public inbox for [email protected]  

help / color / mirror / Atom feed

[PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
25+ messages / 4 participants
[nested] [flat]

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension
@ 2020-12-31 12:20  Luc Vlaming <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Luc Vlaming @ 2020-12-31 12:20 UTC (permalink / raw)

---
 src/backend/storage/buffer/bufmgr.c | 280 +++++++++++++++++++++++++++-
 src/include/storage/buf_internals.h |   2 +-
 2 files changed, 275 insertions(+), 7 deletions(-)

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index d1ce88fee5..14f0923bb5 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -474,6 +474,10 @@ static BufferDesc *BufferAlloc(SMgrRelation smgr,
 							   BlockNumber blockNum,
 							   BufferAccessStrategy strategy,
 							   bool *foundPtr);
+static BufferDesc *BufferAllocExtend(SMgrRelation smgr, 
+			char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock);
 static void FlushBuffer(BufferDesc *buf, SMgrRelation reln);
 static void AtProcExit_Buffers(int code, Datum arg);
 static void CheckForBufferLeaks(void);
@@ -996,7 +1000,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 	BlockNumber	firstBlock,
 				blockNum;
 	Block		bufBlock;
-	bool		found;
+	LWLock*		lastPartitionLock = NULL;
 
 	/* Open it at the smgr level if not already done */
 	RelationOpenSmgr(reln);
@@ -1007,7 +1011,7 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 
 	LockRelationForExtension(reln, ExclusiveLock);
 	firstBlock = smgrnblocks(smgr, MAIN_FORKNUM);
-	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, false, BULK_INSERT_BATCH_SIZE);
+	smgrextend_count(smgr, MAIN_FORKNUM, firstBlock, bistate->empty_buffer, true, BULK_INSERT_BATCH_SIZE);
 	UnlockRelationForExtension(reln, ExclusiveLock);
 
 	for (int i=0; i<BULK_INSERT_BATCH_SIZE; ++i)
@@ -1017,12 +1021,10 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		pgstat_count_buffer_read(reln);
 		/* Make sure we will have room to remember the buffer pin */
 		ResourceOwnerEnlargeBuffers(CurrentResourceOwner);
-		bufHdr = BufferAlloc(smgr, relpersistence, MAIN_FORKNUM, blockNum,
-							 bistate->strategy, &found);
+		bufHdr = BufferAllocExtend(smgr, relpersistence, MAIN_FORKNUM, blockNum,
+							 bistate->strategy, &lastPartitionLock);
 		pgBufferUsage.shared_blks_written++;
 
-		Assert(!found);
-
 		bufBlock = BufHdrGetBlock(bufHdr);
 
 		/* new buffers are zero-filled */
@@ -1033,6 +1035,9 @@ ReadBufferExtendBulk(Relation reln, struct BulkInsertStateData* bistate)
 		bistate->local_buffers[i] = BufferDescriptorGetBuffer(bufHdr);
 	}
 
+	Assert(lastPartitionLock);
+	LWLockRelease(lastPartitionLock);
+
 	bistate->local_buffers_idx = 0;
 
 	VacuumPageMiss += BULK_INSERT_BATCH_SIZE;
@@ -1408,6 +1413,269 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 	return buf;
 }
 
+static BufferDesc *
+BufferAllocExtend(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
+			BlockNumber blockNum, BufferAccessStrategy strategy,
+			LWLock** lastPartitionLock)
+{
+	BufferTag	newTag;			/* identity of requested block */
+	uint32		newHash;		/* hash value for newTag */
+	LWLock	   *newPartitionLock;	/* buffer partition lock for it */
+	BufferTag	oldTag;			/* previous identity of selected buffer */
+	uint32		oldHash;		/* hash value for oldTag */
+	LWLock	   *oldPartitionLock;	/* buffer partition lock for it */
+	uint32		oldFlags;
+	int			buf_id;
+	BufferDesc *buf;
+	uint32		buf_state;
+
+	Assert(lastPartitionLock);
+
+	/* create a tag so we can lookup the buffer */
+	INIT_BUFFERTAG(newTag, smgr->smgr_rnode.node, forkNum, blockNum);
+
+	/* determine its hash code and partition lock ID */
+	newHash = BufTableHashCode(&newTag);
+	newPartitionLock = BufMappingPartitionLock(newHash);
+
+	/* Loop here in case we have to try another victim buffer */
+	for (;;)
+	{
+		/*
+		 * Ensure, while the spinlock's not yet held, that there's a free
+		 * refcount entry.
+		 */
+		ReservePrivateRefCountEntry();
+
+		/*
+		 * Select a victim buffer.  The buffer is returned with its header
+		 * spinlock still held!
+		 */
+		buf = StrategyGetBuffer(strategy, &buf_state);
+
+		Assert(BUF_STATE_GET_REFCOUNT(buf_state) == 0);
+
+		/* Must copy buffer flags while we still hold the spinlock */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+
+		/* Pin the buffer and then release the buffer spinlock */
+		PinBuffer_Locked(buf);
+
+		/*
+		 * If the buffer was dirty, try to write it out.  There is a race
+		 * condition here, in that someone might dirty it after we released it
+		 * above, or even while we are writing it out (since our share-lock
+		 * won't prevent hint-bit updates).  We will recheck the dirty bit
+		 * after re-locking the buffer header.
+		 */
+		if (oldFlags & BM_DIRTY)
+		{
+			/*
+			 * We need a share-lock on the buffer contents to write it out
+			 * (else we might write invalid data, eg because someone else is
+			 * compacting the page contents while we write).  We must use a
+			 * conditional lock acquisition here to avoid deadlock.  Even
+			 * though the buffer was not pinned (and therefore surely not
+			 * locked) when StrategyGetBuffer returned it, someone else could
+			 * have pinned and exclusive-locked it by the time we get here. If
+			 * we try to get the lock unconditionally, we'd block waiting for
+			 * them; if they later block waiting for us, deadlock ensues.
+			 * (This has been observed to happen when two backends are both
+			 * trying to split btree index pages, and the second one just
+			 * happens to be trying to split the page the first one got from
+			 * StrategyGetBuffer.)
+			 */
+			if (LWLockConditionalAcquire(BufferDescriptorGetContentLock(buf),
+										 LW_SHARED))
+			{
+				/*
+				 * If using a nondefault strategy, and writing the buffer
+				 * would require a WAL flush, let the strategy decide whether
+				 * to go ahead and write/reuse the buffer or to choose another
+				 * victim.  We need lock to inspect the page LSN, so this
+				 * can't be done inside StrategyGetBuffer.
+				 */
+				if (strategy != NULL)
+				{
+					XLogRecPtr	lsn;
+
+					/* Read the LSN while holding buffer header lock */
+					buf_state = LockBufHdr(buf);
+					lsn = BufferGetLSN(buf);
+					UnlockBufHdr(buf, buf_state);
+
+					if (XLogNeedsFlush(lsn) &&
+						StrategyRejectBuffer(strategy, buf))
+					{
+						/* Drop lock/pin and loop around for another buffer */
+						LWLockRelease(BufferDescriptorGetContentLock(buf));
+						UnpinBuffer(buf, true);
+						continue;
+					}
+				}
+
+				/* OK, do the I/O */
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_START(forkNum, blockNum,
+														  smgr->smgr_rnode.node.spcNode,
+														  smgr->smgr_rnode.node.dbNode,
+														  smgr->smgr_rnode.node.relNode);
+
+				FlushBuffer(buf, NULL);
+				LWLockRelease(BufferDescriptorGetContentLock(buf));
+
+				ScheduleBufferTagForWriteback(&BackendWritebackContext,
+											  &buf->tag);
+
+				TRACE_POSTGRESQL_BUFFER_WRITE_DIRTY_DONE(forkNum, blockNum,
+														 smgr->smgr_rnode.node.spcNode,
+														 smgr->smgr_rnode.node.dbNode,
+														 smgr->smgr_rnode.node.relNode);
+			}
+			else
+			{
+				/*
+				 * Someone else has locked the buffer, so give it up and loop
+				 * back to get another one.
+				 */
+				UnpinBuffer(buf, true);
+				continue;
+			}
+		}
+
+		/*
+		 * To change the association of a valid buffer, we'll need to have
+		 * exclusive lock on both the old and new mapping partitions.
+		 */
+		if (oldFlags & BM_TAG_VALID)
+		{
+			/*
+			 * Need to compute the old tag's hashcode and partition lock ID.
+			 * XXX is it worth storing the hashcode in BufferDesc so we need
+			 * not recompute it here?  Probably not.
+			 */
+			oldTag = buf->tag;
+			oldHash = BufTableHashCode(&oldTag);
+			oldPartitionLock = BufMappingPartitionLock(oldHash);
+
+			if (*lastPartitionLock && 
+				(*lastPartitionLock != oldPartitionLock || 
+				 *lastPartitionLock != newPartitionLock))
+			{
+				LWLockRelease(*lastPartitionLock);
+				*lastPartitionLock = NULL;
+			}
+
+			/*
+			 * Must lock the lower-numbered partition first to avoid
+			 * deadlocks.
+			 */
+			if (oldPartitionLock < newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+			else if (oldPartitionLock > newPartitionLock)
+			{
+				Assert(*lastPartitionLock == NULL);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+				LWLockAcquire(oldPartitionLock, LW_EXCLUSIVE);
+			}
+			else
+			{
+				/* only one partition, only one lock */
+				if (*lastPartitionLock != newPartitionLock)
+					LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+		}
+		else
+		{
+			/* if it wasn't valid, we need only the new partition */
+			if (*lastPartitionLock != newPartitionLock)
+			{
+				if (*lastPartitionLock)
+					LWLockRelease(*lastPartitionLock);
+				LWLockAcquire(newPartitionLock, LW_EXCLUSIVE);
+			}
+
+			/* remember we have no old-partition lock or tag */
+			oldPartitionLock = NULL;
+			/* keep the compiler quiet about uninitialized variables */
+			oldHash = 0;
+		}
+
+		*lastPartitionLock = newPartitionLock;
+
+		/*
+		 * Try to make a hashtable entry for the buffer under its new tag.
+		 * This could fail because while we were writing someone else
+		 * allocated another buffer for the same block we want to read in.
+		 * Note that we have not yet removed the hashtable entry for the old
+		 * tag.
+		 */
+		buf_id = BufTableInsert(&newTag, newHash, buf->buf_id);
+		Assert(buf_id < 0);
+
+		/*
+		 * Need to lock the buffer header too in order to change its tag.
+		 */
+		buf_state = LockBufHdr(buf);
+
+		/*
+		 * Somebody could have pinned or re-dirtied the buffer while we were
+		 * doing the I/O and making the new hashtable entry.  If so, we can't
+		 * recycle this buffer; we must undo everything we've done and start
+		 * over with a new victim buffer.
+		 */
+		oldFlags = buf_state & BUF_FLAG_MASK;
+		if (BUF_STATE_GET_REFCOUNT(buf_state) == 1 && !(oldFlags & BM_DIRTY))
+			break;
+
+		pg_unreachable();
+	}
+
+	/*
+	 * Okay, it's finally safe to rename the buffer.
+	 *
+	 * Clearing BM_VALID here is necessary, clearing the dirtybits is just
+	 * paranoia.  We also reset the usage_count since any recency of use of
+	 * the old content is no longer relevant.  (The usage_count starts out at
+	 * 1 so that the buffer can survive one clock-sweep pass.)
+	 *
+	 * Make sure BM_PERMANENT is set for buffers that must be written at every
+	 * checkpoint.  Unlogged buffers only need to be written at shutdown
+	 * checkpoints, except for their "init" forks, which need to be treated
+	 * just like permanent relations.
+	 */
+	buf->tag = newTag;
+	buf_state &= ~(BM_VALID | BM_DIRTY | BM_JUST_DIRTIED |
+				   BM_CHECKPOINT_NEEDED | BM_IO_ERROR | BM_PERMANENT |
+				   BUF_USAGECOUNT_MASK);
+	if (relpersistence == RELPERSISTENCE_PERMANENT || forkNum == INIT_FORKNUM)
+		buf_state |= BM_TAG_VALID | BM_PERMANENT | BUF_USAGECOUNT_ONE;
+	else
+		buf_state |= BM_TAG_VALID | BUF_USAGECOUNT_ONE;
+
+	UnlockBufHdr(buf, buf_state);
+
+	if (oldPartitionLock != NULL)
+	{
+		BufTableDelete(&oldTag, oldHash);
+		if (oldPartitionLock != newPartitionLock)
+			LWLockRelease(oldPartitionLock);
+	}
+
+	/*
+	 * Buffer contents are currently invalid.  Try to get the io_in_progress
+	 * lock.  If StartBufferIO returns false, then someone else managed to
+	 * read it before we did, so there's nothing left for BufferAlloc() to do.
+	 */
+	StartBufferIO(buf, true);
+
+	return buf;
+}
+
+
 /*
  * InvalidateBuffer -- mark a shared buffer invalid and return it to the
  * freelist.
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index 3377fa5676..277705f18c 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -124,7 +124,7 @@ typedef struct buftag
  * NB: NUM_BUFFER_PARTITIONS must be a power of 2!
  */
 #define BufTableHashPartition(hashcode) \
-	((hashcode) % NUM_BUFFER_PARTITIONS)
+	((hashcode >> 7) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
 	(&MainLWLockArray[BUFFER_MAPPING_LWLOCK_OFFSET + \
 		BufTableHashPartition(hashcode)].lock)
-- 
2.25.1


--------------065E078E0383D7C686F94091--





^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* Emit "checkpoint skipped because system is idle" message at LOG level if log_checkpoints is set
@ 2022-01-05 04:54  Bharath Rupireddy <[email protected]>
  0 siblings, 1 reply; 25+ messages in thread

From: Bharath Rupireddy @ 2022-01-05 04:54 UTC (permalink / raw)
  To: PostgreSQL Hackers <[email protected]>

Hi,

Postgres server emits a message at DEBUG1 level when it skips a
checkpoint. At times, developers might be surprised after figuring out
from server logs that there were no checkpoints happening at all
during a certain period of time when DEBUG1 messages aren't captured.
How about emitting the message at LOG level if log_checkpoints is set?
Patch attached.

Thoughts?

Regards,
Bharath Rupireddy.


Attachments:

  [application/octet-stream] v1-0001-Emit-checkpoint-skipped-message-at-LOG-level.patch (902B, ../../CALj2ACX9LPadCNACDZ3JykDT-6sfPrPchXTLyNHP3cbwyJYjsg@mail.gmail.com/2-v1-0001-Emit-checkpoint-skipped-message-at-LOG-level.patch)
  download | inline diff:
From 2e560a0bc89285814a75c06d67dd74b57b1a1d94 Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy <[email protected]>
Date: Wed, 5 Jan 2022 04:52:45 +0000
Subject: [PATCH v1] Emit checkpoint skipped message at LOG level

Emit "checkpoint skipped because system is idle" message at LOG
level if log_checkpoints is set.
---
 src/backend/access/transam/xlog.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index 87cd05c945..fcd4b7b801 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -9159,7 +9159,7 @@ CreateCheckPoint(int flags)
 		{
 			WALInsertLockRelease();
 			END_CRIT_SECTION();
-			ereport(DEBUG1,
+			ereport((log_checkpoints ? LOG : DEBUG1),
 					(errmsg_internal("checkpoint skipped because system is idle")));
 			return;
 		}
-- 
2.25.1



^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* Re: Emit "checkpoint skipped because system is idle" message at LOG level if log_checkpoints is set
@ 2022-01-05 05:15  Dilip Kumar <[email protected]>
  parent: Bharath Rupireddy <[email protected]>
  0 siblings, 1 reply; 25+ messages in thread

From: Dilip Kumar @ 2022-01-05 05:15 UTC (permalink / raw)
  To: Bharath Rupireddy <[email protected]>; +Cc: PostgreSQL Hackers <[email protected]>

On Wed, Jan 5, 2022 at 10:24 AM Bharath Rupireddy
<[email protected]> wrote:
>
> Hi,
>
> Postgres server emits a message at DEBUG1 level when it skips a
> checkpoint. At times, developers might be surprised after figuring out
> from server logs that there were no checkpoints happening at all
> during a certain period of time when DEBUG1 messages aren't captured.
> How about emitting the message at LOG level if log_checkpoints is set?
> Patch attached.
>
> Thoughts?

+1 to convert to LOG when log_checkpoints is set.

-- 
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com






^ permalink  raw  reply  [nested|flat] 25+ messages in thread

* Re: Emit "checkpoint skipped because system is idle" message at LOG level if log_checkpoints is set
@ 2022-01-05 23:18  Justin Pryzby <[email protected]>
  parent: Dilip Kumar <[email protected]>
  0 siblings, 0 replies; 25+ messages in thread

From: Justin Pryzby @ 2022-01-05 23:18 UTC (permalink / raw)
  To: Dilip Kumar <[email protected]>; +Cc: Bharath Rupireddy <[email protected]>; [email protected]

On Wed, Jan 05, 2022 at 10:45:06AM +0530, Dilip Kumar wrote:
> On Wed, Jan 5, 2022 at 10:24 AM Bharath Rupireddy <[email protected]> wrote:
> >
> > Postgres server emits a message at DEBUG1 level when it skips a
> > checkpoint. At times, developers might be surprised after figuring out
> > from server logs that there were no checkpoints happening at all
> > during a certain period of time when DEBUG1 messages aren't captured.
> > How about emitting the message at LOG level if log_checkpoints is set?
> > Patch attached.
> 
> +1 to convert to LOG when log_checkpoints is set.

I think it would be odd to write logs of increased severity, for the case where
we did not do anything.  I think it really is a debug log.

I don't think the log level should be changed to avoid "developer" confusion,
as you said (I'm not sure if you mean a postgres developer or an application
developer, though).

Is there any evidence that this has caused user confusion in the last 4 years ?

|commit 6ef2eba3f57f17960b7cd4958e18aa79e357de2f
|Author: Andres Freund <[email protected]>
|Date:   Thu Dec 22 11:31:50 2016 -0800
|
|    Skip checkpoints, archiving on idle systems.

Note that logging a message may not be benign ; I think it could cause the
disks to spin up, that would othewise have been in power saving mode,
especially if you log to syslog, which can issue fsync.  Also, part of the
argument for enabling log_checkpoint by default was that a small, quiescent
instance would not write logs every 5 minutes.

-- 
Justin






^ permalink  raw  reply  [nested|flat] 25+ messages in thread

end of thread, other threads:[~2022-01-05 23:18 UTC | newest]

Thread overview: 25+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2020-12-31 12:20 [PATCH v1 2/2] WIP: buffer alloc specialized for relation extension Luc Vlaming <[email protected]>
2022-01-05 04:54 Emit "checkpoint skipped because system is idle" message at LOG level if log_checkpoints is set Bharath Rupireddy <[email protected]>
2022-01-05 05:15 ` Re: Emit "checkpoint skipped because system is idle" message at LOG level if log_checkpoints is set Dilip Kumar <[email protected]>
2022-01-05 23:18   ` Re: Emit "checkpoint skipped because system is idle" message at LOG level if log_checkpoints is set Justin Pryzby <[email protected]>

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox