Resolve timing issue with logging locks for Hot Standby.

We log AccessExclusiveLocks for replay onto standby nodes,
but because of timing issues on ProcArray it is possible to
log a lock that is still held by a just committed transaction
that is very soon to be removed. To avoid any timing issue we
avoid applying locks made by transactions with InvalidXid.

Simon Riggs, bug report Tom Lane, diagnosis Pavan Deolasee

src/backend/storage/ipc/procarray.c

@@ -466,7 +466,7 @@ ProcArrayApplyRecoveryInfo(RunningTransactions running)
 	 * Remove stale transactions, if any.
 	 */
 	ExpireOldKnownAssignedTransactionIds(running->oldestRunningXid);
-	StandbyReleaseOldLocks(running->oldestRunningXid);
+	StandbyReleaseOldLocks(running->xcnt, running->xids);
 
 	/*
 	 * If our snapshot is already valid, nothing else to do...
@@ -520,12 +520,6 @@ ProcArrayApplyRecoveryInfo(RunningTransactions running)
 	 * OK, we need to initialise from the RunningTransactionsData record
 	 */
 
-	/*
-	 * Release any locks belonging to old transactions that are not running
-	 * according to the running-xacts record.
-	 */
-	StandbyReleaseOldLocks(running->nextXid);
-
 	/*
 	 * Nobody else is running yet, but take locks anyhow
 	 */

src/backend/storage/ipc/standby.c

@@ -524,7 +524,9 @@ StandbyAcquireAccessExclusiveLock(TransactionId xid, Oid dbOid, Oid relOid)
 	LOCKTAG		locktag;
 
 	/* Already processed? */
-	if (TransactionIdDidCommit(xid) || TransactionIdDidAbort(xid))
+	if (!TransactionIdIsValid(xid) ||
+		TransactionIdDidCommit(xid) ||
+		TransactionIdDidAbort(xid))
 		return;
 
 	elog(trace_recovery(DEBUG4),
@@ -606,23 +608,18 @@ StandbyReleaseLockTree(TransactionId xid, int nsubxids, TransactionId *subxids)
 }
 
 /*
- * StandbyReleaseLocksMany
- *		Release standby locks held by XIDs < removeXid
- *
- * If keepPreparedXacts is true, keep prepared transactions even if
- * they're older than removeXid
+ * Called at end of recovery and when we see a shutdown checkpoint.
  */
-static void
-StandbyReleaseLocksMany(TransactionId removeXid, bool keepPreparedXacts)
+void
+StandbyReleaseAllLocks(void)
 {
 	ListCell   *cell,
 			   *prev,
 			   *next;
 	LOCKTAG		locktag;
 
-	/*
-	 * Release all matching locks.
-	 */
+	elog(trace_recovery(DEBUG2), "release all standby locks");
+
 	prev = NULL;
 	for (cell = list_head(RecoveryLockList); cell; cell = next)
 	{
@@ -630,10 +627,6 @@ StandbyReleaseLocksMany(TransactionId removeXid, bool keepPreparedXacts)
 
 		next = lnext(cell);
 
-		if (!TransactionIdIsValid(removeXid) || TransactionIdPrecedes(lock->xid, removeXid))
-		{
-			if (keepPreparedXacts && StandbyTransactionIdIsPrepared(lock->xid))
-				continue;
 		elog(trace_recovery(DEBUG4),
 			 "releasing recovery lock: xid %u db %u rel %u",
 			 lock->xid, lock->dbOid, lock->relOid);
@@ -645,30 +638,70 @@ StandbyReleaseLocksMany(TransactionId removeXid, bool keepPreparedXacts)
 		RecoveryLockList = list_delete_cell(RecoveryLockList, cell, prev);
 		pfree(lock);
 	}
-		else
-			prev = cell;
-	}
 }
 
 /*
- * Called at end of recovery and when we see a shutdown checkpoint.
+ * StandbyReleaseOldLocks
+ *		Release standby locks held by XIDs that aren't running, as long
+ *		as they're not prepared transactions.
  */
 void
-StandbyReleaseAllLocks(void)
+StandbyReleaseOldLocks(int nxids, TransactionId *xids)
 {
-	elog(trace_recovery(DEBUG2), "release all standby locks");
-	StandbyReleaseLocksMany(InvalidTransactionId, false);
+	ListCell   *cell,
+			   *prev,
+			   *next;
+	LOCKTAG		locktag;
+
+	prev = NULL;
+	for (cell = list_head(RecoveryLockList); cell; cell = next)
+	{
+		xl_standby_lock *lock = (xl_standby_lock *) lfirst(cell);
+		bool	remove = false;
+
+		next = lnext(cell);
+
+		Assert(TransactionIdIsValid(lock->xid));
+
+		if (StandbyTransactionIdIsPrepared(lock->xid))
+			remove = false;
+		else
+		{
+			int		i;
+			bool	found = false;
+
+			for (i = 0; i < nxids; i++)
+			{
+				if (lock->xid == xids[i])
+				{
+					found = true;
+					break;
+				}
 			}
 
 			/*
- * StandbyReleaseOldLocks
- *		Release standby locks held by XIDs < removeXid, as long
- *		as they're not prepared transactions.
+			 * If its not a running transaction, remove it.
 			 */
-void
-StandbyReleaseOldLocks(TransactionId removeXid)
+			if (!found)
+				remove = true;
+		}
+
+		if (remove)
 		{
-	StandbyReleaseLocksMany(removeXid, true);
+			elog(trace_recovery(DEBUG4),
+				 "releasing recovery lock: xid %u db %u rel %u",
+				 lock->xid, lock->dbOid, lock->relOid);
+			SET_LOCKTAG_RELATION(locktag, lock->dbOid, lock->relOid);
+			if (!LockRelease(&locktag, AccessExclusiveLock, true))
+				elog(LOG,
+					 "RecoveryLockList contains entry for lock no longer recorded by lock manager: xid %u database %u relation %u",
+					 lock->xid, lock->dbOid, lock->relOid);
+			RecoveryLockList = list_delete_cell(RecoveryLockList, cell, prev);
+			pfree(lock);
+		}
+		else
+			prev = cell;
+	}
 }
 
 /*
@@ -812,6 +845,13 @@ standby_desc(StringInfo buf, uint8 xl_info, char *rec)
  * Later, when we apply the running xact data we must be careful to ignore
  * transactions already committed, since those commits raced ahead when
  * making WAL entries.
+ *
+ * The loose timing also means that locks may be recorded that have a
+ * zero xid, since xids are removed from procs before locks are removed.
+ * So we must prune the lock list down to ensure we hold locks only for
+ * currently running xids, performed by StandbyReleaseOldLocks().
+ * Zero xids should no longer be possible, but we may be replaying WAL
+ * from a time when they were possible.
  */
 void
 LogStandbySnapshot(TransactionId *nextXid)

src/backend/storage/lmgr/lock.c

@@ -2355,8 +2355,18 @@ GetRunningTransactionLocks(int *nlocks)
 		{
 			PGPROC	   *proc = proclock->tag.myProc;
 			LOCK	   *lock = proclock->tag.myLock;
+			TransactionId xid = proc->xid;
 
-			accessExclusiveLocks[index].xid = proc->xid;
+			/*
+			 * Don't record locks for transactions if we know they have already
+			 * issued their WAL record for commit but not yet released lock.
+			 * It is still possible that we see locks held by already complete
+			 * transactions, if they haven't yet zeroed their xids.
+			 */
+			if (!TransactionIdIsValid(xid))
+				continue;
+
+			accessExclusiveLocks[index].xid = xid;
 			accessExclusiveLocks[index].dbOid = lock->tag.locktag_field1;
 			accessExclusiveLocks[index].relOid = lock->tag.locktag_field2;
 

src/include/storage/standby.h

@@ -48,7 +48,7 @@ extern void StandbyAcquireAccessExclusiveLock(TransactionId xid, Oid dbOid, Oid
 extern void StandbyReleaseLockTree(TransactionId xid,
 					   int nsubxids, TransactionId *subxids);
 extern void StandbyReleaseAllLocks(void);
-extern void StandbyReleaseOldLocks(TransactionId removeXid);
+extern void StandbyReleaseOldLocks(int nxids, TransactionId *xids);
 
 /*
  * XLOG message types