_HEAD;
967 2 : max = QUEUE_TAIL;
968 2 : if (QUEUE_POS_PAGE(max) != QUEUE_POS_PAGE(head))
969 : {
970 0 : for (i = 1; i <= MaxBackends; i++)
971 : {
972 0 : if (QUEUE_BACKEND_DBOID(i) == MyDatabaseId)
973 0 : max = QUEUE_POS_MAX(max, QUEUE_BACKEND_POS(i));
974 : }
975 : }
976 2 : QUEUE_BACKEND_POS(MyBackendId) = max;
977 2 : QUEUE_BACKEND_PID(MyBackendId) = MyProcPid;
978 2 : QUEUE_BACKEND_DBOID(MyBackendId) = MyDatabaseId;
979 2 : LWLockRelease(AsyncQueueLock);
980 :
981 : /* Now we are listed in the global array, so remember we're listening */
982 2 : amRegisteredListener = true;
983 :
984 : /*
985 : * Try to move our pointer forward as far as possible. This will skip over
986 : * already-committed notifications. Still, we could get notifications that
987 : * have already committed before we started to LISTEN.
988 : *
989 : * Note that we are not yet listening on anything, so we won't deliver any
990 : * notification to the frontend.
991 : *
992 : * This will also advance the global tail pointer if possible.
993 : */
994 2 : if (!QUEUE_POS_EQUAL(max, head))
995 0 : asyncQueueReadAllNotifications();
996 : }
997 :
998 : /*
999 : * Exec_ListenCommit --- subroutine for AtCommit_Notify
1000 : *
1001 : * Add the channel to the list of channels we are listening on.
1002 : */
1003 : static void
1004 2 : Exec_ListenCommit(const char *channel)
1005 : {
1006 : MemoryContext oldcontext;
1007 :
1008 : /* Do nothing if we are already listening on this channel */
1009 2 : if (IsListeningOn(channel))
1010 2 : return;
1011 :
1012 : /*
1013 : * Add the new channel name to listenChannels.
1014 : *
1015 : * XXX It is theoretically possible to get an out-of-memory failure here,
1016 : * which would be bad because we already committed. For the moment it
1017 : * doesn't seem worth trying to guard against that, but maybe improve this
1018 : * later.
1019 : */
1020 2 : oldcontext = MemoryContextSwitchTo(TopMemoryContext);
1021 2 : listenChannels = lappend(listenChannels, pstrdup(channel));
1022 2 : MemoryContextSwitchTo(oldcontext);
1023 : }
1024 :
1025 : /*
1026 : * Exec_UnlistenCommit --- subroutine for AtCommit_Notify
1027 : *
1028 : * Remove the specified channel name from listenChannels.
1029 : */
1030 : static void
1031 1 : Exec_UnlistenCommit(const char *channel)
1032 : {
1033 : ListCell *q;
1034 : ListCell *prev;
1035 :
1036 1 : if (Trace_notify)
1037 0 : elog(DEBUG1, "Exec_UnlistenCommit(%s,%d)", channel, MyProcPid);
1038 :
1039 1 : prev = NULL;
1040 1 : foreach(q, listenChannels)
1041 : {
1042 1 : char *lchan = (char *) lfirst(q);
1043 :
1044 1 : if (strcmp(lchan, channel) == 0)
1045 : {
1046 1 : listenChannels = list_delete_cell(listenChannels, q, prev);
1047 1 : pfree(lchan);
1048 1 : break;
1049 : }
1050 0 : prev = q;
1051 : }
1052 :
1053 : /*
1054 : * We do not complain about unlistening something not being listened;
1055 : * should we?
1056 : */
1057 1 : }
1058 :
1059 : /*
1060 : * Exec_UnlistenAllCommit --- subroutine for AtCommit_Notify
1061 : *
1062 : * Unlisten on all channels for this backend.
1063 : */
1064 : static void
1065 4 : Exec_UnlistenAllCommit(void)
1066 : {
1067 4 : if (Trace_notify)
1068 0 : elog(DEBUG1, "Exec_UnlistenAllCommit(%d)", MyProcPid);
1069 :
1070 4 : list_free_deep(listenChannels);
1071 4 : listenChannels = NIL;
1072 4 : }
1073 :
1074 : /*
1075 : * ProcessCompletedNotifies --- send out signals and self-notifies
1076 : *
1077 : * This is called from postgres.c just before going idle at the completion
1078 : * of a transaction. If we issued any notifications in the just-completed
1079 : * transaction, send signals to other backends to process them, and also
1080 : * process the queue ourselves to send messages to our own frontend.
1081 : *
1082 : * The reason that this is not done in AtCommit_Notify is that there is
1083 : * a nonzero chance of errors here (for example, encoding conversion errors
1084 : * while trying to format messages to our frontend). An error during
1085 : * AtCommit_Notify would be a PANIC condition. The timing is also arranged
1086 : * to ensure that a transaction's self-notifies are delivered to the frontend
1087 : * before it gets the terminating ReadyForQuery message.
1088 : *
1089 : * Note that we send signals and process the queue even if the transaction
1090 : * eventually aborted. This is because we need to clean out whatever got
1091 : * added to the queue.
1092 : *
1093 : * NOTE: we are outside of any transaction here.
1094 : */
1095 : void
1096 24949 : ProcessCompletedNotifies(void)
1097 : {
1098 : MemoryContext caller_context;
1099 : bool signalled;
1100 :
1101 : /* Nothing to do if we didn't send any notifications */
1102 24949 : if (!backendHasSentNotifications)
1103 49894 : return;
1104 :
1105 : /*
1106 : * We reset the flag immediately; otherwise, if any sort of error occurs
1107 : * below, we'd be locked up in an infinite loop, because control will come
1108 : * right back here after error cleanup.
1109 : */
1110 4 : backendHasSentNotifications = false;
1111 :
1112 : /*
1113 : * We must preserve the caller's memory context (probably MessageContext)
1114 : * across the transaction we do here.
1115 : */
1116 4 : caller_context = CurrentMemoryContext;
1117 :
1118 4 : if (Trace_notify)
1119 0 : elog(DEBUG1, "ProcessCompletedNotifies");
1120 :
1121 : /*
1122 : * We must run asyncQueueReadAllNotifications inside a transaction, else
1123 : * bad things happen if it gets an error.
1124 : */
1125 4 : StartTransactionCommand();
1126 :
1127 : /* Send signals to other backends */
1128 4 : signalled = SignalBackends();
1129 :
1130 4 : if (listenChannels != NIL)
1131 : {
1132 : /* Read the queue ourselves, and send relevant stuff to the frontend */
1133 0 : asyncQueueReadAllNotifications();
1134 : }
1135 4 : else if (!signalled)
1136 : {
1137 : /*
1138 : * If we found no other listening backends, and we aren't listening
1139 : * ourselves, then we must execute asyncQueueAdvanceTail to flush the
1140 : * queue, because ain't nobody else gonna do it. This prevents queue
1141 : * overflow when we're sending useless notifies to nobody. (A new
1142 : * listener could have joined since we looked, but if so this is
1143 : * harmless.)
1144 : */
1145 4 : asyncQueueAdvanceTail();
1146 : }
1147 :
1148 4 : CommitTransactionCommand();
1149 :
1150 4 : MemoryContextSwitchTo(caller_context);
1151 :
1152 : /* We don't need pq_flush() here since postgres.c will do one shortly */
1153 : }
1154 :
1155 : /*
1156 : * Test whether we are actively listening on the given channel name.
1157 : *
1158 : * Note: this function is executed for every notification found in the queue.
1159 : * Perhaps it is worth further optimization, eg convert the list to a sorted
1160 : * array so we can binary-search it. In practice the list is likely to be
1161 : * fairly short, though.
1162 : */
1163 : static bool
1164 2 : IsListeningOn(const char *channel)
1165 : {
1166 : ListCell *p;
1167 :
1168 2 : foreach(p, listenChannels)
1169 : {
1170 0 : char *lchan = (char *) lfirst(p);
1171 :
1172 0 : if (strcmp(lchan, channel) == 0)
1173 0 : return true;
1174 : }
1175 2 : return false;
1176 : }
1177 :
1178 : /*
1179 : * Remove our entry from the listeners array when we are no longer listening
1180 : * on any channel. NB: must not fail if we're already not listening.
1181 : */
1182 : static void
1183 4 : asyncQueueUnregister(void)
1184 : {
1185 : bool advanceTail;
1186 :
1187 4 : Assert(listenChannels == NIL); /* else caller error */
1188 :
1189 4 : if (!amRegisteredListener) /* nothing to do */
1190 6 : return;
1191 :
1192 2 : LWLockAcquire(AsyncQueueLock, LW_SHARED);
1193 : /* check if entry is valid and oldest ... */
1194 6 : advanceTail = (MyProcPid == QUEUE_BACKEND_PID(MyBackendId)) &&
1195 4 : QUEUE_POS_EQUAL(QUEUE_BACKEND_POS(MyBackendId), QUEUE_TAIL);
1196 : /* ... then mark it invalid */
1197 2 : QUEUE_BACKEND_PID(MyBackendId) = InvalidPid;
1198 2 : QUEUE_BACKEND_DBOID(MyBackendId) = InvalidOid;
1199 2 : LWLockRelease(AsyncQueueLock);
1200 :
1201 : /* mark ourselves as no longer listed in the global array */
1202 2 : amRegisteredListener = false;
1203 :
1204 : /* If we were the laziest backend, try to advance the tail pointer */
1205 2 : if (advanceTail)
1206 2 : asyncQueueAdvanceTail();
1207 : }
1208 :
1209 : /*
1210 : * Test whether there is room to insert more notification messages.
1211 : *
1212 : * Caller must hold at least shared AsyncQueueLock.
1213 : */
1214 : static bool
1215 4 : asyncQueueIsFull(void)
1216 : {
1217 : int nexthead;
1218 : int boundary;
1219 :
1220 : /*
1221 : * The queue is full if creating a new head page would create a page that
1222 : * logically precedes the current global tail pointer, ie, the head
1223 : * pointer would wrap around compared to the tail. We cannot create such
1224 : * a head page for fear of confusing slru.c. For safety we round the tail
1225 : * pointer back to a segment boundary (compare the truncation logic in
1226 : * asyncQueueAdvanceTail).
1227 : *
1228 : * Note that this test is *not* dependent on how much space there is on
1229 : * the current head page. This is necessary because asyncQueueAddEntries
1230 : * might try to create the next head page in any case.
1231 : */
1232 4 : nexthead = QUEUE_POS_PAGE(QUEUE_HEAD) + 1;
1233 4 : if (nexthead > QUEUE_MAX_PAGE)
1234 0 : nexthead = 0; /* wrap around */
1235 4 : boundary = QUEUE_POS_PAGE(QUEUE_TAIL);
1236 4 : boundary -= boundary % SLRU_PAGES_PER_SEGMENT;
1237 4 : return asyncQueuePagePrecedes(nexthead, boundary);
1238 : }
1239 :
1240 : /*
1241 : * Advance the QueuePosition to the next entry, assuming that the current
1242 : * entry is of length entryLength. If we jump to a new page the function
1243 : * returns true, else false.
1244 : */
1245 : static bool
1246 4 : asyncQueueAdvance(volatile QueuePosition *position, int entryLength)
1247 : {
1248 4 : int pageno = QUEUE_POS_PAGE(*position);
1249 4 : int offset = QUEUE_POS_OFFSET(*position);
1250 4 : bool pageJump = false;
1251 :
1252 : /*
1253 : * Move to the next writing position: First jump over what we have just
1254 : * written or read.
1255 : */
1256 4 : offset += entryLength;
1257 4 : Assert(offset <= QUEUE_PAGESIZE);
1258 :
1259 : /*
1260 : * In a second step check if another entry can possibly be written to the
1261 : * page. If so, stay here, we have reached the next position. If not, then
1262 : * we need to move on to the next page.
1263 : */
1264 4 : if (offset + QUEUEALIGN(AsyncQueueEntryEmptySize) > QUEUE_PAGESIZE)
1265 : {
1266 0 : pageno++;
1267 0 : if (pageno > QUEUE_MAX_PAGE)
1268 0 : pageno = 0; /* wrap around */
1269 0 : offset = 0;
1270 0 : pageJump = true;
1271 : }
1272 :
1273 4 : SET_QUEUE_POS(*position, pageno, offset);
1274 4 : return pageJump;
1275 : }
1276 :
1277 : /*
1278 : * Fill the AsyncQueueEntry at *qe with an outbound notification message.
1279 : */
1280 : static void
1281 4 : asyncQueueNotificationToEntry(Notification *n, AsyncQueueEntry *qe)
1282 : {
1283 4 : size_t channellen = strlen(n->channel);
1284 4 : size_t payloadlen = strlen(n->payload);
1285 : int entryLength;
1286 :
1287 4 : Assert(channellen < NAMEDATALEN);
1288 4 : Assert(payloadlen < NOTIFY_PAYLOAD_MAX_LENGTH);
1289 :
1290 : /* The terminators are already included in AsyncQueueEntryEmptySize */
1291 4 : entryLength = AsyncQueueEntryEmptySize + payloadlen + channellen;
1292 4 : entryLength = QUEUEALIGN(entryLength);
1293 4 : qe->length = entryLength;
1294 4 : qe->dboid = MyDatabaseId;
1295 4 : qe->xid = GetCurrentTransactionId();
1296 4 : qe->srcPid = MyProcPid;
1297 4 : memcpy(qe->data, n->channel, channellen + 1);
1298 4 : memcpy(qe->data + channellen + 1, n->payload, payloadlen + 1);
1299 4 : }
1300 :
1301 : /*
1302 : * Add pending notifications to the queue.
1303 : *
1304 : * We go page by page here, i.e. we stop once we have to go to a new page but
1305 : * we will be called again and then fill that next page. If an entry does not
1306 : * fit into the current page, we write a dummy entry with an InvalidOid as the
1307 : * database OID in order to fill the page. So every page is always used up to
1308 : * the last byte which simplifies reading the page later.
1309 : *
1310 : * We are passed the list cell containing the next notification to write
1311 : * and return the first still-unwritten cell back. Eventually we will return
1312 : * NULL indicating all is done.
1313 : *
1314 : * We are holding AsyncQueueLock already from the caller and grab AsyncCtlLock
1315 : * locally in this function.
1316 : */
1317 : static ListCell *
1318 4 : asyncQueueAddEntries(ListCell *nextNotify)
1319 : {
1320 : AsyncQueueEntry qe;
1321 : QueuePosition queue_head;
1322 : int pageno;
1323 : int offset;
1324 : int slotno;
1325 :
1326 : /* We hold both AsyncQueueLock and AsyncCtlLock during this operation */
1327 4 : LWLockAcquire(AsyncCtlLock, LW_EXCLUSIVE);
1328 :
1329 : /*
1330 : * We work with a local copy of QUEUE_HEAD, which we write back to shared
1331 : * memory upon exiting. The reason for this is that if we have to advance
1332 : * to a new page, SimpleLruZeroPage might fail (out of disk space, for
1333 : * instance), and we must not advance QUEUE_HEAD if it does. (Otherwise,
1334 : * subsequent insertions would try to put entries into a page that slru.c
1335 : * thinks doesn't exist yet.) So, use a local position variable. Note
1336 : * that if we do fail, any already-inserted queue entries are forgotten;
1337 : * this is okay, since they'd be useless anyway after our transaction
1338 : * rolls back.
1339 : */
1340 4 : queue_head = QUEUE_HEAD;
1341 :
1342 : /* Fetch the current page */
1343 4 : pageno = QUEUE_POS_PAGE(queue_head);
1344 4 : slotno = SimpleLruReadPage(AsyncCtl, pageno, true, InvalidTransactionId);
1345 : /* Note we mark the page dirty before writing in it */
1346 4 : AsyncCtl->shared->page_dirty[slotno] = true;
1347 :
1348 12 : while (nextNotify != NULL)
1349 : {
1350 4 : Notification *n = (Notification *) lfirst(nextNotify);
1351 :
1352 : /* Construct a valid queue entry in local variable qe */
1353 4 : asyncQueueNotificationToEntry(n, &qe);
1354 :
1355 4 : offset = QUEUE_POS_OFFSET(queue_head);
1356 :
1357 : /* Check whether the entry really fits on the current page */
1358 4 : if (offset + qe.length <= QUEUE_PAGESIZE)
1359 : {
1360 : /* OK, so advance nextNotify past this item */
1361 4 : nextNotify = lnext(nextNotify);
1362 : }
1363 : else
1364 : {
1365 : /*
1366 : * Write a dummy entry to fill up the page. Actually readers will
1367 : * only check dboid and since it won't match any reader's database
1368 : * OID, they will ignore this entry and move on.
1369 : */
1370 0 : qe.length = QUEUE_PAGESIZE - offset;
1371 0 : qe.dboid = InvalidOid;
1372 0 : qe.data[0] = '\0'; /* empty channel */
1373 0 : qe.data[1] = '\0'; /* empty payload */
1374 : }
1375 :
1376 : /* Now copy qe into the shared buffer page */
1377 4 : memcpy(AsyncCtl->shared->page_buffer[slotno] + offset,
1378 : &qe,
1379 4 : qe.length);
1380 :
1381 : /* Advance queue_head appropriately, and detect if page is full */
1382 4 : if (asyncQueueAdvance(&(queue_head), qe.length))
1383 : {
1384 : /*
1385 : * Page is full, so we're done here, but first fill the next page
1386 : * with zeroes. The reason to do this is to ensure that slru.c's
1387 : * idea of the head page is always the same as ours, which avoids
1388 : * boundary problems in SimpleLruTruncate. The test in
1389 : * asyncQueueIsFull() ensured that there is room to create this
1390 : * page without overrunning the queue.
1391 : */
1392 0 : slotno = SimpleLruZeroPage(AsyncCtl, QUEUE_POS_PAGE(queue_head));
1393 : /* And exit the loop */
1394 0 : break;
1395 : }
1396 : }
1397 :
1398 : /* Success, so update the global QUEUE_HEAD */
1399 4 : QUEUE_HEAD = queue_head;
1400 :
1401 4 : LWLockRelease(AsyncCtlLock);
1402 :
1403 4 : return nextNotify;
1404 : }
1405 :
1406 : /*
1407 : * SQL function to return the fraction of the notification queue currently
1408 : * occupied.
1409 : */
1410 : Datum
1411 1 : pg_notification_queue_usage(PG_FUNCTION_ARGS)
1412 : {
1413 : double usage;
1414 :
1415 1 : LWLockAcquire(AsyncQueueLock, LW_SHARED);
1416 1 : usage = asyncQueueUsage();
1417 1 : LWLockRelease(AsyncQueueLock);
1418 :
1419 1 : PG_RETURN_FLOAT8(usage);
1420 : }
1421 :
1422 : /*
1423 : * Return the fraction of the queue that is currently occupied.
1424 : *
1425 : * The caller must hold AsyncQueueLock in (at least) shared mode.
1426 : */
1427 : static double
1428 5 : asyncQueueUsage(void)
1429 : {
1430 5 : int headPage = QUEUE_POS_PAGE(QUEUE_HEAD);
1431 5 : int tailPage = QUEUE_POS_PAGE(QUEUE_TAIL);
1432 : int occupied;
1433 :
1434 5 : occupied = headPage - tailPage;
1435 :
1436 5 : if (occupied == 0)
1437 5 : return (double) 0; /* fast exit for common case */
1438 :
1439 0 : if (occupied < 0)
1440 : {
1441 : /* head has wrapped around, tail not yet */
1442 0 : occupied += QUEUE_MAX_PAGE + 1;
1443 : }
1444 :
1445 0 : return (double) occupied / (double) ((QUEUE_MAX_PAGE + 1) / 2);
1446 : }
1447 :
1448 : /*
1449 : * Check whether the queue is at least half full, and emit a warning if so.
1450 : *
1451 : * This is unlikely given the size of the queue, but possible.
1452 : * The warnings show up at most once every QUEUE_FULL_WARN_INTERVAL.
1453 : *
1454 : * Caller must hold exclusive AsyncQueueLock.
1455 : */
1456 : static void
1457 4 : asyncQueueFillWarning(void)
1458 : {
1459 : double fillDegree;
1460 : TimestampTz t;
1461 :
1462 4 : fillDegree = asyncQueueUsage();
1463 4 : if (fillDegree < 0.5)
1464 8 : return;
1465 :
1466 0 : t = GetCurrentTimestamp();
1467 :
1468 0 : if (TimestampDifferenceExceeds(asyncQueueControl->lastQueueFillWarn,
1469 : t, QUEUE_FULL_WARN_INTERVAL))
1470 : {
1471 0 : QueuePosition min = QUEUE_HEAD;
1472 0 : int32 minPid = InvalidPid;
1473 : int i;
1474 :
1475 0 : for (i = 1; i <= MaxBackends; i++)
1476 : {
1477 0 : if (QUEUE_BACKEND_PID(i) != InvalidPid)
1478 : {
1479 0 : min = QUEUE_POS_MIN(min, QUEUE_BACKEND_POS(i));
1480 0 : if (QUEUE_POS_EQUAL(min, QUEUE_BACKEND_POS(i)))
1481 0 : minPid = QUEUE_BACKEND_PID(i);
1482 : }
1483 : }
1484 :
1485 0 : ereport(WARNING,
1486 : (errmsg("NOTIFY queue is %.0f%% full", fillDegree * 100),
1487 : (minPid != InvalidPid ?
1488 : errdetail("The server process with PID %d is among those with the oldest transactions.", minPid)
1489 : : 0),
1490 : (minPid != InvalidPid ?
1491 : errhint("The NOTIFY queue cannot be emptied until that process ends its current transaction.")
1492 : : 0)));
1493 :
1494 0 : asyncQueueControl->lastQueueFillWarn = t;
1495 : }
1496 : }
1497 :
1498 : /*
1499 : * Send signals to all listening backends (except our own).
1500 : *
1501 : * Returns true if we sent at least one signal.
1502 : *
1503 : * Since we need EXCLUSIVE lock anyway we also check the position of the other
1504 : * backends and in case one is already up-to-date we don't signal it.
1505 : * This can happen if concurrent notifying transactions have sent a signal and
1506 : * the signaled backend has read the other notifications and ours in the same
1507 : * step.
1508 : *
1509 : * Since we know the BackendId and the Pid the signalling is quite cheap.
1510 : */
1511 : static bool
1512 4 : SignalBackends(void)
1513 : {
1514 4 : bool signalled = false;
1515 : int32 *pids;
1516 : BackendId *ids;
1517 : int count;
1518 : int i;
1519 : int32 pid;
1520 :
1521 : /*
1522 : * Identify all backends that are listening and not already up-to-date. We
1523 : * don't want to send signals while holding the AsyncQueueLock, so we just
1524 : * build a list of target PIDs.
1525 : *
1526 : * XXX in principle these pallocs could fail, which would be bad. Maybe
1527 : * preallocate the arrays? But in practice this is only run in trivial
1528 : * transactions, so there should surely be space available.
1529 : */
1530 4 : pids = (int32 *) palloc(MaxBackends * sizeof(int32));
1531 4 : ids = (BackendId *) palloc(MaxBackends * sizeof(BackendId));
1532 4 : count = 0;
1533 :
1534 4 : LWLockAcquire(AsyncQueueLock, LW_EXCLUSIVE);
1535 452 : for (i = 1; i <= MaxBackends; i++)
1536 : {
1537 448 : pid = QUEUE_BACKEND_PID(i);
1538 448 : if (pid != InvalidPid && pid != MyProcPid)
1539 : {
1540 0 : QueuePosition pos = QUEUE_BACKEND_POS(i);
1541 :
1542 0 : if (!QUEUE_POS_EQUAL(pos, QUEUE_HEAD))
1543 : {
1544 0 : pids[count] = pid;
1545 0 : ids[count] = i;
1546 0 : count++;
1547 : }
1548 : }
1549 : }
1550 4 : LWLockRelease(AsyncQueueLock);
1551 :
1552 : /* Now send signals */
1553 4 : for (i = 0; i < count; i++)
1554 : {
1555 0 : pid = pids[i];
1556 :
1557 : /*
1558 : * Note: assuming things aren't broken, a signal failure here could
1559 : * only occur if the target backend exited since we released
1560 : * AsyncQueueLock; which is unlikely but certainly possible. So we
1561 : * just log a low-level debug message if it happens.
1562 : */
1563 0 : if (SendProcSignal(pid, PROCSIG_NOTIFY_INTERRUPT, ids[i]) < 0)
1564 0 : elog(DEBUG3, "could not signal backend with PID %d: %m", pid);
1565 : else
1566 0 : signalled = true;
1567 : }
1568 :
1569 4 : pfree(pids);
1570 4 : pfree(ids);
1571 :
1572 4 : return signalled;
1573 : }
1574 :
1575 : /*
1576 : * AtAbort_Notify
1577 : *
1578 : * This is called at transaction abort.
1579 : *
1580 : * Gets rid of pending actions and outbound notifies that we would have
1581 : * executed if the transaction got committed.
1582 : */
1583 : void
1584 3306 : AtAbort_Notify(void)
1585 : {
1586 : /*
1587 : * If we LISTEN but then roll back the transaction after PreCommit_Notify,
1588 : * we have registered as a listener but have not made any entry in
1589 : * listenChannels. In that case, deregister again.
1590 : */
1591 3306 : if (amRegisteredListener && listenChannels == NIL)
1592 0 : asyncQueueUnregister();
1593 :
1594 : /* And clean up */
1595 3306 : ClearPendingActionsAndNotifies();
1596 3306 : }
1597 :
1598 : /*
1599 : * AtSubStart_Notify() --- Take care of subtransaction start.
1600 : *
1601 : * Push empty state for the new subtransaction.
1602 : */
1603 : void
1604 372 : AtSubStart_Notify(void)
1605 : {
1606 : MemoryContext old_cxt;
1607 :
1608 : /* Keep the list-of-lists in TopTransactionContext for simplicity */
1609 372 : old_cxt = MemoryContextSwitchTo(TopTransactionContext);
1610 :
1611 372 : upperPendingActions = lcons(pendingActions, upperPendingActions);
1612 :
1613 372 : Assert(list_length(upperPendingActions) ==
1614 : GetCurrentTransactionNestLevel() - 1);
1615 :
1616 372 : pendingActions = NIL;
1617 :
1618 372 : upperPendingNotifies = lcons(pendingNotifies, upperPendingNotifies);
1619 :
1620 372 : Assert(list_length(upperPendingNotifies) ==
1621 : GetCurrentTransactionNestLevel() - 1);
1622 :
1623 372 : pendingNotifies = NIL;
1624 :
1625 372 : MemoryContextSwitchTo(old_cxt);
1626 372 : }
1627 :
1628 : /*
1629 : * AtSubCommit_Notify() --- Take care of subtransaction commit.
1630 : *
1631 : * Reassign all items in the pending lists to the parent transaction.
1632 : */
1633 : void
1634 49 : AtSubCommit_Notify(void)
1635 : {
1636 : List *parentPendingActions;
1637 : List *parentPendingNotifies;
1638 :
1639 49 : parentPendingActions = linitial_node(List, upperPendingActions);
1640 49 : upperPendingActions = list_delete_first(upperPendingActions);
1641 :
1642 49 : Assert(list_length(upperPendingActions) ==
1643 : GetCurrentTransactionNestLevel() - 2);
1644 :
1645 : /*
1646 : * Mustn't try to eliminate duplicates here --- see queue_listen()
1647 : */
1648 49 : pendingActions = list_concat(parentPendingActions, pendingActions);
1649 :
1650 49 : parentPendingNotifies = linitial_node(List, upperPendingNotifies);
1651 49 : upperPendingNotifies = list_delete_first(upperPendingNotifies);
1652 :
1653 49 : Assert(list_length(upperPendingNotifies) ==
1654 : GetCurrentTransactionNestLevel() - 2);
1655 :
1656 : /*
1657 : * We could try to eliminate duplicates here, but it seems not worthwhile.
1658 : */
1659 49 : pendingNotifies = list_concat(parentPendingNotifies, pendingNotifies);
1660 49 : }
1661 :
1662 : /*
1663 : * AtSubAbort_Notify() --- Take care of subtransaction abort.
1664 : */
1665 : void
1666 323 : AtSubAbort_Notify(void)
1667 : {
1668 323 : int my_level = GetCurrentTransactionNestLevel();
1669 :
1670 : /*
1671 : * All we have to do is pop the stack --- the actions/notifies made in
1672 : * this subxact are no longer interesting, and the space will be freed
1673 : * when CurTransactionContext is recycled.
1674 : *
1675 : * This routine could be called more than once at a given nesting level if
1676 : * there is trouble during subxact abort. Avoid dumping core by using
1677 : * GetCurrentTransactionNestLevel as the indicator of how far we need to
1678 : * prune the list.
1679 : */
1680 969 : while (list_length(upperPendingActions) > my_level - 2)
1681 : {
1682 323 : pendingActions = linitial_node(List, upperPendingActions);
1683 323 : upperPendingActions = list_delete_first(upperPendingActions);
1684 : }
1685 :
1686 969 : while (list_length(upperPendingNotifies) > my_level - 2)
1687 : {
1688 323 : pendingNotifies = linitial_node(List, upperPendingNotifies);
1689 323 : upperPendingNotifies = list_delete_first(upperPendingNotifies);
1690 : }
1691 323 : }
1692 :
1693 : /*
1694 : * HandleNotifyInterrupt
1695 : *
1696 : * Signal handler portion of interrupt handling. Let the backend know
1697 : * that there's a pending notify interrupt. If we're currently reading
1698 : * from the client, this will interrupt the read and
1699 : * ProcessClientReadInterrupt() will call ProcessNotifyInterrupt().
1700 : */
1701 : void
1702 0 : HandleNotifyInterrupt(void)
1703 : {
1704 : /*
1705 : * Note: this is called by a SIGNAL HANDLER. You must be very wary what
1706 : * you do here.
1707 : */
1708 :
1709 : /* signal that work needs to be done */
1710 0 : notifyInterruptPending = true;
1711 :
1712 : /* make sure the event is processed in due course */
1713 0 : SetLatch(MyLatch);
1714 0 : }
1715 :
1716 : /*
1717 : * ProcessNotifyInterrupt
1718 : *
1719 : * This is called just after waiting for a frontend command. If a
1720 : * interrupt arrives (via HandleNotifyInterrupt()) while reading, the
1721 : * read will be interrupted via the process's latch, and this routine
1722 : * will get called. If we are truly idle (ie, *not* inside a transaction
1723 : * block), process the incoming notifies.
1724 : */
1725 : void
1726 0 : ProcessNotifyInterrupt(void)
1727 : {
1728 0 : if (IsTransactionOrTransactionBlock())
1729 0 : return; /* not really idle */
1730 :
1731 0 : while (notifyInterruptPending)
1732 0 : ProcessIncomingNotify();
1733 : }
1734 :
1735 :
1736 : /*
1737 : * Read all pending notifications from the queue, and deliver appropriate
1738 : * ones to my frontend. Stop when we reach queue head or an uncommitted
1739 : * notification.
1740 : */
1741 : static void
1742 0 : asyncQueueReadAllNotifications(void)
1743 : {
1744 : volatile QueuePosition pos;
1745 : QueuePosition oldpos;
1746 : QueuePosition head;
1747 : bool advanceTail;
1748 :
1749 : /* page_buffer must be adequately aligned, so use a union */
1750 : union
1751 : {
1752 : char buf[QUEUE_PAGESIZE];
1753 : AsyncQueueEntry align;
1754 : } page_buffer;
1755 :
1756 : /* Fetch current state */
1757 0 : LWLockAcquire(AsyncQueueLock, LW_SHARED);
1758 : /* Assert checks that we have a valid state entry */
1759 0 : Assert(MyProcPid == QUEUE_BACKEND_PID(MyBackendId));
1760 0 : pos = oldpos = QUEUE_BACKEND_POS(MyBackendId);
1761 0 : head = QUEUE_HEAD;
1762 0 : LWLockRelease(AsyncQueueLock);
1763 :
1764 0 : if (QUEUE_POS_EQUAL(pos, head))
1765 : {
1766 : /* Nothing to do, we have read all notifications already. */
1767 0 : return;
1768 : }
1769 :
1770 : /*----------
1771 : * Note that we deliver everything that we see in the queue and that
1772 : * matches our _current_ listening state.
1773 : * Especially we do not take into account different commit times.
1774 : * Consider the following example:
1775 : *
1776 : * Backend 1: Backend 2:
1777 : *
1778 : * transaction starts
1779 : * NOTIFY foo;
1780 : * commit starts
1781 : * transaction starts
1782 : * LISTEN foo;
1783 : * commit starts
1784 : * commit to clog
1785 : * commit to clog
1786 : *
1787 : * It could happen that backend 2 sees the notification from backend 1 in
1788 : * the queue. Even though the notifying transaction committed before
1789 : * the listening transaction, we still deliver the notification.
1790 : *
1791 : * The idea is that an additional notification does not do any harm, we
1792 : * just need to make sure that we do not miss a notification.
1793 : *
1794 : * It is possible that we fail while trying to send a message to our
1795 : * frontend (for example, because of encoding conversion failure).
1796 : * If that happens it is critical that we not try to send the same
1797 : * message over and over again. Therefore, we place a PG_TRY block
1798 : * here that will forcibly advance our backend position before we lose
1799 : * control to an error. (We could alternatively retake AsyncQueueLock
1800 : * and move the position before handling each individual message, but
1801 : * that seems like too much lock traffic.)
1802 : *----------
1803 : */
1804 0 : PG_TRY();
1805 : {
1806 : bool reachedStop;
1807 :
1808 : do
1809 : {
1810 0 : int curpage = QUEUE_POS_PAGE(pos);
1811 0 : int curoffset = QUEUE_POS_OFFSET(pos);
1812 : int slotno;
1813 : int copysize;
1814 :
1815 : /*
1816 : * We copy the data from SLRU into a local buffer, so as to avoid
1817 : * holding the AsyncCtlLock while we are examining the entries and
1818 : * possibly transmitting them to our frontend. Copy only the part
1819 : * of the page we will actually inspect.
1820 : */
1821 0 : slotno = SimpleLruReadPage_ReadOnly(AsyncCtl, curpage,
1822 : InvalidTransactionId);
1823 0 : if (curpage == QUEUE_POS_PAGE(head))
1824 : {
1825 : /* we only want to read as far as head */
1826 0 : copysize = QUEUE_POS_OFFSET(head) - curoffset;
1827 0 : if (copysize < 0)
1828 0 : copysize = 0; /* just for safety */
1829 : }
1830 : else
1831 : {
1832 : /* fetch all the rest of the page */
1833 0 : copysize = QUEUE_PAGESIZE - curoffset;
1834 : }
1835 0 : memcpy(page_buffer.buf + curoffset,
1836 0 : AsyncCtl->shared->page_buffer[slotno] + curoffset,
1837 : copysize);
1838 : /* Release lock that we got from SimpleLruReadPage_ReadOnly() */
1839 0 : LWLockRelease(AsyncCtlLock);
1840 :
1841 : /*
1842 : * Process messages up to the stop position, end of page, or an
1843 : * uncommitted message.
1844 : *
1845 : * Our stop position is what we found to be the head's position
1846 : * when we entered this function. It might have changed already.
1847 : * But if it has, we will receive (or have already received and
1848 : * queued) another signal and come here again.
1849 : *
1850 : * We are not holding AsyncQueueLock here! The queue can only
1851 : * extend beyond the head pointer (see above) and we leave our
1852 : * backend's pointer where it is so nobody will truncate or
1853 : * rewrite pages under us. Especially we don't want to hold a lock
1854 : * while sending the notifications to the frontend.
1855 : */
1856 0 : reachedStop = asyncQueueProcessPageEntries(&pos, head,
1857 : page_buffer.buf);
1858 0 : } while (!reachedStop);
1859 : }
1860 0 : PG_CATCH();
1861 : {
1862 : /* Update shared state */
1863 0 : LWLockAcquire(AsyncQueueLock, LW_SHARED);
1864 0 : QUEUE_BACKEND_POS(MyBackendId) = pos;
1865 0 : advanceTail = QUEUE_POS_EQUAL(oldpos, QUEUE_TAIL);
1866 0 : LWLockRelease(AsyncQueueLock);
1867 :
1868 : /* If we were the laziest backend, try to advance the tail pointer */
1869 0 : if (advanceTail)
1870 0 : asyncQueueAdvanceTail();
1871 :
1872 0 : PG_RE_THROW();
1873 : }
1874 0 : PG_END_TRY();
1875 :
1876 : /* Update shared state */
1877 0 : LWLockAcquire(AsyncQueueLock, LW_SHARED);
1878 0 : QUEUE_BACKEND_POS(MyBackendId) = pos;
1879 0 : advanceTail = QUEUE_POS_EQUAL(oldpos, QUEUE_TAIL);
1880 0 : LWLockRelease(AsyncQueueLock);
1881 :
1882 : /* If we were the laziest backend, try to advance the tail pointer */
1883 0 : if (advanceTail)
1884 0 : asyncQueueAdvanceTail();
1885 : }
1886 :
1887 : /*
1888 : * Fetch notifications from the shared queue, beginning at position current,
1889 : * and deliver relevant ones to my frontend.
1890 : *
1891 : * The current page must have been fetched into page_buffer from shared
1892 : * memory. (We could access the page right in shared memory, but that
1893 : * would imply holding the AsyncCtlLock throughout this routine.)
1894 : *
1895 : * We stop if we reach the "stop" position, or reach a notification from an
1896 : * uncommitted transaction, or reach the end of the page.
1897 : *
1898 : * The function returns true once we have reached the stop position or an
1899 : * uncommitted notification, and false if we have finished with the page.
1900 : * In other words: once it returns true there is no need to look further.
1901 : * The QueuePosition *current is advanced past all processed messages.
1902 : */
1903 : static bool
1904 0 : asyncQueueProcessPageEntries(volatile QueuePosition *current,
1905 : QueuePosition stop,
1906 : char *page_buffer)
1907 : {
1908 0 : bool reachedStop = false;
1909 : bool reachedEndOfPage;
1910 : AsyncQueueEntry *qe;
1911 :
1912 : do
1913 : {
1914 0 : QueuePosition thisentry = *current;
1915 :
1916 0 : if (QUEUE_POS_EQUAL(thisentry, stop))
1917 0 : break;
1918 :
1919 0 : qe = (AsyncQueueEntry *) (page_buffer + QUEUE_POS_OFFSET(thisentry));
1920 :
1921 : /*
1922 : * Advance *current over this message, possibly to the next page. As
1923 : * noted in the comments for asyncQueueReadAllNotifications, we must
1924 : * do this before possibly failing while processing the message.
1925 : */
1926 0 : reachedEndOfPage = asyncQueueAdvance(current, qe->length);
1927 :
1928 : /* Ignore messages destined for other databases */
1929 0 : if (qe->dboid == MyDatabaseId)
1930 : {
1931 0 : if (TransactionIdIsInProgress(qe->xid))
1932 : {
1933 : /*
1934 : * The source transaction is still in progress, so we can't
1935 : * process this message yet. Break out of the loop, but first
1936 : * back up *current so we will reprocess the message next
1937 : * time. (Note: it is unlikely but not impossible for
1938 : * TransactionIdDidCommit to fail, so we can't really avoid
1939 : * this advance-then-back-up behavior when dealing with an
1940 : * uncommitted message.)
1941 : *
1942 : * Note that we must test TransactionIdIsInProgress before we
1943 : * test TransactionIdDidCommit, else we might return a message
1944 : * from a transaction that is not yet visible to snapshots;
1945 : * compare the comments at the head of tqual.c.
1946 : */
1947 0 : *current = thisentry;
1948 0 : reachedStop = true;
1949 0 : break;
1950 : }
1951 0 : else if (TransactionIdDidCommit(qe->xid))
1952 : {
1953 : /* qe->data is the null-terminated channel name */
1954 0 : char *channel = qe->data;
1955 :
1956 0 : if (IsListeningOn(channel))
1957 : {
1958 : /* payload follows channel name */
1959 0 : char *payload = qe->data + strlen(channel) + 1;
1960 :
1961 0 : NotifyMyFrontEnd(channel, payload, qe->srcPid);
1962 : }
1963 : }
1964 : else
1965 : {
1966 : /*
1967 : * The source transaction aborted or crashed, so we just
1968 : * ignore its notifications.
1969 : */
1970 : }
1971 : }
1972 :
1973 : /* Loop back if we're not at end of page */
1974 0 : } while (!reachedEndOfPage);
1975 :
1976 0 : if (QUEUE_POS_EQUAL(*current, stop))
1977 0 : reachedStop = true;
1978 :
1979 0 : return reachedStop;
1980 : }
1981 :
1982 : /*
1983 : * Advance the shared queue tail variable to the minimum of all the
1984 : * per-backend tail pointers. Truncate pg_notify space if possible.
1985 : */
1986 : static void
1987 6 : asyncQueueAdvanceTail(void)
1988 : {
1989 : QueuePosition min;
1990 : int i;
1991 : int oldtailpage;
1992 : int newtailpage;
1993 : int boundary;
1994 :
1995 6 : LWLockAcquire(AsyncQueueLock, LW_EXCLUSIVE);
1996 6 : min = QUEUE_HEAD;
1997 678 : for (i = 1; i <= MaxBackends; i++)
1998 : {
1999 672 : if (QUEUE_BACKEND_PID(i) != InvalidPid)
2000 0 : min = QUEUE_POS_MIN(min, QUEUE_BACKEND_POS(i));
2001 : }
2002 6 : oldtailpage = QUEUE_POS_PAGE(QUEUE_TAIL);
2003 6 : QUEUE_TAIL = min;
2004 6 : LWLockRelease(AsyncQueueLock);
2005 :
2006 : /*
2007 : * We can truncate something if the global tail advanced across an SLRU
2008 : * segment boundary.
2009 : *
2010 : * XXX it might be better to truncate only once every several segments, to
2011 : * reduce the number of directory scans.
2012 : */
2013 6 : newtailpage = QUEUE_POS_PAGE(min);
2014 6 : boundary = newtailpage - (newtailpage % SLRU_PAGES_PER_SEGMENT);
2015 6 : if (asyncQueuePagePrecedes(oldtailpage, boundary))
2016 : {
2017 : /*
2018 : * SimpleLruTruncate() will ask for AsyncCtlLock but will also release
2019 : * the lock again.
2020 : */
2021 0 : SimpleLruTruncate(AsyncCtl, newtailpage);
2022 : }
2023 6 : }
2024 :
2025 : /*
2026 : * ProcessIncomingNotify
2027 : *
2028 : * Deal with arriving NOTIFYs from other backends as soon as it's safe to
2029 : * do so. This used to be called from the PROCSIG_NOTIFY_INTERRUPT
2030 : * signal handler, but isn't anymore.
2031 : *
2032 : * Scan the queue for arriving notifications and report them to my front
2033 : * end.
2034 : *
2035 : * NOTE: since we are outside any transaction, we must create our own.
2036 : */
2037 : static void
2038 0 : ProcessIncomingNotify(void)
2039 : {
2040 : /* We *must* reset the flag */
2041 0 : notifyInterruptPending = false;
2042 :
2043 : /* Do nothing else if we aren't actively listening */
2044 0 : if (listenChannels == NIL)
2045 0 : return;
2046 :
2047 0 : if (Trace_notify)
2048 0 : elog(DEBUG1, "ProcessIncomingNotify");
2049 :
2050 0 : set_ps_display("notify interrupt", false);
2051 :
2052 : /*
2053 : * We must run asyncQueueReadAllNotifications inside a transaction, else
2054 : * bad things happen if it gets an error.
2055 : */
2056 0 : StartTransactionCommand();
2057 :
2058 0 : asyncQueueReadAllNotifications();
2059 :
2060 0 : CommitTransactionCommand();
2061 :
2062 : /*
2063 : * Must flush the notify messages to ensure frontend gets them promptly.
2064 : */
2065 0 : pq_flush();
2066 :
2067 0 : set_ps_display("idle", false);
2068 :
2069 0 : if (Trace_notify)
2070 0 : elog(DEBUG1, "ProcessIncomingNotify: done");
2071 : }
2072 :
2073 : /*
2074 : * Send NOTIFY message to my front end.
2075 : */
2076 : void
2077 0 : NotifyMyFrontEnd(const char *channel, const char *payload, int32 srcPid)
2078 : {
2079 0 : if (whereToSendOutput == DestRemote)
2080 : {
2081 : StringInfoData buf;
2082 :
2083 0 : pq_beginmessage(&buf, 'A');
2084 0 : pq_sendint(&buf, srcPid, sizeof(int32));
2085 0 : pq_sendstring(&buf, channel);
2086 0 : if (PG_PROTOCOL_MAJOR(FrontendProtocol) >= 3)
2087 0 : pq_sendstring(&buf, payload);
2088 0 : pq_endmessage(&buf);
2089 :
2090 : /*
2091 : * NOTE: we do not do pq_flush() here. For a self-notify, it will
2092 : * happen at the end of the transaction, and for incoming notifies
2093 : * ProcessIncomingNotify will do it after finding all the notifies.
2094 : */
2095 : }
2096 : else
2097 0 : elog(INFO, "NOTIFY for \"%s\" payload \"%s\"", channel, payload);
2098 0 : }
2099 :
2100 : /* Does pendingNotifies include the given channel/payload? */
2101 : static bool
2102 4 : AsyncExistsPendingNotify(const char *channel, const char *payload)
2103 : {
2104 : ListCell *p;
2105 : Notification *n;
2106 :
2107 4 : if (pendingNotifies == NIL)
2108 4 : return false;
2109 :
2110 0 : if (payload == NULL)
2111 0 : payload = "";
2112 :
2113 : /*----------
2114 : * We need to append new elements to the end of the list in order to keep
2115 : * the order. However, on the other hand we'd like to check the list
2116 : * backwards in order to make duplicate-elimination a tad faster when the
2117 : * same condition is signaled many times in a row. So as a compromise we
2118 : * check the tail element first which we can access directly. If this
2119 : * doesn't match, we check the whole list.
2120 : *
2121 : * As we are not checking our parents' lists, we can still get duplicates
2122 : * in combination with subtransactions, like in:
2123 : *
2124 : * begin;
2125 : * notify foo '1';
2126 : * savepoint foo;
2127 : * notify foo '1';
2128 : * commit;
2129 : *----------
2130 : */
2131 0 : n = (Notification *) llast(pendingNotifies);
2132 0 : if (strcmp(n->channel, channel) == 0 &&
2133 0 : strcmp(n->payload, payload) == 0)
2134 0 : return true;
2135 :
2136 0 : foreach(p, pendingNotifies)
2137 : {
2138 0 : n = (Notification *) lfirst(p);
2139 :
2140 0 : if (strcmp(n->channel, channel) == 0 &&
2141 0 : strcmp(n->payload, payload) == 0)
2142 0 : return true;
2143 : }
2144 :
2145 0 : return false;
2146 : }
2147 :
2148 : /* Clear the pendingActions and pendingNotifies lists. */
2149 : static void
2150 3315 : ClearPendingActionsAndNotifies(void)
2151 : {
2152 : /*
2153 : * We used to have to explicitly deallocate the list members and nodes,
2154 : * because they were malloc'd. Now, since we know they are palloc'd in
2155 : * CurTransactionContext, we need not do that --- they'll go away
2156 : * automatically at transaction exit. We need only reset the list head
2157 : * pointers.
2158 : */
2159 3315 : pendingActions = NIL;
2160 3315 : pendingNotifies = NIL;
2161 3315 : }
|
