Subject: Re: COPY FREEZE and setting PD_ALL_VISIBLE/visibility map bits
To: Tomas Vondra <tomas.vondra@enterprisedb.com>,
 Tomas Vondra <tomas.vondra@2ndquadrant.com>
Cc: Tatsuo Ishii <ishii@sraoss.co.jp>, lubennikovaav@gmail.com,
 pgsql-hackers@lists.postgresql.org, pavan.deolasee@gmail.com,
 ibrar.ahmad@gmail.com
References: 
 <CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com>
 <CALtqXTfpEOiyJchYQoUKw9pvEPhVgnkxL=JvKNj57BTxva8bTA@mail.gmail.com>
 <160382617166.1204.475752573208579406.pgcf@coridan.postgresql.org>
 <20201028.144653.966121790856478893.t-ishii@sraoss.co.jp>
 <20201030004252.d63rwclys4bniluu@development>
 <1d2e0f4a-6ab8-10f1-c177-af97475df001@postgrespro.ru>
 <20201102144453.tumlk7qhw3gzn2dk@development>
 <7e6a006f-c59a-93fc-3ddf-f9c661319bc8@2ndquadrant.com>
 <75f3760f-1077-07c1-aea4-975d5c127d6d@postgrespro.ru>
 <f6e0af44-b147-d27f-cc21-2404b4f11f0e@enterprisedb.com>
 <93b767d6-4cde-7326-8de4-d71d81817c16@postgrespro.ru>
 <8669d452-d078-bdf9-5a62-6f1acd55f3c6@enterprisedb.com>
From: Anastasia Lubennikova <a.lubennikova@postgrespro.ru>
Message-ID: <58359211-c505-1ee6-c5fd-78db84afb46f@postgrespro.ru>
Date: Sat, 16 Jan 2021 18:11:29 +0300
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.10.0
MIME-Version: 1.0
In-Reply-To: <8669d452-d078-bdf9-5a62-6f1acd55f3c6@enterprisedb.com>
Content-Type: multipart/mixed;
 boundary="------------85FACDF4FECA0056FC4211D0"
Content-Language: en-US
Precedence: bulk

This is a multi-part message in MIME format.
--------------85FACDF4FECA0056FC4211D0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit

On 12.01.2021 22:30, Tomas Vondra wrote:
> Thanks. These patches seem to resolve the TOAST table issue, freezing 
> it as expected. I think the code duplication is not an issue, but I 
> wonder why heap_insert uses this condition:
>
>     /*
>      * ...
>      *
>      * No need to update the visibilitymap if it had all_frozen bit set
>      * before this insertion.
>      */
>     if (all_frozen_set && ((vmstatus & VISIBILITYMAP_ALL_FROZEN) == 0))
>
> while heap_multi_insert only does this:
>
>     if (all_frozen_set) { ... }
>
> I haven't looked at the details, but shouldn't both do the same thing?


I decided to add this check for heap_insert() to avoid unneeded calls of 
visibilitymap_set(). If we insert tuples one by one, we can only call 
this once per page.
In my understanding, heap_multi_insert() inserts tuples in batches, so 
it doesn't need this optimization.

>
>
> However, I've also repeated the test counting all-frozen pages in both 
> the main table and TOAST table, and I get this:
>
> patched
> =======
>
> select count(*) from pg_visibility((select reltoastrelid from pg_class 
> where relname = 't'));
>
>  count
> --------
>  100002
> (1 row)
>
>
> select count(*) from pg_visibility((select reltoastrelid from pg_class 
> where relname = 't')) where not all_visible;
>
>  count
> --------
>       0
> (1 row)
>
> That is - all TOAST pages are frozen (as expected, which is good). But 
> now there are 100002 pages, not just 100000 pages. That is, we're now 
> creating 2 extra pages, for some reason. I recall Pavan reported 
> similar issue with every 32768-th page not being properly filled, but 
> I'm not sure if that's the same issue.
>
>
> regards
>

As Pavan correctly figured it out before the problem is that 
RelationGetBufferForTuple() moves to the next page, losing free space in 
the block:

 > ... I see that a relcache invalidation arrives
 > after 1st and then after every 32672th block is filled. That clears the
 > rel->rd_smgr field and we lose the information about the saved target
 > block. The code then moves to extend the relation again and thus 
skips the
 > previously less-than-half filled block, losing the free space in that 
block.

The reason of this cache invalidation is vm_extend() call, which happens 
every 32762 blocks.

RelationGetBufferForTuple() tries to use the last page, but for some 
reason this code is under 'use_fsm' check. And COPY FROM doesn't use fsm 
(see TABLE_INSERT_SKIP_FSM).


         /*
          * If the FSM knows nothing of the rel, try the last page before we
          * give up and extend.  This avoids one-tuple-per-page syndrome 
during
          * bootstrapping or in a recently-started system.
          */
         if (targetBlock == InvalidBlockNumber)
         {
             BlockNumber nblocks = RelationGetNumberOfBlocks(relation);
             if (nblocks > 0)
                 targetBlock = nblocks - 1;
         }


I think we can use this code without regard to 'use_fsm'. With this 
change, the number of toast rel pages is correct. The patch is attached.

-- 
Anastasia Lubennikova
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company


--------------85FACDF4FECA0056FC4211D0
Content-Type: text/x-patch; charset=UTF-8;
 name="0001-Set-PD_ALL_VISIBLE-and-visibility-map-bits-in-COP-v7.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
 filename*0="0001-Set-PD_ALL_VISIBLE-and-visibility-map-bits-in-COP-v7.pa";
 filename*1="tch"

diff --git a/contrib/pg_visibility/expected/pg_visibility.out b/contrib/pg_visibility/expected/pg_visibility.out
index ca4b6e186b..0017e3415c 100644
--- a/contrib/pg_visibility/expected/pg_visibility.out
+++ b/contrib/pg_visibility/expected/pg_visibility.out
@@ -179,6 +179,69 @@ select pg_truncate_visibility_map('test_partition');
  
 (1 row)
 
+-- test copy freeze
+create table copyfreeze (a int, b char(1500));
+-- load all rows via COPY FREEZE and ensure that all pages are set all-visible
+-- and all-frozen.
+begin;
+truncate copyfreeze;
+copy copyfreeze from stdin freeze;
+commit;
+select * from pg_visibility_map('copyfreeze');
+ blkno | all_visible | all_frozen 
+-------+-------------+------------
+     0 | t           | t
+     1 | t           | t
+     2 | t           | t
+(3 rows)
+
+select * from pg_check_frozen('copyfreeze');
+ t_ctid 
+--------
+(0 rows)
+
+-- load half the rows via regular COPY and rest via COPY FREEZE. The pages
+-- which are touched by regular COPY must not be set all-visible/all-frozen. On
+-- the other hand, pages allocated by COPY FREEZE should be marked
+-- all-frozen/all-visible.
+begin;
+truncate copyfreeze;
+copy copyfreeze from stdin;
+copy copyfreeze from stdin freeze;
+commit;
+select * from pg_visibility_map('copyfreeze');
+ blkno | all_visible | all_frozen 
+-------+-------------+------------
+     0 | f           | f
+     1 | f           | f
+     2 | t           | t
+(3 rows)
+
+select * from pg_check_frozen('copyfreeze');
+ t_ctid 
+--------
+(0 rows)
+
+-- Try a mix of regular COPY and COPY FREEZE.
+begin;
+truncate copyfreeze;
+copy copyfreeze from stdin freeze;
+copy copyfreeze from stdin;
+copy copyfreeze from stdin freeze;
+commit;
+select * from pg_visibility_map('copyfreeze');
+ blkno | all_visible | all_frozen 
+-------+-------------+------------
+     0 | t           | t
+     1 | f           | f
+     2 | t           | t
+(3 rows)
+
+select * from pg_check_frozen('copyfreeze');
+ t_ctid 
+--------
+(0 rows)
+
 -- cleanup
 drop table test_partitioned;
 drop view test_view;
@@ -188,3 +251,4 @@ drop server dummy_server;
 drop foreign data wrapper dummy;
 drop materialized view matview_visibility_test;
 drop table regular_table;
+drop table copyfreeze;
diff --git a/contrib/pg_visibility/sql/pg_visibility.sql b/contrib/pg_visibility/sql/pg_visibility.sql
index f79b54480b..ec1afd4906 100644
--- a/contrib/pg_visibility/sql/pg_visibility.sql
+++ b/contrib/pg_visibility/sql/pg_visibility.sql
@@ -94,6 +94,82 @@ select count(*) > 0 from pg_visibility_map_summary('test_partition');
 select * from pg_check_frozen('test_partition'); -- hopefully none
 select pg_truncate_visibility_map('test_partition');
 
+-- test copy freeze
+create table copyfreeze (a int, b char(1500));
+
+-- load all rows via COPY FREEZE and ensure that all pages are set all-visible
+-- and all-frozen.
+begin;
+truncate copyfreeze;
+copy copyfreeze from stdin freeze;
+1	'1'
+2	'2'
+3	'3'
+4	'4'
+5	'5'
+6	'6'
+7	'7'
+8	'8'
+9	'9'
+10	'10'
+11	'11'
+12	'12'
+\.
+commit;
+select * from pg_visibility_map('copyfreeze');
+select * from pg_check_frozen('copyfreeze');
+
+-- load half the rows via regular COPY and rest via COPY FREEZE. The pages
+-- which are touched by regular COPY must not be set all-visible/all-frozen. On
+-- the other hand, pages allocated by COPY FREEZE should be marked
+-- all-frozen/all-visible.
+begin;
+truncate copyfreeze;
+copy copyfreeze from stdin;
+1	'1'
+2	'2'
+3	'3'
+4	'4'
+5	'5'
+6	'6'
+\.
+copy copyfreeze from stdin freeze;
+7	'7'
+8	'8'
+9	'9'
+10	'10'
+11	'11'
+12	'12'
+\.
+commit;
+select * from pg_visibility_map('copyfreeze');
+select * from pg_check_frozen('copyfreeze');
+
+-- Try a mix of regular COPY and COPY FREEZE.
+begin;
+truncate copyfreeze;
+copy copyfreeze from stdin freeze;
+1	'1'
+2	'2'
+3	'3'
+4	'4'
+5	'5'
+\.
+copy copyfreeze from stdin;
+6	'6'
+\.
+copy copyfreeze from stdin freeze;
+7	'7'
+8	'8'
+9	'9'
+10	'10'
+11	'11'
+12	'12'
+\.
+commit;
+select * from pg_visibility_map('copyfreeze');
+select * from pg_check_frozen('copyfreeze');
+
 -- cleanup
 drop table test_partitioned;
 drop view test_view;
@@ -103,3 +179,4 @@ drop server dummy_server;
 drop foreign data wrapper dummy;
 drop materialized view matview_visibility_test;
 drop table regular_table;
+drop table copyfreeze;
diff --git a/src/backend/access/heap/heapam.c b/src/backend/access/heap/heapam.c
index 5b9cfb26cf..3158a4de0d 100644
--- a/src/backend/access/heap/heapam.c
+++ b/src/backend/access/heap/heapam.c
@@ -1880,8 +1880,12 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid,
 	TransactionId xid = GetCurrentTransactionId();
 	HeapTuple	heaptup;
 	Buffer		buffer;
+	Page		page;
 	Buffer		vmbuffer = InvalidBuffer;
+	bool		starting_with_empty_page;
 	bool		all_visible_cleared = false;
+	bool		all_frozen_set = false;
+	uint8		vmstatus = 0;
 
 	/*
 	 * Fill in tuple header fields and toast the tuple if necessary.
@@ -1894,11 +1898,36 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid,
 	/*
 	 * Find buffer to insert this tuple into.  If the page is all visible,
 	 * this will also pin the requisite visibility map page.
+	 *
+	 * Also pin visibility map page if COPY FREEZE inserts tuples into an
+	 * empty page. See all_frozen_set below.
 	 */
 	buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
 									   InvalidBuffer, options, bistate,
 									   &vmbuffer, NULL);
 
+
+	/*
+	 * If we're inserting frozen entry into an empty page,
+	 * set visibility map bits and PageAllVisible() hint.
+	 *
+	 * If we're inserting frozen entry into already all_frozen page,
+	 * preserve this state.
+	 */
+	if (options & HEAP_INSERT_FROZEN)
+	{
+		page = BufferGetPage(buffer);
+
+		starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
+
+		if (visibilitymap_pin_ok(BufferGetBlockNumber(buffer), vmbuffer))
+			vmstatus = visibilitymap_get_status(relation,
+								 BufferGetBlockNumber(buffer), &vmbuffer);
+
+		if ((starting_with_empty_page || vmstatus & VISIBILITYMAP_ALL_FROZEN))
+			all_frozen_set = true;
+	}
+
 	/*
 	 * We're about to do the actual insert -- but check for conflict first, to
 	 * avoid possibly having to roll back work we've just done.
@@ -1922,7 +1951,14 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid,
 	RelationPutHeapTuple(relation, buffer, heaptup,
 						 (options & HEAP_INSERT_SPECULATIVE) != 0);
 
-	if (PageIsAllVisible(BufferGetPage(buffer)))
+	/*
+	 * If the page is all visible, need to clear that, unless we're only
+	 * going to add further frozen rows to it.
+	 *
+	 * If we're only adding already frozen rows to a previously empty
+	 * or all visible page, mark it as all-visible.
+	 */
+	if (PageIsAllVisible(BufferGetPage(buffer)) && !(options & HEAP_INSERT_FROZEN))
 	{
 		all_visible_cleared = true;
 		PageClearAllVisible(BufferGetPage(buffer));
@@ -1930,6 +1966,8 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid,
 							ItemPointerGetBlockNumber(&(heaptup->t_self)),
 							vmbuffer, VISIBILITYMAP_VALID_BITS);
 	}
+	else if (all_frozen_set)
+		PageSetAllVisible(page);
 
 	/*
 	 * XXX Should we set PageSetPrunable on this page ?
@@ -1977,6 +2015,8 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid,
 		xlrec.flags = 0;
 		if (all_visible_cleared)
 			xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
+		if (all_frozen_set)
+			xlrec.flags = XLH_INSERT_ALL_FROZEN_SET;
 		if (options & HEAP_INSERT_SPECULATIVE)
 			xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
 		Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
@@ -2025,6 +2065,29 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid,
 
 	END_CRIT_SECTION();
 
+	/*
+	 * If we've frozen everything on the page, update the visibilitymap.
+	 * We're already holding pin on the vmbuffer.
+	 *
+	 * No need to update the visibilitymap if it had all_frozen bit set
+	 * before this insertion.
+	 */
+	if (all_frozen_set && ((vmstatus & VISIBILITYMAP_ALL_FROZEN) == 0))
+	{
+		Assert(PageIsAllVisible(page));
+		Assert(visibilitymap_pin_ok(BufferGetBlockNumber(buffer), vmbuffer));
+
+		/*
+		 * It's fine to use InvalidTransactionId here - this is only used
+		 * when HEAP_INSERT_FROZEN is specified, which intentionally
+		 * violates visibility rules.
+		 */
+		visibilitymap_set(relation, BufferGetBlockNumber(buffer), buffer,
+							InvalidXLogRecPtr, vmbuffer,
+							InvalidTransactionId,
+							VISIBILITYMAP_ALL_VISIBLE | VISIBILITYMAP_ALL_FROZEN);
+	}
+
 	UnlockReleaseBuffer(buffer);
 	if (vmbuffer != InvalidBuffer)
 		ReleaseBuffer(vmbuffer);
@@ -2121,6 +2184,7 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
 	int			ndone;
 	PGAlignedBlock scratch;
 	Page		page;
+	Buffer		vmbuffer = InvalidBuffer;
 	bool		needwal;
 	Size		saveFreeSpace;
 	bool		need_tuple_data = RelationIsLogicallyLogged(relation);
@@ -2175,8 +2239,9 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
 	while (ndone < ntuples)
 	{
 		Buffer		buffer;
-		Buffer		vmbuffer = InvalidBuffer;
+		bool		starting_with_empty_page;
 		bool		all_visible_cleared = false;
+		bool		all_frozen_set = false;
 		int			nthispage;
 
 		CHECK_FOR_INTERRUPTS();
@@ -2184,12 +2249,20 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
 		/*
 		 * Find buffer where at least the next tuple will fit.  If the page is
 		 * all-visible, this will also pin the requisite visibility map page.
+		 *
+		 * Also pin visibility map page if COPY FREEZE inserts tuples into an
+		 * empty page. See all_frozen_set below.
 		 */
 		buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
 										   InvalidBuffer, options, bistate,
 										   &vmbuffer, NULL);
 		page = BufferGetPage(buffer);
 
+		starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
+
+		if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
+			all_frozen_set = true;
+
 		/* NO EREPORT(ERROR) from here till changes are logged */
 		START_CRIT_SECTION();
 
@@ -2223,7 +2296,14 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
 				log_heap_new_cid(relation, heaptup);
 		}
 
-		if (PageIsAllVisible(page))
+		/*
+		 * If the page is all visible, need to clear that, unless we're only
+		 * going to add further frozen rows to it.
+		 *
+		 * If we're only adding already frozen rows to a previously empty
+		 * page, mark it as all-visible.
+		 */
+		if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
 		{
 			all_visible_cleared = true;
 			PageClearAllVisible(page);
@@ -2231,6 +2311,8 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
 								BufferGetBlockNumber(buffer),
 								vmbuffer, VISIBILITYMAP_VALID_BITS);
 		}
+		else if (all_frozen_set)
+			PageSetAllVisible(page);
 
 		/*
 		 * XXX Should we set PageSetPrunable on this page ? See heap_insert()
@@ -2254,8 +2336,7 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
 			 * If the page was previously empty, we can reinit the page
 			 * instead of restoring the whole thing.
 			 */
-			init = (ItemPointerGetOffsetNumber(&(heaptuples[ndone]->t_self)) == FirstOffsetNumber &&
-					PageGetMaxOffsetNumber(page) == FirstOffsetNumber + nthispage - 1);
+			init = starting_with_empty_page;
 
 			/* allocate xl_heap_multi_insert struct from the scratch area */
 			xlrec = (xl_heap_multi_insert *) scratchptr;
@@ -2273,7 +2354,15 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
 			/* the rest of the scratch space is used for tuple data */
 			tupledata = scratchptr;
 
-			xlrec->flags = all_visible_cleared ? XLH_INSERT_ALL_VISIBLE_CLEARED : 0;
+			/* check that the mutually exclusive flags are not both set */
+			Assert (!(all_visible_cleared && all_frozen_set));
+
+			xlrec->flags = 0;
+			if (all_visible_cleared)
+				xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
+			if (all_frozen_set)
+				xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
+
 			xlrec->ntuples = nthispage;
 
 			/*
@@ -2347,13 +2436,39 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
 
 		END_CRIT_SECTION();
 
-		UnlockReleaseBuffer(buffer);
-		if (vmbuffer != InvalidBuffer)
-			ReleaseBuffer(vmbuffer);
+		/*
+		 * If we've frozen everything on the page, update the visibilitymap.
+		 * We're already holding pin on the vmbuffer.
+		 */
+		if (all_frozen_set)
+		{
+			Assert(PageIsAllVisible(page));
+			Assert(visibilitymap_pin_ok(BufferGetBlockNumber(buffer), vmbuffer));
+
+			/*
+			 * It's fine to use InvalidTransactionId here - this is only used
+			 * when HEAP_INSERT_FROZEN is specified, which intentionally
+			 * violates visibility rules.
+			 */
+			visibilitymap_set(relation, BufferGetBlockNumber(buffer), buffer,
+							  InvalidXLogRecPtr, vmbuffer,
+							  InvalidTransactionId,
+							  VISIBILITYMAP_ALL_VISIBLE | VISIBILITYMAP_ALL_FROZEN);
+		}
 
+		UnlockReleaseBuffer(buffer);
 		ndone += nthispage;
+
+		/*
+		 * NB: Only release vmbuffer after inserting all tuples - it's fairly
+		 * likely that we'll insert into subsequent heap pages that are likely
+		 * to use the same vm page.
+		 */
 	}
 
+	if (vmbuffer != InvalidBuffer)
+		ReleaseBuffer(vmbuffer);
+
 	/*
 	 * We're done with the actual inserts.  Check for conflicts again, to
 	 * ensure that all rw-conflicts in to these inserts are detected.  Without
@@ -8655,6 +8770,10 @@ heap_xlog_insert(XLogReaderState *record)
 	ItemPointerSetBlockNumber(&target_tid, blkno);
 	ItemPointerSetOffsetNumber(&target_tid, xlrec->offnum);
 
+	/* check that the mutually exclusive flags are not both set */
+	Assert (!((xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED) &&
+			  (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)));
+
 	/*
 	 * The visibility map may need to be fixed even if the heap page is
 	 * already up-to-date.
@@ -8725,6 +8844,11 @@ heap_xlog_insert(XLogReaderState *record)
 		if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
 			PageClearAllVisible(page);
 
+		/* XLH_INSERT_ALL_FROZEN_SET implies that all tuples are visible */
+		if (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)
+			PageSetAllVisible(page);
+
+
 		MarkBufferDirty(buffer);
 	}
 	if (BufferIsValid(buffer))
@@ -8775,6 +8899,10 @@ heap_xlog_multi_insert(XLogReaderState *record)
 
 	XLogRecGetBlockTag(record, 0, &rnode, NULL, &blkno);
 
+	/* check that the mutually exclusive flags are not both set */
+	Assert (!((xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED) &&
+			  (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)));
+
 	/*
 	 * The visibility map may need to be fixed even if the heap page is
 	 * already up-to-date.
@@ -8864,6 +8992,10 @@ heap_xlog_multi_insert(XLogReaderState *record)
 		if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
 			PageClearAllVisible(page);
 
+		/* XLH_INSERT_ALL_FROZEN_SET implies that all tuples are visible */
+		if (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)
+			PageSetAllVisible(page);
+
 		MarkBufferDirty(buffer);
 	}
 	if (BufferIsValid(buffer))
diff --git a/src/backend/access/heap/hio.c b/src/backend/access/heap/hio.c
index fac3b8e9ff..fb7ad0bab4 100644
--- a/src/backend/access/heap/hio.c
+++ b/src/backend/access/heap/hio.c
@@ -396,19 +396,19 @@ RelationGetBufferForTuple(Relation relation, Size len,
 		 * target.
 		 */
 		targetBlock = GetPageWithFreeSpace(relation, len + saveFreeSpace);
+	}
 
-		/*
-		 * If the FSM knows nothing of the rel, try the last page before we
-		 * give up and extend.  This avoids one-tuple-per-page syndrome during
-		 * bootstrapping or in a recently-started system.
-		 */
-		if (targetBlock == InvalidBlockNumber)
-		{
-			BlockNumber nblocks = RelationGetNumberOfBlocks(relation);
+	/*
+	 * If the FSM knows nothing of the rel, try the last page before we
+	 * give up and extend.  This avoids one-tuple-per-page syndrome during
+	 * bootstrapping or in a recently-started system.
+	 */
+	if (targetBlock == InvalidBlockNumber)
+	{
+		BlockNumber nblocks = RelationGetNumberOfBlocks(relation);
 
-			if (nblocks > 0)
-				targetBlock = nblocks - 1;
-		}
+		if (nblocks > 0)
+			targetBlock = nblocks - 1;
 	}
 
 loop:
@@ -433,6 +433,14 @@ loop:
 			buffer = ReadBufferBI(relation, targetBlock, RBM_NORMAL, bistate);
 			if (PageIsAllVisible(BufferGetPage(buffer)))
 				visibilitymap_pin(relation, targetBlock, vmbuffer);
+
+			/*
+			 * If the page is empty, pin vmbuffer to set all_frozen bit later.
+			 */
+			if ((options & HEAP_INSERT_FROZEN) &&
+				(PageGetMaxOffsetNumber(BufferGetPage(buffer)) == 0))
+				visibilitymap_pin(relation, targetBlock, vmbuffer);
+
 			LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
 		}
 		else if (otherBlock == targetBlock)
@@ -619,6 +627,15 @@ loop:
 	PageInit(page, BufferGetPageSize(buffer), 0);
 	MarkBufferDirty(buffer);
 
+	/*
+	 * The page is empty, pin vmbuffer to set all_frozen bit.
+	 */
+	if (options & HEAP_INSERT_FROZEN)
+	{
+		Assert(PageGetMaxOffsetNumber(BufferGetPage(buffer)) == 0);
+		visibilitymap_pin(relation, BufferGetBlockNumber(buffer), vmbuffer);
+	}
+
 	/*
 	 * Release the file-extension lock; it's now OK for someone else to extend
 	 * the relation some more.
diff --git a/src/include/access/heapam_xlog.h b/src/include/access/heapam_xlog.h
index 51586b883d..178d49710a 100644
--- a/src/include/access/heapam_xlog.h
+++ b/src/include/access/heapam_xlog.h
@@ -69,6 +69,9 @@
 #define XLH_INSERT_CONTAINS_NEW_TUPLE			(1<<3)
 #define XLH_INSERT_ON_TOAST_RELATION			(1<<4)
 
+/* all_frozen_set always implies all_visible_set */
+#define XLH_INSERT_ALL_FROZEN_SET				(1<<5)
+
 /*
  * xl_heap_update flag values, 8 bits are available.
  */

--------------85FACDF4FECA0056FC4211D0--