Skip to content

Commit eba917d

Browse files
hlinnakahlinnaka
committed
Fix bug where we truncated CLOG that was still needed by LISTEN/NOTIFY
The async notification queue contains the XID of the sender, and when processing notifications we call TransactionIdDidCommit() on the XID. But we had no safeguards to prevent the CLOG segments containing those XIDs from being truncated away. As a result, if a backend didn't for some reason process its notifications for a long time, or when a new backend issued LISTEN, you could get an error like: test=# listen c21; ERROR: 58P01: could not access status of transaction 14279685 DETAIL: Could not open file "pg_xact/000D": No such file or directory. LOCATION: SlruReportIOError, slru.c:1087 To fix, make VACUUM "freeze" the XIDs in the async notification queue before truncating the CLOG. Old XIDs are replaced with FrozenTransactionId or InvalidTransactionId. Note: This commit is not a full fix. A race condition remains, where a backend is executing asyncQueueReadAllNotifications() and has just made a local copy of an async SLRU page which contains old XIDs, while vacuum concurrently truncates the CLOG covering those XIDs. When the backend then calls TransactionIdDidCommit() on those XIDs from the local copy, you still get the error. The next commit will fix that remaining race condition. This was first reported by Sergey Zhuravlev in 2021, with many other people hitting the same issue later. Thanks to: - Alexandra Wang, Daniil Davydov, Andrei Varashen and Jacques Combrink for investigating and providing reproducable test cases, - Matheus Alcantara and Arseniy Mukhin for review and earlier proposed patches to fix this, - Álvaro Herrera and Masahiko Sawada for reviews, - Yura Sokolov aka funny-falcon for the idea of marking transactions as committed in the notification queue, and - Joel Jacobson for the final patch version. I hope I didn't forget anyone. Backpatch to all supported versions. I believe the bug goes back all the way to commit d1e0272, which introduced the SLRU-based async notification queue. Discussion: https://www.postgresql.org/message-id/16961-25f29f95b3604a8a@postgresql.org Discussion: https://www.postgresql.org/message-id/18804-bccbbde5e77a68c2@postgresql.org Discussion: https://www.postgresql.org/message-id/CAK98qZ3wZLE-RZJN_Y%2BTFjiTRPPFPBwNBpBi5K5CU8hUHkzDpw@mail.gmail.com Backpatch-through: 14
1 parent 7cb05dd commit eba917d

File tree

3 files changed

+121
-0
lines changed

3 files changed

+121
-0
lines changed

‎src/backend/commands/async.c‎

Lines changed: 111 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2244,6 +2244,117 @@ asyncQueueAdvanceTail(void)
22442244
LWLockRelease(NotifyQueueTailLock);
22452245
}
22462246

2247+
/*
2248+
* AsyncNotifyFreezeXids
2249+
*
2250+
* Prepare the async notification queue for CLOG truncation by freezing
2251+
* transaction IDs that are about to become inaccessible.
2252+
*
2253+
* This function is called by VACUUM before advancing datfrozenxid. It scans
2254+
* the notification queue and replaces XIDs that would become inaccessible
2255+
* after CLOG truncation with special markers:
2256+
* - Committed transactions are set to FrozenTransactionId
2257+
* - Aborted/crashed transactions are set to InvalidTransactionId
2258+
*
2259+
* Only XIDs < newFrozenXid are processed, as those are the ones whose CLOG
2260+
* pages will be truncated. If XID < newFrozenXid, it cannot still be running
2261+
* (or it would have held back newFrozenXid through ProcArray).
2262+
* Therefore, if TransactionIdDidCommit returns false, we know the transaction
2263+
* either aborted explicitly or crashed, and we can safely mark it invalid.
2264+
*/
2265+
void
2266+
AsyncNotifyFreezeXids(TransactionId newFrozenXid)
2267+
{
2268+
QueuePosition pos;
2269+
QueuePosition head;
2270+
int64 curpage = -1;
2271+
int slotno = -1;
2272+
char *page_buffer = NULL;
2273+
bool page_dirty = false;
2274+
2275+
/*
2276+
* Acquire locks in the correct order to avoid deadlocks. As per the
2277+
* locking protocol: NotifyQueueTailLock, then NotifyQueueLock, then
2278+
* NotifySLRULock.
2279+
*
2280+
* We only need SHARED mode since we're just reading the head/tail
2281+
* positions, not modifying them.
2282+
*/
2283+
LWLockAcquire(NotifyQueueTailLock, LW_SHARED);
2284+
LWLockAcquire(NotifyQueueLock, LW_SHARED);
2285+
2286+
pos = QUEUE_TAIL;
2287+
head = QUEUE_HEAD;
2288+
2289+
/* Release NotifyQueueLock early, we only needed to read the positions */
2290+
LWLockRelease(NotifyQueueLock);
2291+
2292+
/*
2293+
* Scan the queue from tail to head, freezing XIDs as needed. We hold
2294+
* NotifyQueueTailLock throughout to ensure the tail doesn't move while
2295+
* we're working.
2296+
*/
2297+
while (!QUEUE_POS_EQUAL(pos, head))
2298+
{
2299+
AsyncQueueEntry *qe;
2300+
TransactionId xid;
2301+
int64 pageno = QUEUE_POS_PAGE(pos);
2302+
int offset = QUEUE_POS_OFFSET(pos);
2303+
2304+
/* If we need a different page, release old lock and get new one */
2305+
if (pageno != curpage)
2306+
{
2307+
/* Release previous page if any */
2308+
if (slotno >= 0)
2309+
{
2310+
if (page_dirty)
2311+
{
2312+
NotifyCtl->shared->page_dirty[slotno] = true;
2313+
page_dirty = false;
2314+
}
2315+
LWLockRelease(NotifySLRULock);
2316+
}
2317+
2318+
LWLockAcquire(NotifySLRULock, LW_EXCLUSIVE);
2319+
slotno = SimpleLruReadPage(NotifyCtl, pageno, true,
2320+
InvalidTransactionId);
2321+
page_buffer = NotifyCtl->shared->page_buffer[slotno];
2322+
curpage = pageno;
2323+
}
2324+
2325+
qe = (AsyncQueueEntry *) (page_buffer + offset);
2326+
xid = qe->xid;
2327+
2328+
if (TransactionIdIsNormal(xid) &&
2329+
TransactionIdPrecedes(xid, newFrozenXid))
2330+
{
2331+
if (TransactionIdDidCommit(xid))
2332+
{
2333+
qe->xid = FrozenTransactionId;
2334+
page_dirty = true;
2335+
}
2336+
else
2337+
{
2338+
qe->xid = InvalidTransactionId;
2339+
page_dirty = true;
2340+
}
2341+
}
2342+
2343+
/* Advance to next entry */
2344+
asyncQueueAdvance(&pos, qe->length);
2345+
}
2346+
2347+
/* Release final page lock if we acquired one */
2348+
if (slotno >= 0)
2349+
{
2350+
if (page_dirty)
2351+
NotifyCtl->shared->page_dirty[slotno] = true;
2352+
LWLockRelease(NotifySLRULock);
2353+
}
2354+
2355+
LWLockRelease(NotifyQueueTailLock);
2356+
}
2357+
22472358
/*
22482359
* ProcessIncomingNotify
22492360
*

‎src/backend/commands/vacuum.c‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -35,6 +35,7 @@
3535
#include "catalog/pg_database.h"
3636
#include "catalog/pg_inherits.h"
3737
#include "catalog/pg_namespace.h"
38+
#include "commands/async.h"
3839
#include "commands/cluster.h"
3940
#include "commands/defrem.h"
4041
#include "commands/vacuum.h"
@@ -1788,6 +1789,12 @@ vac_truncate_clog(TransactionId frozenXID,
17881789
return;
17891790
}
17901791

1792+
/*
1793+
* Freeze any old transaction IDs in the async notification queue before
1794+
* CLOG truncation.
1795+
*/
1796+
AsyncNotifyFreezeXids(frozenXID);
1797+
17911798
/*
17921799
* Advance the oldest value for commit timestamps before truncating, so
17931800
* that if a user requests a timestamp for a transaction we're truncating

‎src/include/commands/async.h‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -51,4 +51,7 @@ extern void HandleNotifyInterrupt(void);
5151
/* process interrupts */
5252
extern void ProcessNotifyInterrupt(bool flush);
5353

54+
/* freeze old transaction IDs in notify queue (called by VACUUM) */
55+
extern void AsyncNotifyFreezeXids(TransactionId newFrozenXid);
56+
5457
#endif /* ASYNC_H */

0 commit comments

Comments
 (0)