Skip to content

Commit

Permalink
Fix visibility check when XID is committed in CLOG but not in procarray.
Browse files Browse the repository at this point in the history
TransactionIdIsInProgress had a fast path to return 'false' if the
single-item CLOG cache said that the transaction was known to be
committed. However, that was wrong, because a transaction is first
marked as committed in the CLOG but doesn't become visible to others
until it has removed its XID from the proc array. That could lead to an
error:

    ERROR:  t_xmin is uncommitted in tuple to be updated

or for an UPDATE to go ahead without blocking, before the previous
UPDATE on the same row was made visible.

The window is usually very short, but synchronous replication makes it
much wider, because the wait for synchronous replica happens in that
window.

Another thing that makes it hard to hit is that it's hard to get such
a commit-in-progress transaction into the single item CLOG cache.
Normally, if you call TransactionIdIsInProgress on such a transaction,
it determines that the XID is in progress without checking the CLOG
and without populating the cache. One way to prime the cache is to
explicitly call pg_xact_status() on the XID. Another way is to use a
lot of subtransactions, so that the subxid cache in the proc array is
overflown, making TransactionIdIsInProgress rely on pg_subtrans and
CLOG checks.

This has been broken ever since it was introduced in 2008, but the race
condition is very hard to hit, especially without synchronous
replication. There were a couple of reports of the error starting from
summer 2021, but no one was able to find the root cause then.

TransactionIdIsKnownCompleted() is now unused. In 'master', remove it,
but I left it in place in backbranches in case it's used by extensions.

Also change pg_xact_status() to check TransactionIdIsInProgress().
Previously, it only checked the CLOG, and returned "committed" before
the transaction was actually made visible to other queries. Note that
this also means that you cannot use pg_xact_status() to reproduce the
bug anymore, even if the code wasn't fixed.

Report and analysis by Konstantin Knizhnik. Patch by Simon Riggs, with
the pg_xact_status() change added by me.

Author: Simon Riggs
Reviewed-by: Andres Freund
Discussion: https://www.postgresql.org/message-id/flat/4da7913d-398c-e2ad-d777-f752cf7f0bbb%40garret.ru
  • Loading branch information
hlinnaka committed Jun 27, 2022
1 parent ca590a4 commit 4822b46
Show file tree
Hide file tree
Showing 3 changed files with 32 additions and 23 deletions.
13 changes: 9 additions & 4 deletions src/backend/access/transam/transam.c
Expand Up @@ -226,10 +226,15 @@ TransactionIdDidAbort(TransactionId transactionId)
*
* This does NOT look into pg_xact but merely probes our local cache
* (and so it's not named TransactionIdDidComplete, which would be the
* appropriate name for a function that worked that way). The intended
* use is just to short-circuit TransactionIdIsInProgress calls when doing
* repeated tqual.c checks for the same XID. If this isn't extremely fast
* then it will be counterproductive.
* appropriate name for a function that worked that way).
*
* NB: This is unused, and will be removed in v15. This was used to
* short-circuit TransactionIdIsInProgress, but that was wrong for a
* transaction that was known to be marked as committed in CLOG but not
* yet removed from the proc array. This is kept in backbranches just in
* case it is still used by extensions. However, extensions doing
* something similar to tuple visibility checks should also be careful to
* check the proc array first!
*
* Note:
* Assumes transaction identifier is valid.
Expand Down
12 changes: 11 additions & 1 deletion src/backend/storage/ipc/procarray.c
Expand Up @@ -100,6 +100,11 @@ static ProcArrayStruct *procArray;
static PGPROC *allProcs;
static PGXACT *allPgXact;

/*
* Cache to reduce overhead of repeated calls to TransactionIdIsInProgress()
*/
static TransactionId cachedXidIsNotInProgress = InvalidTransactionId;

/*
* Bookkeeping for tracking emulated transactions in recovery
*/
Expand Down Expand Up @@ -1051,7 +1056,7 @@ TransactionIdIsInProgress(TransactionId xid)
* already known to be completed, we can fall out without any access to
* shared memory.
*/
if (TransactionIdIsKnownCompleted(xid))
if (TransactionIdEquals(cachedXidIsNotInProgress, xid))
{
xc_by_known_xact_inc();
return false;
Expand Down Expand Up @@ -1198,6 +1203,7 @@ TransactionIdIsInProgress(TransactionId xid)
if (nxids == 0)
{
xc_no_overflow_inc();
cachedXidIsNotInProgress = xid;
return false;
}

Expand All @@ -1212,7 +1218,10 @@ TransactionIdIsInProgress(TransactionId xid)
xc_slow_answer_inc();

if (TransactionIdDidAbort(xid))
{
cachedXidIsNotInProgress = xid;
return false;
}

/*
* It isn't aborted, so check whether the transaction tree it belongs to
Expand All @@ -1230,6 +1239,7 @@ TransactionIdIsInProgress(TransactionId xid)
}
}

cachedXidIsNotInProgress = xid;
return false;
}

Expand Down
30 changes: 12 additions & 18 deletions src/backend/utils/adt/txid.c
Expand Up @@ -30,6 +30,7 @@
#include "libpq/pqformat.h"
#include "postmaster/postmaster.h"
#include "storage/lwlock.h"
#include "storage/procarray.h"
#include "utils/builtins.h"
#include "utils/memutils.h"
#include "utils/snapmgr.h"
Expand Down Expand Up @@ -754,29 +755,22 @@ txid_status(PG_FUNCTION_ARGS)
{
Assert(TransactionIdIsValid(xid));

if (TransactionIdIsCurrentTransactionId(xid))
/*
* Like when doing visiblity checks on a row, check whether the
* transaction is still in progress before looking into the CLOG.
* Otherwise we would incorrectly return "committed" for a transaction
* that is committing and has already updated the CLOG, but hasn't
* removed its XID from the proc array yet. (See comment on that race
* condition at the top of heapam_visibility.c)
*/
if (TransactionIdIsInProgress(xid))
status = "in progress";
else if (TransactionIdDidCommit(xid))
status = "committed";
else if (TransactionIdDidAbort(xid))
status = "aborted";
else
{
/*
* The xact is not marked as either committed or aborted in clog.
*
* It could be a transaction that ended without updating clog or
* writing an abort record due to a crash. We can safely assume
* it's aborted if it isn't committed and is older than our
* snapshot xmin.
*
* Otherwise it must be in-progress (or have been at the time we
* checked commit/abort status).
*/
if (TransactionIdPrecedes(xid, GetActiveSnapshot()->xmin))
status = "aborted";
else
status = "in progress";
/* it must have aborted or crashed */
status = "aborted";
}
}
else
Expand Down

0 comments on commit 4822b46

Please sign in to comment.