Skip to content

Commit 7db0cde

Browse files
author
Amit Kapila
committed
Revert "Avoid the creation of the free space map for small heap relations".
This feature was using a process local map to track the first few blocks in the relation. The map was reset each time we get the block with enough freespace. It was discussed that it would be better to track this map on a per-relation basis in relcache and then invalidate the same whenever vacuum frees up some space in the page or when FSM is created. The new design would be better both in terms of API design and performance. List of commits reverted, in reverse chronological order: 06c8a50 Improve code comments in b0eaa4c. 13e8643 During pg_upgrade, conditionally skip transfer of FSMs. 6f91815 Add more tests for FSM. 9c32e4c Clear the local map when not used. 29d108c Update the documentation for FSM behavior.. 08ecdfe Make FSM test portable. b0eaa4c Avoid creation of the free space map for small heap relations. Discussion: https://postgr.es/m/20190416180452.3pm6uegx54iitbt5@alap3.anarazel.de
1 parent af82f95 commit 7db0cde

File tree

23 files changed

+107
-686
lines changed

23 files changed

+107
-686
lines changed

‎contrib/pageinspect/expected/page.out

Lines changed: 38 additions & 26 deletions
Original file line numberDiff line numberDiff line change
@@ -1,56 +1,48 @@
11
CREATE EXTENSION pageinspect;
2-
CREATE TABLE test_rel_forks (a int);
3-
-- Make sure there are enough blocks in the heap for the FSM to be created.
4-
INSERT INTO test_rel_forks SELECT i from generate_series(1,2000) i;
5-
-- set up FSM and VM
6-
VACUUM test_rel_forks;
2+
CREATE TABLE test1 (a int, b int);
3+
INSERT INTO test1 VALUES (16777217, 131584);
4+
VACUUM test1; -- set up FSM
75
-- The page contents can vary, so just test that it can be read
86
-- successfully, but don't keep the output.
9-
SELECT octet_length(get_raw_page('test_rel_forks', 'main', 0)) AS main_0;
7+
SELECT octet_length(get_raw_page('test1', 'main', 0)) AS main_0;
108
main_0
119
--------
1210
8192
1311
(1 row)
1412

15-
SELECT octet_length(get_raw_page('test_rel_forks', 'main', 100)) AS main_100;
16-
ERROR: block number 100 is out of range for relation "test_rel_forks"
17-
SELECT octet_length(get_raw_page('test_rel_forks', 'fsm', 0)) AS fsm_0;
13+
SELECT octet_length(get_raw_page('test1', 'main', 1)) AS main_1;
14+
ERROR: block number 1 is out of range for relation "test1"
15+
SELECT octet_length(get_raw_page('test1', 'fsm', 0)) AS fsm_0;
1816
fsm_0
1917
-------
2018
8192
2119
(1 row)
2220

23-
SELECT octet_length(get_raw_page('test_rel_forks', 'fsm', 20)) AS fsm_20;
24-
ERROR: block number 20 is out of range for relation "test_rel_forks"
25-
SELECT octet_length(get_raw_page('test_rel_forks', 'vm', 0)) AS vm_0;
21+
SELECT octet_length(get_raw_page('test1', 'fsm', 1)) AS fsm_1;
22+
fsm_1
23+
-------
24+
8192
25+
(1 row)
26+
27+
SELECT octet_length(get_raw_page('test1', 'vm', 0)) AS vm_0;
2628
vm_0
2729
------
2830
8192
2931
(1 row)
3032

31-
SELECT octet_length(get_raw_page('test_rel_forks', 'vm', 1)) AS vm_1;
32-
ERROR: block number 1 is out of range for relation "test_rel_forks"
33+
SELECT octet_length(get_raw_page('test1', 'vm', 1)) AS vm_1;
34+
ERROR: block number 1 is out of range for relation "test1"
3335
SELECT octet_length(get_raw_page('xxx', 'main', 0));
3436
ERROR: relation "xxx" does not exist
35-
SELECT octet_length(get_raw_page('test_rel_forks', 'xxx', 0));
37+
SELECT octet_length(get_raw_page('test1', 'xxx', 0));
3638
ERROR: invalid fork name
3739
HINT: Valid fork names are "main", "fsm", "vm", and "init".
38-
EXPLAIN (costs off, analyze on, timing off, summary off) SELECT * FROM
39-
fsm_page_contents(get_raw_page('test_rel_forks', 'fsm', 0));
40-
QUERY PLAN
41-
------------------------------------------------------------
42-
Function Scan on fsm_page_contents (actual rows=1 loops=1)
43-
(1 row)
44-
45-
SELECT get_raw_page('test_rel_forks', 0) = get_raw_page('test_rel_forks', 'main', 0);
40+
SELECT get_raw_page('test1', 0) = get_raw_page('test1', 'main', 0);
4641
?column?
4742
----------
4843
t
4944
(1 row)
5045

51-
DROP TABLE test_rel_forks;
52-
CREATE TABLE test1 (a int, b int);
53-
INSERT INTO test1 VALUES (16777217, 131584);
5446
SELECT pagesize, version FROM page_header(get_raw_page('test1', 0));
5547
pagesize | version
5648
----------+---------
@@ -70,6 +62,26 @@ SELECT tuple_data_split('test1'::regclass, t_data, t_infomask, t_infomask2, t_bi
7062
{"\\x01000001","\\x00020200"}
7163
(1 row)
7264

65+
SELECT * FROM fsm_page_contents(get_raw_page('test1', 'fsm', 0));
66+
fsm_page_contents
67+
-------------------
68+
0: 254 +
69+
1: 254 +
70+
3: 254 +
71+
7: 254 +
72+
15: 254 +
73+
31: 254 +
74+
63: 254 +
75+
127: 254 +
76+
255: 254 +
77+
511: 254 +
78+
1023: 254 +
79+
2047: 254 +
80+
4095: 254 +
81+
fp_next_slot: 0 +
82+
83+
(1 row)
84+
7385
DROP TABLE test1;
7486
-- check that using any of these functions with a partitioned table or index
7587
-- would fail

‎contrib/pageinspect/sql/page.sql

Lines changed: 13 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -1,36 +1,26 @@
11
CREATE EXTENSION pageinspect;
22

3-
CREATE TABLE test_rel_forks (a int);
4-
-- Make sure there are enough blocks in the heap for the FSM to be created.
5-
INSERT INTO test_rel_forks SELECT i from generate_series(1,2000) i;
3+
CREATE TABLE test1 (a int, b int);
4+
INSERT INTO test1 VALUES (16777217, 131584);
65

7-
-- set up FSM and VM
8-
VACUUM test_rel_forks;
6+
VACUUM test1; -- set up FSM
97

108
-- The page contents can vary, so just test that it can be read
119
-- successfully, but don't keep the output.
1210

13-
SELECT octet_length(get_raw_page('test_rel_forks', 'main', 0)) AS main_0;
14-
SELECT octet_length(get_raw_page('test_rel_forks', 'main', 100)) AS main_100;
11+
SELECT octet_length(get_raw_page('test1', 'main', 0)) AS main_0;
12+
SELECT octet_length(get_raw_page('test1', 'main', 1)) AS main_1;
1513

16-
SELECT octet_length(get_raw_page('test_rel_forks', 'fsm', 0)) AS fsm_0;
17-
SELECT octet_length(get_raw_page('test_rel_forks', 'fsm', 20)) AS fsm_20;
14+
SELECT octet_length(get_raw_page('test1', 'fsm', 0)) AS fsm_0;
15+
SELECT octet_length(get_raw_page('test1', 'fsm', 1)) AS fsm_1;
1816

19-
SELECT octet_length(get_raw_page('test_rel_forks', 'vm', 0)) AS vm_0;
20-
SELECT octet_length(get_raw_page('test_rel_forks', 'vm', 1)) AS vm_1;
17+
SELECT octet_length(get_raw_page('test1', 'vm', 0)) AS vm_0;
18+
SELECT octet_length(get_raw_page('test1', 'vm', 1)) AS vm_1;
2119

2220
SELECT octet_length(get_raw_page('xxx', 'main', 0));
23-
SELECT octet_length(get_raw_page('test_rel_forks', 'xxx', 0));
24-
25-
EXPLAIN (costs off, analyze on, timing off, summary off) SELECT * FROM
26-
fsm_page_contents(get_raw_page('test_rel_forks', 'fsm', 0));
27-
28-
SELECT get_raw_page('test_rel_forks', 0) = get_raw_page('test_rel_forks', 'main', 0);
21+
SELECT octet_length(get_raw_page('test1', 'xxx', 0));
2922

30-
DROP TABLE test_rel_forks;
31-
32-
CREATE TABLE test1 (a int, b int);
33-
INSERT INTO test1 VALUES (16777217, 131584);
23+
SELECT get_raw_page('test1', 0) = get_raw_page('test1', 'main', 0);
3424

3525
SELECT pagesize, version FROM page_header(get_raw_page('test1', 0));
3626

@@ -39,6 +29,8 @@ SELECT page_checksum(get_raw_page('test1', 0), 0) IS NOT NULL AS silly_checksum_
3929
SELECT tuple_data_split('test1'::regclass, t_data, t_infomask, t_infomask2, t_bits)
4030
FROM heap_page_items(get_raw_page('test1', 0));
4131

32+
SELECT * FROM fsm_page_contents(get_raw_page('test1', 'fsm', 0));
33+
4234
DROP TABLE test1;
4335

4436
-- check that using any of these functions with a partitioned table or index

‎contrib/pgstattuple/pgstatapprox.c

Lines changed: 0 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -90,9 +90,6 @@ statapprox_heap(Relation rel, output_type *stat)
9090
/*
9191
* If the page has only visible tuples, then we can find out the free
9292
* space from the FSM and move on.
93-
*
94-
* Note: If a relation has no FSM, GetRecordedFreeSpace() will report
95-
* zero free space. This is fine for the purposes of approximation.
9693
*/
9794
if (VM_ALL_VISIBLE(rel, blkno, &vmbuffer))
9895
{

‎doc/src/sgml/pgfreespacemap.sgml

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -61,8 +61,6 @@
6161
The values stored in the free space map are not exact. They're rounded
6262
to precision of 1/256th of <symbol>BLCKSZ</symbol> (32 bytes with default <symbol>BLCKSZ</symbol>), and
6363
they're not kept fully up-to-date as tuples are inserted and updated.
64-
In addition, small tables don't have a free space map, so these functions
65-
will return zero even if free space is available.
6664
</para>
6765

6866
<para>

‎doc/src/sgml/pgstattuple.sgml

Lines changed: 1 addition & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -527,9 +527,7 @@ approx_free_percent | 2.09
527527
bit set, then it is assumed to contain no dead tuples). For such
528528
pages, it derives the free space value from the free space map, and
529529
assumes that the rest of the space on the page is taken up by live
530-
tuples. Small tables don't have a free space map, so in that case
531-
this function will report zero free space, likewise inflating the
532-
approximate tuple length.
530+
tuples.
533531
</para>
534532

535533
<para>

‎doc/src/sgml/ref/pgupgrade.sgml

Lines changed: 0 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -812,13 +812,6 @@ psql --username=postgres --file=script.sql postgres
812812
is down.
813813
</para>
814814

815-
<para>
816-
In <productname>PostgreSQL</productname> 12 and later small tables by
817-
default don't have a free space map, as a space optimization. If you are
818-
upgrading a pre-12 cluster, the free space maps of small tables will
819-
likewise not be transferred to the new cluster.
820-
</para>
821-
822815
</refsect1>
823816

824817
<refsect1>

‎doc/src/sgml/storage.sgml

Lines changed: 6 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -598,13 +598,12 @@ tuple would otherwise be too big.
598598
<indexterm><primary>FSM</primary><see>Free Space Map</see></indexterm>
599599

600600
<para>
601-
Each heap relation, unless it is very small, and each index relation, except
602-
for hash indexes, has a Free Space Map (FSM) to keep track of available
603-
space in the relation. It's stored alongside the main relation data in a
604-
separate relation fork, named after the filenode number of the relation, plus
605-
a <literal>_fsm</literal> suffix. For example, if the filenode of a relation
606-
is 12345, the FSM is stored in a file called <filename>12345_fsm</filename>,
607-
in the same directory as the main relation file.
601+
Each heap and index relation, except for hash indexes, has a Free Space Map
602+
(FSM) to keep track of available space in the relation. It's stored
603+
alongside the main relation data in a separate relation fork, named after the
604+
filenode number of the relation, plus a <literal>_fsm</literal> suffix. For example,
605+
if the filenode of a relation is 12345, the FSM is stored in a file called
606+
<filename>12345_fsm</filename>, in the same directory as the main relation file.
608607
</para>
609608

610609
<para>

‎src/backend/access/brin/brin.c

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1152,7 +1152,7 @@ terminate_brin_buildstate(BrinBuildState *state)
11521152
freespace = PageGetFreeSpace(page);
11531153
blk = BufferGetBlockNumber(state->bs_currentInsertBuf);
11541154
ReleaseBuffer(state->bs_currentInsertBuf);
1155-
RecordPageWithFreeSpace(state->bs_irel, blk, freespace, InvalidBlockNumber);
1155+
RecordPageWithFreeSpace(state->bs_irel, blk, freespace);
11561156
FreeSpaceMapVacuumRange(state->bs_irel, blk, blk + 1);
11571157
}
11581158

‎src/backend/access/brin/brin_pageops.c

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -310,7 +310,7 @@ brin_doupdate(Relation idxrel, BlockNumber pagesPerRange,
310310

311311
if (extended)
312312
{
313-
RecordPageWithFreeSpace(idxrel, newblk, freespace, InvalidBlockNumber);
313+
RecordPageWithFreeSpace(idxrel, newblk, freespace);
314314
FreeSpaceMapVacuumRange(idxrel, newblk, newblk + 1);
315315
}
316316

@@ -461,7 +461,7 @@ brin_doinsert(Relation idxrel, BlockNumber pagesPerRange,
461461

462462
if (extended)
463463
{
464-
RecordPageWithFreeSpace(idxrel, blk, freespace, InvalidBlockNumber);
464+
RecordPageWithFreeSpace(idxrel, blk, freespace);
465465
FreeSpaceMapVacuumRange(idxrel, blk, blk + 1);
466466
}
467467

@@ -654,7 +654,7 @@ brin_page_cleanup(Relation idxrel, Buffer buf)
654654

655655
/* Measure free space and record it */
656656
RecordPageWithFreeSpace(idxrel, BufferGetBlockNumber(buf),
657-
br_page_get_freespace(page), InvalidBlockNumber);
657+
br_page_get_freespace(page));
658658
}
659659

660660
/*
@@ -703,7 +703,7 @@ brin_getinsertbuffer(Relation irel, Buffer oldbuf, Size itemsz,
703703
/* Choose initial target page, re-using existing target if known */
704704
newblk = RelationGetTargetBlock(irel);
705705
if (newblk == InvalidBlockNumber)
706-
newblk = GetPageWithFreeSpace(irel, itemsz, true);
706+
newblk = GetPageWithFreeSpace(irel, itemsz);
707707

708708
/*
709709
* 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)
895895
* pages whose FSM records were forgotten in a crash.
896896
*/
897897
RecordPageWithFreeSpace(idxrel, BufferGetBlockNumber(buffer),
898-
br_page_get_freespace(page), InvalidBlockNumber);
898+
br_page_get_freespace(page));
899899
}
900900

901901

‎src/backend/access/heap/hio.c

Lines changed: 17 additions & 25 deletions
Original file line numberDiff line numberDiff line change
@@ -246,14 +246,8 @@ RelationAddExtraBlocks(Relation relation, BulkInsertState bistate)
246246
* Immediately update the bottom level of the FSM. This has a good
247247
* chance of making this page visible to other concurrently inserting
248248
* backends, and we want that to happen without delay.
249-
*
250-
* Since we know the table will end up with extraBlocks additional
251-
* pages, we pass the final number to avoid possible unnecessary
252-
* system calls and to make sure the FSM is created when we add the
253-
* first new page.
254249
*/
255-
RecordPageWithFreeSpace(relation, blockNum, freespace,
256-
firstBlock + extraBlocks);
250+
RecordPageWithFreeSpace(relation, blockNum, freespace);
257251
}
258252
while (--extraBlocks > 0);
259253

@@ -390,9 +384,20 @@ RelationGetBufferForTuple(Relation relation, Size len,
390384
* We have no cached target page, so ask the FSM for an initial
391385
* target.
392386
*/
393-
targetBlock = GetPageWithFreeSpace(relation,
394-
len + saveFreeSpace,
395-
false);
387+
targetBlock = GetPageWithFreeSpace(relation, len + saveFreeSpace);
388+
389+
/*
390+
* If the FSM knows nothing of the rel, try the last page before we
391+
* give up and extend. This avoids one-tuple-per-page syndrome during
392+
* bootstrapping or in a recently-started system.
393+
*/
394+
if (targetBlock == InvalidBlockNumber)
395+
{
396+
BlockNumber nblocks = RelationGetNumberOfBlocks(relation);
397+
398+
if (nblocks > 0)
399+
targetBlock = nblocks - 1;
400+
}
396401
}
397402

398403
loop:
@@ -499,13 +504,6 @@ RelationGetBufferForTuple(Relation relation, Size len,
499504
{
500505
/* use this page as future insert target, too */
501506
RelationSetTargetBlock(relation, targetBlock);
502-
503-
/*
504-
* In case we used an in-memory map of available blocks, reset it
505-
* for next use.
506-
*/
507-
FSMClearLocalMap();
508-
509507
return buffer;
510508
}
511509

@@ -565,12 +563,9 @@ RelationGetBufferForTuple(Relation relation, Size len,
565563

566564
/*
567565
* Check if some other backend has extended a block for us while
568-
* we were waiting on the lock. We only check the FSM -- if there
569-
* isn't one we don't recheck the number of blocks.
566+
* we were waiting on the lock.
570567
*/
571-
targetBlock = GetPageWithFreeSpace(relation,
572-
len + saveFreeSpace,
573-
true);
568+
targetBlock = GetPageWithFreeSpace(relation, len + saveFreeSpace);
574569

575570
/*
576571
* If some other waiter has already extended the relation, we
@@ -675,8 +670,5 @@ RelationGetBufferForTuple(Relation relation, Size len,
675670
*/
676671
RelationSetTargetBlock(relation, BufferGetBlockNumber(buffer));
677672

678-
/* This should already be cleared by now, but make sure it is. */
679-
FSMClearLocalMap();
680-
681673
return buffer;
682674
}

0 commit comments

Comments
 (0)