Avoid creation of the free space map for small heap relations.

Previously, all heaps had FSMs. For very small tables, this means that the
FSM took up more space than the heap did. This is wasteful, so now we
refrain from creating the FSM for heaps with 4 pages or fewer. If the last
known target block has insufficient space, we still try to insert into some
other page before giving up and extending the relation, since doing
otherwise leads to table bloat. Testing showed that trying every page
penalized performance slightly, so we compromise and try every other page.
This way, we visit at most two pages. Any pages with wasted free space
become visible at next relation extension, so we still control table bloat.
As a bonus, directly attempting one or two pages can even be faster than
consulting the FSM would have been.

Once the FSM is created for a heap we don't remove it even if somebody
deletes all the rows from the corresponding relation.  We don't think it is
a useful optimization as it is quite likely that relation will again grow
to the same size.

Author: John Naylor with design inputs and some code contribution by Amit Kapila
Reviewed-by: Amit Kapila
Tested-by: Mithun C Y
Discussion: https://www.postgresql.org/message-id/CAJVSVGWvB13PzpbLEecFuGFc5V2fsO736BsdTakPiPAcdMM5tQ@mail.gmail.com

Author: Amit Kapila

contrib/pageinspect/expected/page.out

@@ -1,48 +1,69 @@
 CREATE EXTENSION pageinspect;
-CREATE TABLE test1 (a int, b int);
-INSERT INTO test1 VALUES (16777217, 131584);
-VACUUM test1;  -- set up FSM
+CREATE TABLE test_rel_forks (a int);
+-- Make sure there are enough blocks in the heap for the FSM to be created.
+INSERT INTO test_rel_forks SELECT i from generate_series(1,1000) i;
+-- set up FSM and VM
+VACUUM test_rel_forks;
 -- The page contents can vary, so just test that it can be read
 -- successfully, but don't keep the output.
-SELECT octet_length(get_raw_page('test1', 'main', 0)) AS main_0;
+SELECT octet_length(get_raw_page('test_rel_forks', 'main', 0)) AS main_0;
  main_0 
 --------
    8192
 (1 row)
 
-SELECT octet_length(get_raw_page('test1', 'main', 1)) AS main_1;
-ERROR:  block number 1 is out of range for relation "test1"
-SELECT octet_length(get_raw_page('test1', 'fsm', 0)) AS fsm_0;
+SELECT octet_length(get_raw_page('test_rel_forks', 'main', 100)) AS main_100;
+ERROR:  block number 100 is out of range for relation "test_rel_forks"
+SELECT octet_length(get_raw_page('test_rel_forks', 'fsm', 0)) AS fsm_0;
  fsm_0 
 -------
   8192
 (1 row)
 
-SELECT octet_length(get_raw_page('test1', 'fsm', 1)) AS fsm_1;
- fsm_1 
--------
-  8192
-(1 row)
-
-SELECT octet_length(get_raw_page('test1', 'vm', 0)) AS vm_0;
+SELECT octet_length(get_raw_page('test_rel_forks', 'fsm', 10)) AS fsm_10;
+ERROR:  block number 10 is out of range for relation "test_rel_forks"
+SELECT octet_length(get_raw_page('test_rel_forks', 'vm', 0)) AS vm_0;
  vm_0 
 ------
  8192
 (1 row)
 
-SELECT octet_length(get_raw_page('test1', 'vm', 1)) AS vm_1;
-ERROR:  block number 1 is out of range for relation "test1"
+SELECT octet_length(get_raw_page('test_rel_forks', 'vm', 1)) AS vm_1;
+ERROR:  block number 1 is out of range for relation "test_rel_forks"
 SELECT octet_length(get_raw_page('xxx', 'main', 0));
 ERROR:  relation "xxx" does not exist
-SELECT octet_length(get_raw_page('test1', 'xxx', 0));
+SELECT octet_length(get_raw_page('test_rel_forks', 'xxx', 0));
 ERROR:  invalid fork name
 HINT:  Valid fork names are "main", "fsm", "vm", and "init".
-SELECT get_raw_page('test1', 0) = get_raw_page('test1', 'main', 0);
+SELECT * FROM fsm_page_contents(get_raw_page('test_rel_forks', 'fsm', 0));
+ fsm_page_contents 
+-------------------
+ 0: 147           +
+ 1: 147           +
+ 3: 147           +
+ 7: 147           +
+ 15: 147          +
+ 31: 147          +
+ 63: 147          +
+ 127: 147         +
+ 255: 147         +
+ 511: 147         +
+ 1023: 147        +
+ 2047: 147        +
+ 4095: 147        +
+ fp_next_slot: 0  +
+ 
+(1 row)
+
+SELECT get_raw_page('test_rel_forks', 0) = get_raw_page('test_rel_forks', 'main', 0);
  ?column? 
 ----------
  t
 (1 row)
 
+DROP TABLE test_rel_forks;
+CREATE TABLE test1 (a int, b int);
+INSERT INTO test1 VALUES (16777217, 131584);
 SELECT pagesize, version FROM page_header(get_raw_page('test1', 0));
  pagesize | version 
 ----------+---------
@@ -62,26 +83,6 @@ SELECT tuple_data_split('test1'::regclass, t_data, t_infomask, t_infomask2, t_bi
  {"\\x01000001","\\x00020200"}
 (1 row)
 
-SELECT * FROM fsm_page_contents(get_raw_page('test1', 'fsm', 0));
- fsm_page_contents 
--------------------
- 0: 254           +
- 1: 254           +
- 3: 254           +
- 7: 254           +
- 15: 254          +
- 31: 254          +
- 63: 254          +
- 127: 254         +
- 255: 254         +
- 511: 254         +
- 1023: 254        +
- 2047: 254        +
- 4095: 254        +
- fp_next_slot: 0  +
- 
-(1 row)
-
 DROP TABLE test1;
 -- check that using any of these functions with a partitioned table or index
 -- would fail

contrib/pageinspect/sql/page.sql

@@ -1,26 +1,35 @@
 CREATE EXTENSION pageinspect;
 
-CREATE TABLE test1 (a int, b int);
-INSERT INTO test1 VALUES (16777217, 131584);
+CREATE TABLE test_rel_forks (a int);
+-- Make sure there are enough blocks in the heap for the FSM to be created.
+INSERT INTO test_rel_forks SELECT i from generate_series(1,1000) i;
 
-VACUUM test1;  -- set up FSM
+-- set up FSM and VM
+VACUUM test_rel_forks;
 
 -- The page contents can vary, so just test that it can be read
 -- successfully, but don't keep the output.
 
-SELECT octet_length(get_raw_page('test1', 'main', 0)) AS main_0;
-SELECT octet_length(get_raw_page('test1', 'main', 1)) AS main_1;
+SELECT octet_length(get_raw_page('test_rel_forks', 'main', 0)) AS main_0;
+SELECT octet_length(get_raw_page('test_rel_forks', 'main', 100)) AS main_100;
 
-SELECT octet_length(get_raw_page('test1', 'fsm', 0)) AS fsm_0;
-SELECT octet_length(get_raw_page('test1', 'fsm', 1)) AS fsm_1;
+SELECT octet_length(get_raw_page('test_rel_forks', 'fsm', 0)) AS fsm_0;
+SELECT octet_length(get_raw_page('test_rel_forks', 'fsm', 10)) AS fsm_10;
 
-SELECT octet_length(get_raw_page('test1', 'vm', 0)) AS vm_0;
-SELECT octet_length(get_raw_page('test1', 'vm', 1)) AS vm_1;
+SELECT octet_length(get_raw_page('test_rel_forks', 'vm', 0)) AS vm_0;
+SELECT octet_length(get_raw_page('test_rel_forks', 'vm', 1)) AS vm_1;
 
 SELECT octet_length(get_raw_page('xxx', 'main', 0));
-SELECT octet_length(get_raw_page('test1', 'xxx', 0));
+SELECT octet_length(get_raw_page('test_rel_forks', 'xxx', 0));
+
+SELECT * FROM fsm_page_contents(get_raw_page('test_rel_forks', 'fsm', 0));
+
+SELECT get_raw_page('test_rel_forks', 0) = get_raw_page('test_rel_forks', 'main', 0);
 
-SELECT get_raw_page('test1', 0) = get_raw_page('test1', 'main', 0);
+DROP TABLE test_rel_forks;
+
+CREATE TABLE test1 (a int, b int);
+INSERT INTO test1 VALUES (16777217, 131584);
 
 SELECT pagesize, version FROM page_header(get_raw_page('test1', 0));
 
@@ -29,8 +38,6 @@ SELECT page_checksum(get_raw_page('test1', 0), 0) IS NOT NULL AS silly_checksum_
 SELECT tuple_data_split('test1'::regclass, t_data, t_infomask, t_infomask2, t_bits)
     FROM heap_page_items(get_raw_page('test1', 0));
 
-SELECT * FROM fsm_page_contents(get_raw_page('test1', 'fsm', 0));
-
 DROP TABLE test1;
 
 -- check that using any of these functions with a partitioned table or index

doc/src/sgml/storage.sgml

@@ -590,12 +590,13 @@ tuple would otherwise be too big.
 <indexterm><primary>FSM</primary><see>Free Space Map</see></indexterm>
 
 <para>
-Each heap and index relation, except for hash indexes, has a Free Space Map
-(FSM) to keep track of available space in the relation. It's stored
-alongside the main relation data in a separate relation fork, named after the
-filenode number of the relation, plus a <literal>_fsm</literal> suffix. For example,
-if the filenode of a relation is 12345, the FSM is stored in a file called
-<filename>12345_fsm</filename>, in the same directory as the main relation file.
+Each heap relation, unless it is very small, and each index relation, except
+for hash indexes, has a Free Space Map (FSM) to keep track of available
+space in the relation. It's stored alongside the main relation data in a
+separate relation fork, named after the filenode number of the relation, plus
+a <literal>_fsm</literal> suffix. For example, if the filenode of a relation
+is 12345, the FSM is stored in a file called <filename>12345_fsm</filename>,
+in the same directory as the main relation file.
 </para>
 
 <para>

src/backend/access/brin/brin.c

@@ -1150,7 +1150,7 @@ terminate_brin_buildstate(BrinBuildState *state)
 		freespace = PageGetFreeSpace(page);
 		blk = BufferGetBlockNumber(state->bs_currentInsertBuf);
 		ReleaseBuffer(state->bs_currentInsertBuf);
-		RecordPageWithFreeSpace(state->bs_irel, blk, freespace);
+		RecordPageWithFreeSpace(state->bs_irel, blk, freespace, InvalidBlockNumber);
 		FreeSpaceMapVacuumRange(state->bs_irel, blk, blk + 1);
 	}
 

src/backend/access/brin/brin_pageops.c

@@ -310,7 +310,7 @@ brin_doupdate(Relation idxrel, BlockNumber pagesPerRange,
 
 		if (extended)
 		{
-			RecordPageWithFreeSpace(idxrel, newblk, freespace);
+			RecordPageWithFreeSpace(idxrel, newblk, freespace, InvalidBlockNumber);
 			FreeSpaceMapVacuumRange(idxrel, newblk, newblk + 1);
 		}
 
@@ -461,7 +461,7 @@ brin_doinsert(Relation idxrel, BlockNumber pagesPerRange,
 
 	if (extended)
 	{
-		RecordPageWithFreeSpace(idxrel, blk, freespace);
+		RecordPageWithFreeSpace(idxrel, blk, freespace, InvalidBlockNumber);
 		FreeSpaceMapVacuumRange(idxrel, blk, blk + 1);
 	}
 
@@ -654,7 +654,7 @@ brin_page_cleanup(Relation idxrel, Buffer buf)
 
 	/* Measure free space and record it */
 	RecordPageWithFreeSpace(idxrel, BufferGetBlockNumber(buf),
-							br_page_get_freespace(page));
+							br_page_get_freespace(page), InvalidBlockNumber);
 }
 
 /*
@@ -703,7 +703,7 @@ brin_getinsertbuffer(Relation irel, Buffer oldbuf, Size itemsz,
 	/* Choose initial target page, re-using existing target if known */
 	newblk = RelationGetTargetBlock(irel);
 	if (newblk == InvalidBlockNumber)
-		newblk = GetPageWithFreeSpace(irel, itemsz);
+		newblk = GetPageWithFreeSpace(irel, itemsz, true);
 
 	/*
 	 * Loop until we find a page with sufficient free space.  By the time we
@@ -895,7 +895,7 @@ brin_initialize_empty_new_buffer(Relation idxrel, Buffer buffer)
 	 * pages whose FSM records were forgotten in a crash.
 	 */
 	RecordPageWithFreeSpace(idxrel, BufferGetBlockNumber(buffer),
-							br_page_get_freespace(page));
+							br_page_get_freespace(page), InvalidBlockNumber);
 }
 
 

src/backend/access/heap/hio.c

@@ -239,8 +239,14 @@ RelationAddExtraBlocks(Relation relation, BulkInsertState bistate)
 		 * Immediately update the bottom level of the FSM.  This has a good
 		 * chance of making this page visible to other concurrently inserting
 		 * backends, and we want that to happen without delay.
+		 *
+		 * Since we know the table will end up with extraBlocks additional
+		 * pages, we pass the final number to avoid possible unnecessary
+		 * system calls and to make sure the FSM is created when we add the
+		 * first new page.
 		 */
-		RecordPageWithFreeSpace(relation, blockNum, freespace);
+		RecordPageWithFreeSpace(relation, blockNum, freespace,
+								firstBlock + extraBlocks);
 	}
 	while (--extraBlocks > 0);
 
@@ -377,20 +383,9 @@ RelationGetBufferForTuple(Relation relation, Size len,
 		 * We have no cached target page, so ask the FSM for an initial
 		 * 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 (nblocks > 0)
-				targetBlock = nblocks - 1;
-		}
+		targetBlock = GetPageWithFreeSpace(relation,
+										   len + saveFreeSpace,
+										   false);
 	}
 
 loop:
@@ -484,6 +479,14 @@ RelationGetBufferForTuple(Relation relation, Size len,
 		{
 			/* use this page as future insert target, too */
 			RelationSetTargetBlock(relation, targetBlock);
+
+			/*
+			 * In case we used an in-memory map of available blocks, reset it
+			 * for next use.
+			 */
+			if (targetBlock < HEAP_FSM_CREATION_THRESHOLD)
+				FSMClearLocalMap();
+
 			return buffer;
 		}
 
@@ -543,9 +546,12 @@ RelationGetBufferForTuple(Relation relation, Size len,
 
 			/*
 			 * Check if some other backend has extended a block for us while
-			 * we were waiting on the lock.
+			 * we were waiting on the lock.  We only check the FSM -- if there
+			 * isn't one we don't recheck the number of blocks.
 			 */
-			targetBlock = GetPageWithFreeSpace(relation, len + saveFreeSpace);
+			targetBlock = GetPageWithFreeSpace(relation,
+											   len + saveFreeSpace,
+											   true);
 
 			/*
 			 * If some other waiter has already extended the relation, we
@@ -625,5 +631,12 @@ RelationGetBufferForTuple(Relation relation, Size len,
 	 */
 	RelationSetTargetBlock(relation, BufferGetBlockNumber(buffer));
 
+	/*
+	 * In case we used an in-memory map of available blocks, reset it for next
+	 * use.  We do this unconditionally since after relation extension we
+	 * can't skip this based on the targetBlock.
+	 */
+	FSMClearLocalMap();
+
 	return buffer;
 }