Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * async.c
4 : * Asynchronous notification: NOTIFY, LISTEN, UNLISTEN
5 : *
6 : * Portions Copyright (c) 1996-2026, 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 v19:
17 : *
18 : * 1. Multiple backends on same machine. Multiple backends may be listening
19 : * on each of several channels.
20 : *
21 : * 2. There is one central queue in disk-based storage (directory pg_notify/),
22 : * with actively-used pages mapped into shared memory by the slru.c module.
23 : * All notification messages are placed in the queue and later read out
24 : * by listening backends. The single queue allows us to guarantee that
25 : * notifications are received in commit order.
26 : *
27 : * Although there is only one queue, notifications are treated as being
28 : * database-local; this is done by including the sender's database OID
29 : * in each notification message. Listening backends ignore messages
30 : * that don't match their database OID. This is important because it
31 : * ensures senders and receivers have the same database encoding and won't
32 : * misinterpret non-ASCII text in the channel name or payload string.
33 : *
34 : * Since notifications are not expected to survive database crashes,
35 : * we can simply clean out the pg_notify data at any reboot, and there
36 : * is no need for WAL support or fsync'ing.
37 : *
38 : * 3. Every backend that is listening on at least one channel registers by
39 : * entering its PID into the array in AsyncQueueControl. It then scans all
40 : * incoming notifications in the central queue and first compares the
41 : * database OID of the notification with its own database OID and then
42 : * compares the notified channel with the list of channels that it listens
43 : * to. In case there is a match it delivers the notification event to its
44 : * frontend. Non-matching events are simply skipped.
45 : *
46 : * 4. The NOTIFY statement (routine Async_Notify) stores the notification in
47 : * a backend-local list which will not be processed until transaction end.
48 : *
49 : * Duplicate notifications from the same transaction are sent out as one
50 : * notification only. This is done to save work when for example a trigger
51 : * on a 2 million row table fires a notification for each row that has been
52 : * changed. If the application needs to receive every single notification
53 : * that has been sent, it can easily add some unique string into the extra
54 : * payload parameter.
55 : *
56 : * When the transaction is ready to commit, PreCommit_Notify() adds the
57 : * pending notifications to the head of the queue. The head pointer of the
58 : * queue always points to the next free position and a position is just a
59 : * page number and the offset in that page. This is done before marking the
60 : * transaction as committed in clog. If we run into problems writing the
61 : * notifications, we can still call elog(ERROR, ...) and the transaction
62 : * will roll back safely.
63 : *
64 : * Once we have put all of the notifications into the queue, we return to
65 : * CommitTransaction() which will then do the actual transaction commit.
66 : *
67 : * After commit we are called another time (AtCommit_Notify()). Here we
68 : * make any required updates to the effective listen state (see below).
69 : * Then we signal any backends that may be interested in our messages
70 : * (including our own backend, if listening). This is done by
71 : * SignalBackends(), which sends a PROCSIG_NOTIFY_INTERRUPT signal to
72 : * each relevant backend, as described below.
73 : *
74 : * Finally, after we are out of the transaction altogether and about to go
75 : * idle, we scan the queue for messages that need to be sent to our
76 : * frontend (which might be notifies from other backends, or self-notifies
77 : * from our own). This step is not part of the CommitTransaction sequence
78 : * for two important reasons. First, we could get errors while sending
79 : * data to our frontend, and it's really bad for errors to happen in
80 : * post-commit cleanup. Second, in cases where a procedure issues commits
81 : * within a single frontend command, we don't want to send notifies to our
82 : * frontend until the command is done; but notifies to other backends
83 : * should go out immediately after each commit.
84 : *
85 : * 5. Upon receipt of a PROCSIG_NOTIFY_INTERRUPT signal, the signal handler
86 : * sets the process's latch, which triggers the event to be processed
87 : * immediately if this backend is idle (i.e., it is waiting for a frontend
88 : * command and is not within a transaction block. C.f.
89 : * ProcessClientReadInterrupt()). Otherwise the handler may only set a
90 : * flag, which will cause the processing to occur just before we next go
91 : * idle.
92 : *
93 : * Inbound-notify processing consists of reading all of the notifications
94 : * that have arrived since scanning last time. We read every notification
95 : * until we reach either a notification from an uncommitted transaction or
96 : * the head pointer's position.
97 : *
98 : * 6. To limit disk space consumption, the tail pointer needs to be advanced
99 : * so that old pages can be truncated. This is relatively expensive
100 : * (notably, it requires an exclusive lock), so we don't want to do it
101 : * often. We make sending backends do this work if they advanced the queue
102 : * head into a new page, but only once every QUEUE_CLEANUP_DELAY pages.
103 : *
104 : * 7. So far we have not discussed how backends change their listening state,
105 : * nor how notification senders know which backends to awaken. To handle
106 : * the latter, we maintain a global channel table (implemented as a dynamic
107 : * shared hash table, or dshash) that maps channel names to the set of
108 : * backends listening on each channel. This table is created lazily on the
109 : * first LISTEN command and grows dynamically as needed. There is also a
110 : * local channel table (a plain dynahash table) in each listening backend,
111 : * tracking which channels that backend is listening to. The local table
112 : * serves to reduce the number of accesses needed to the shared table.
113 : *
114 : * If the current transaction has executed any LISTEN/UNLISTEN actions,
115 : * PreCommit_Notify() prepares to commit those. For LISTEN, it
116 : * pre-allocates entries in both the per-backend localChannelTable and the
117 : * shared globalChannelTable (with listening=false so that these entries
118 : * are no-ops for the moment). It also records the final per-channel
119 : * intent in pendingListenActions, so post-commit/abort processing can
120 : * apply that in a single step. Since all these allocations happen before
121 : * committing to clog, we can safely abort the transaction on failure.
122 : *
123 : * After commit, AtCommit_Notify() runs through pendingListenActions and
124 : * updates the backend's per-channel listening flags to activate or
125 : * deactivate listening. This happens before sending signals.
126 : *
127 : * SignalBackends() consults the shared global channel table to identify
128 : * listeners for the channels that the current transaction sent
129 : * notification(s) to. Each selected backend is marked as having a wakeup
130 : * pending to avoid duplicate signals, and a PROCSIG_NOTIFY_INTERRUPT
131 : * signal is sent to it.
132 : *
133 : * 8. While writing notifications, PreCommit_Notify() records the queue head
134 : * position both before and after the write. Because all writers serialize
135 : * on a cluster-wide heavyweight lock, no other backend can insert entries
136 : * between these two points. SignalBackends() uses this fact to directly
137 : * advance the queue pointer for any backend that is still positioned at
138 : * the old head, or within the range written, but is not interested in any
139 : * of our notifications. This avoids unnecessary wakeups for idle
140 : * listeners that have nothing to read. Backends that are not interested
141 : * in our notifications, but cannot be directly advanced, are signaled only
142 : * if they are far behind the current queue head; that is to ensure that
143 : * we can advance the queue tail without undue delay.
144 : *
145 : * An application that listens on the same channel it notifies will get
146 : * NOTIFY messages for its own NOTIFYs. These can be ignored, if not useful,
147 : * by comparing be_pid in the NOTIFY message to the application's own backend's
148 : * PID. (As of FE/BE protocol 2.0, the backend's PID is provided to the
149 : * frontend during startup.) The above design guarantees that notifies from
150 : * other backends will never be missed by ignoring self-notifies.
151 : *
152 : * The amount of shared memory used for notify management (notify_buffers)
153 : * can be varied without affecting anything but performance. The maximum
154 : * amount of notification data that can be queued at one time is determined
155 : * by the max_notify_queue_pages GUC.
156 : *-------------------------------------------------------------------------
157 : */
158 :
159 : #include "postgres.h"
160 :
161 : #include <limits.h>
162 : #include <unistd.h>
163 : #include <signal.h>
164 :
165 : #include "access/parallel.h"
166 : #include "access/slru.h"
167 : #include "access/transam.h"
168 : #include "access/xact.h"
169 : #include "catalog/pg_database.h"
170 : #include "commands/async.h"
171 : #include "common/hashfn.h"
172 : #include "funcapi.h"
173 : #include "lib/dshash.h"
174 : #include "libpq/libpq.h"
175 : #include "libpq/pqformat.h"
176 : #include "miscadmin.h"
177 : #include "storage/dsm_registry.h"
178 : #include "storage/ipc.h"
179 : #include "storage/lmgr.h"
180 : #include "storage/procsignal.h"
181 : #include "tcop/tcopprot.h"
182 : #include "utils/builtins.h"
183 : #include "utils/dsa.h"
184 : #include "utils/guc_hooks.h"
185 : #include "utils/memutils.h"
186 : #include "utils/ps_status.h"
187 : #include "utils/snapmgr.h"
188 : #include "utils/timestamp.h"
189 :
190 :
191 : /*
192 : * Maximum size of a NOTIFY payload, including terminating NULL. This
193 : * must be kept small enough so that a notification message fits on one
194 : * SLRU page. The magic fudge factor here is noncritical as long as it's
195 : * more than AsyncQueueEntryEmptySize --- we make it significantly bigger
196 : * than that, so changes in that data structure won't affect user-visible
197 : * restrictions.
198 : */
199 : #define NOTIFY_PAYLOAD_MAX_LENGTH (BLCKSZ - NAMEDATALEN - 128)
200 :
201 : /*
202 : * Struct representing an entry in the global notify queue
203 : *
204 : * This struct declaration has the maximal length, but in a real queue entry
205 : * the data area is only big enough for the actual channel and payload strings
206 : * (each null-terminated). AsyncQueueEntryEmptySize is the minimum possible
207 : * entry size, if both channel and payload strings are empty (but note it
208 : * doesn't include alignment padding).
209 : *
210 : * The "length" field should always be rounded up to the next QUEUEALIGN
211 : * multiple so that all fields are properly aligned.
212 : */
213 : typedef struct AsyncQueueEntry
214 : {
215 : int length; /* total allocated length of entry */
216 : Oid dboid; /* sender's database OID */
217 : TransactionId xid; /* sender's XID */
218 : int32 srcPid; /* sender's PID */
219 : char data[NAMEDATALEN + NOTIFY_PAYLOAD_MAX_LENGTH];
220 : } AsyncQueueEntry;
221 :
222 : /* Currently, no field of AsyncQueueEntry requires more than int alignment */
223 : #define QUEUEALIGN(len) INTALIGN(len)
224 :
225 : #define AsyncQueueEntryEmptySize (offsetof(AsyncQueueEntry, data) + 2)
226 :
227 : /*
228 : * Struct describing a queue position, and assorted macros for working with it
229 : */
230 : typedef struct QueuePosition
231 : {
232 : int64 page; /* SLRU page number */
233 : int offset; /* byte offset within page */
234 : } QueuePosition;
235 :
236 : #define QUEUE_POS_PAGE(x) ((x).page)
237 : #define QUEUE_POS_OFFSET(x) ((x).offset)
238 :
239 : #define SET_QUEUE_POS(x,y,z) \
240 : do { \
241 : (x).page = (y); \
242 : (x).offset = (z); \
243 : } while (0)
244 :
245 : #define QUEUE_POS_EQUAL(x,y) \
246 : ((x).page == (y).page && (x).offset == (y).offset)
247 :
248 : #define QUEUE_POS_IS_ZERO(x) \
249 : ((x).page == 0 && (x).offset == 0)
250 :
251 : /* choose logically smaller QueuePosition */
252 : #define QUEUE_POS_MIN(x,y) \
253 : (asyncQueuePagePrecedes((x).page, (y).page) ? (x) : \
254 : (x).page != (y).page ? (y) : \
255 : (x).offset < (y).offset ? (x) : (y))
256 :
257 : /* choose logically larger QueuePosition */
258 : #define QUEUE_POS_MAX(x,y) \
259 : (asyncQueuePagePrecedes((x).page, (y).page) ? (y) : \
260 : (x).page != (y).page ? (x) : \
261 : (x).offset > (y).offset ? (x) : (y))
262 :
263 : /* returns true if x comes before y in queue order */
264 : #define QUEUE_POS_PRECEDES(x,y) \
265 : (asyncQueuePagePrecedes((x).page, (y).page) || \
266 : ((x).page == (y).page && (x).offset < (y).offset))
267 :
268 : /*
269 : * Parameter determining how often we try to advance the tail pointer:
270 : * we do that after every QUEUE_CLEANUP_DELAY pages of NOTIFY data. This is
271 : * also the distance by which a backend that's not interested in our
272 : * notifications needs to be behind before we'll decide we need to wake it
273 : * up so it can advance its pointer.
274 : *
275 : * Resist the temptation to make this really large. While that would save
276 : * work in some places, it would add cost in others. In particular, this
277 : * should likely be less than notify_buffers, to ensure that backends
278 : * catch up before the pages they'll need to read fall out of SLRU cache.
279 : */
280 : #define QUEUE_CLEANUP_DELAY 4
281 :
282 : /*
283 : * Struct describing a listening backend's status
284 : */
285 : typedef struct QueueBackendStatus
286 : {
287 : int32 pid; /* either a PID or InvalidPid */
288 : Oid dboid; /* backend's database OID, or InvalidOid */
289 : ProcNumber nextListener; /* id of next listener, or INVALID_PROC_NUMBER */
290 : QueuePosition pos; /* backend has read queue up to here */
291 : bool wakeupPending; /* signal sent to backend, not yet processed */
292 : bool isAdvancing; /* backend is advancing its position */
293 : } QueueBackendStatus;
294 :
295 : /*
296 : * Shared memory state for LISTEN/NOTIFY (excluding its SLRU stuff)
297 : *
298 : * The AsyncQueueControl structure is protected by the NotifyQueueLock and
299 : * NotifyQueueTailLock.
300 : *
301 : * When holding NotifyQueueLock in SHARED mode, backends may only inspect
302 : * their own entries as well as the head and tail pointers. Consequently we
303 : * can allow a backend to update its own record while holding only SHARED lock
304 : * (since no other backend will inspect it).
305 : *
306 : * When holding NotifyQueueLock in EXCLUSIVE mode, backends can inspect the
307 : * entries of other backends and also change the head pointer. They can
308 : * also advance other backends' queue positions, unless the other backend
309 : * has isAdvancing set (i.e., is in process of doing that itself).
310 : *
311 : * When holding both NotifyQueueLock and NotifyQueueTailLock in EXCLUSIVE
312 : * mode, backends can change the tail pointers.
313 : *
314 : * SLRU buffer pool is divided in banks and bank wise SLRU lock is used as
315 : * the control lock for the pg_notify SLRU buffers.
316 : * In order to avoid deadlocks, whenever we need multiple locks, we first get
317 : * NotifyQueueTailLock, then NotifyQueueLock, then SLRU bank lock, and lastly
318 : * globalChannelTable partition locks.
319 : *
320 : * Each backend uses the backend[] array entry with index equal to its
321 : * ProcNumber. We rely on this to make SendProcSignal fast.
322 : *
323 : * The backend[] array entries for actively-listening backends are threaded
324 : * together using firstListener and the nextListener links, so that we can
325 : * scan them without having to iterate over inactive entries. We keep this
326 : * list in order by ProcNumber so that the scan is cache-friendly when there
327 : * are many active entries.
328 : */
329 : typedef struct AsyncQueueControl
330 : {
331 : QueuePosition head; /* head points to the next free location */
332 : QueuePosition tail; /* tail must be <= the queue position of every
333 : * listening backend */
334 : int64 stopPage; /* oldest unrecycled page; must be <=
335 : * tail.page */
336 : ProcNumber firstListener; /* id of first listener, or
337 : * INVALID_PROC_NUMBER */
338 : TimestampTz lastQueueFillWarn; /* time of last queue-full msg */
339 : dsa_handle globalChannelTableDSA; /* global channel table's DSA handle */
340 : dshash_table_handle globalChannelTableDSH; /* and its dshash handle */
341 : /* Array with room for MaxBackends entries: */
342 : QueueBackendStatus backend[FLEXIBLE_ARRAY_MEMBER];
343 : } AsyncQueueControl;
344 :
345 : static AsyncQueueControl *asyncQueueControl;
346 :
347 : #define QUEUE_HEAD (asyncQueueControl->head)
348 : #define QUEUE_TAIL (asyncQueueControl->tail)
349 : #define QUEUE_STOP_PAGE (asyncQueueControl->stopPage)
350 : #define QUEUE_FIRST_LISTENER (asyncQueueControl->firstListener)
351 : #define QUEUE_BACKEND_PID(i) (asyncQueueControl->backend[i].pid)
352 : #define QUEUE_BACKEND_DBOID(i) (asyncQueueControl->backend[i].dboid)
353 : #define QUEUE_NEXT_LISTENER(i) (asyncQueueControl->backend[i].nextListener)
354 : #define QUEUE_BACKEND_POS(i) (asyncQueueControl->backend[i].pos)
355 : #define QUEUE_BACKEND_WAKEUP_PENDING(i) (asyncQueueControl->backend[i].wakeupPending)
356 : #define QUEUE_BACKEND_IS_ADVANCING(i) (asyncQueueControl->backend[i].isAdvancing)
357 :
358 : /*
359 : * The SLRU buffer area through which we access the notification queue
360 : */
361 : static SlruCtlData NotifyCtlData;
362 :
363 : #define NotifyCtl (&NotifyCtlData)
364 : #define QUEUE_PAGESIZE BLCKSZ
365 :
366 : #define QUEUE_FULL_WARN_INTERVAL 5000 /* warn at most once every 5s */
367 :
368 : /*
369 : * Global channel table definitions
370 : *
371 : * This hash table maps (database OID, channel name) keys to arrays of
372 : * ProcNumbers representing the backends listening or about to listen
373 : * on each channel. The "listening" flags allow us to create hash table
374 : * entries pre-commit and not have to assume that creating them post-commit
375 : * will succeed.
376 : */
377 : #define INITIAL_LISTENERS_ARRAY_SIZE 4
378 :
379 : typedef struct GlobalChannelKey
380 : {
381 : Oid dboid;
382 : char channel[NAMEDATALEN];
383 : } GlobalChannelKey;
384 :
385 : typedef struct ListenerEntry
386 : {
387 : ProcNumber procNo; /* listener's ProcNumber */
388 : bool listening; /* true if committed listener */
389 : } ListenerEntry;
390 :
391 : typedef struct GlobalChannelEntry
392 : {
393 : GlobalChannelKey key; /* hash key */
394 : dsa_pointer listenersArray; /* DSA pointer to ListenerEntry array */
395 : int numListeners; /* Number of listeners currently stored */
396 : int allocatedListeners; /* Allocated size of array */
397 : } GlobalChannelEntry;
398 :
399 : static dshash_table *globalChannelTable = NULL;
400 : static dsa_area *globalChannelDSA = NULL;
401 :
402 : /*
403 : * localChannelTable caches the channel names this backend is listening on
404 : * (including those we have staged to be listened on, but not yet committed).
405 : * Used by IsListeningOn() for fast lookups when reading notifications.
406 : */
407 : static HTAB *localChannelTable = NULL;
408 :
409 : /* We test this condition to detect that we're not listening at all */
410 : #define LocalChannelTableIsEmpty() \
411 : (localChannelTable == NULL || hash_get_num_entries(localChannelTable) == 0)
412 :
413 : /*
414 : * State for pending LISTEN/UNLISTEN actions consists of an ordered list of
415 : * all actions requested in the current transaction. As explained above,
416 : * we don't actually change listen state until we reach transaction commit.
417 : *
418 : * The list is kept in CurTransactionContext. In subtransactions, each
419 : * subtransaction has its own list in its own CurTransactionContext, but
420 : * successful subtransactions attach their lists to their parent's list.
421 : * Failed subtransactions simply discard their lists.
422 : */
423 : typedef enum
424 : {
425 : LISTEN_LISTEN,
426 : LISTEN_UNLISTEN,
427 : LISTEN_UNLISTEN_ALL,
428 : } ListenActionKind;
429 :
430 : typedef struct
431 : {
432 : ListenActionKind action;
433 : char channel[FLEXIBLE_ARRAY_MEMBER]; /* nul-terminated string */
434 : } ListenAction;
435 :
436 : typedef struct ActionList
437 : {
438 : int nestingLevel; /* current transaction nesting depth */
439 : List *actions; /* list of ListenAction structs */
440 : struct ActionList *upper; /* details for upper transaction levels */
441 : } ActionList;
442 :
443 : static ActionList *pendingActions = NULL;
444 :
445 : /*
446 : * Hash table recording the final listen/unlisten intent per channel for
447 : * the current transaction. Key is channel name, value is PENDING_LISTEN or
448 : * PENDING_UNLISTEN. This keeps critical commit/abort processing to one step
449 : * per channel instead of replaying every action. This is built from the
450 : * pendingActions list by PreCommit_Notify, then used by AtCommit_Notify or
451 : * AtAbort_Notify.
452 : */
453 : typedef enum
454 : {
455 : PENDING_LISTEN,
456 : PENDING_UNLISTEN,
457 : } PendingListenAction;
458 :
459 : typedef struct PendingListenEntry
460 : {
461 : char channel[NAMEDATALEN]; /* hash key */
462 : PendingListenAction action; /* which action should we perform? */
463 : } PendingListenEntry;
464 :
465 : static HTAB *pendingListenActions = NULL;
466 :
467 : /*
468 : * State for outbound notifies consists of a list of all channels+payloads
469 : * NOTIFYed in the current transaction. We do not actually perform a NOTIFY
470 : * until and unless the transaction commits. pendingNotifies is NULL if no
471 : * NOTIFYs have been done in the current (sub) transaction.
472 : *
473 : * We discard duplicate notify events issued in the same transaction.
474 : * Hence, in addition to the list proper (which we need to track the order
475 : * of the events, since we guarantee to deliver them in order), we build a
476 : * hash table which we can probe to detect duplicates. Since building the
477 : * hash table is somewhat expensive, we do so only once we have at least
478 : * MIN_HASHABLE_NOTIFIES events queued in the current (sub) transaction;
479 : * before that we just scan the events linearly.
480 : *
481 : * The list is kept in CurTransactionContext. In subtransactions, each
482 : * subtransaction has its own list in its own CurTransactionContext, but
483 : * successful subtransactions add their entries to their parent's list.
484 : * Failed subtransactions simply discard their lists. Since these lists
485 : * are independent, there may be notify events in a subtransaction's list
486 : * that duplicate events in some ancestor (sub) transaction; we get rid of
487 : * the dups when merging the subtransaction's list into its parent's.
488 : *
489 : * Note: the action and notify lists do not interact within a transaction.
490 : * In particular, if a transaction does NOTIFY and then LISTEN on the same
491 : * condition name, it will get a self-notify at commit. This is a bit odd
492 : * but is consistent with our historical behavior.
493 : */
494 : typedef struct Notification
495 : {
496 : uint16 channel_len; /* length of channel-name string */
497 : uint16 payload_len; /* length of payload string */
498 : /* null-terminated channel name, then null-terminated payload follow */
499 : char data[FLEXIBLE_ARRAY_MEMBER];
500 : } Notification;
501 :
502 : typedef struct NotificationList
503 : {
504 : int nestingLevel; /* current transaction nesting depth */
505 : List *events; /* list of Notification structs */
506 : HTAB *hashtab; /* hash of NotificationHash structs, or NULL */
507 : List *uniqueChannelNames; /* unique channel names being notified */
508 : HTAB *uniqueChannelHash; /* hash of unique channel names, or NULL */
509 : struct NotificationList *upper; /* details for upper transaction levels */
510 : } NotificationList;
511 :
512 : #define MIN_HASHABLE_NOTIFIES 16 /* threshold to build hashtab */
513 :
514 : struct NotificationHash
515 : {
516 : Notification *event; /* => the actual Notification struct */
517 : };
518 :
519 : static NotificationList *pendingNotifies = NULL;
520 :
521 : /*
522 : * Hash entry in NotificationList.uniqueChannelHash or localChannelTable
523 : * (both just carry the channel name, with no payload).
524 : */
525 : typedef struct ChannelName
526 : {
527 : char channel[NAMEDATALEN]; /* hash key */
528 : } ChannelName;
529 :
530 : /*
531 : * Inbound notifications are initially processed by HandleNotifyInterrupt(),
532 : * called from inside a signal handler. That just sets the
533 : * notifyInterruptPending flag and sets the process
534 : * latch. ProcessNotifyInterrupt() will then be called whenever it's safe to
535 : * actually deal with the interrupt.
536 : */
537 : volatile sig_atomic_t notifyInterruptPending = false;
538 :
539 : /* True if we've registered an on_shmem_exit cleanup */
540 : static bool unlistenExitRegistered = false;
541 :
542 : /* True if we're currently registered as a listener in asyncQueueControl */
543 : static bool amRegisteredListener = false;
544 :
545 : /*
546 : * Queue head positions for direct advancement.
547 : * These are captured during PreCommit_Notify while holding the heavyweight
548 : * lock on database 0, ensuring no other backend can insert notifications
549 : * between them. SignalBackends uses these to advance idle backends.
550 : */
551 : static QueuePosition queueHeadBeforeWrite;
552 : static QueuePosition queueHeadAfterWrite;
553 :
554 : /*
555 : * Workspace arrays for SignalBackends. These are preallocated in
556 : * PreCommit_Notify to avoid needing memory allocation after committing to
557 : * clog.
558 : */
559 : static int32 *signalPids = NULL;
560 : static ProcNumber *signalProcnos = NULL;
561 :
562 : /* have we advanced to a page that's a multiple of QUEUE_CLEANUP_DELAY? */
563 : static bool tryAdvanceTail = false;
564 :
565 : /* GUC parameters */
566 : bool Trace_notify = false;
567 :
568 : /* For 8 KB pages this gives 8 GB of disk space */
569 : int max_notify_queue_pages = 1048576;
570 :
571 : /* local function prototypes */
572 : static inline int64 asyncQueuePageDiff(int64 p, int64 q);
573 : static inline bool asyncQueuePagePrecedes(int64 p, int64 q);
574 : static inline void GlobalChannelKeyInit(GlobalChannelKey *key, Oid dboid,
575 : const char *channel);
576 : static dshash_hash globalChannelTableHash(const void *key, size_t size,
577 : void *arg);
578 : static void initGlobalChannelTable(void);
579 : static void initLocalChannelTable(void);
580 : static void queue_listen(ListenActionKind action, const char *channel);
581 : static void Async_UnlistenOnExit(int code, Datum arg);
582 : static void BecomeRegisteredListener(void);
583 : static void PrepareTableEntriesForListen(const char *channel);
584 : static void PrepareTableEntriesForUnlisten(const char *channel);
585 : static void PrepareTableEntriesForUnlistenAll(void);
586 : static void RemoveListenerFromChannel(GlobalChannelEntry **entry_ptr,
587 : ListenerEntry *listeners,
588 : int idx);
589 : static void ApplyPendingListenActions(bool isCommit);
590 : static void CleanupListenersOnExit(void);
591 : static bool IsListeningOn(const char *channel);
592 : static void asyncQueueUnregister(void);
593 : static bool asyncQueueIsFull(void);
594 : static bool asyncQueueAdvance(volatile QueuePosition *position, int entryLength);
595 : static void asyncQueueNotificationToEntry(Notification *n, AsyncQueueEntry *qe);
596 : static ListCell *asyncQueueAddEntries(ListCell *nextNotify);
597 : static double asyncQueueUsage(void);
598 : static void asyncQueueFillWarning(void);
599 : static void SignalBackends(void);
600 : static void asyncQueueReadAllNotifications(void);
601 : static bool asyncQueueProcessPageEntries(QueuePosition *current,
602 : QueuePosition stop,
603 : Snapshot snapshot);
604 : static void asyncQueueAdvanceTail(void);
605 : static void ProcessIncomingNotify(bool flush);
606 : static bool AsyncExistsPendingNotify(Notification *n);
607 : static void AddEventToPendingNotifies(Notification *n);
608 : static uint32 notification_hash(const void *key, Size keysize);
609 : static int notification_match(const void *key1, const void *key2, Size keysize);
610 : static void ClearPendingActionsAndNotifies(void);
611 :
612 : /*
613 : * Compute the difference between two queue page numbers.
614 : * Previously this function accounted for a wraparound.
615 : */
616 : static inline int64
617 0 : asyncQueuePageDiff(int64 p, int64 q)
618 : {
619 0 : return p - q;
620 : }
621 :
622 : /*
623 : * Determines whether p precedes q.
624 : * Previously this function accounted for a wraparound.
625 : */
626 : static inline bool
627 288 : asyncQueuePagePrecedes(int64 p, int64 q)
628 : {
629 288 : return p < q;
630 : }
631 :
632 : /*
633 : * GlobalChannelKeyInit
634 : * Prepare a global channel table key for hashing.
635 : */
636 : static inline void
637 432 : GlobalChannelKeyInit(GlobalChannelKey *key, Oid dboid, const char *channel)
638 : {
639 432 : memset(key, 0, sizeof(GlobalChannelKey));
640 432 : key->dboid = dboid;
641 432 : strlcpy(key->channel, channel, NAMEDATALEN);
642 432 : }
643 :
644 : /*
645 : * globalChannelTableHash
646 : * Hash function for global channel table keys.
647 : */
648 : static dshash_hash
649 432 : globalChannelTableHash(const void *key, size_t size, void *arg)
650 : {
651 432 : const GlobalChannelKey *k = (const GlobalChannelKey *) key;
652 : dshash_hash h;
653 :
654 432 : h = DatumGetUInt32(hash_uint32(k->dboid));
655 432 : h ^= DatumGetUInt32(hash_any((const unsigned char *) k->channel,
656 432 : strnlen(k->channel, NAMEDATALEN)));
657 :
658 432 : return h;
659 : }
660 :
661 : /* parameters for the global channel table */
662 : static const dshash_parameters globalChannelTableDSHParams = {
663 : sizeof(GlobalChannelKey),
664 : sizeof(GlobalChannelEntry),
665 : dshash_memcmp,
666 : globalChannelTableHash,
667 : dshash_memcpy,
668 : LWTRANCHE_NOTIFY_CHANNEL_HASH
669 : };
670 :
671 : /*
672 : * initGlobalChannelTable
673 : * Lazy initialization of the global channel table.
674 : */
675 : static void
676 312 : initGlobalChannelTable(void)
677 : {
678 : MemoryContext oldcontext;
679 :
680 : /* Quick exit if we already did this */
681 312 : if (asyncQueueControl->globalChannelTableDSH != DSHASH_HANDLE_INVALID &&
682 298 : globalChannelTable != NULL)
683 254 : return;
684 :
685 : /* Otherwise, use a lock to ensure only one process creates the table */
686 58 : LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
687 :
688 : /* Be sure any local memory allocated by DSA routines is persistent */
689 58 : oldcontext = MemoryContextSwitchTo(TopMemoryContext);
690 :
691 58 : if (asyncQueueControl->globalChannelTableDSH == DSHASH_HANDLE_INVALID)
692 : {
693 : /* Initialize dynamic shared hash table for global channels */
694 14 : globalChannelDSA = dsa_create(LWTRANCHE_NOTIFY_CHANNEL_HASH);
695 14 : dsa_pin(globalChannelDSA);
696 14 : dsa_pin_mapping(globalChannelDSA);
697 14 : globalChannelTable = dshash_create(globalChannelDSA,
698 : &globalChannelTableDSHParams,
699 : NULL);
700 :
701 : /* Store handles in shared memory for other backends to use */
702 14 : asyncQueueControl->globalChannelTableDSA = dsa_get_handle(globalChannelDSA);
703 14 : asyncQueueControl->globalChannelTableDSH =
704 14 : dshash_get_hash_table_handle(globalChannelTable);
705 : }
706 44 : else if (!globalChannelTable)
707 : {
708 : /* Attach to existing dynamic shared hash table */
709 44 : globalChannelDSA = dsa_attach(asyncQueueControl->globalChannelTableDSA);
710 44 : dsa_pin_mapping(globalChannelDSA);
711 44 : globalChannelTable = dshash_attach(globalChannelDSA,
712 : &globalChannelTableDSHParams,
713 44 : asyncQueueControl->globalChannelTableDSH,
714 : NULL);
715 : }
716 :
717 58 : MemoryContextSwitchTo(oldcontext);
718 58 : LWLockRelease(NotifyQueueLock);
719 : }
720 :
721 : /*
722 : * initLocalChannelTable
723 : * Lazy initialization of the local channel table.
724 : * Once created, this table lasts for the life of the session.
725 : */
726 : static void
727 182 : initLocalChannelTable(void)
728 : {
729 : HASHCTL hash_ctl;
730 :
731 : /* Quick exit if we already did this */
732 182 : if (localChannelTable != NULL)
733 148 : return;
734 :
735 : /* Initialize local hash table for this backend's listened channels */
736 34 : hash_ctl.keysize = NAMEDATALEN;
737 34 : hash_ctl.entrysize = sizeof(ChannelName);
738 :
739 34 : localChannelTable =
740 34 : hash_create("Local Listen Channels",
741 : 64,
742 : &hash_ctl,
743 : HASH_ELEM | HASH_STRINGS);
744 : }
745 :
746 : /*
747 : * initPendingListenActions
748 : * Lazy initialization of the pending listen actions hash table.
749 : * This is allocated in CurTransactionContext during PreCommit_Notify,
750 : * and destroyed at transaction end.
751 : */
752 : static void
753 182 : initPendingListenActions(void)
754 : {
755 : HASHCTL hash_ctl;
756 :
757 182 : if (pendingListenActions != NULL)
758 0 : return;
759 :
760 182 : hash_ctl.keysize = NAMEDATALEN;
761 182 : hash_ctl.entrysize = sizeof(PendingListenEntry);
762 182 : hash_ctl.hcxt = CurTransactionContext;
763 :
764 182 : pendingListenActions =
765 182 : hash_create("Pending Listen Actions",
766 182 : list_length(pendingActions->actions),
767 : &hash_ctl,
768 : HASH_ELEM | HASH_STRINGS | HASH_CONTEXT);
769 : }
770 :
771 : /*
772 : * Report space needed for our shared memory area
773 : */
774 : Size
775 4254 : AsyncShmemSize(void)
776 : {
777 : Size size;
778 :
779 : /* This had better match AsyncShmemInit */
780 4254 : size = mul_size(MaxBackends, sizeof(QueueBackendStatus));
781 4254 : size = add_size(size, offsetof(AsyncQueueControl, backend));
782 :
783 4254 : size = add_size(size, SimpleLruShmemSize(notify_buffers, 0));
784 :
785 4254 : return size;
786 : }
787 :
788 : /*
789 : * Initialize our shared memory area
790 : */
791 : void
792 2280 : AsyncShmemInit(void)
793 : {
794 : bool found;
795 : Size size;
796 :
797 : /*
798 : * Create or attach to the AsyncQueueControl structure.
799 : */
800 2280 : size = mul_size(MaxBackends, sizeof(QueueBackendStatus));
801 2280 : size = add_size(size, offsetof(AsyncQueueControl, backend));
802 :
803 2280 : asyncQueueControl = (AsyncQueueControl *)
804 2280 : ShmemInitStruct("Async Queue Control", size, &found);
805 :
806 2280 : if (!found)
807 : {
808 : /* First time through, so initialize it */
809 2280 : SET_QUEUE_POS(QUEUE_HEAD, 0, 0);
810 2280 : SET_QUEUE_POS(QUEUE_TAIL, 0, 0);
811 2280 : QUEUE_STOP_PAGE = 0;
812 2280 : QUEUE_FIRST_LISTENER = INVALID_PROC_NUMBER;
813 2280 : asyncQueueControl->lastQueueFillWarn = 0;
814 2280 : asyncQueueControl->globalChannelTableDSA = DSA_HANDLE_INVALID;
815 2280 : asyncQueueControl->globalChannelTableDSH = DSHASH_HANDLE_INVALID;
816 211480 : for (int i = 0; i < MaxBackends; i++)
817 : {
818 209200 : QUEUE_BACKEND_PID(i) = InvalidPid;
819 209200 : QUEUE_BACKEND_DBOID(i) = InvalidOid;
820 209200 : QUEUE_NEXT_LISTENER(i) = INVALID_PROC_NUMBER;
821 209200 : SET_QUEUE_POS(QUEUE_BACKEND_POS(i), 0, 0);
822 209200 : QUEUE_BACKEND_WAKEUP_PENDING(i) = false;
823 209200 : QUEUE_BACKEND_IS_ADVANCING(i) = false;
824 : }
825 : }
826 :
827 : /*
828 : * Set up SLRU management of the pg_notify data. Note that long segment
829 : * names are used in order to avoid wraparound.
830 : */
831 2280 : NotifyCtl->PagePrecedes = asyncQueuePagePrecedes;
832 2280 : SimpleLruInit(NotifyCtl, "notify", notify_buffers, 0,
833 : "pg_notify", LWTRANCHE_NOTIFY_BUFFER, LWTRANCHE_NOTIFY_SLRU,
834 : SYNC_HANDLER_NONE, true);
835 :
836 2280 : if (!found)
837 : {
838 : /*
839 : * During start or reboot, clean out the pg_notify directory.
840 : */
841 2280 : (void) SlruScanDirectory(NotifyCtl, SlruScanDirCbDeleteAll, NULL);
842 : }
843 2280 : }
844 :
845 :
846 : /*
847 : * pg_notify -
848 : * SQL function to send a notification event
849 : */
850 : Datum
851 2144 : pg_notify(PG_FUNCTION_ARGS)
852 : {
853 : const char *channel;
854 : const char *payload;
855 :
856 2144 : if (PG_ARGISNULL(0))
857 6 : channel = "";
858 : else
859 2138 : channel = text_to_cstring(PG_GETARG_TEXT_PP(0));
860 :
861 2144 : if (PG_ARGISNULL(1))
862 12 : payload = "";
863 : else
864 2132 : payload = text_to_cstring(PG_GETARG_TEXT_PP(1));
865 :
866 : /* For NOTIFY as a statement, this is checked in ProcessUtility */
867 2144 : PreventCommandDuringRecovery("NOTIFY");
868 :
869 2144 : Async_Notify(channel, payload);
870 :
871 2126 : PG_RETURN_VOID();
872 : }
873 :
874 :
875 : /*
876 : * Async_Notify
877 : *
878 : * This is executed by the SQL notify command.
879 : *
880 : * Adds the message to the list of pending notifies.
881 : * Actual notification happens during transaction commit.
882 : * ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
883 : */
884 : void
885 2276 : Async_Notify(const char *channel, const char *payload)
886 : {
887 2276 : int my_level = GetCurrentTransactionNestLevel();
888 : size_t channel_len;
889 : size_t payload_len;
890 : Notification *n;
891 : MemoryContext oldcontext;
892 :
893 2276 : if (IsParallelWorker())
894 0 : elog(ERROR, "cannot send notifications from a parallel worker");
895 :
896 2276 : if (Trace_notify)
897 0 : elog(DEBUG1, "Async_Notify(%s)", channel);
898 :
899 2276 : channel_len = channel ? strlen(channel) : 0;
900 2276 : payload_len = payload ? strlen(payload) : 0;
901 :
902 : /* a channel name must be specified */
903 2276 : if (channel_len == 0)
904 12 : ereport(ERROR,
905 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
906 : errmsg("channel name cannot be empty")));
907 :
908 : /* enforce length limits */
909 2264 : if (channel_len >= NAMEDATALEN)
910 6 : ereport(ERROR,
911 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
912 : errmsg("channel name too long")));
913 :
914 2258 : if (payload_len >= NOTIFY_PAYLOAD_MAX_LENGTH)
915 0 : ereport(ERROR,
916 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
917 : errmsg("payload string too long")));
918 :
919 : /*
920 : * We must construct the Notification entry, even if we end up not using
921 : * it, in order to compare it cheaply to existing list entries.
922 : *
923 : * The notification list needs to live until end of transaction, so store
924 : * it in the transaction context.
925 : */
926 2258 : oldcontext = MemoryContextSwitchTo(CurTransactionContext);
927 :
928 2258 : n = (Notification *) palloc(offsetof(Notification, data) +
929 2258 : channel_len + payload_len + 2);
930 2258 : n->channel_len = channel_len;
931 2258 : n->payload_len = payload_len;
932 2258 : strcpy(n->data, channel);
933 2258 : if (payload)
934 2230 : strcpy(n->data + channel_len + 1, payload);
935 : else
936 28 : n->data[channel_len + 1] = '\0';
937 :
938 2258 : if (pendingNotifies == NULL || my_level > pendingNotifies->nestingLevel)
939 134 : {
940 : NotificationList *notifies;
941 :
942 : /*
943 : * First notify event in current (sub)xact. Note that we allocate the
944 : * NotificationList in TopTransactionContext; the nestingLevel might
945 : * get changed later by AtSubCommit_Notify.
946 : */
947 : notifies = (NotificationList *)
948 134 : MemoryContextAlloc(TopTransactionContext,
949 : sizeof(NotificationList));
950 134 : notifies->nestingLevel = my_level;
951 134 : notifies->events = list_make1(n);
952 : /* We certainly don't need a hashtable yet */
953 134 : notifies->hashtab = NULL;
954 : /* We won't build uniqueChannelNames/Hash till later, either */
955 134 : notifies->uniqueChannelNames = NIL;
956 134 : notifies->uniqueChannelHash = NULL;
957 134 : notifies->upper = pendingNotifies;
958 134 : pendingNotifies = notifies;
959 : }
960 : else
961 : {
962 : /* Now check for duplicates */
963 2124 : if (AsyncExistsPendingNotify(n))
964 : {
965 : /* It's a dup, so forget it */
966 26 : pfree(n);
967 26 : MemoryContextSwitchTo(oldcontext);
968 26 : return;
969 : }
970 :
971 : /* Append more events to existing list */
972 2098 : AddEventToPendingNotifies(n);
973 : }
974 :
975 2232 : MemoryContextSwitchTo(oldcontext);
976 : }
977 :
978 : /*
979 : * queue_listen
980 : * Common code for listen, unlisten, unlisten all commands.
981 : *
982 : * Adds the request to the list of pending actions.
983 : * Actual update of localChannelTable and globalChannelTable happens during
984 : * PreCommit_Notify, with staged changes committed in AtCommit_Notify.
985 : */
986 : static void
987 220 : queue_listen(ListenActionKind action, const char *channel)
988 : {
989 : MemoryContext oldcontext;
990 : ListenAction *actrec;
991 220 : int my_level = GetCurrentTransactionNestLevel();
992 :
993 : /*
994 : * Unlike Async_Notify, we don't try to collapse out duplicates here. We
995 : * keep the ordered list to preserve interactions like UNLISTEN ALL; the
996 : * final per-channel intent is computed during PreCommit_Notify.
997 : */
998 220 : oldcontext = MemoryContextSwitchTo(CurTransactionContext);
999 :
1000 : /* space for terminating null is included in sizeof(ListenAction) */
1001 220 : actrec = (ListenAction *) palloc(offsetof(ListenAction, channel) +
1002 220 : strlen(channel) + 1);
1003 220 : actrec->action = action;
1004 220 : strcpy(actrec->channel, channel);
1005 :
1006 220 : if (pendingActions == NULL || my_level > pendingActions->nestingLevel)
1007 188 : {
1008 : ActionList *actions;
1009 :
1010 : /*
1011 : * First action in current sub(xact). Note that we allocate the
1012 : * ActionList in TopTransactionContext; the nestingLevel might get
1013 : * changed later by AtSubCommit_Notify.
1014 : */
1015 : actions = (ActionList *)
1016 188 : MemoryContextAlloc(TopTransactionContext, sizeof(ActionList));
1017 188 : actions->nestingLevel = my_level;
1018 188 : actions->actions = list_make1(actrec);
1019 188 : actions->upper = pendingActions;
1020 188 : pendingActions = actions;
1021 : }
1022 : else
1023 32 : pendingActions->actions = lappend(pendingActions->actions, actrec);
1024 :
1025 220 : MemoryContextSwitchTo(oldcontext);
1026 220 : }
1027 :
1028 : /*
1029 : * Async_Listen
1030 : *
1031 : * This is executed by the SQL listen command.
1032 : */
1033 : void
1034 116 : Async_Listen(const char *channel)
1035 : {
1036 116 : if (Trace_notify)
1037 0 : elog(DEBUG1, "Async_Listen(%s,%d)", channel, MyProcPid);
1038 :
1039 116 : queue_listen(LISTEN_LISTEN, channel);
1040 116 : }
1041 :
1042 : /*
1043 : * Async_Unlisten
1044 : *
1045 : * This is executed by the SQL unlisten command.
1046 : */
1047 : void
1048 6 : Async_Unlisten(const char *channel)
1049 : {
1050 6 : if (Trace_notify)
1051 0 : elog(DEBUG1, "Async_Unlisten(%s,%d)", channel, MyProcPid);
1052 :
1053 : /* If we couldn't possibly be listening, no need to queue anything */
1054 6 : if (pendingActions == NULL && !unlistenExitRegistered)
1055 0 : return;
1056 :
1057 6 : queue_listen(LISTEN_UNLISTEN, channel);
1058 : }
1059 :
1060 : /*
1061 : * Async_UnlistenAll
1062 : *
1063 : * This is invoked by UNLISTEN * command, and also at backend exit.
1064 : */
1065 : void
1066 156 : Async_UnlistenAll(void)
1067 : {
1068 156 : if (Trace_notify)
1069 0 : elog(DEBUG1, "Async_UnlistenAll(%d)", MyProcPid);
1070 :
1071 : /* If we couldn't possibly be listening, no need to queue anything */
1072 156 : if (pendingActions == NULL && !unlistenExitRegistered)
1073 58 : return;
1074 :
1075 98 : queue_listen(LISTEN_UNLISTEN_ALL, "");
1076 : }
1077 :
1078 : /*
1079 : * SQL function: return a set of the channel names this backend is actively
1080 : * listening to.
1081 : *
1082 : * Note: this coding relies on the fact that the localChannelTable cannot
1083 : * change within a transaction.
1084 : */
1085 : Datum
1086 18 : pg_listening_channels(PG_FUNCTION_ARGS)
1087 : {
1088 : FuncCallContext *funcctx;
1089 : HASH_SEQ_STATUS *status;
1090 :
1091 : /* stuff done only on the first call of the function */
1092 18 : if (SRF_IS_FIRSTCALL())
1093 : {
1094 : /* create a function context for cross-call persistence */
1095 12 : funcctx = SRF_FIRSTCALL_INIT();
1096 :
1097 : /* Initialize hash table iteration if we have any channels */
1098 12 : if (localChannelTable != NULL)
1099 : {
1100 : MemoryContext oldcontext;
1101 :
1102 12 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
1103 12 : status = (HASH_SEQ_STATUS *) palloc(sizeof(HASH_SEQ_STATUS));
1104 12 : hash_seq_init(status, localChannelTable);
1105 12 : funcctx->user_fctx = status;
1106 12 : MemoryContextSwitchTo(oldcontext);
1107 : }
1108 : else
1109 : {
1110 0 : funcctx->user_fctx = NULL;
1111 : }
1112 : }
1113 :
1114 : /* stuff done on every call of the function */
1115 18 : funcctx = SRF_PERCALL_SETUP();
1116 18 : status = (HASH_SEQ_STATUS *) funcctx->user_fctx;
1117 :
1118 18 : if (status != NULL)
1119 : {
1120 : ChannelName *entry;
1121 :
1122 18 : entry = (ChannelName *) hash_seq_search(status);
1123 18 : if (entry != NULL)
1124 6 : SRF_RETURN_NEXT(funcctx, CStringGetTextDatum(entry->channel));
1125 : }
1126 :
1127 12 : SRF_RETURN_DONE(funcctx);
1128 : }
1129 :
1130 : /*
1131 : * Async_UnlistenOnExit
1132 : *
1133 : * This is executed at backend exit if we have done any LISTENs in this
1134 : * backend. It might not be necessary anymore, if the user UNLISTENed
1135 : * everything, but we don't try to detect that case.
1136 : */
1137 : static void
1138 34 : Async_UnlistenOnExit(int code, Datum arg)
1139 : {
1140 34 : CleanupListenersOnExit();
1141 34 : asyncQueueUnregister();
1142 34 : }
1143 :
1144 : /*
1145 : * AtPrepare_Notify
1146 : *
1147 : * This is called at the prepare phase of a two-phase
1148 : * transaction. Save the state for possible commit later.
1149 : */
1150 : void
1151 592 : AtPrepare_Notify(void)
1152 : {
1153 : /* It's not allowed to have any pending LISTEN/UNLISTEN/NOTIFY actions */
1154 592 : if (pendingActions || pendingNotifies)
1155 0 : ereport(ERROR,
1156 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1157 : errmsg("cannot PREPARE a transaction that has executed LISTEN, UNLISTEN, or NOTIFY")));
1158 592 : }
1159 :
1160 : /*
1161 : * PreCommit_Notify
1162 : *
1163 : * This is called at transaction commit, before actually committing to
1164 : * clog.
1165 : *
1166 : * If there are pending LISTEN actions, make sure we are listed in the
1167 : * shared-memory listener array. This must happen before commit to
1168 : * ensure we don't miss any notifies from transactions that commit
1169 : * just after ours.
1170 : *
1171 : * If there are outbound notify requests in the pendingNotifies list,
1172 : * add them to the global queue. We do that before commit so that
1173 : * we can still throw error if we run out of queue space.
1174 : */
1175 : void
1176 951058 : PreCommit_Notify(void)
1177 : {
1178 : ListCell *p;
1179 :
1180 951058 : if (!pendingActions && !pendingNotifies)
1181 950746 : return; /* no relevant statements in this xact */
1182 :
1183 312 : if (Trace_notify)
1184 0 : elog(DEBUG1, "PreCommit_Notify");
1185 :
1186 : /* Preflight for any pending listen/unlisten actions */
1187 312 : initGlobalChannelTable();
1188 :
1189 312 : if (pendingActions != NULL)
1190 : {
1191 : /* Ensure we have a local channel table */
1192 182 : initLocalChannelTable();
1193 : /* Create pendingListenActions hash table for this transaction */
1194 182 : initPendingListenActions();
1195 :
1196 : /* Stage all the actions this transaction wants to perform */
1197 396 : foreach(p, pendingActions->actions)
1198 : {
1199 214 : ListenAction *actrec = (ListenAction *) lfirst(p);
1200 :
1201 214 : switch (actrec->action)
1202 : {
1203 112 : case LISTEN_LISTEN:
1204 112 : BecomeRegisteredListener();
1205 112 : PrepareTableEntriesForListen(actrec->channel);
1206 112 : break;
1207 6 : case LISTEN_UNLISTEN:
1208 6 : PrepareTableEntriesForUnlisten(actrec->channel);
1209 6 : break;
1210 96 : case LISTEN_UNLISTEN_ALL:
1211 96 : PrepareTableEntriesForUnlistenAll();
1212 96 : break;
1213 : }
1214 : }
1215 : }
1216 :
1217 : /* Queue any pending notifies (must happen after the above) */
1218 312 : if (pendingNotifies)
1219 : {
1220 : ListCell *nextNotify;
1221 130 : bool firstIteration = true;
1222 :
1223 : /*
1224 : * Build list of unique channel names being notified for use by
1225 : * SignalBackends().
1226 : *
1227 : * If uniqueChannelHash is available, use it to efficiently get the
1228 : * unique channels. Otherwise, fall back to the O(N^2) approach.
1229 : */
1230 130 : pendingNotifies->uniqueChannelNames = NIL;
1231 130 : if (pendingNotifies->uniqueChannelHash != NULL)
1232 : {
1233 : HASH_SEQ_STATUS status;
1234 : ChannelName *channelEntry;
1235 :
1236 4 : hash_seq_init(&status, pendingNotifies->uniqueChannelHash);
1237 8 : while ((channelEntry = (ChannelName *) hash_seq_search(&status)) != NULL)
1238 4 : pendingNotifies->uniqueChannelNames =
1239 4 : lappend(pendingNotifies->uniqueChannelNames,
1240 4 : channelEntry->channel);
1241 : }
1242 : else
1243 : {
1244 : /* O(N^2) approach is better for small number of notifications */
1245 438 : foreach_ptr(Notification, n, pendingNotifies->events)
1246 : {
1247 186 : char *channel = n->data;
1248 186 : bool found = false;
1249 :
1250 : /* Name present in list? */
1251 380 : foreach_ptr(char, oldchan, pendingNotifies->uniqueChannelNames)
1252 : {
1253 62 : if (strcmp(oldchan, channel) == 0)
1254 : {
1255 54 : found = true;
1256 54 : break;
1257 : }
1258 : }
1259 : /* Add if not already in list */
1260 186 : if (!found)
1261 132 : pendingNotifies->uniqueChannelNames =
1262 132 : lappend(pendingNotifies->uniqueChannelNames,
1263 : channel);
1264 : }
1265 : }
1266 :
1267 : /* Preallocate workspace that will be needed by SignalBackends() */
1268 130 : if (signalPids == NULL)
1269 44 : signalPids = MemoryContextAlloc(TopMemoryContext,
1270 : MaxBackends * sizeof(int32));
1271 :
1272 130 : if (signalProcnos == NULL)
1273 44 : signalProcnos = MemoryContextAlloc(TopMemoryContext,
1274 : MaxBackends * sizeof(ProcNumber));
1275 :
1276 : /*
1277 : * Make sure that we have an XID assigned to the current transaction.
1278 : * GetCurrentTransactionId is cheap if we already have an XID, but not
1279 : * so cheap if we don't, and we'd prefer not to do that work while
1280 : * holding NotifyQueueLock.
1281 : */
1282 130 : (void) GetCurrentTransactionId();
1283 :
1284 : /*
1285 : * Serialize writers by acquiring a special lock that we hold till
1286 : * after commit. This ensures that queue entries appear in commit
1287 : * order, and in particular that there are never uncommitted queue
1288 : * entries ahead of committed ones, so an uncommitted transaction
1289 : * can't block delivery of deliverable notifications.
1290 : *
1291 : * We use a heavyweight lock so that it'll automatically be released
1292 : * after either commit or abort. This also allows deadlocks to be
1293 : * detected, though really a deadlock shouldn't be possible here.
1294 : *
1295 : * The lock is on "database 0", which is pretty ugly but it doesn't
1296 : * seem worth inventing a special locktag category just for this.
1297 : * (Historical note: before PG 9.0, a similar lock on "database 0" was
1298 : * used by the flatfiles mechanism.)
1299 : */
1300 130 : LockSharedObject(DatabaseRelationId, InvalidOid, 0,
1301 : AccessExclusiveLock);
1302 :
1303 : /*
1304 : * For the direct advancement optimization in SignalBackends(), we
1305 : * need to ensure that no other backend can insert queue entries
1306 : * between queueHeadBeforeWrite and queueHeadAfterWrite. The
1307 : * heavyweight lock above provides this guarantee, since it serializes
1308 : * all writers.
1309 : *
1310 : * Note: if the heavyweight lock were ever removed for scalability
1311 : * reasons, we could achieve the same guarantee by holding
1312 : * NotifyQueueLock in EXCLUSIVE mode across all our insertions, rather
1313 : * than releasing and reacquiring it for each page as we do below.
1314 : */
1315 :
1316 : /* Initialize values to a safe default in case list is empty */
1317 130 : SET_QUEUE_POS(queueHeadBeforeWrite, 0, 0);
1318 130 : SET_QUEUE_POS(queueHeadAfterWrite, 0, 0);
1319 :
1320 : /* Now push the notifications into the queue */
1321 130 : nextNotify = list_head(pendingNotifies->events);
1322 330 : while (nextNotify != NULL)
1323 : {
1324 : /*
1325 : * Add the pending notifications to the queue. We acquire and
1326 : * release NotifyQueueLock once per page, which might be overkill
1327 : * but it does allow readers to get in while we're doing this.
1328 : *
1329 : * A full queue is very uncommon and should really not happen,
1330 : * given that we have so much space available in the SLRU pages.
1331 : * Nevertheless we need to deal with this possibility. Note that
1332 : * when we get here we are in the process of committing our
1333 : * transaction, but we have not yet committed to clog, so at this
1334 : * point in time we can still roll the transaction back.
1335 : */
1336 200 : LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
1337 200 : if (firstIteration)
1338 : {
1339 130 : queueHeadBeforeWrite = QUEUE_HEAD;
1340 130 : firstIteration = false;
1341 : }
1342 200 : asyncQueueFillWarning();
1343 200 : if (asyncQueueIsFull())
1344 0 : ereport(ERROR,
1345 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1346 : errmsg("too many notifications in the NOTIFY queue")));
1347 200 : nextNotify = asyncQueueAddEntries(nextNotify);
1348 200 : queueHeadAfterWrite = QUEUE_HEAD;
1349 200 : LWLockRelease(NotifyQueueLock);
1350 : }
1351 :
1352 : /* Note that we don't clear pendingNotifies; AtCommit_Notify will. */
1353 : }
1354 : }
1355 :
1356 : /*
1357 : * AtCommit_Notify
1358 : *
1359 : * This is called at transaction commit, after committing to clog.
1360 : *
1361 : * Apply pending listen/unlisten changes and clear transaction-local state.
1362 : *
1363 : * If we issued any notifications in the transaction, send signals to
1364 : * listening backends (possibly including ourselves) to process them.
1365 : * Also, if we filled enough queue pages with new notifies, try to
1366 : * advance the queue tail pointer.
1367 : */
1368 : void
1369 950748 : AtCommit_Notify(void)
1370 : {
1371 : /*
1372 : * Allow transactions that have not executed LISTEN/UNLISTEN/NOTIFY to
1373 : * return as soon as possible
1374 : */
1375 950748 : if (!pendingActions && !pendingNotifies)
1376 950436 : return;
1377 :
1378 312 : if (Trace_notify)
1379 0 : elog(DEBUG1, "AtCommit_Notify");
1380 :
1381 : /* Apply staged listen/unlisten changes */
1382 312 : ApplyPendingListenActions(true);
1383 :
1384 : /* If no longer listening to anything, get out of listener array */
1385 312 : if (amRegisteredListener && LocalChannelTableIsEmpty())
1386 46 : asyncQueueUnregister();
1387 :
1388 : /*
1389 : * Send signals to listening backends. We need do this only if there are
1390 : * pending notifies, which were previously added to the shared queue by
1391 : * PreCommit_Notify().
1392 : */
1393 312 : if (pendingNotifies != NULL)
1394 130 : SignalBackends();
1395 :
1396 : /*
1397 : * If it's time to try to advance the global tail pointer, do that.
1398 : *
1399 : * (It might seem odd to do this in the sender, when more than likely the
1400 : * listeners won't yet have read the messages we just sent. However,
1401 : * there's less contention if only the sender does it, and there is little
1402 : * need for urgency in advancing the global tail. So this typically will
1403 : * be clearing out messages that were sent some time ago.)
1404 : */
1405 312 : if (tryAdvanceTail)
1406 : {
1407 16 : tryAdvanceTail = false;
1408 16 : asyncQueueAdvanceTail();
1409 : }
1410 :
1411 : /* And clean up */
1412 312 : ClearPendingActionsAndNotifies();
1413 : }
1414 :
1415 : /*
1416 : * BecomeRegisteredListener --- subroutine for PreCommit_Notify
1417 : *
1418 : * This function must make sure we are ready to catch any incoming messages.
1419 : */
1420 : static void
1421 112 : BecomeRegisteredListener(void)
1422 : {
1423 : QueuePosition head;
1424 : QueuePosition max;
1425 : ProcNumber prevListener;
1426 :
1427 : /*
1428 : * Nothing to do if we are already listening to something, nor if we
1429 : * already ran this routine in this transaction.
1430 : */
1431 112 : if (amRegisteredListener)
1432 54 : return;
1433 :
1434 58 : if (Trace_notify)
1435 0 : elog(DEBUG1, "BecomeRegisteredListener(%d)", MyProcPid);
1436 :
1437 : /*
1438 : * Before registering, make sure we will unlisten before dying. (Note:
1439 : * this action does not get undone if we abort later.)
1440 : */
1441 58 : if (!unlistenExitRegistered)
1442 : {
1443 34 : before_shmem_exit(Async_UnlistenOnExit, 0);
1444 34 : unlistenExitRegistered = true;
1445 : }
1446 :
1447 : /*
1448 : * This is our first LISTEN, so establish our pointer.
1449 : *
1450 : * We set our pointer to the global tail pointer and then move it forward
1451 : * over already-committed notifications. This ensures we cannot miss any
1452 : * not-yet-committed notifications. We might get a few more but that
1453 : * doesn't hurt.
1454 : *
1455 : * In some scenarios there might be a lot of committed notifications that
1456 : * have not yet been pruned away (because some backend is being lazy about
1457 : * reading them). To reduce our startup time, we can look at other
1458 : * backends and adopt the maximum "pos" pointer of any backend that's in
1459 : * our database; any notifications it's already advanced over are surely
1460 : * committed and need not be re-examined by us. (We must consider only
1461 : * backends connected to our DB, because others will not have bothered to
1462 : * check committed-ness of notifications in our DB.)
1463 : *
1464 : * We need exclusive lock here so we can look at other backends' entries
1465 : * and manipulate the list links.
1466 : */
1467 58 : LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
1468 58 : head = QUEUE_HEAD;
1469 58 : max = QUEUE_TAIL;
1470 58 : prevListener = INVALID_PROC_NUMBER;
1471 88 : for (ProcNumber i = QUEUE_FIRST_LISTENER; i != INVALID_PROC_NUMBER; i = QUEUE_NEXT_LISTENER(i))
1472 : {
1473 30 : if (QUEUE_BACKEND_DBOID(i) == MyDatabaseId)
1474 30 : max = QUEUE_POS_MAX(max, QUEUE_BACKEND_POS(i));
1475 : /* Also find last listening backend before this one */
1476 30 : if (i < MyProcNumber)
1477 18 : prevListener = i;
1478 : }
1479 58 : QUEUE_BACKEND_POS(MyProcNumber) = max;
1480 58 : QUEUE_BACKEND_PID(MyProcNumber) = MyProcPid;
1481 58 : QUEUE_BACKEND_DBOID(MyProcNumber) = MyDatabaseId;
1482 58 : QUEUE_BACKEND_WAKEUP_PENDING(MyProcNumber) = false;
1483 58 : QUEUE_BACKEND_IS_ADVANCING(MyProcNumber) = false;
1484 : /* Insert backend into list of listeners at correct position */
1485 58 : if (prevListener != INVALID_PROC_NUMBER)
1486 : {
1487 12 : QUEUE_NEXT_LISTENER(MyProcNumber) = QUEUE_NEXT_LISTENER(prevListener);
1488 12 : QUEUE_NEXT_LISTENER(prevListener) = MyProcNumber;
1489 : }
1490 : else
1491 : {
1492 46 : QUEUE_NEXT_LISTENER(MyProcNumber) = QUEUE_FIRST_LISTENER;
1493 46 : QUEUE_FIRST_LISTENER = MyProcNumber;
1494 : }
1495 58 : LWLockRelease(NotifyQueueLock);
1496 :
1497 : /* Now we are listed in the global array, so remember we're listening */
1498 58 : amRegisteredListener = true;
1499 :
1500 : /*
1501 : * Try to move our pointer forward as far as possible. This will skip
1502 : * over already-committed notifications, which we want to do because they
1503 : * might be quite stale. Note that we are not yet listening on anything,
1504 : * so we won't deliver such notifications to our frontend. Also, although
1505 : * our transaction might have executed NOTIFY, those message(s) aren't
1506 : * queued yet so we won't skip them here.
1507 : */
1508 58 : if (!QUEUE_POS_EQUAL(max, head))
1509 30 : asyncQueueReadAllNotifications();
1510 : }
1511 :
1512 : /*
1513 : * PrepareTableEntriesForListen --- subroutine for PreCommit_Notify
1514 : *
1515 : * Prepare a LISTEN by recording it in pendingListenActions, pre-allocating
1516 : * an entry in localChannelTable, and pre-allocating an entry in the shared
1517 : * globalChannelTable with listening=false. The listening flag will be set
1518 : * to true in AtCommit_Notify. If we abort later, unwanted table entries
1519 : * will be removed.
1520 : */
1521 : static void
1522 112 : PrepareTableEntriesForListen(const char *channel)
1523 : {
1524 : GlobalChannelKey key;
1525 : GlobalChannelEntry *entry;
1526 : bool found;
1527 : ListenerEntry *listeners;
1528 : PendingListenEntry *pending;
1529 :
1530 : /*
1531 : * Record in local pending hash that we want to LISTEN, overwriting any
1532 : * earlier attempt to UNLISTEN.
1533 : */
1534 : pending = (PendingListenEntry *)
1535 112 : hash_search(pendingListenActions, channel, HASH_ENTER, NULL);
1536 112 : pending->action = PENDING_LISTEN;
1537 :
1538 : /*
1539 : * Ensure that there is an entry for the channel in localChannelTable.
1540 : * (Should this fail, we can just roll back.) If the transaction fails
1541 : * after this point, we will remove the entry if appropriate during
1542 : * ApplyPendingListenActions. Note that this entry allows IsListeningOn()
1543 : * to return TRUE; we assume nothing is going to consult that before
1544 : * AtCommit_Notify/AtAbort_Notify. However, if later actions attempt to
1545 : * UNLISTEN this channel or UNLISTEN *, we need to have the local entry
1546 : * present to ensure they do the right things; see
1547 : * PrepareTableEntriesForUnlisten and PrepareTableEntriesForUnlistenAll.
1548 : */
1549 112 : (void) hash_search(localChannelTable, channel, HASH_ENTER, NULL);
1550 :
1551 : /* Pre-allocate entry in shared globalChannelTable with listening=false */
1552 112 : GlobalChannelKeyInit(&key, MyDatabaseId, channel);
1553 112 : entry = dshash_find_or_insert(globalChannelTable, &key, &found);
1554 :
1555 112 : if (!found)
1556 : {
1557 : /* New channel entry, so initialize it to a safe state */
1558 68 : entry->listenersArray = InvalidDsaPointer;
1559 68 : entry->numListeners = 0;
1560 68 : entry->allocatedListeners = 0;
1561 : }
1562 :
1563 : /*
1564 : * Create listenersArray if entry doesn't have one. It's tempting to fold
1565 : * this into the !found case, but this coding allows us to cope in case
1566 : * dsa_allocate() failed in an earlier attempt.
1567 : */
1568 112 : if (!DsaPointerIsValid(entry->listenersArray))
1569 : {
1570 68 : entry->listenersArray = dsa_allocate(globalChannelDSA,
1571 : sizeof(ListenerEntry) * INITIAL_LISTENERS_ARRAY_SIZE);
1572 68 : entry->allocatedListeners = INITIAL_LISTENERS_ARRAY_SIZE;
1573 : }
1574 :
1575 : listeners = (ListenerEntry *)
1576 112 : dsa_get_address(globalChannelDSA, entry->listenersArray);
1577 :
1578 : /*
1579 : * Check if we already have a ListenerEntry (possibly from earlier in this
1580 : * transaction)
1581 : */
1582 150 : for (int i = 0; i < entry->numListeners; i++)
1583 : {
1584 60 : if (listeners[i].procNo == MyProcNumber)
1585 : {
1586 : /* Already have an entry; listening flag stays as-is until commit */
1587 22 : dshash_release_lock(globalChannelTable, entry);
1588 22 : return;
1589 : }
1590 : }
1591 :
1592 : /* Need to add a new entry; grow array if necessary */
1593 90 : if (entry->numListeners >= entry->allocatedListeners)
1594 : {
1595 2 : int new_size = entry->allocatedListeners * 2;
1596 2 : dsa_pointer old_array = entry->listenersArray;
1597 2 : dsa_pointer new_array = dsa_allocate(globalChannelDSA,
1598 : sizeof(ListenerEntry) * new_size);
1599 2 : ListenerEntry *new_listeners = (ListenerEntry *) dsa_get_address(globalChannelDSA, new_array);
1600 :
1601 2 : memcpy(new_listeners, listeners, sizeof(ListenerEntry) * entry->numListeners);
1602 2 : entry->listenersArray = new_array;
1603 2 : entry->allocatedListeners = new_size;
1604 2 : dsa_free(globalChannelDSA, old_array);
1605 2 : listeners = new_listeners;
1606 : }
1607 :
1608 90 : listeners[entry->numListeners].procNo = MyProcNumber;
1609 90 : listeners[entry->numListeners].listening = false; /* staged, not yet
1610 : * committed */
1611 90 : entry->numListeners++;
1612 :
1613 90 : dshash_release_lock(globalChannelTable, entry);
1614 : }
1615 :
1616 : /*
1617 : * PrepareTableEntriesForUnlisten --- subroutine for PreCommit_Notify
1618 : *
1619 : * Prepare an UNLISTEN by recording it in pendingListenActions, but only if
1620 : * we're currently listening (committed or staged). We don't touch
1621 : * globalChannelTable yet - the listener keeps receiving signals until
1622 : * commit, when the entry is removed.
1623 : */
1624 : static void
1625 6 : PrepareTableEntriesForUnlisten(const char *channel)
1626 : {
1627 : PendingListenEntry *pending;
1628 :
1629 : /*
1630 : * If the channel name is not in localChannelTable, then we are neither
1631 : * listening on it nor preparing to listen on it, so we don't need to
1632 : * record an UNLISTEN action.
1633 : */
1634 : Assert(localChannelTable != NULL);
1635 6 : if (hash_search(localChannelTable, channel, HASH_FIND, NULL) == NULL)
1636 0 : return;
1637 :
1638 : /*
1639 : * Record in local pending hash that we want to UNLISTEN, overwriting any
1640 : * earlier attempt to LISTEN. Don't touch localChannelTable or
1641 : * globalChannelTable yet - we keep receiving signals until commit.
1642 : */
1643 : pending = (PendingListenEntry *)
1644 6 : hash_search(pendingListenActions, channel, HASH_ENTER, NULL);
1645 6 : pending->action = PENDING_UNLISTEN;
1646 : }
1647 :
1648 : /*
1649 : * PrepareTableEntriesForUnlistenAll --- subroutine for PreCommit_Notify
1650 : *
1651 : * Prepare UNLISTEN * by recording an UNLISTEN for all listened or
1652 : * about-to-be-listened channels in pendingListenActions.
1653 : */
1654 : static void
1655 96 : PrepareTableEntriesForUnlistenAll(void)
1656 : {
1657 : HASH_SEQ_STATUS seq;
1658 : ChannelName *channelEntry;
1659 : PendingListenEntry *pending;
1660 :
1661 : /*
1662 : * Scan localChannelTable, which will have the names of all channels that
1663 : * we are listening on or have prepared to listen on. Record an UNLISTEN
1664 : * action for each one, overwriting any earlier attempt to LISTEN.
1665 : */
1666 96 : hash_seq_init(&seq, localChannelTable);
1667 164 : while ((channelEntry = (ChannelName *) hash_seq_search(&seq)) != NULL)
1668 : {
1669 : pending = (PendingListenEntry *)
1670 68 : hash_search(pendingListenActions, channelEntry->channel, HASH_ENTER, NULL);
1671 68 : pending->action = PENDING_UNLISTEN;
1672 : }
1673 96 : }
1674 :
1675 : /*
1676 : * RemoveListenerFromChannel --- remove idx'th listener from global channel entry
1677 : *
1678 : * Decrements numListeners, compacts the array, and frees the entry if empty.
1679 : * Sets *entry_ptr to NULL if the entry was deleted.
1680 : *
1681 : * We could get the listeners pointer from the entry, but all callers
1682 : * already have it at hand.
1683 : */
1684 : static void
1685 74 : RemoveListenerFromChannel(GlobalChannelEntry **entry_ptr,
1686 : ListenerEntry *listeners,
1687 : int idx)
1688 : {
1689 74 : GlobalChannelEntry *entry = *entry_ptr;
1690 :
1691 74 : entry->numListeners--;
1692 74 : if (idx < entry->numListeners)
1693 16 : memmove(&listeners[idx], &listeners[idx + 1],
1694 16 : sizeof(ListenerEntry) * (entry->numListeners - idx));
1695 :
1696 74 : if (entry->numListeners == 0)
1697 : {
1698 52 : dsa_free(globalChannelDSA, entry->listenersArray);
1699 52 : dshash_delete_entry(globalChannelTable, entry);
1700 : /* tells caller not to release the entry's lock: */
1701 52 : *entry_ptr = NULL;
1702 : }
1703 74 : }
1704 :
1705 : /*
1706 : * ApplyPendingListenActions
1707 : *
1708 : * Apply, or revert, staged listen/unlisten changes to the local and global
1709 : * hash tables.
1710 : */
1711 : static void
1712 52592 : ApplyPendingListenActions(bool isCommit)
1713 : {
1714 : HASH_SEQ_STATUS seq;
1715 : PendingListenEntry *pending;
1716 :
1717 : /* Quick exit if nothing to do */
1718 52592 : if (pendingListenActions == NULL)
1719 52410 : return;
1720 :
1721 : /* We made a globalChannelTable before building pendingListenActions */
1722 182 : if (globalChannelTable == NULL)
1723 0 : elog(PANIC, "global channel table missing post-commit/abort");
1724 :
1725 : /* For each staged action ... */
1726 182 : hash_seq_init(&seq, pendingListenActions);
1727 366 : while ((pending = (PendingListenEntry *) hash_seq_search(&seq)) != NULL)
1728 : {
1729 : GlobalChannelKey key;
1730 : GlobalChannelEntry *entry;
1731 184 : bool removeLocal = true;
1732 184 : bool foundListener = false;
1733 :
1734 : /*
1735 : * Find the global entry for this channel. If isCommit, it had better
1736 : * exist (it was created in PreCommit). In an abort, it might not
1737 : * exist, in which case we are not listening and should discard any
1738 : * local entry that PreCommit may have managed to create.
1739 : */
1740 184 : GlobalChannelKeyInit(&key, MyDatabaseId, pending->channel);
1741 184 : entry = dshash_find(globalChannelTable, &key, true);
1742 184 : if (entry != NULL)
1743 : {
1744 : /* Scan entry to find the ListenerEntry for this backend */
1745 : ListenerEntry *listeners;
1746 :
1747 : listeners = (ListenerEntry *)
1748 184 : dsa_get_address(globalChannelDSA, entry->listenersArray);
1749 :
1750 232 : for (int i = 0; i < entry->numListeners; i++)
1751 : {
1752 232 : if (listeners[i].procNo != MyProcNumber)
1753 48 : continue;
1754 184 : foundListener = true;
1755 184 : if (isCommit)
1756 : {
1757 184 : if (pending->action == PENDING_LISTEN)
1758 : {
1759 : /*
1760 : * LISTEN being committed: set listening=true.
1761 : * localChannelTable entry was created during
1762 : * PreCommit and should be kept.
1763 : */
1764 110 : listeners[i].listening = true;
1765 110 : removeLocal = false;
1766 : }
1767 : else
1768 : {
1769 : /*
1770 : * UNLISTEN being committed: remove pre-allocated
1771 : * entries from both tables.
1772 : */
1773 74 : RemoveListenerFromChannel(&entry, listeners, i);
1774 : }
1775 : }
1776 : else
1777 : {
1778 : /*
1779 : * Note: this part is reachable only if the transaction
1780 : * aborts after PreCommit_Notify() has made some
1781 : * pendingListenActions entries, so it's pretty hard to
1782 : * test.
1783 : */
1784 0 : if (!listeners[i].listening)
1785 : {
1786 : /*
1787 : * Staged LISTEN (or LISTEN+UNLISTEN) being aborted,
1788 : * and we weren't listening before, so remove
1789 : * pre-allocated entries from both tables.
1790 : */
1791 0 : RemoveListenerFromChannel(&entry, listeners, i);
1792 : }
1793 : else
1794 : {
1795 : /*
1796 : * We're aborting, but the previous state was that
1797 : * we're listening, so keep localChannelTable entry.
1798 : */
1799 0 : removeLocal = false;
1800 : }
1801 : }
1802 184 : break; /* there shouldn't be another match */
1803 : }
1804 :
1805 : /* We might have already released the entry by removing it */
1806 184 : if (entry != NULL)
1807 132 : dshash_release_lock(globalChannelTable, entry);
1808 : }
1809 :
1810 : /*
1811 : * If we're committing a LISTEN action, we should have found a
1812 : * matching ListenerEntry, but otherwise it's okay if we didn't.
1813 : */
1814 184 : if (isCommit && pending->action == PENDING_LISTEN && !foundListener)
1815 0 : elog(PANIC, "could not find listener entry for channel \"%s\" backend %d",
1816 : pending->channel, MyProcNumber);
1817 :
1818 : /*
1819 : * If we did not find a globalChannelTable entry for our backend, or
1820 : * if we are unlistening, remove any localChannelTable entry that may
1821 : * exist. (Note in particular that this cleans up if we created a
1822 : * localChannelTable entry and then failed while trying to create a
1823 : * globalChannelTable entry.)
1824 : */
1825 184 : if (removeLocal && localChannelTable != NULL)
1826 74 : (void) hash_search(localChannelTable, pending->channel,
1827 : HASH_REMOVE, NULL);
1828 : }
1829 : }
1830 :
1831 : /*
1832 : * CleanupListenersOnExit --- called from Async_UnlistenOnExit
1833 : *
1834 : * Remove this backend from all channels in the shared global table.
1835 : */
1836 : static void
1837 34 : CleanupListenersOnExit(void)
1838 : {
1839 : dshash_seq_status status;
1840 : GlobalChannelEntry *entry;
1841 :
1842 34 : if (Trace_notify)
1843 0 : elog(DEBUG1, "CleanupListenersOnExit(%d)", MyProcPid);
1844 :
1845 : /* Clear our local cache (not really necessary, but be consistent) */
1846 34 : if (localChannelTable != NULL)
1847 : {
1848 34 : hash_destroy(localChannelTable);
1849 34 : localChannelTable = NULL;
1850 : }
1851 :
1852 : /* Now remove our entries from the shared globalChannelTable */
1853 34 : if (globalChannelTable == NULL)
1854 0 : return;
1855 :
1856 34 : dshash_seq_init(&status, globalChannelTable, true);
1857 66 : while ((entry = dshash_seq_next(&status)) != NULL)
1858 : {
1859 : ListenerEntry *listeners;
1860 :
1861 32 : if (entry->key.dboid != MyDatabaseId)
1862 0 : continue; /* not relevant */
1863 :
1864 : listeners = (ListenerEntry *)
1865 32 : dsa_get_address(globalChannelDSA, entry->listenersArray);
1866 :
1867 48 : for (int i = 0; i < entry->numListeners; i++)
1868 : {
1869 32 : if (listeners[i].procNo == MyProcNumber)
1870 : {
1871 16 : entry->numListeners--;
1872 16 : if (i < entry->numListeners)
1873 0 : memmove(&listeners[i], &listeners[i + 1],
1874 0 : sizeof(ListenerEntry) * (entry->numListeners - i));
1875 :
1876 16 : if (entry->numListeners == 0)
1877 : {
1878 16 : dsa_free(globalChannelDSA, entry->listenersArray);
1879 16 : dshash_delete_current(&status);
1880 : }
1881 16 : break;
1882 : }
1883 : }
1884 : }
1885 34 : dshash_seq_term(&status);
1886 : }
1887 :
1888 : /*
1889 : * Test whether we are actively listening on the given channel name.
1890 : *
1891 : * Note: this function is executed for every notification found in the queue.
1892 : */
1893 : static bool
1894 110 : IsListeningOn(const char *channel)
1895 : {
1896 110 : if (localChannelTable == NULL)
1897 0 : return false;
1898 :
1899 110 : return (hash_search(localChannelTable, channel, HASH_FIND, NULL) != NULL);
1900 : }
1901 :
1902 : /*
1903 : * Remove our entry from the listeners array when we are no longer listening
1904 : * on any channel. NB: must not fail if we're already not listening.
1905 : */
1906 : static void
1907 80 : asyncQueueUnregister(void)
1908 : {
1909 : Assert(LocalChannelTableIsEmpty()); /* else caller error */
1910 :
1911 80 : if (!amRegisteredListener) /* nothing to do */
1912 22 : return;
1913 :
1914 : /*
1915 : * Need exclusive lock here to manipulate list links.
1916 : */
1917 58 : LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
1918 : /* Mark our entry as invalid */
1919 58 : QUEUE_BACKEND_PID(MyProcNumber) = InvalidPid;
1920 58 : QUEUE_BACKEND_DBOID(MyProcNumber) = InvalidOid;
1921 58 : QUEUE_BACKEND_WAKEUP_PENDING(MyProcNumber) = false;
1922 58 : QUEUE_BACKEND_IS_ADVANCING(MyProcNumber) = false;
1923 : /* and remove it from the list */
1924 58 : if (QUEUE_FIRST_LISTENER == MyProcNumber)
1925 44 : QUEUE_FIRST_LISTENER = QUEUE_NEXT_LISTENER(MyProcNumber);
1926 : else
1927 : {
1928 22 : for (ProcNumber i = QUEUE_FIRST_LISTENER; i != INVALID_PROC_NUMBER; i = QUEUE_NEXT_LISTENER(i))
1929 : {
1930 22 : if (QUEUE_NEXT_LISTENER(i) == MyProcNumber)
1931 : {
1932 14 : QUEUE_NEXT_LISTENER(i) = QUEUE_NEXT_LISTENER(MyProcNumber);
1933 14 : break;
1934 : }
1935 : }
1936 : }
1937 58 : QUEUE_NEXT_LISTENER(MyProcNumber) = INVALID_PROC_NUMBER;
1938 58 : LWLockRelease(NotifyQueueLock);
1939 :
1940 : /* mark ourselves as no longer listed in the global array */
1941 58 : amRegisteredListener = false;
1942 : }
1943 :
1944 : /*
1945 : * Test whether there is room to insert more notification messages.
1946 : *
1947 : * Caller must hold at least shared NotifyQueueLock.
1948 : */
1949 : static bool
1950 200 : asyncQueueIsFull(void)
1951 : {
1952 200 : int64 headPage = QUEUE_POS_PAGE(QUEUE_HEAD);
1953 200 : int64 tailPage = QUEUE_POS_PAGE(QUEUE_TAIL);
1954 200 : int64 occupied = headPage - tailPage;
1955 :
1956 200 : return occupied >= max_notify_queue_pages;
1957 : }
1958 :
1959 : /*
1960 : * Advance the QueuePosition to the next entry, assuming that the current
1961 : * entry is of length entryLength. If we jump to a new page the function
1962 : * returns true, else false.
1963 : */
1964 : static bool
1965 4906 : asyncQueueAdvance(volatile QueuePosition *position, int entryLength)
1966 : {
1967 4906 : int64 pageno = QUEUE_POS_PAGE(*position);
1968 4906 : int offset = QUEUE_POS_OFFSET(*position);
1969 4906 : bool pageJump = false;
1970 :
1971 : /*
1972 : * Move to the next writing position: First jump over what we have just
1973 : * written or read.
1974 : */
1975 4906 : offset += entryLength;
1976 : Assert(offset <= QUEUE_PAGESIZE);
1977 :
1978 : /*
1979 : * In a second step check if another entry can possibly be written to the
1980 : * page. If so, stay here, we have reached the next position. If not, then
1981 : * we need to move on to the next page.
1982 : */
1983 4906 : if (offset + QUEUEALIGN(AsyncQueueEntryEmptySize) > QUEUE_PAGESIZE)
1984 : {
1985 76 : pageno++;
1986 76 : offset = 0;
1987 76 : pageJump = true;
1988 : }
1989 :
1990 4906 : SET_QUEUE_POS(*position, pageno, offset);
1991 4906 : return pageJump;
1992 : }
1993 :
1994 : /*
1995 : * Fill the AsyncQueueEntry at *qe with an outbound notification message.
1996 : */
1997 : static void
1998 2284 : asyncQueueNotificationToEntry(Notification *n, AsyncQueueEntry *qe)
1999 : {
2000 2284 : size_t channellen = n->channel_len;
2001 2284 : size_t payloadlen = n->payload_len;
2002 : int entryLength;
2003 :
2004 : Assert(channellen < NAMEDATALEN);
2005 : Assert(payloadlen < NOTIFY_PAYLOAD_MAX_LENGTH);
2006 :
2007 : /* The terminators are already included in AsyncQueueEntryEmptySize */
2008 2284 : entryLength = AsyncQueueEntryEmptySize + payloadlen + channellen;
2009 2284 : entryLength = QUEUEALIGN(entryLength);
2010 2284 : qe->length = entryLength;
2011 2284 : qe->dboid = MyDatabaseId;
2012 2284 : qe->xid = GetCurrentTransactionId();
2013 2284 : qe->srcPid = MyProcPid;
2014 2284 : memcpy(qe->data, n->data, channellen + payloadlen + 2);
2015 2284 : }
2016 :
2017 : /*
2018 : * Add pending notifications to the queue.
2019 : *
2020 : * We go page by page here, i.e. we stop once we have to go to a new page but
2021 : * we will be called again and then fill that next page. If an entry does not
2022 : * fit into the current page, we write a dummy entry with an InvalidOid as the
2023 : * database OID in order to fill the page. So every page is always used up to
2024 : * the last byte which simplifies reading the page later.
2025 : *
2026 : * We are passed the list cell (in pendingNotifies->events) containing the next
2027 : * notification to write and return the first still-unwritten cell back.
2028 : * Eventually we will return NULL indicating all is done.
2029 : *
2030 : * We are holding NotifyQueueLock already from the caller and grab
2031 : * page specific SLRU bank lock locally in this function.
2032 : */
2033 : static ListCell *
2034 200 : asyncQueueAddEntries(ListCell *nextNotify)
2035 : {
2036 : AsyncQueueEntry qe;
2037 : QueuePosition queue_head;
2038 : int64 pageno;
2039 : int offset;
2040 : int slotno;
2041 : LWLock *prevlock;
2042 :
2043 : /*
2044 : * We work with a local copy of QUEUE_HEAD, which we write back to shared
2045 : * memory upon exiting. The reason for this is that if we have to advance
2046 : * to a new page, SimpleLruZeroPage might fail (out of disk space, for
2047 : * instance), and we must not advance QUEUE_HEAD if it does. (Otherwise,
2048 : * subsequent insertions would try to put entries into a page that slru.c
2049 : * thinks doesn't exist yet.) So, use a local position variable. Note
2050 : * that if we do fail, any already-inserted queue entries are forgotten;
2051 : * this is okay, since they'd be useless anyway after our transaction
2052 : * rolls back.
2053 : */
2054 200 : queue_head = QUEUE_HEAD;
2055 :
2056 : /*
2057 : * If this is the first write since the postmaster started, we need to
2058 : * initialize the first page of the async SLRU. Otherwise, the current
2059 : * page should be initialized already, so just fetch it.
2060 : */
2061 200 : pageno = QUEUE_POS_PAGE(queue_head);
2062 200 : prevlock = SimpleLruGetBankLock(NotifyCtl, pageno);
2063 :
2064 : /* We hold both NotifyQueueLock and SLRU bank lock during this operation */
2065 200 : LWLockAcquire(prevlock, LW_EXCLUSIVE);
2066 :
2067 200 : if (QUEUE_POS_IS_ZERO(queue_head))
2068 16 : slotno = SimpleLruZeroPage(NotifyCtl, pageno);
2069 : else
2070 184 : slotno = SimpleLruReadPage(NotifyCtl, pageno, true,
2071 : InvalidTransactionId);
2072 :
2073 : /* Note we mark the page dirty before writing in it */
2074 200 : NotifyCtl->shared->page_dirty[slotno] = true;
2075 :
2076 2414 : while (nextNotify != NULL)
2077 : {
2078 2284 : Notification *n = (Notification *) lfirst(nextNotify);
2079 :
2080 : /* Construct a valid queue entry in local variable qe */
2081 2284 : asyncQueueNotificationToEntry(n, &qe);
2082 :
2083 2284 : offset = QUEUE_POS_OFFSET(queue_head);
2084 :
2085 : /* Check whether the entry really fits on the current page */
2086 2284 : if (offset + qe.length <= QUEUE_PAGESIZE)
2087 : {
2088 : /* OK, so advance nextNotify past this item */
2089 2220 : nextNotify = lnext(pendingNotifies->events, nextNotify);
2090 : }
2091 : else
2092 : {
2093 : /*
2094 : * Write a dummy entry to fill up the page. Actually readers will
2095 : * only check dboid and since it won't match any reader's database
2096 : * OID, they will ignore this entry and move on.
2097 : */
2098 64 : qe.length = QUEUE_PAGESIZE - offset;
2099 64 : qe.dboid = InvalidOid;
2100 64 : qe.xid = InvalidTransactionId;
2101 64 : qe.data[0] = '\0'; /* empty channel */
2102 64 : qe.data[1] = '\0'; /* empty payload */
2103 : }
2104 :
2105 : /* Now copy qe into the shared buffer page */
2106 2284 : memcpy(NotifyCtl->shared->page_buffer[slotno] + offset,
2107 : &qe,
2108 2284 : qe.length);
2109 :
2110 : /* Advance queue_head appropriately, and detect if page is full */
2111 2284 : if (asyncQueueAdvance(&(queue_head), qe.length))
2112 : {
2113 : LWLock *lock;
2114 :
2115 70 : pageno = QUEUE_POS_PAGE(queue_head);
2116 70 : lock = SimpleLruGetBankLock(NotifyCtl, pageno);
2117 70 : if (lock != prevlock)
2118 : {
2119 0 : LWLockRelease(prevlock);
2120 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
2121 0 : prevlock = lock;
2122 : }
2123 :
2124 : /*
2125 : * Page is full, so we're done here, but first fill the next page
2126 : * with zeroes. The reason to do this is to ensure that slru.c's
2127 : * idea of the head page is always the same as ours, which avoids
2128 : * boundary problems in SimpleLruTruncate. The test in
2129 : * asyncQueueIsFull() ensured that there is room to create this
2130 : * page without overrunning the queue.
2131 : */
2132 70 : slotno = SimpleLruZeroPage(NotifyCtl, QUEUE_POS_PAGE(queue_head));
2133 :
2134 : /*
2135 : * If the new page address is a multiple of QUEUE_CLEANUP_DELAY,
2136 : * set flag to remember that we should try to advance the tail
2137 : * pointer (we don't want to actually do that right here).
2138 : */
2139 70 : if (QUEUE_POS_PAGE(queue_head) % QUEUE_CLEANUP_DELAY == 0)
2140 16 : tryAdvanceTail = true;
2141 :
2142 : /* And exit the loop */
2143 70 : break;
2144 : }
2145 : }
2146 :
2147 : /* Success, so update the global QUEUE_HEAD */
2148 200 : QUEUE_HEAD = queue_head;
2149 :
2150 200 : LWLockRelease(prevlock);
2151 :
2152 200 : return nextNotify;
2153 : }
2154 :
2155 : /*
2156 : * SQL function to return the fraction of the notification queue currently
2157 : * occupied.
2158 : */
2159 : Datum
2160 10 : pg_notification_queue_usage(PG_FUNCTION_ARGS)
2161 : {
2162 : double usage;
2163 :
2164 : /* Advance the queue tail so we don't report a too-large result */
2165 10 : asyncQueueAdvanceTail();
2166 :
2167 10 : LWLockAcquire(NotifyQueueLock, LW_SHARED);
2168 10 : usage = asyncQueueUsage();
2169 10 : LWLockRelease(NotifyQueueLock);
2170 :
2171 10 : PG_RETURN_FLOAT8(usage);
2172 : }
2173 :
2174 : /*
2175 : * Return the fraction of the queue that is currently occupied.
2176 : *
2177 : * The caller must hold NotifyQueueLock in (at least) shared mode.
2178 : *
2179 : * Note: we measure the distance to the logical tail page, not the physical
2180 : * tail page. In some sense that's wrong, but the relative position of the
2181 : * physical tail is affected by details such as SLRU segment boundaries,
2182 : * so that a result based on that is unpleasantly unstable.
2183 : */
2184 : static double
2185 210 : asyncQueueUsage(void)
2186 : {
2187 210 : int64 headPage = QUEUE_POS_PAGE(QUEUE_HEAD);
2188 210 : int64 tailPage = QUEUE_POS_PAGE(QUEUE_TAIL);
2189 210 : int64 occupied = headPage - tailPage;
2190 :
2191 210 : if (occupied == 0)
2192 132 : return (double) 0; /* fast exit for common case */
2193 :
2194 78 : return (double) occupied / (double) max_notify_queue_pages;
2195 : }
2196 :
2197 : /*
2198 : * Check whether the queue is at least half full, and emit a warning if so.
2199 : *
2200 : * This is unlikely given the size of the queue, but possible.
2201 : * The warnings show up at most once every QUEUE_FULL_WARN_INTERVAL.
2202 : *
2203 : * Caller must hold exclusive NotifyQueueLock.
2204 : */
2205 : static void
2206 200 : asyncQueueFillWarning(void)
2207 : {
2208 : double fillDegree;
2209 : TimestampTz t;
2210 :
2211 200 : fillDegree = asyncQueueUsage();
2212 200 : if (fillDegree < 0.5)
2213 200 : return;
2214 :
2215 0 : t = GetCurrentTimestamp();
2216 :
2217 0 : if (TimestampDifferenceExceeds(asyncQueueControl->lastQueueFillWarn,
2218 : t, QUEUE_FULL_WARN_INTERVAL))
2219 : {
2220 0 : QueuePosition min = QUEUE_HEAD;
2221 0 : int32 minPid = InvalidPid;
2222 :
2223 0 : for (ProcNumber i = QUEUE_FIRST_LISTENER; i != INVALID_PROC_NUMBER; i = QUEUE_NEXT_LISTENER(i))
2224 : {
2225 : Assert(QUEUE_BACKEND_PID(i) != InvalidPid);
2226 0 : min = QUEUE_POS_MIN(min, QUEUE_BACKEND_POS(i));
2227 0 : if (QUEUE_POS_EQUAL(min, QUEUE_BACKEND_POS(i)))
2228 0 : minPid = QUEUE_BACKEND_PID(i);
2229 : }
2230 :
2231 0 : ereport(WARNING,
2232 : (errmsg("NOTIFY queue is %.0f%% full", fillDegree * 100),
2233 : (minPid != InvalidPid ?
2234 : errdetail("The server process with PID %d is among those with the oldest transactions.", minPid)
2235 : : 0),
2236 : (minPid != InvalidPid ?
2237 : errhint("The NOTIFY queue cannot be emptied until that process ends its current transaction.")
2238 : : 0)));
2239 :
2240 0 : asyncQueueControl->lastQueueFillWarn = t;
2241 : }
2242 : }
2243 :
2244 : /*
2245 : * Send signals to listening backends.
2246 : *
2247 : * Normally we signal only backends that are interested in the notifies that
2248 : * we just sent. However, that will leave idle listeners falling further and
2249 : * further behind. Waken them anyway if they're far enough behind, so they'll
2250 : * advance their queue position pointers, allowing the global tail to advance.
2251 : *
2252 : * Since we know the ProcNumber and the Pid the signaling is quite cheap.
2253 : *
2254 : * This is called during CommitTransaction(), so it's important for it
2255 : * to have very low probability of failure.
2256 : */
2257 : static void
2258 130 : SignalBackends(void)
2259 : {
2260 : int count;
2261 :
2262 : /* Can't get here without PreCommit_Notify having made the global table */
2263 : Assert(globalChannelTable != NULL);
2264 :
2265 : /* It should have set up these arrays, too */
2266 : Assert(signalPids != NULL && signalProcnos != NULL);
2267 :
2268 : /*
2269 : * Identify backends that we need to signal. We don't want to send
2270 : * signals while holding the NotifyQueueLock, so this part just builds a
2271 : * list of target PIDs in signalPids[] and signalProcnos[].
2272 : */
2273 130 : count = 0;
2274 :
2275 130 : LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
2276 :
2277 : /* Scan each channel name that we notified in this transaction */
2278 396 : foreach_ptr(char, channel, pendingNotifies->uniqueChannelNames)
2279 : {
2280 : GlobalChannelKey key;
2281 : GlobalChannelEntry *entry;
2282 : ListenerEntry *listeners;
2283 :
2284 136 : GlobalChannelKeyInit(&key, MyDatabaseId, channel);
2285 136 : entry = dshash_find(globalChannelTable, &key, false);
2286 136 : if (entry == NULL)
2287 56 : continue; /* nobody is listening */
2288 :
2289 80 : listeners = (ListenerEntry *) dsa_get_address(globalChannelDSA,
2290 : entry->listenersArray);
2291 :
2292 : /* Identify listeners that now need waking, add them to arrays */
2293 170 : for (int j = 0; j < entry->numListeners; j++)
2294 : {
2295 : ProcNumber i;
2296 : int32 pid;
2297 : QueuePosition pos;
2298 :
2299 90 : if (!listeners[j].listening)
2300 24 : continue; /* ignore not-yet-committed listeners */
2301 :
2302 90 : i = listeners[j].procNo;
2303 :
2304 90 : if (QUEUE_BACKEND_WAKEUP_PENDING(i))
2305 24 : continue; /* already signaled, no need to repeat */
2306 :
2307 66 : pid = QUEUE_BACKEND_PID(i);
2308 66 : pos = QUEUE_BACKEND_POS(i);
2309 :
2310 66 : if (QUEUE_POS_EQUAL(pos, QUEUE_HEAD))
2311 0 : continue; /* it's fully caught up already */
2312 :
2313 : Assert(pid != InvalidPid);
2314 :
2315 66 : QUEUE_BACKEND_WAKEUP_PENDING(i) = true;
2316 66 : signalPids[count] = pid;
2317 66 : signalProcnos[count] = i;
2318 66 : count++;
2319 : }
2320 :
2321 80 : dshash_release_lock(globalChannelTable, entry);
2322 : }
2323 :
2324 : /*
2325 : * Scan all listeners. Any that are not already pending wakeup must not
2326 : * be interested in our notifications (else we'd have set their wakeup
2327 : * flags above). Check to see if we can directly advance their queue
2328 : * pointers to save a wakeup. Otherwise, if they are far behind, wake
2329 : * them anyway so they will catch up.
2330 : */
2331 256 : for (ProcNumber i = QUEUE_FIRST_LISTENER; i != INVALID_PROC_NUMBER; i = QUEUE_NEXT_LISTENER(i))
2332 : {
2333 : int32 pid;
2334 : QueuePosition pos;
2335 :
2336 126 : if (QUEUE_BACKEND_WAKEUP_PENDING(i))
2337 84 : continue;
2338 :
2339 : /* If it's currently advancing, we should not touch it */
2340 42 : if (QUEUE_BACKEND_IS_ADVANCING(i))
2341 0 : continue;
2342 :
2343 42 : pid = QUEUE_BACKEND_PID(i);
2344 42 : pos = QUEUE_BACKEND_POS(i);
2345 :
2346 : /*
2347 : * We can directly advance the other backend's queue pointer if it's
2348 : * not currently advancing (else there are race conditions), and its
2349 : * current pointer is not behind queueHeadBeforeWrite (else we'd make
2350 : * it miss some older messages), and we'd not be moving the pointer
2351 : * backward.
2352 : */
2353 84 : if (!QUEUE_POS_PRECEDES(pos, queueHeadBeforeWrite) &&
2354 62 : QUEUE_POS_PRECEDES(pos, queueHeadAfterWrite))
2355 : {
2356 : /* We can directly advance its pointer past what we wrote */
2357 42 : QUEUE_BACKEND_POS(i) = queueHeadAfterWrite;
2358 : }
2359 0 : else if (asyncQueuePageDiff(QUEUE_POS_PAGE(QUEUE_HEAD),
2360 : QUEUE_POS_PAGE(pos)) >= QUEUE_CLEANUP_DELAY)
2361 : {
2362 : /* It's idle and far behind, so wake it up */
2363 : Assert(pid != InvalidPid);
2364 :
2365 0 : QUEUE_BACKEND_WAKEUP_PENDING(i) = true;
2366 0 : signalPids[count] = pid;
2367 0 : signalProcnos[count] = i;
2368 0 : count++;
2369 : }
2370 : }
2371 :
2372 130 : LWLockRelease(NotifyQueueLock);
2373 :
2374 : /* Now send signals */
2375 196 : for (int i = 0; i < count; i++)
2376 : {
2377 66 : int32 pid = signalPids[i];
2378 :
2379 : /*
2380 : * If we are signaling our own process, no need to involve the kernel;
2381 : * just set the flag directly.
2382 : */
2383 66 : if (pid == MyProcPid)
2384 : {
2385 42 : notifyInterruptPending = true;
2386 42 : continue;
2387 : }
2388 :
2389 : /*
2390 : * Note: assuming things aren't broken, a signal failure here could
2391 : * only occur if the target backend exited since we released
2392 : * NotifyQueueLock; which is unlikely but certainly possible. So we
2393 : * just log a low-level debug message if it happens.
2394 : */
2395 24 : if (SendProcSignal(pid, PROCSIG_NOTIFY_INTERRUPT, signalProcnos[i]) < 0)
2396 0 : elog(DEBUG3, "could not signal backend with PID %d: %m", pid);
2397 : }
2398 130 : }
2399 :
2400 : /*
2401 : * AtAbort_Notify
2402 : *
2403 : * This is called at transaction abort.
2404 : *
2405 : * Revert any staged listen/unlisten changes and clean up transaction state.
2406 : * This only does anything if we abort after PreCommit_Notify has staged
2407 : * some entries.
2408 : */
2409 : void
2410 52280 : AtAbort_Notify(void)
2411 : {
2412 : /* Revert staged listen/unlisten changes */
2413 52280 : ApplyPendingListenActions(false);
2414 :
2415 : /* If we're no longer listening on anything, unregister */
2416 52280 : if (amRegisteredListener && LocalChannelTableIsEmpty())
2417 0 : asyncQueueUnregister();
2418 :
2419 : /* And clean up */
2420 52280 : ClearPendingActionsAndNotifies();
2421 52280 : }
2422 :
2423 : /*
2424 : * AtSubCommit_Notify() --- Take care of subtransaction commit.
2425 : *
2426 : * Reassign all items in the pending lists to the parent transaction.
2427 : */
2428 : void
2429 10706 : AtSubCommit_Notify(void)
2430 : {
2431 10706 : int my_level = GetCurrentTransactionNestLevel();
2432 :
2433 : /* If there are actions at our nesting level, we must reparent them. */
2434 10706 : if (pendingActions != NULL &&
2435 4 : pendingActions->nestingLevel >= my_level)
2436 : {
2437 4 : if (pendingActions->upper == NULL ||
2438 2 : pendingActions->upper->nestingLevel < my_level - 1)
2439 : {
2440 : /* nothing to merge; give the whole thing to the parent */
2441 2 : --pendingActions->nestingLevel;
2442 : }
2443 : else
2444 : {
2445 2 : ActionList *childPendingActions = pendingActions;
2446 :
2447 2 : pendingActions = pendingActions->upper;
2448 :
2449 : /*
2450 : * Mustn't try to eliminate duplicates here --- see queue_listen()
2451 : */
2452 4 : pendingActions->actions =
2453 2 : list_concat(pendingActions->actions,
2454 2 : childPendingActions->actions);
2455 2 : pfree(childPendingActions);
2456 : }
2457 : }
2458 :
2459 : /* If there are notifies at our nesting level, we must reparent them. */
2460 10706 : if (pendingNotifies != NULL &&
2461 6 : pendingNotifies->nestingLevel >= my_level)
2462 : {
2463 : Assert(pendingNotifies->nestingLevel == my_level);
2464 :
2465 4 : if (pendingNotifies->upper == NULL ||
2466 2 : pendingNotifies->upper->nestingLevel < my_level - 1)
2467 : {
2468 : /* nothing to merge; give the whole thing to the parent */
2469 2 : --pendingNotifies->nestingLevel;
2470 : }
2471 : else
2472 : {
2473 : /*
2474 : * Formerly, we didn't bother to eliminate duplicates here, but
2475 : * now we must, else we fall foul of "Assert(!found)", either here
2476 : * or during a later attempt to build the parent-level hashtable.
2477 : */
2478 2 : NotificationList *childPendingNotifies = pendingNotifies;
2479 : ListCell *l;
2480 :
2481 2 : pendingNotifies = pendingNotifies->upper;
2482 : /* Insert all the subxact's events into parent, except for dups */
2483 10 : foreach(l, childPendingNotifies->events)
2484 : {
2485 8 : Notification *childn = (Notification *) lfirst(l);
2486 :
2487 8 : if (!AsyncExistsPendingNotify(childn))
2488 4 : AddEventToPendingNotifies(childn);
2489 : }
2490 2 : pfree(childPendingNotifies);
2491 : }
2492 : }
2493 10706 : }
2494 :
2495 : /*
2496 : * AtSubAbort_Notify() --- Take care of subtransaction abort.
2497 : */
2498 : void
2499 9404 : AtSubAbort_Notify(void)
2500 : {
2501 9404 : int my_level = GetCurrentTransactionNestLevel();
2502 :
2503 : /*
2504 : * All we have to do is pop the stack --- the actions/notifies made in
2505 : * this subxact are no longer interesting, and the space will be freed
2506 : * when CurTransactionContext is recycled. We still have to free the
2507 : * ActionList and NotificationList objects themselves, though, because
2508 : * those are allocated in TopTransactionContext.
2509 : *
2510 : * Note that there might be no entries at all, or no entries for the
2511 : * current subtransaction level, either because none were ever created, or
2512 : * because we reentered this routine due to trouble during subxact abort.
2513 : */
2514 9406 : while (pendingActions != NULL &&
2515 2 : pendingActions->nestingLevel >= my_level)
2516 : {
2517 2 : ActionList *childPendingActions = pendingActions;
2518 :
2519 2 : pendingActions = pendingActions->upper;
2520 2 : pfree(childPendingActions);
2521 : }
2522 :
2523 9406 : while (pendingNotifies != NULL &&
2524 4 : pendingNotifies->nestingLevel >= my_level)
2525 : {
2526 2 : NotificationList *childPendingNotifies = pendingNotifies;
2527 :
2528 2 : pendingNotifies = pendingNotifies->upper;
2529 2 : pfree(childPendingNotifies);
2530 : }
2531 9404 : }
2532 :
2533 : /*
2534 : * HandleNotifyInterrupt
2535 : *
2536 : * Signal handler portion of interrupt handling. Let the backend know
2537 : * that there's a pending notify interrupt. If we're currently reading
2538 : * from the client, this will interrupt the read and
2539 : * ProcessClientReadInterrupt() will call ProcessNotifyInterrupt().
2540 : */
2541 : void
2542 22 : HandleNotifyInterrupt(void)
2543 : {
2544 : /*
2545 : * Note: this is called by a SIGNAL HANDLER. You must be very wary what
2546 : * you do here.
2547 : */
2548 :
2549 : /* signal that work needs to be done */
2550 22 : notifyInterruptPending = true;
2551 :
2552 : /* make sure the event is processed in due course */
2553 22 : SetLatch(MyLatch);
2554 22 : }
2555 :
2556 : /*
2557 : * ProcessNotifyInterrupt
2558 : *
2559 : * This is called if we see notifyInterruptPending set, just before
2560 : * transmitting ReadyForQuery at the end of a frontend command, and
2561 : * also if a notify signal occurs while reading from the frontend.
2562 : * HandleNotifyInterrupt() will cause the read to be interrupted
2563 : * via the process's latch, and this routine will get called.
2564 : * If we are truly idle (ie, *not* inside a transaction block),
2565 : * process the incoming notifies.
2566 : *
2567 : * If "flush" is true, force any frontend messages out immediately.
2568 : * This can be false when being called at the end of a frontend command,
2569 : * since we'll flush after sending ReadyForQuery.
2570 : */
2571 : void
2572 74 : ProcessNotifyInterrupt(bool flush)
2573 : {
2574 74 : if (IsTransactionOrTransactionBlock())
2575 12 : return; /* not really idle */
2576 :
2577 : /* Loop in case another signal arrives while sending messages */
2578 124 : while (notifyInterruptPending)
2579 62 : ProcessIncomingNotify(flush);
2580 : }
2581 :
2582 :
2583 : /*
2584 : * Read all pending notifications from the queue, and deliver appropriate
2585 : * ones to my frontend. Stop when we reach queue head or an uncommitted
2586 : * notification.
2587 : */
2588 : static void
2589 92 : asyncQueueReadAllNotifications(void)
2590 : {
2591 : QueuePosition pos;
2592 : QueuePosition head;
2593 : Snapshot snapshot;
2594 :
2595 : /*
2596 : * Fetch current state, indicate to others that we have woken up, and that
2597 : * we are in process of advancing our position.
2598 : */
2599 92 : LWLockAcquire(NotifyQueueLock, LW_SHARED);
2600 : /* Assert checks that we have a valid state entry */
2601 : Assert(MyProcPid == QUEUE_BACKEND_PID(MyProcNumber));
2602 92 : QUEUE_BACKEND_WAKEUP_PENDING(MyProcNumber) = false;
2603 92 : pos = QUEUE_BACKEND_POS(MyProcNumber);
2604 92 : head = QUEUE_HEAD;
2605 :
2606 92 : if (QUEUE_POS_EQUAL(pos, head))
2607 : {
2608 : /* Nothing to do, we have read all notifications already. */
2609 0 : LWLockRelease(NotifyQueueLock);
2610 0 : return;
2611 : }
2612 :
2613 92 : QUEUE_BACKEND_IS_ADVANCING(MyProcNumber) = true;
2614 92 : LWLockRelease(NotifyQueueLock);
2615 :
2616 : /*----------
2617 : * Get snapshot we'll use to decide which xacts are still in progress.
2618 : * This is trickier than it might seem, because of race conditions.
2619 : * Consider the following example:
2620 : *
2621 : * Backend 1: Backend 2:
2622 : *
2623 : * transaction starts
2624 : * UPDATE foo SET ...;
2625 : * NOTIFY foo;
2626 : * commit starts
2627 : * queue the notify message
2628 : * transaction starts
2629 : * LISTEN foo; -- first LISTEN in session
2630 : * SELECT * FROM foo WHERE ...;
2631 : * commit to clog
2632 : * commit starts
2633 : * add backend 2 to array of listeners
2634 : * advance to queue head (this code)
2635 : * commit to clog
2636 : *
2637 : * Transaction 2's SELECT has not seen the UPDATE's effects, since that
2638 : * wasn't committed yet. Ideally we'd ensure that client 2 would
2639 : * eventually get transaction 1's notify message, but there's no way
2640 : * to do that; until we're in the listener array, there's no guarantee
2641 : * that the notify message doesn't get removed from the queue.
2642 : *
2643 : * Therefore the coding technique transaction 2 is using is unsafe:
2644 : * applications must commit a LISTEN before inspecting database state,
2645 : * if they want to ensure they will see notifications about subsequent
2646 : * changes to that state.
2647 : *
2648 : * What we do guarantee is that we'll see all notifications from
2649 : * transactions committing after the snapshot we take here.
2650 : * BecomeRegisteredListener has already added us to the listener array,
2651 : * so no not-yet-committed messages can be removed from the queue
2652 : * before we see them.
2653 : *----------
2654 : */
2655 92 : snapshot = RegisterSnapshot(GetLatestSnapshot());
2656 :
2657 : /*
2658 : * It is possible that we fail while trying to send a message to our
2659 : * frontend (for example, because of encoding conversion failure). If
2660 : * that happens it is critical that we not try to send the same message
2661 : * over and over again. Therefore, we set ExitOnAnyError to upgrade any
2662 : * ERRORs to FATAL, causing the client connection to be closed on error.
2663 : *
2664 : * We used to only skip over the offending message and try to soldier on,
2665 : * but it was somewhat questionable to lose a notification and give the
2666 : * client an ERROR instead. A client application is not be prepared for
2667 : * that and can't tell that a notification was missed. It was also not
2668 : * very useful in practice because notifications are often processed while
2669 : * a connection is idle and reading a message from the client, and in that
2670 : * state, any error is upgraded to FATAL anyway. Closing the connection
2671 : * is a clear signal to the application that it might have missed
2672 : * notifications.
2673 : */
2674 : {
2675 92 : bool save_ExitOnAnyError = ExitOnAnyError;
2676 : bool reachedStop;
2677 :
2678 92 : ExitOnAnyError = true;
2679 :
2680 : do
2681 : {
2682 : /*
2683 : * Process messages up to the stop position, end of page, or an
2684 : * uncommitted message.
2685 : *
2686 : * Our stop position is what we found to be the head's position
2687 : * when we entered this function. It might have changed already.
2688 : * But if it has, we will receive (or have already received and
2689 : * queued) another signal and come here again.
2690 : *
2691 : * We are not holding NotifyQueueLock here! The queue can only
2692 : * extend beyond the head pointer (see above) and we leave our
2693 : * backend's pointer where it is so nobody will truncate or
2694 : * rewrite pages under us. Especially we don't want to hold a lock
2695 : * while sending the notifications to the frontend.
2696 : */
2697 98 : reachedStop = asyncQueueProcessPageEntries(&pos, head, snapshot);
2698 98 : } while (!reachedStop);
2699 :
2700 : /* Update shared state */
2701 92 : LWLockAcquire(NotifyQueueLock, LW_SHARED);
2702 92 : QUEUE_BACKEND_POS(MyProcNumber) = pos;
2703 92 : QUEUE_BACKEND_IS_ADVANCING(MyProcNumber) = false;
2704 92 : LWLockRelease(NotifyQueueLock);
2705 :
2706 92 : ExitOnAnyError = save_ExitOnAnyError;
2707 : }
2708 :
2709 : /* Done with snapshot */
2710 92 : UnregisterSnapshot(snapshot);
2711 : }
2712 :
2713 : /*
2714 : * Fetch notifications from the shared queue, beginning at position current,
2715 : * and deliver relevant ones to my frontend.
2716 : *
2717 : * The function returns true once we have reached the stop position or an
2718 : * uncommitted notification, and false if we have finished with the page.
2719 : * In other words: once it returns true there is no need to look further.
2720 : * The QueuePosition *current is advanced past all processed messages.
2721 : */
2722 : static bool
2723 98 : asyncQueueProcessPageEntries(QueuePosition *current,
2724 : QueuePosition stop,
2725 : Snapshot snapshot)
2726 : {
2727 98 : int64 curpage = QUEUE_POS_PAGE(*current);
2728 : int slotno;
2729 : char *page_buffer;
2730 98 : bool reachedStop = false;
2731 : bool reachedEndOfPage;
2732 :
2733 : /*
2734 : * We copy the entries into a local buffer to avoid holding the SLRU lock
2735 : * while we transmit them to our frontend. The local buffer must be
2736 : * adequately aligned.
2737 : */
2738 : alignas(AsyncQueueEntry) char local_buf[QUEUE_PAGESIZE];
2739 98 : char *local_buf_end = local_buf;
2740 :
2741 98 : slotno = SimpleLruReadPage_ReadOnly(NotifyCtl, curpage,
2742 : InvalidTransactionId);
2743 98 : page_buffer = NotifyCtl->shared->page_buffer[slotno];
2744 :
2745 : do
2746 : {
2747 2654 : QueuePosition thisentry = *current;
2748 : AsyncQueueEntry *qe;
2749 :
2750 2654 : if (QUEUE_POS_EQUAL(thisentry, stop))
2751 92 : break;
2752 :
2753 2562 : qe = (AsyncQueueEntry *) (page_buffer + QUEUE_POS_OFFSET(thisentry));
2754 :
2755 : /*
2756 : * Advance *current over this message, possibly to the next page. As
2757 : * noted in the comments for asyncQueueReadAllNotifications, we must
2758 : * do this before possibly failing while processing the message.
2759 : */
2760 2562 : reachedEndOfPage = asyncQueueAdvance(current, qe->length);
2761 :
2762 : /* Ignore messages destined for other databases */
2763 2562 : if (qe->dboid == MyDatabaseId)
2764 : {
2765 2562 : if (XidInMVCCSnapshot(qe->xid, snapshot))
2766 : {
2767 : /*
2768 : * The source transaction is still in progress, so we can't
2769 : * process this message yet. Break out of the loop, but first
2770 : * back up *current so we will reprocess the message next
2771 : * time. (Note: it is unlikely but not impossible for
2772 : * TransactionIdDidCommit to fail, so we can't really avoid
2773 : * this advance-then-back-up behavior when dealing with an
2774 : * uncommitted message.)
2775 : *
2776 : * Note that we must test XidInMVCCSnapshot before we test
2777 : * TransactionIdDidCommit, else we might return a message from
2778 : * a transaction that is not yet visible to snapshots; compare
2779 : * the comments at the head of heapam_visibility.c.
2780 : *
2781 : * Also, while our own xact won't be listed in the snapshot,
2782 : * we need not check for TransactionIdIsCurrentTransactionId
2783 : * because our transaction cannot (yet) have queued any
2784 : * messages.
2785 : */
2786 0 : *current = thisentry;
2787 0 : reachedStop = true;
2788 0 : break;
2789 : }
2790 :
2791 : /*
2792 : * Quick check for the case that we're not listening on any
2793 : * channels, before calling TransactionIdDidCommit(). This makes
2794 : * that case a little faster, but more importantly, it ensures
2795 : * that if there's a bad entry in the queue for which
2796 : * TransactionIdDidCommit() fails for some reason, we can skip
2797 : * over it on the first LISTEN in a session, and not get stuck on
2798 : * it indefinitely. (This is a little trickier than it looks: it
2799 : * works because BecomeRegisteredListener runs this code before we
2800 : * have made the first entry in localChannelTable.)
2801 : */
2802 2562 : if (LocalChannelTableIsEmpty())
2803 2452 : continue;
2804 :
2805 110 : if (TransactionIdDidCommit(qe->xid))
2806 : {
2807 110 : memcpy(local_buf_end, qe, qe->length);
2808 110 : local_buf_end += qe->length;
2809 : }
2810 : else
2811 : {
2812 : /*
2813 : * The source transaction aborted or crashed, so we just
2814 : * ignore its notifications.
2815 : */
2816 : }
2817 : }
2818 :
2819 : /* Loop back if we're not at end of page */
2820 2562 : } while (!reachedEndOfPage);
2821 :
2822 : /* Release lock that we got from SimpleLruReadPage_ReadOnly() */
2823 98 : LWLockRelease(SimpleLruGetBankLock(NotifyCtl, curpage));
2824 :
2825 : /*
2826 : * Now that we have let go of the SLRU bank lock, send the notifications
2827 : * to our backend
2828 : */
2829 : Assert(local_buf_end - local_buf <= BLCKSZ);
2830 208 : for (char *p = local_buf; p < local_buf_end;)
2831 : {
2832 110 : AsyncQueueEntry *qe = (AsyncQueueEntry *) p;
2833 :
2834 : /* qe->data is the null-terminated channel name */
2835 110 : char *channel = qe->data;
2836 :
2837 110 : if (IsListeningOn(channel))
2838 : {
2839 : /* payload follows channel name */
2840 110 : char *payload = qe->data + strlen(channel) + 1;
2841 :
2842 110 : NotifyMyFrontEnd(channel, payload, qe->srcPid);
2843 : }
2844 :
2845 110 : p += qe->length;
2846 : }
2847 :
2848 98 : if (QUEUE_POS_EQUAL(*current, stop))
2849 92 : reachedStop = true;
2850 :
2851 98 : return reachedStop;
2852 : }
2853 :
2854 : /*
2855 : * Advance the shared queue tail variable to the minimum of all the
2856 : * per-backend tail pointers. Truncate pg_notify space if possible.
2857 : *
2858 : * This is (usually) called during CommitTransaction(), so it's important for
2859 : * it to have very low probability of failure.
2860 : */
2861 : static void
2862 26 : asyncQueueAdvanceTail(void)
2863 : {
2864 : QueuePosition min;
2865 : int64 oldtailpage;
2866 : int64 newtailpage;
2867 : int64 boundary;
2868 :
2869 : /* Restrict task to one backend per cluster; see SimpleLruTruncate(). */
2870 26 : LWLockAcquire(NotifyQueueTailLock, LW_EXCLUSIVE);
2871 :
2872 : /*
2873 : * Compute the new tail. Pre-v13, it's essential that QUEUE_TAIL be exact
2874 : * (ie, exactly match at least one backend's queue position), so it must
2875 : * be updated atomically with the actual computation. Since v13, we could
2876 : * get away with not doing it like that, but it seems prudent to keep it
2877 : * so.
2878 : *
2879 : * Also, because incoming backends will scan forward from QUEUE_TAIL, that
2880 : * must be advanced before we can truncate any data. Thus, QUEUE_TAIL is
2881 : * the logical tail, while QUEUE_STOP_PAGE is the physical tail, or oldest
2882 : * un-truncated page. When QUEUE_STOP_PAGE != QUEUE_POS_PAGE(QUEUE_TAIL),
2883 : * there are pages we can truncate but haven't yet finished doing so.
2884 : *
2885 : * For concurrency's sake, we don't want to hold NotifyQueueLock while
2886 : * performing SimpleLruTruncate. This is OK because no backend will try
2887 : * to access the pages we are in the midst of truncating.
2888 : */
2889 26 : LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
2890 26 : min = QUEUE_HEAD;
2891 46 : for (ProcNumber i = QUEUE_FIRST_LISTENER; i != INVALID_PROC_NUMBER; i = QUEUE_NEXT_LISTENER(i))
2892 : {
2893 : Assert(QUEUE_BACKEND_PID(i) != InvalidPid);
2894 20 : min = QUEUE_POS_MIN(min, QUEUE_BACKEND_POS(i));
2895 : }
2896 26 : QUEUE_TAIL = min;
2897 26 : oldtailpage = QUEUE_STOP_PAGE;
2898 26 : LWLockRelease(NotifyQueueLock);
2899 :
2900 : /*
2901 : * We can truncate something if the global tail advanced across an SLRU
2902 : * segment boundary.
2903 : *
2904 : * XXX it might be better to truncate only once every several segments, to
2905 : * reduce the number of directory scans.
2906 : */
2907 26 : newtailpage = QUEUE_POS_PAGE(min);
2908 26 : boundary = newtailpage - (newtailpage % SLRU_PAGES_PER_SEGMENT);
2909 26 : if (asyncQueuePagePrecedes(oldtailpage, boundary))
2910 : {
2911 : /*
2912 : * SimpleLruTruncate() will ask for SLRU bank locks but will also
2913 : * release the lock again.
2914 : */
2915 2 : SimpleLruTruncate(NotifyCtl, newtailpage);
2916 :
2917 2 : LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
2918 2 : QUEUE_STOP_PAGE = newtailpage;
2919 2 : LWLockRelease(NotifyQueueLock);
2920 : }
2921 :
2922 26 : LWLockRelease(NotifyQueueTailLock);
2923 26 : }
2924 :
2925 : /*
2926 : * AsyncNotifyFreezeXids
2927 : *
2928 : * Prepare the async notification queue for CLOG truncation by freezing
2929 : * transaction IDs that are about to become inaccessible.
2930 : *
2931 : * This function is called by VACUUM before advancing datfrozenxid. It scans
2932 : * the notification queue and replaces XIDs that would become inaccessible
2933 : * after CLOG truncation with special markers:
2934 : * - Committed transactions are set to FrozenTransactionId
2935 : * - Aborted/crashed transactions are set to InvalidTransactionId
2936 : *
2937 : * Only XIDs < newFrozenXid are processed, as those are the ones whose CLOG
2938 : * pages will be truncated. If XID < newFrozenXid, it cannot still be running
2939 : * (or it would have held back newFrozenXid through ProcArray).
2940 : * Therefore, if TransactionIdDidCommit returns false, we know the transaction
2941 : * either aborted explicitly or crashed, and we can safely mark it invalid.
2942 : */
2943 : void
2944 1880 : AsyncNotifyFreezeXids(TransactionId newFrozenXid)
2945 : {
2946 : QueuePosition pos;
2947 : QueuePosition head;
2948 1880 : int64 curpage = -1;
2949 1880 : int slotno = -1;
2950 1880 : char *page_buffer = NULL;
2951 1880 : bool page_dirty = false;
2952 :
2953 : /*
2954 : * Acquire locks in the correct order to avoid deadlocks. As per the
2955 : * locking protocol: NotifyQueueTailLock, then NotifyQueueLock, then SLRU
2956 : * bank locks.
2957 : *
2958 : * We only need SHARED mode since we're just reading the head/tail
2959 : * positions, not modifying them.
2960 : */
2961 1880 : LWLockAcquire(NotifyQueueTailLock, LW_SHARED);
2962 1880 : LWLockAcquire(NotifyQueueLock, LW_SHARED);
2963 :
2964 1880 : pos = QUEUE_TAIL;
2965 1880 : head = QUEUE_HEAD;
2966 :
2967 : /* Release NotifyQueueLock early, we only needed to read the positions */
2968 1880 : LWLockRelease(NotifyQueueLock);
2969 :
2970 : /*
2971 : * Scan the queue from tail to head, freezing XIDs as needed. We hold
2972 : * NotifyQueueTailLock throughout to ensure the tail doesn't move while
2973 : * we're working.
2974 : */
2975 1940 : while (!QUEUE_POS_EQUAL(pos, head))
2976 : {
2977 : AsyncQueueEntry *qe;
2978 : TransactionId xid;
2979 60 : int64 pageno = QUEUE_POS_PAGE(pos);
2980 60 : int offset = QUEUE_POS_OFFSET(pos);
2981 :
2982 : /* If we need a different page, release old lock and get new one */
2983 60 : if (pageno != curpage)
2984 : {
2985 : LWLock *lock;
2986 :
2987 : /* Release previous page if any */
2988 6 : if (slotno >= 0)
2989 : {
2990 0 : if (page_dirty)
2991 : {
2992 0 : NotifyCtl->shared->page_dirty[slotno] = true;
2993 0 : page_dirty = false;
2994 : }
2995 0 : LWLockRelease(SimpleLruGetBankLock(NotifyCtl, curpage));
2996 : }
2997 :
2998 6 : lock = SimpleLruGetBankLock(NotifyCtl, pageno);
2999 6 : LWLockAcquire(lock, LW_EXCLUSIVE);
3000 6 : slotno = SimpleLruReadPage(NotifyCtl, pageno, true,
3001 : InvalidTransactionId);
3002 6 : page_buffer = NotifyCtl->shared->page_buffer[slotno];
3003 6 : curpage = pageno;
3004 : }
3005 :
3006 60 : qe = (AsyncQueueEntry *) (page_buffer + offset);
3007 60 : xid = qe->xid;
3008 :
3009 120 : if (TransactionIdIsNormal(xid) &&
3010 60 : TransactionIdPrecedes(xid, newFrozenXid))
3011 : {
3012 20 : if (TransactionIdDidCommit(xid))
3013 : {
3014 20 : qe->xid = FrozenTransactionId;
3015 20 : page_dirty = true;
3016 : }
3017 : else
3018 : {
3019 0 : qe->xid = InvalidTransactionId;
3020 0 : page_dirty = true;
3021 : }
3022 : }
3023 :
3024 : /* Advance to next entry */
3025 60 : asyncQueueAdvance(&pos, qe->length);
3026 : }
3027 :
3028 : /* Release final page lock if we acquired one */
3029 1880 : if (slotno >= 0)
3030 : {
3031 6 : if (page_dirty)
3032 2 : NotifyCtl->shared->page_dirty[slotno] = true;
3033 6 : LWLockRelease(SimpleLruGetBankLock(NotifyCtl, curpage));
3034 : }
3035 :
3036 1880 : LWLockRelease(NotifyQueueTailLock);
3037 1880 : }
3038 :
3039 : /*
3040 : * ProcessIncomingNotify
3041 : *
3042 : * Scan the queue for arriving notifications and report them to the front
3043 : * end. The notifications might be from other sessions, or our own;
3044 : * there's no need to distinguish here.
3045 : *
3046 : * If "flush" is true, force any frontend messages out immediately.
3047 : *
3048 : * NOTE: since we are outside any transaction, we must create our own.
3049 : */
3050 : static void
3051 62 : ProcessIncomingNotify(bool flush)
3052 : {
3053 : /* We *must* reset the flag */
3054 62 : notifyInterruptPending = false;
3055 :
3056 : /* Do nothing else if we aren't actively listening */
3057 62 : if (LocalChannelTableIsEmpty())
3058 0 : return;
3059 :
3060 62 : if (Trace_notify)
3061 0 : elog(DEBUG1, "ProcessIncomingNotify");
3062 :
3063 62 : set_ps_display("notify interrupt");
3064 :
3065 : /*
3066 : * We must run asyncQueueReadAllNotifications inside a transaction, else
3067 : * bad things happen if it gets an error.
3068 : */
3069 62 : StartTransactionCommand();
3070 :
3071 62 : asyncQueueReadAllNotifications();
3072 :
3073 62 : CommitTransactionCommand();
3074 :
3075 : /*
3076 : * If this isn't an end-of-command case, we must flush the notify messages
3077 : * to ensure frontend gets them promptly.
3078 : */
3079 62 : if (flush)
3080 18 : pq_flush();
3081 :
3082 62 : set_ps_display("idle");
3083 :
3084 62 : if (Trace_notify)
3085 0 : elog(DEBUG1, "ProcessIncomingNotify: done");
3086 : }
3087 :
3088 : /*
3089 : * Send NOTIFY message to my front end.
3090 : */
3091 : void
3092 110 : NotifyMyFrontEnd(const char *channel, const char *payload, int32 srcPid)
3093 : {
3094 110 : if (whereToSendOutput == DestRemote)
3095 : {
3096 : StringInfoData buf;
3097 :
3098 110 : pq_beginmessage(&buf, PqMsg_NotificationResponse);
3099 110 : pq_sendint32(&buf, srcPid);
3100 110 : pq_sendstring(&buf, channel);
3101 110 : pq_sendstring(&buf, payload);
3102 110 : pq_endmessage(&buf);
3103 :
3104 : /*
3105 : * NOTE: we do not do pq_flush() here. Some level of caller will
3106 : * handle it later, allowing this message to be combined into a packet
3107 : * with other ones.
3108 : */
3109 : }
3110 : else
3111 0 : elog(INFO, "NOTIFY for \"%s\" payload \"%s\"", channel, payload);
3112 110 : }
3113 :
3114 : /* Does pendingNotifies include a match for the given event? */
3115 : static bool
3116 2132 : AsyncExistsPendingNotify(Notification *n)
3117 : {
3118 2132 : if (pendingNotifies == NULL)
3119 0 : return false;
3120 :
3121 2132 : if (pendingNotifies->hashtab != NULL)
3122 : {
3123 : /* Use the hash table to probe for a match */
3124 1968 : if (hash_search(pendingNotifies->hashtab,
3125 : &n,
3126 : HASH_FIND,
3127 : NULL))
3128 2 : return true;
3129 : }
3130 : else
3131 : {
3132 : /* Must scan the event list */
3133 : ListCell *l;
3134 :
3135 850 : foreach(l, pendingNotifies->events)
3136 : {
3137 714 : Notification *oldn = (Notification *) lfirst(l);
3138 :
3139 714 : if (n->channel_len == oldn->channel_len &&
3140 714 : n->payload_len == oldn->payload_len &&
3141 380 : memcmp(n->data, oldn->data,
3142 380 : n->channel_len + n->payload_len + 2) == 0)
3143 28 : return true;
3144 : }
3145 : }
3146 :
3147 2102 : return false;
3148 : }
3149 :
3150 : /*
3151 : * Add a notification event to a pre-existing pendingNotifies list.
3152 : *
3153 : * Because pendingNotifies->events is already nonempty, this works
3154 : * correctly no matter what CurrentMemoryContext is.
3155 : */
3156 : static void
3157 2102 : AddEventToPendingNotifies(Notification *n)
3158 : {
3159 : Assert(pendingNotifies->events != NIL);
3160 :
3161 : /* Create the hash tables if it's time to */
3162 2102 : if (list_length(pendingNotifies->events) >= MIN_HASHABLE_NOTIFIES &&
3163 1970 : pendingNotifies->hashtab == NULL)
3164 : {
3165 : HASHCTL hash_ctl;
3166 : ListCell *l;
3167 :
3168 : /* Create the hash table */
3169 4 : hash_ctl.keysize = sizeof(Notification *);
3170 4 : hash_ctl.entrysize = sizeof(struct NotificationHash);
3171 4 : hash_ctl.hash = notification_hash;
3172 4 : hash_ctl.match = notification_match;
3173 4 : hash_ctl.hcxt = CurTransactionContext;
3174 8 : pendingNotifies->hashtab =
3175 4 : hash_create("Pending Notifies",
3176 : 256L,
3177 : &hash_ctl,
3178 : HASH_ELEM | HASH_FUNCTION | HASH_COMPARE | HASH_CONTEXT);
3179 :
3180 : /* Create the unique channel name table */
3181 : Assert(pendingNotifies->uniqueChannelHash == NULL);
3182 4 : hash_ctl.keysize = NAMEDATALEN;
3183 4 : hash_ctl.entrysize = sizeof(ChannelName);
3184 4 : hash_ctl.hcxt = CurTransactionContext;
3185 8 : pendingNotifies->uniqueChannelHash =
3186 4 : hash_create("Pending Notify Channel Names",
3187 : 64L,
3188 : &hash_ctl,
3189 : HASH_ELEM | HASH_STRINGS | HASH_CONTEXT);
3190 :
3191 : /* Insert all the already-existing events */
3192 68 : foreach(l, pendingNotifies->events)
3193 : {
3194 64 : Notification *oldn = (Notification *) lfirst(l);
3195 64 : char *channel = oldn->data;
3196 : bool found;
3197 :
3198 64 : (void) hash_search(pendingNotifies->hashtab,
3199 : &oldn,
3200 : HASH_ENTER,
3201 : &found);
3202 : Assert(!found);
3203 :
3204 : /* Add channel name to uniqueChannelHash; might be there already */
3205 64 : (void) hash_search(pendingNotifies->uniqueChannelHash,
3206 : channel,
3207 : HASH_ENTER,
3208 : NULL);
3209 : }
3210 : }
3211 :
3212 : /* Add new event to the list, in order */
3213 2102 : pendingNotifies->events = lappend(pendingNotifies->events, n);
3214 :
3215 : /* Add event to the hash tables if needed */
3216 2102 : if (pendingNotifies->hashtab != NULL)
3217 : {
3218 1970 : char *channel = n->data;
3219 : bool found;
3220 :
3221 1970 : (void) hash_search(pendingNotifies->hashtab,
3222 : &n,
3223 : HASH_ENTER,
3224 : &found);
3225 : Assert(!found);
3226 :
3227 : /* Add channel name to uniqueChannelHash; might be there already */
3228 1970 : (void) hash_search(pendingNotifies->uniqueChannelHash,
3229 : channel,
3230 : HASH_ENTER,
3231 : NULL);
3232 : }
3233 2102 : }
3234 :
3235 : /*
3236 : * notification_hash: hash function for notification hash table
3237 : *
3238 : * The hash "keys" are pointers to Notification structs.
3239 : */
3240 : static uint32
3241 4002 : notification_hash(const void *key, Size keysize)
3242 : {
3243 4002 : const Notification *k = *(const Notification *const *) key;
3244 :
3245 : Assert(keysize == sizeof(Notification *));
3246 : /* We don't bother to include the payload's trailing null in the hash */
3247 4002 : return DatumGetUInt32(hash_any((const unsigned char *) k->data,
3248 4002 : k->channel_len + k->payload_len + 1));
3249 : }
3250 :
3251 : /*
3252 : * notification_match: match function to use with notification_hash
3253 : */
3254 : static int
3255 2 : notification_match(const void *key1, const void *key2, Size keysize)
3256 : {
3257 2 : const Notification *k1 = *(const Notification *const *) key1;
3258 2 : const Notification *k2 = *(const Notification *const *) key2;
3259 :
3260 : Assert(keysize == sizeof(Notification *));
3261 2 : if (k1->channel_len == k2->channel_len &&
3262 2 : k1->payload_len == k2->payload_len &&
3263 2 : memcmp(k1->data, k2->data,
3264 2 : k1->channel_len + k1->payload_len + 2) == 0)
3265 2 : return 0; /* equal */
3266 0 : return 1; /* not equal */
3267 : }
3268 :
3269 : /* Clear the pendingActions and pendingNotifies lists. */
3270 : static void
3271 52592 : ClearPendingActionsAndNotifies(void)
3272 : {
3273 : /*
3274 : * Everything's allocated in either TopTransactionContext or the context
3275 : * for the subtransaction to which it corresponds. So, there's nothing to
3276 : * do here except reset the pointers; the space will be reclaimed when the
3277 : * contexts are deleted.
3278 : */
3279 52592 : pendingActions = NULL;
3280 52592 : pendingNotifies = NULL;
3281 : /* Also clear pendingListenActions, which is derived from pendingActions */
3282 52592 : pendingListenActions = NULL;
3283 52592 : }
3284 :
3285 : /*
3286 : * GUC check_hook for notify_buffers
3287 : */
3288 : bool
3289 2354 : check_notify_buffers(int *newval, void **extra, GucSource source)
3290 : {
3291 2354 : return check_slru_buffers("notify_buffers", newval);
3292 : }
|
