On Fri, 2022-04-08 at 08:47 +0900, Michael Paquier wrote:
> On Thu, Apr 07, 2022 at 11:19:15AM -0400, Robert Haas wrote:
> > Here are patches for master and v14 to do things this way.
> > Comments?
> 
> Thanks for the patches.  They look correct.

+1, looks good to me and addresses my specific original concern.

> For ~14, I'd rather avoid
> the code duplication done by GetVirtualXIDsDelayingChkptEnd() and
> HaveVirtualXIDsDelayingChkpt() that could be avoided with an extra
> bool argument to the existing routine.  The same kind of duplication
> happens with GetVirtualXIDsDelayingChkpt() and
> GetVirtualXIDsDelayingChkptEnd().

I agree with Michael, it would be nice to not duplicate the code, but
use a common underlying method.  A modified patch is attached.

Best Regards

Markus

From: Robert Haas <rh...@postgresql.org>
Date: Thu, 7 Apr 2022 11:15:07 -0400
Subject: [PATCH] Rethink the delay-checkpoint-end mechanism in the
 back-branches.

The back-patch of commit bbace5697df12398e87ffd9879171c39d27f5b33 had
the unfortunate effect of changing the layout of PGPROC in the
back-branches, which could break extensions. This happened because it
changed the delayChkpt from type bool to type int. So, change it back,
and add a new bool delayChkptEnd field instead. The new field should
fall within what used to be padding space within the struct, and so
hopefully won't cause any extensions to break.

Per report from Markus Wanner and discussion with Tom Lane and others.
---
 backend/access/transam/multixact.c  |   6 +-
 backend/access/transam/twophase.c   |  13 ++--
 backend/access/transam/xact.c       |   6 +-
 backend/access/transam/xlog.c       |  10 +--
 backend/access/transam/xloginsert.c |   2 
 backend/catalog/storage.c           |   6 +-
 backend/storage/buffer/bufmgr.c     |   6 +-
 backend/storage/ipc/procarray.c     | 100 +++++++++++++++++++++++++++---------
 include/storage/proc.h              |  35 +-----------
 include/storage/procarray.h         |   7 +-
 10 files changed, 107 insertions(+), 84 deletions(-)

diff --git a/src/backend/access/transam/multixact.c b/src/backend/access/transam/multixact.c
index 50d8bab9e21..b643564f16a 100644
--- a/src/backend/access/transam/multixact.c
+++ b/src/backend/access/transam/multixact.c
@@ -3075,8 +3075,8 @@ TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
 	 * crash/basebackup, even though the state of the data directory would
 	 * require it.
 	 */
-	Assert((MyProc->delayChkpt & DELAY_CHKPT_START) == 0);
-	MyProc->delayChkpt |= DELAY_CHKPT_START;
+	Assert(!MyProc->delayChkpt);
+	MyProc->delayChkpt = true;
 
 	/* WAL log truncation */
 	WriteMTruncateXlogRec(newOldestMultiDB,
@@ -3102,7 +3102,7 @@ TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
 	/* Then offsets */
 	PerformOffsetsTruncation(oldestMulti, newOldestMulti);
 
-	MyProc->delayChkpt &= ~DELAY_CHKPT_START;
+	MyProc->delayChkpt = false;
 
 	END_CRIT_SECTION();
 	LWLockRelease(MultiXactTruncationLock);
diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c
index dea3f485f7a..633a6f1747f 100644
--- a/src/backend/access/transam/twophase.c
+++ b/src/backend/access/transam/twophase.c
@@ -474,8 +474,9 @@ MarkAsPreparingGuts(GlobalTransaction gxact, TransactionId xid, const char *gid,
 	}
 	proc->xid = xid;
 	Assert(proc->xmin == InvalidTransactionId);
-	proc->delayChkpt = 0;
+	proc->delayChkpt = false;
 	proc->statusFlags = 0;
+	proc->delayChkptEnd = false;
 	proc->pid = 0;
 	proc->databaseId = databaseid;
 	proc->roleId = owner;
@@ -1165,8 +1166,7 @@ EndPrepare(GlobalTransaction gxact)
 
 	START_CRIT_SECTION();
 
-	Assert((MyProc->delayChkpt & DELAY_CHKPT_START) == 0);
-	MyProc->delayChkpt |= DELAY_CHKPT_START;
+	MyProc->delayChkpt = true;
 
 	XLogBeginInsert();
 	for (record = records.head; record != NULL; record = record->next)
@@ -1209,7 +1209,7 @@ EndPrepare(GlobalTransaction gxact)
 	 * checkpoint starting after this will certainly see the gxact as a
 	 * candidate for fsyncing.
 	 */
-	MyProc->delayChkpt &= ~DELAY_CHKPT_START;
+	MyProc->delayChkpt = false;
 
 	/*
 	 * Remember that we have this GlobalTransaction entry locked for us.  If
@@ -2276,8 +2276,7 @@ RecordTransactionCommitPrepared(TransactionId xid,
 	START_CRIT_SECTION();
 
 	/* See notes in RecordTransactionCommit */
-	Assert((MyProc->delayChkpt & DELAY_CHKPT_START) == 0);
-	MyProc->delayChkpt |= DELAY_CHKPT_START;
+	MyProc->delayChkpt = true;
 
 	/*
 	 * Emit the XLOG commit record. Note that we mark 2PC commits as
@@ -2325,7 +2324,7 @@ RecordTransactionCommitPrepared(TransactionId xid,
 	TransactionIdCommitTree(xid, nchildren, children);
 
 	/* Checkpoint can proceed now */
-	MyProc->delayChkpt &= ~DELAY_CHKPT_START;
+	MyProc->delayChkpt = false;
 
 	END_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xact.c b/src/backend/access/transam/xact.c
index c5e7261921d..e8523693c0a 100644
--- a/src/backend/access/transam/xact.c
+++ b/src/backend/access/transam/xact.c
@@ -1335,9 +1335,9 @@ RecordTransactionCommit(void)
 		 * This makes checkpoint's determination of which xacts are delayChkpt
 		 * a bit fuzzy, but it doesn't matter.
 		 */
-		Assert((MyProc->delayChkpt & DELAY_CHKPT_START) == 0);
+		Assert(!MyProc->delayChkpt);
 		START_CRIT_SECTION();
-		MyProc->delayChkpt |= DELAY_CHKPT_START;
+		MyProc->delayChkpt = true;
 
 		SetCurrentTransactionStopTimestamp();
 
@@ -1438,7 +1438,7 @@ RecordTransactionCommit(void)
 	 */
 	if (markXidCommitted)
 	{
-		MyProc->delayChkpt &= ~DELAY_CHKPT_START;
+		MyProc->delayChkpt = false;
 		END_CRIT_SECTION();
 	}
 
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index d889d69387e..8edf0fa6461 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -9228,27 +9228,25 @@ CreateCheckPoint(int flags)
 	 * and we will correctly flush the update below.  So we cannot miss any
 	 * xacts we need to wait for.
 	 */
-	vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_START);
+	vxids = GetVirtualXIDsDelayingChkpt(&nvxids);
 	if (nvxids > 0)
 	{
 		do
 		{
 			pg_usleep(10000L);	/* wait for 10 msec */
-		} while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
-											  DELAY_CHKPT_START));
+		} while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids));
 	}
 	pfree(vxids);
 
 	CheckPointGuts(checkPoint.redo, flags);
 
-	vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_COMPLETE);
+	vxids = GetVirtualXIDsDelayingChkptEnd(&nvxids);
 	if (nvxids > 0)
 	{
 		do
 		{
 			pg_usleep(10000L);	/* wait for 10 msec */
-		} while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
-											  DELAY_CHKPT_COMPLETE));
+		} while (HaveVirtualXIDsDelayingChkptEnd(vxids, nvxids));
 	}
 	pfree(vxids);
 
diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c
index 1af4a90c419..b153fad594d 100644
--- a/src/backend/access/transam/xloginsert.c
+++ b/src/backend/access/transam/xloginsert.c
@@ -925,7 +925,7 @@ XLogSaveBufferForHint(Buffer buffer, bool buffer_std)
 	/*
 	 * Ensure no checkpoint can change our view of RedoRecPtr.
 	 */
-	Assert((MyProc->delayChkpt & DELAY_CHKPT_START) != 0);
+	Assert(MyProc->delayChkpt);
 
 	/*
 	 * Update RedoRecPtr so that we can make the right decision
diff --git a/src/backend/catalog/storage.c b/src/backend/catalog/storage.c
index fa5682dce80..b7fc491a684 100644
--- a/src/backend/catalog/storage.c
+++ b/src/backend/catalog/storage.c
@@ -338,8 +338,8 @@ RelationTruncate(Relation rel, BlockNumber nblocks)
 	 * the blocks to not exist on disk at all, but not for them to have the
 	 * wrong contents.
 	 */
-	Assert((MyProc->delayChkpt & DELAY_CHKPT_COMPLETE) == 0);
-	MyProc->delayChkpt |= DELAY_CHKPT_COMPLETE;
+	Assert(!MyProc->delayChkptEnd);
+	MyProc->delayChkptEnd = true;
 
 	/*
 	 * We WAL-log the truncation before actually truncating, which means
@@ -387,7 +387,7 @@ RelationTruncate(Relation rel, BlockNumber nblocks)
 	smgrtruncate(rel->rd_smgr, forks, nforks, blocks);
 
 	/* We've done all the critical work, so checkpoints are OK now. */
-	MyProc->delayChkpt &= ~DELAY_CHKPT_COMPLETE;
+	MyProc->delayChkptEnd = false;
 
 	/*
 	 * Update upper-level FSM pages to account for the truncation. This is
diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index a55545a1875..49561ae794d 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -3946,8 +3946,8 @@ MarkBufferDirtyHint(Buffer buffer, bool buffer_std)
 			 * essential that CreateCheckpoint waits for virtual transactions
 			 * rather than full transactionids.
 			 */
-			Assert((MyProc->delayChkpt & DELAY_CHKPT_START) == 0);
-			MyProc->delayChkpt |= DELAY_CHKPT_START;
+			Assert(!MyProc->delayChkpt);
+			MyProc->delayChkpt = true;
 			delayChkpt = true;
 			lsn = XLogSaveBufferForHint(buffer, buffer_std);
 		}
@@ -3981,7 +3981,7 @@ MarkBufferDirtyHint(Buffer buffer, bool buffer_std)
 		UnlockBufHdr(bufHdr, buf_state);
 
 		if (delayChkpt)
-			MyProc->delayChkpt &= ~DELAY_CHKPT_START;
+			MyProc->delayChkpt = false;
 
 		if (dirtied)
 		{
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index ae71d7538be..54fd8615ad9 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -330,6 +330,11 @@ static void DisplayXidCache(void);
 #define xc_slow_answer_inc()		((void) 0)
 #endif							/* XIDCACHE_DEBUG */
 
+static VirtualTransactionId *GetVirtualXIDsDelayingChkptGuts(int *nvxids,
+															 int type);
+static bool HaveVirtualXIDsDelayingChkptGuts(VirtualTransactionId *vxids,
+											 int nvxids, int type);
+
 /* Primitives for KnownAssignedXids array handling for standby */
 static void KnownAssignedXidsCompress(bool force);
 static void KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
@@ -690,8 +695,9 @@ ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
 		proc->lxid = InvalidLocalTransactionId;
 		proc->xmin = InvalidTransactionId;
 
-		/* be sure this is cleared in abort */
-		proc->delayChkpt = 0;
+		/* be sure these are cleared in abort */
+		proc->delayChkpt = false;
+		proc->delayChkptEnd = false;
 
 		proc->recoveryConflictPending = false;
 
@@ -732,8 +738,9 @@ ProcArrayEndTransactionInternal(PGPROC *proc, TransactionId latestXid)
 	proc->lxid = InvalidLocalTransactionId;
 	proc->xmin = InvalidTransactionId;
 
-	/* be sure this is cleared in abort */
-	proc->delayChkpt = 0;
+	/* be sure these are cleared in abort */
+	proc->delayChkpt = false;
+	proc->delayChkptEnd = false;
 
 	proc->recoveryConflictPending = false;
 
@@ -3045,34 +3052,34 @@ GetOldestSafeDecodingTransactionId(bool catalogOnly)
 }
 
 /*
- * GetVirtualXIDsDelayingChkpt -- Get the VXIDs of transactions that are
- * delaying checkpoint because they have critical actions in progress.
+ * GetVirtualXIDsDelayingChkptGuts -- Get the VXIDs of transactions that are
+ * delaying the start or end of a checkpoint because they have critical
+ * actions in progress.
  *
  * Constructs an array of VXIDs of transactions that are currently in commit
- * critical sections, as shown by having specified delayChkpt bits set in their
- * PGPROC.
+ * critical sections, as shown by having delayChkpt or delayChkptEnd set in
+ * their PGPROC.
  *
  * Returns a palloc'd array that should be freed by the caller.
  * *nvxids is the number of valid entries.
  *
- * Note that because backends set or clear delayChkpt without holding any lock,
- * the result is somewhat indeterminate, but we don't really care.  Even in
- * a multiprocessor with delayed writes to shared memory, it should be certain
- * that setting of delayChkpt will propagate to shared memory when the backend
- * takes a lock, so we cannot fail to see a virtual xact as delayChkpt if
- * it's already inserted its commit record.  Whether it takes a little while
- * for clearing of delayChkpt to propagate is unimportant for correctness.
+ * Note that because backends set or clear delayChkpt and delayChkptEnd
+ * without holding any lock, the result is somewhat indeterminate, but we
+ * don't really care.  Even in a multiprocessor with delayed writes to
+ * shared memory, it should be certain that setting of delayChkpt will
+ * propagate to shared memory when the backend takes a lock, so we cannot
+ * fail to see a virtual xact as delayChkpt if it's already inserted its
+ * commit record.  Whether it takes a little while for clearing of
+ * delayChkpt to propagate is unimportant for correctness.
  */
-VirtualTransactionId *
-GetVirtualXIDsDelayingChkpt(int *nvxids, int type)
+static VirtualTransactionId *
+GetVirtualXIDsDelayingChkptGuts(int *nvxids, int type)
 {
 	VirtualTransactionId *vxids;
 	ProcArrayStruct *arrayP = procArray;
 	int			count = 0;
 	int			index;
 
-	Assert(type != 0);
-
 	/* allocate what's certainly enough result space */
 	vxids = (VirtualTransactionId *)
 		palloc(sizeof(VirtualTransactionId) * arrayP->maxProcs);
@@ -3084,7 +3091,8 @@ GetVirtualXIDsDelayingChkpt(int *nvxids, int type)
 		int			pgprocno = arrayP->pgprocnos[index];
 		PGPROC	   *proc = &allProcs[pgprocno];
 
-		if ((proc->delayChkpt & type) != 0)
+		if (((type & DELAY_CHKPT_START) && proc->delayChkpt) ||
+			((type & DELAY_CHKPT_COMPLETE) && proc->delayChkptEnd))
 		{
 			VirtualTransactionId vxid;
 
@@ -3100,6 +3108,26 @@ GetVirtualXIDsDelayingChkpt(int *nvxids, int type)
 	return vxids;
 }
 
+/*
+ * GetVirtualXIDsDelayingChkpt - Get the VXIDs of transactions that are
+ * delaying the start of a checkpoint.
+ */
+VirtualTransactionId *
+GetVirtualXIDsDelayingChkpt(int *nvxids)
+{
+	return GetVirtualXIDsDelayingChkptGuts(nvxids, DELAY_CHKPT_START);
+}
+
+/*
+ * GetVirtualXIDsDelayingChkptEnd - Get the VXIDs of transactions that are
+ * delaying the end of a checkpoint.
+ */
+VirtualTransactionId *
+GetVirtualXIDsDelayingChkptEnd(int *nvxids)
+{
+	return GetVirtualXIDsDelayingChkptGuts(nvxids, DELAY_CHKPT_COMPLETE);
+}
+
 /*
  * HaveVirtualXIDsDelayingChkpt -- Are any of the specified VXIDs delaying?
  *
@@ -3109,15 +3137,14 @@ GetVirtualXIDsDelayingChkpt(int *nvxids, int type)
  * Note: this is O(N^2) in the number of vxacts that are/were delaying, but
  * those numbers should be small enough for it not to be a problem.
  */
-bool
-HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids, int nvxids, int type)
+static bool
+HaveVirtualXIDsDelayingChkptGuts(VirtualTransactionId *vxids, int nvxids,
+								 int type)
 {
 	bool		result = false;
 	ProcArrayStruct *arrayP = procArray;
 	int			index;
 
-	Assert(type != 0);
-
 	LWLockAcquire(ProcArrayLock, LW_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
@@ -3128,7 +3155,8 @@ HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids, int nvxids, int type)
 
 		GET_VXID_FROM_PGPROC(vxid, *proc);
 
-		if ((proc->delayChkpt & type) != 0 &&
+		if ((((type & DELAY_CHKPT_START) && proc->delayChkpt) ||
+			 ((type & DELAY_CHKPT_COMPLETE) && proc->delayChkptEnd)) &&
 			VirtualTransactionIdIsValid(vxid))
 		{
 			int			i;
@@ -3151,6 +3179,28 @@ HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids, int nvxids, int type)
 	return result;
 }
 
+/*
+ * HaveVirtualXIDsDelayingChkpt -- Are any of the specified VXIDs delaying
+ * the start of a checkpoint?
+ */
+bool
+HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids, int nvxids)
+{
+	return HaveVirtualXIDsDelayingChkptGuts(vxids, nvxids,
+											DELAY_CHKPT_START);
+}
+
+/*
+ * HaveVirtualXIDsDelayingChkptEnd -- Are any of the specified VXIDs delaying
+ * the end of a checkpoint?
+ */
+bool
+HaveVirtualXIDsDelayingChkptEnd(VirtualTransactionId *vxids, int nvxids)
+{
+	return HaveVirtualXIDsDelayingChkptGuts(vxids, nvxids,
+											DELAY_CHKPT_COMPLETE);
+}
+
 /*
  * BackendPidGetProc -- get a backend's PGPROC given its PID
  *
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index b78012ec2b6..11b514c9ae8 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -87,36 +87,8 @@ struct XidCache
 #define INVALID_PGPROCNO		PG_INT32_MAX
 
 /*
- * Flags for PGPROC.delayChkpt
- *
- * These flags can be used to delay the start or completion of a checkpoint
- * for short periods. A flag is in effect if the corresponding bit is set in
- * the PGPROC of any backend.
- *
- * For our purposes here, a checkpoint has three phases: (1) determine the
- * location to which the redo pointer will be moved, (2) write all the
- * data durably to disk, and (3) WAL-log the checkpoint.
- *
- * Setting DELAY_CHKPT_START prevents the system from moving from phase 1
- * to phase 2. This is useful when we are performing a WAL-logged modification
- * of data that will be flushed to disk in phase 2. By setting this flag
- * before writing WAL and clearing it after we've both written WAL and
- * performed the corresponding modification, we ensure that if the WAL record
- * is inserted prior to the new redo point, the corresponding data changes will
- * also be flushed to disk before the checkpoint can complete. (In the
- * extremely common case where the data being modified is in shared buffers
- * and we acquire an exclusive content lock on the relevant buffers before
- * writing WAL, this mechanism is not needed, because phase 2 will block
- * until we release the content lock and then flush the modified data to
- * disk.)
- *
- * Setting DELAY_CHKPT_COMPLETE prevents the system from moving from phase 2
- * to phase 3. This is useful if we are performing a WAL-logged operation that
- * might invalidate buffers, such as relation truncation. In this case, we need
- * to ensure that any buffers which were invalidated and thus not flushed by
- * the checkpoint are actaully destroyed on disk. Replay can cope with a file
- * or block that doesn't exist, but not with a block that has the wrong
- * contents.
+ * Flags used only for type of internal functions
+ * GetVirtualXIDsDelayingChkptGuts and HaveVirtualXIDsDelayingChkptGuts.
  */
 #define DELAY_CHKPT_START		(1<<0)
 #define DELAY_CHKPT_COMPLETE	(1<<1)
@@ -226,11 +198,12 @@ struct PGPROC
 	pg_atomic_uint64 waitStart; /* time at which wait for lock acquisition
 								 * started */
 
-	int			delayChkpt;		/* for DELAY_CHKPT_* flags */
+	bool		delayChkpt;		/* true if this proc delays checkpoint start */
 
 	uint8		statusFlags;	/* this backend's status flags, see PROC_*
 								 * above. mirrored in
 								 * ProcGlobal->statusFlags[pgxactoff] */
+	bool		delayChkptEnd;	/* true if this proc delays checkpoint end */
 
 	/*
 	 * Info to allow us to wait for synchronous replication, if needed.
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index 93de230a329..f5d7d410986 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -59,9 +59,12 @@ extern TransactionId GetOldestActiveTransactionId(void);
 extern TransactionId GetOldestSafeDecodingTransactionId(bool catalogOnly);
 extern void GetReplicationHorizons(TransactionId *slot_xmin, TransactionId *catalog_xmin);
 
-extern VirtualTransactionId *GetVirtualXIDsDelayingChkpt(int *nvxids, int type);
+extern VirtualTransactionId *GetVirtualXIDsDelayingChkpt(int *nvxids);
+extern VirtualTransactionId *GetVirtualXIDsDelayingChkptEnd(int *nvxids);
 extern bool HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids,
-										 int nvxids, int type);
+										 int nvxids);
+extern bool HaveVirtualXIDsDelayingChkptEnd(VirtualTransactionId *vxids,
+											int nvxids);
 
 extern PGPROC *BackendPidGetProc(int pid);
 extern PGPROC *BackendPidGetProcWithLock(int pid);

Reply via email to