snapshot scalability: Move PGXACT->vacuumFlags to ProcGlobal->vacuumFlags.

Similar to the previous commit this increases the chance that data
frequently needed by GetSnapshotData() stays in l2 cache. As we now
take care to not unnecessarily write to ProcGlobal->vacuumFlags, there
should be very few modifications to the ProcGlobal->vacuumFlags array.

Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de

9 files changed, 75 insertions, 42 deletions

diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c
index a0398bf3a3e..744b8a7f393 100644
--- a/src/backend/access/transam/twophase.c
+++ b/src/backend/access/transam/twophase.c
@@ -466,7 +466,7 @@ MarkAsPreparingGuts(GlobalTransaction gxact, TransactionId xid, const char *gid,
 	proc->xid = xid;
 	Assert(proc->xmin == InvalidTransactionId);
 	proc->delayChkpt = false;
-	pgxact->vacuumFlags = 0;
+	proc->vacuumFlags = 0;
 	proc->pid = 0;
 	proc->backendId = InvalidBackendId;
 	proc->databaseId = databaseid;
diff --git a/src/backend/commands/vacuum.c b/src/backend/commands/vacuum.c
index 648e12c78d8..aba13c31d1b 100644
--- a/src/backend/commands/vacuum.c
+++ b/src/backend/commands/vacuum.c
@@ -1728,9 +1728,10 @@ vacuum_rel(Oid relid, RangeVar *relation, VacuumParams *params)
 		 * might appear to go backwards, which is probably Not Good.
 		 */
 		LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
-		MyPgXact->vacuumFlags |= PROC_IN_VACUUM;
+		MyProc->vacuumFlags |= PROC_IN_VACUUM;
 		if (params->is_wraparound)
-			MyPgXact->vacuumFlags |= PROC_VACUUM_FOR_WRAPAROUND;
+			MyProc->vacuumFlags |= PROC_VACUUM_FOR_WRAPAROUND;
+		ProcGlobal->vacuumFlags[MyProc->pgxactoff] = MyProc->vacuumFlags;
 		LWLockRelease(ProcArrayLock);
 	}
 
diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c
index ac97e28be19..c6ec657a936 100644
--- a/src/backend/postmaster/autovacuum.c
+++ b/src/backend/postmaster/autovacuum.c
@@ -2493,7 +2493,7 @@ do_autovacuum(void)
 						   tab->at_datname, tab->at_nspname, tab->at_relname);
 			EmitErrorReport();
 
-			/* this resets the PGXACT flags too */
+			/* this resets ProcGlobal->vacuumFlags[i] too */
 			AbortOutOfAnyTransaction();
 			FlushErrorState();
 			MemoryContextResetAndDeleteChildren(PortalContext);
@@ -2509,7 +2509,7 @@ do_autovacuum(void)
 
 		did_vacuum = true;
 
-		/* the PGXACT flags are reset at the next end of transaction */
+		/* ProcGlobal->vacuumFlags[i] are reset at the next end of xact */
 
 		/* be tidy */
 deleted:
@@ -2686,7 +2686,7 @@ perform_work_item(AutoVacuumWorkItem *workitem)
 				   cur_datname, cur_nspname, cur_relname);
 		EmitErrorReport();
 
-		/* this resets the PGXACT flags too */
+		/* this resets ProcGlobal->vacuumFlags[i] too */
 		AbortOutOfAnyTransaction();
 		FlushErrorState();
 		MemoryContextResetAndDeleteChildren(PortalContext);
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 57c5b513ccf..0f6af952f93 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -181,7 +181,8 @@ StartupDecodingContext(List *output_plugin_options,
 	if (!IsTransactionOrTransactionBlock())
 	{
 		LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
-		MyPgXact->vacuumFlags |= PROC_IN_LOGICAL_DECODING;
+		MyProc->vacuumFlags |= PROC_IN_LOGICAL_DECODING;
+		ProcGlobal->vacuumFlags[MyProc->pgxactoff] = MyProc->vacuumFlags;
 		LWLockRelease(ProcArrayLock);
 	}
 
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 3dc01b6df22..42c78eabd4e 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -520,7 +520,8 @@ ReplicationSlotRelease(void)
 
 	/* might not have been set when we've been a plain slot */
 	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
-	MyPgXact->vacuumFlags &= ~PROC_IN_LOGICAL_DECODING;
+	MyProc->vacuumFlags &= ~PROC_IN_LOGICAL_DECODING;
+	ProcGlobal->vacuumFlags[MyProc->pgxactoff] = MyProc->vacuumFlags;
 	LWLockRelease(ProcArrayLock);
 }
 
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index 0bf20a49375..224da4f9510 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -476,9 +476,12 @@ ProcArrayAdd(PGPROC *proc)
 			(arrayP->numProcs - index) * sizeof(*arrayP->pgprocnos));
 	memmove(&ProcGlobal->xids[index + 1], &ProcGlobal->xids[index],
 			(arrayP->numProcs - index) * sizeof(*ProcGlobal->xids));
+	memmove(&ProcGlobal->vacuumFlags[index + 1], &ProcGlobal->vacuumFlags[index],
+			(arrayP->numProcs - index) * sizeof(*ProcGlobal->vacuumFlags));
 
 	arrayP->pgprocnos[index] = proc->pgprocno;
 	ProcGlobal->xids[index] = proc->xid;
+	ProcGlobal->vacuumFlags[index] = proc->vacuumFlags;
 
 	arrayP->numProcs++;
 
@@ -539,6 +542,7 @@ ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
 	}
 
 	Assert(TransactionIdIsValid(ProcGlobal->xids[proc->pgxactoff] == 0));
+	ProcGlobal->vacuumFlags[proc->pgxactoff] = 0;
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -549,6 +553,8 @@ ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
 					(arrayP->numProcs - index - 1) * sizeof(*arrayP->pgprocnos));
 			memmove(&ProcGlobal->xids[index], &ProcGlobal->xids[index + 1],
 					(arrayP->numProcs - index - 1) * sizeof(*ProcGlobal->xids));
+			memmove(&ProcGlobal->vacuumFlags[index], &ProcGlobal->vacuumFlags[index + 1],
+					(arrayP->numProcs - index - 1) * sizeof(*ProcGlobal->vacuumFlags));
 
 			arrayP->pgprocnos[arrayP->numProcs - 1] = -1;	/* for debugging */
 			arrayP->numProcs--;
@@ -626,14 +632,24 @@ ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
 		Assert(!TransactionIdIsValid(proc->xid));
 
 		proc->lxid = InvalidLocalTransactionId;
-		/* must be cleared with xid/xmin: */
-		pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
 		proc->xmin = InvalidTransactionId;
 		proc->delayChkpt = false;	/* be sure this is cleared in abort */
 		proc->recoveryConflictPending = false;
 
 		Assert(pgxact->nxids == 0);
 		Assert(pgxact->overflowed == false);
+
+		/* must be cleared with xid/xmin: */
+		/* avoid unnecessarily dirtying shared cachelines */
+		if (proc->vacuumFlags & PROC_VACUUM_STATE_MASK)
+		{
+			Assert(!LWLockHeldByMe(ProcArrayLock));
+			LWLockAcquire(ProcArrayLock, LW_SHARED);
+			Assert(proc->vacuumFlags == ProcGlobal->vacuumFlags[proc->pgxactoff]);
+			proc->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
+			ProcGlobal->vacuumFlags[proc->pgxactoff] = proc->vacuumFlags;
+			LWLockRelease(ProcArrayLock);
+		}
 	}
 }
 
@@ -654,12 +670,18 @@ ProcArrayEndTransactionInternal(PGPROC *proc, PGXACT *pgxact,
 	ProcGlobal->xids[pgxactoff] = InvalidTransactionId;
 	proc->xid = InvalidTransactionId;
 	proc->lxid = InvalidLocalTransactionId;
-	/* must be cleared with xid/xmin: */
-	pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
 	proc->xmin = InvalidTransactionId;
 	proc->delayChkpt = false;	/* be sure this is cleared in abort */
 	proc->recoveryConflictPending = false;
 
+	/* must be cleared with xid/xmin: */
+	/* avoid unnecessarily dirtying shared cachelines */
+	if (proc->vacuumFlags & PROC_VACUUM_STATE_MASK)
+	{
+		proc->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
+		ProcGlobal->vacuumFlags[proc->pgxactoff] = proc->vacuumFlags;
+	}
+
 	/* Clear the subtransaction-XID cache too while holding the lock */
 	pgxact->nxids = 0;
 	pgxact->overflowed = false;
@@ -819,9 +841,8 @@ ProcArrayClearTransaction(PGPROC *proc)
 	proc->xmin = InvalidTransactionId;
 	proc->recoveryConflictPending = false;
 
-	/* redundant, but just in case */
-	pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
-	proc->delayChkpt = false;
+	Assert(!(proc->vacuumFlags & PROC_VACUUM_STATE_MASK));
+	Assert(!proc->delayChkpt);
 
 	/* Clear the subtransaction-XID cache too */
 	pgxact->nxids = 0;
@@ -1623,7 +1644,7 @@ ComputeXidHorizons(ComputeXidHorizonsResult *h)
 	{
 		int			pgprocno = arrayP->pgprocnos[index];
 		PGPROC	   *proc = &allProcs[pgprocno];
-		PGXACT	   *pgxact = &allPgXact[pgprocno];
+		int8		vacuumFlags = ProcGlobal->vacuumFlags[index];
 		TransactionId xid;
 		TransactionId xmin;
 
@@ -1640,8 +1661,8 @@ ComputeXidHorizons(ComputeXidHorizonsResult *h)
 		 */
 		xmin = TransactionIdOlder(xmin, xid);
 
-		/* if neither is set, this proc doesn't influence the horizon */
-		if (!TransactionIdIsValid(xmin))
+        /* if neither is set, this proc doesn't influence the horizon */
+        if (!TransactionIdIsValid(xmin))
 			continue;
 
 		/*
@@ -1658,7 +1679,7 @@ ComputeXidHorizons(ComputeXidHorizonsResult *h)
 		 * removed, as long as pg_subtrans is not truncated) or doing logical
 		 * decoding (which manages xmin separately, check below).
 		 */
-		if (pgxact->vacuumFlags & (PROC_IN_VACUUM | PROC_IN_LOGICAL_DECODING))
+		if (vacuumFlags & (PROC_IN_VACUUM | PROC_IN_LOGICAL_DECODING))
 			continue;
 
 		/* shared tables need to take backends in all database into account */
@@ -1998,6 +2019,7 @@ GetSnapshotData(Snapshot snapshot)
 		size_t		numProcs = arrayP->numProcs;
 		TransactionId *xip = snapshot->xip;
 		int		   *pgprocnos = arrayP->pgprocnos;
+		uint8	   *allVacuumFlags = ProcGlobal->vacuumFlags;
 
 		/*
 		 * First collect set of pgxactoff/xids that need to be included in the
@@ -2007,8 +2029,6 @@ GetSnapshotData(Snapshot snapshot)
 		{
 			/* Fetch xid just once - see GetNewTransactionId */
 			TransactionId xid = UINT32_ACCESS_ONCE(other_xids[pgxactoff]);
-			int			pgprocno;
-			PGXACT	   *pgxact;
 			uint8		vacuumFlags;
 
 			Assert(allProcs[arrayP->pgprocnos[pgxactoff]].pgxactoff == pgxactoff);
@@ -2044,14 +2064,11 @@ GetSnapshotData(Snapshot snapshot)
 			if (!NormalTransactionIdPrecedes(xid, xmax))
 				continue;
 
-			pgprocno = pgprocnos[pgxactoff];
-			pgxact = &allPgXact[pgprocno];
-			vacuumFlags = pgxact->vacuumFlags;
-
 			/*
 			 * Skip over backends doing logical decoding which manages xmin
 			 * separately (check below) and ones running LAZY VACUUM.
 			 */
+			vacuumFlags = allVacuumFlags[pgxactoff];
 			if (vacuumFlags & (PROC_IN_LOGICAL_DECODING | PROC_IN_VACUUM))
 				continue;
 
@@ -2078,6 +2095,9 @@ GetSnapshotData(Snapshot snapshot)
 			 */
 			if (!suboverflowed)
 			{
+				int			pgprocno = pgprocnos[pgxactoff];
+				PGXACT	   *pgxact = &allPgXact[pgprocno];
+
 				if (pgxact->overflowed)
 					suboverflowed = true;
 				else
@@ -2296,11 +2316,11 @@ ProcArrayInstallImportedXmin(TransactionId xmin,
 	{
 		int			pgprocno = arrayP->pgprocnos[index];
 		PGPROC	   *proc = &allProcs[pgprocno];
-		PGXACT	   *pgxact = &allPgXact[pgprocno];
+		int			vacuumFlags = ProcGlobal->vacuumFlags[index];
 		TransactionId xid;
 
 		/* Ignore procs running LAZY VACUUM */
-		if (pgxact->vacuumFlags & PROC_IN_VACUUM)
+		if (vacuumFlags & PROC_IN_VACUUM)
 			continue;
 
 		/* We are only interested in the specific virtual transaction. */
@@ -2990,12 +3010,12 @@ GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
 	{
 		int			pgprocno = arrayP->pgprocnos[index];
 		PGPROC	   *proc = &allProcs[pgprocno];
-		PGXACT	   *pgxact = &allPgXact[pgprocno];
+		uint8		vacuumFlags = ProcGlobal->vacuumFlags[index];
 
 		if (proc == MyProc)
 			continue;
 
-		if (excludeVacuum & pgxact->vacuumFlags)
+		if (excludeVacuum & vacuumFlags)
 			continue;
 
 		if (allDbs || proc->databaseId == MyDatabaseId)
@@ -3410,7 +3430,7 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
 		{
 			int			pgprocno = arrayP->pgprocnos[index];
 			PGPROC	   *proc = &allProcs[pgprocno];
-			PGXACT	   *pgxact = &allPgXact[pgprocno];
+			uint8		vacuumFlags = ProcGlobal->vacuumFlags[index];
 
 			if (proc->databaseId != databaseId)
 				continue;
@@ -3424,7 +3444,7 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
 			else
 			{
 				(*nbackends)++;
-				if ((pgxact->vacuumFlags & PROC_IS_AUTOVACUUM) &&
+				if ((vacuumFlags & PROC_IS_AUTOVACUUM) &&
 					nautovacs < MAXAUTOVACPIDS)
 					autovac_pids[nautovacs++] = proc->pid;
 			}
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index beedc7947db..e1246b8a4da 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -544,7 +544,6 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 {
 	PGPROC	   *proc;
 	LOCK	   *lock = checkProc->waitLock;
-	PGXACT	   *pgxact;
 	PROCLOCK   *proclock;
 	SHM_QUEUE  *procLocks;
 	LockMethod	lockMethodTable;
@@ -582,7 +581,6 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 		PGPROC	   *leader;
 
 		proc = proclock->tag.myProc;
-		pgxact = &ProcGlobal->allPgXact[proc->pgprocno];
 		leader = proc->lockGroupLeader == NULL ? proc : proc->lockGroupLeader;
 
 		/* A proc never blocks itself or any other lock group member */
@@ -630,7 +628,7 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 					 * ProcArrayLock.
 					 */
 					if (checkProc == MyProc &&
-						pgxact->vacuumFlags & PROC_IS_AUTOVACUUM)
+						proc->vacuumFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
 					/* We're done looking at this proclock */
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index 7fad49544ce..f6113b2d243 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -114,6 +114,7 @@ ProcGlobalShmemSize(void)
 	size = add_size(size, mul_size(NUM_AUXILIARY_PROCS, sizeof(PGXACT)));
 	size = add_size(size, mul_size(max_prepared_xacts, sizeof(PGXACT)));
 	size = add_size(size, mul_size(TotalProcs, sizeof(*ProcGlobal->xids)));
+	size = add_size(size, mul_size(TotalProcs, sizeof(*ProcGlobal->vacuumFlags)));
 
 	return size;
 }
@@ -223,6 +224,8 @@ InitProcGlobal(void)
 	ProcGlobal->xids =
 		(TransactionId *) ShmemAlloc(TotalProcs * sizeof(*ProcGlobal->xids));
 	MemSet(ProcGlobal->xids, 0, TotalProcs * sizeof(*ProcGlobal->xids));
+	ProcGlobal->vacuumFlags = (uint8 *) ShmemAlloc(TotalProcs * sizeof(*ProcGlobal->vacuumFlags));
+	MemSet(ProcGlobal->vacuumFlags, 0, TotalProcs * sizeof(*ProcGlobal->vacuumFlags));
 
 	for (i = 0; i < TotalProcs; i++)
 	{
@@ -405,10 +408,10 @@ InitProcess(void)
 	MyProc->tempNamespaceId = InvalidOid;
 	MyProc->isBackgroundWorker = IsBackgroundWorker;
 	MyProc->delayChkpt = false;
-	MyPgXact->vacuumFlags = 0;
+	MyProc->vacuumFlags = 0;
 	/* NB -- autovac launcher intentionally does not set IS_AUTOVACUUM */
 	if (IsAutoVacuumWorkerProcess())
-		MyPgXact->vacuumFlags |= PROC_IS_AUTOVACUUM;
+		MyProc->vacuumFlags |= PROC_IS_AUTOVACUUM;
 	MyProc->lwWaiting = false;
 	MyProc->lwWaitMode = 0;
 	MyProc->waitLock = NULL;
@@ -587,7 +590,7 @@ InitAuxiliaryProcess(void)
 	MyProc->tempNamespaceId = InvalidOid;
 	MyProc->isBackgroundWorker = IsBackgroundWorker;
 	MyProc->delayChkpt = false;
-	MyPgXact->vacuumFlags = 0;
+	MyProc->vacuumFlags = 0;
 	MyProc->lwWaiting = false;
 	MyProc->lwWaitMode = 0;
 	MyProc->waitLock = NULL;
@@ -1323,7 +1326,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 		if (deadlock_state == DS_BLOCKED_BY_AUTOVACUUM && allow_autovacuum_cancel)
 		{
 			PGPROC	   *autovac = GetBlockingAutoVacuumPgproc();
-			PGXACT	   *autovac_pgxact = &ProcGlobal->allPgXact[autovac->pgprocno];
+			uint8		vacuumFlags;
 
 			LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
 
@@ -1331,8 +1334,9 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 			 * Only do it if the worker is not working to protect against Xid
 			 * wraparound.
 			 */
-			if ((autovac_pgxact->vacuumFlags & PROC_IS_AUTOVACUUM) &&
-				!(autovac_pgxact->vacuumFlags & PROC_VACUUM_FOR_WRAPAROUND))
+			vacuumFlags = ProcGlobal->vacuumFlags[proc->pgxactoff];
+			if ((vacuumFlags & PROC_IS_AUTOVACUUM) &&
+				!(vacuumFlags & PROC_VACUUM_FOR_WRAPAROUND))
 			{
 				int			pid = autovac->pid;
 				StringInfoData locktagbuf;
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index e29ed85e53d..9f3a8b518eb 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -41,7 +41,7 @@ struct XidCache
 };
 
 /*
- * Flags for PGXACT->vacuumFlags
+ * Flags for ProcGlobal->vacuumFlags[]
  */
 #define		PROC_IS_AUTOVACUUM	0x01	/* is it an autovac worker? */
 #define		PROC_IN_VACUUM		0x02	/* currently running lazy vacuum */
@@ -167,6 +167,9 @@ struct PGPROC
 
 	bool		delayChkpt;		/* true if this proc delays checkpoint start */
 
+	uint8		vacuumFlags;    /* this backend's vacuum flags, see PROC_*
+								 * above. mirrored in
+								 * ProcGlobal->vacuumFlags[pgxactoff] */
 	/*
 	 * Info to allow us to wait for synchronous replication, if needed.
 	 * waitLSN is InvalidXLogRecPtr if not waiting; set only by user backend.
@@ -244,7 +247,6 @@ extern PGDLLIMPORT struct PGXACT *MyPgXact;
  */
 typedef struct PGXACT
 {
-	uint8		vacuumFlags;	/* vacuum-related flags, see above */
 	bool		overflowed;
 
 	uint8		nxids;
@@ -314,6 +316,12 @@ typedef struct PROC_HDR
 	/* Array mirroring PGPROC.xid for each PGPROC currently in the procarray */
 	TransactionId *xids;
 
+	/*
+	 * Array mirroring PGPROC.vacuumFlags for each PGPROC currently in the
+	 * procarray.
+	 */
+	uint8	   *vacuumFlags;
+
 	/* Length of allProcs array */
 	uint32		allProcCount;
 	/* Head of list of free PGPROC structures */