From b0eaa4c51bbff3e3c600b11e5d104d6feb9ca77f Mon Sep 17 00:00:00 2001 From: Amit Kapila Date: Mon, 4 Feb 2019 07:49:15 +0530 Subject: [PATCH] Avoid creation of the free space map for small heap relations, take 2. 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, Amit Kapila Reviewed-by: Amit Kapila Tested-by: Mithun C Y Discussion: https://www.postgresql.org/message-id/CAJVSVGWvB13PzpbLEecFuGFc5V2fsO736BsdTakPiPAcdMM5tQ@mail.gmail.com --- contrib/pageinspect/expected/page.out | 77 +++--- contrib/pageinspect/sql/page.sql | 33 ++- doc/src/sgml/storage.sgml | 13 +- src/backend/access/brin/brin.c | 2 +- src/backend/access/brin/brin_pageops.c | 10 +- src/backend/access/heap/hio.c | 45 ++-- src/backend/access/heap/vacuumlazy.c | 17 +- src/backend/access/transam/xact.c | 14 ++ src/backend/storage/freespace/README | 38 ++- src/backend/storage/freespace/freespace.c | 285 +++++++++++++++++++++- src/backend/storage/freespace/indexfsm.c | 6 +- src/include/storage/freespace.h | 9 +- src/test/regress/expected/fsm.out | 48 ++++ src/test/regress/parallel_schedule | 6 + src/test/regress/serial_schedule | 1 + src/test/regress/sql/fsm.sql | 41 ++++ 16 files changed, 543 insertions(+), 102 deletions(-) create mode 100644 src/test/regress/expected/fsm.out create mode 100644 src/test/regress/sql/fsm.sql diff --git a/contrib/pageinspect/expected/page.out b/contrib/pageinspect/expected/page.out index 3fcd9fbe6d..89b73ca991 100644 --- a/contrib/pageinspect/expected/page.out +++ b/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(1,2000) 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', 20)) AS fsm_20; +ERROR: block number 20 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: 39 + + 1: 39 + + 3: 39 + + 7: 39 + + 15: 39 + + 31: 39 + + 63: 39 + + 127: 39 + + 255: 39 + + 511: 39 + + 1023: 39 + + 2047: 39 + + 4095: 39 + + 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 diff --git a/contrib/pageinspect/sql/page.sql b/contrib/pageinspect/sql/page.sql index 8ac9991837..67166ef54c 100644 --- a/contrib/pageinspect/sql/page.sql +++ b/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(1,2000) 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', 20)) AS fsm_20; -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 get_raw_page('test1', 0) = get_raw_page('test1', 'main', 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); + +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 diff --git a/doc/src/sgml/storage.sgml b/doc/src/sgml/storage.sgml index 8ef2ac8010..cbdad0c3fb 100644 --- a/doc/src/sgml/storage.sgml +++ b/doc/src/sgml/storage.sgml @@ -590,12 +590,13 @@ tuple would otherwise be too big. FSMFree Space Map -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 _fsm suffix. For example, -if the filenode of a relation is 12345, the FSM is stored in a file called -12345_fsm, 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 _fsm suffix. For example, if the filenode of a relation +is 12345, the FSM is stored in a file called 12345_fsm, +in the same directory as the main relation file. diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c index 467d91e681..8f008dd008 100644 --- a/src/backend/access/brin/brin.c +++ b/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); } diff --git a/src/backend/access/brin/brin_pageops.c b/src/backend/access/brin/brin_pageops.c index 164a468155..2eb354f948 100644 --- a/src/backend/access/brin/brin_pageops.c +++ b/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); } diff --git a/src/backend/access/heap/hio.c b/src/backend/access/heap/hio.c index d41d318eef..a9c8ec43a7 100644 --- a/src/backend/access/heap/hio.c +++ b/src/backend/access/heap/hio.c @@ -246,8 +246,14 @@ 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); @@ -384,20 +390,9 @@ 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: @@ -504,6 +499,13 @@ loop: { /* 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. + */ + FSMClearLocalMap(); + return buffer; } @@ -563,9 +565,12 @@ loop: /* * 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 @@ -670,5 +675,11 @@ loop: */ RelationSetTargetBlock(relation, BufferGetBlockNumber(buffer)); + /* + * In case we used an in-memory map of available blocks, reset it for next + * use. + */ + FSMClearLocalMap(); + return buffer; } diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c index 26dfb0c7e0..9416c31889 100644 --- a/src/backend/access/heap/vacuumlazy.c +++ b/src/backend/access/heap/vacuumlazy.c @@ -153,7 +153,7 @@ static BufferAccessStrategy vac_strategy; static void lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats, Relation *Irel, int nindexes, bool aggressive); -static void lazy_vacuum_heap(Relation onerel, LVRelStats *vacrelstats); +static void lazy_vacuum_heap(Relation onerel, LVRelStats *vacrelstats, BlockNumber nblocks); static bool lazy_check_needs_freeze(Buffer buf, bool *hastup); static void lazy_vacuum_index(Relation indrel, IndexBulkDeleteResult **stats, @@ -758,7 +758,7 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats, pgstat_progress_update_multi_param(2, hvp_index, hvp_val); /* Remove tuples from heap */ - lazy_vacuum_heap(onerel, vacrelstats); + lazy_vacuum_heap(onerel, vacrelstats, nblocks); /* * Forget the now-vacuumed tuples, and press on, but be careful @@ -897,7 +897,7 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats, Size freespace; freespace = BufferGetPageSize(buf) - SizeOfPageHeaderData; - RecordPageWithFreeSpace(onerel, blkno, freespace); + RecordPageWithFreeSpace(onerel, blkno, freespace, nblocks); } } continue; @@ -941,7 +941,7 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats, } UnlockReleaseBuffer(buf); - RecordPageWithFreeSpace(onerel, blkno, freespace); + RecordPageWithFreeSpace(onerel, blkno, freespace, nblocks); continue; } @@ -1338,7 +1338,7 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats, * taken if there are no indexes.) */ if (vacrelstats->num_dead_tuples == prev_dead_count) - RecordPageWithFreeSpace(onerel, blkno, freespace); + RecordPageWithFreeSpace(onerel, blkno, freespace, nblocks); } /* report that everything is scanned and vacuumed */ @@ -1400,7 +1400,7 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats, /* Remove tuples from heap */ pgstat_progress_update_param(PROGRESS_VACUUM_PHASE, PROGRESS_VACUUM_PHASE_VACUUM_HEAP); - lazy_vacuum_heap(onerel, vacrelstats); + lazy_vacuum_heap(onerel, vacrelstats, nblocks); vacrelstats->num_index_scans++; } @@ -1471,9 +1471,10 @@ lazy_scan_heap(Relation onerel, int options, LVRelStats *vacrelstats, * Note: the reason for doing this as a second pass is we cannot remove * the tuples until we've removed their index entries, and we want to * process index entry removal in batches as large as possible. + * Note: nblocks is passed as an optimization for RecordPageWithFreeSpace(). */ static void -lazy_vacuum_heap(Relation onerel, LVRelStats *vacrelstats) +lazy_vacuum_heap(Relation onerel, LVRelStats *vacrelstats, BlockNumber nblocks) { int tupindex; int npages; @@ -1510,7 +1511,7 @@ lazy_vacuum_heap(Relation onerel, LVRelStats *vacrelstats) freespace = PageGetHeapFreeSpace(page); UnlockReleaseBuffer(buf); - RecordPageWithFreeSpace(onerel, tblk, freespace); + RecordPageWithFreeSpace(onerel, tblk, freespace, nblocks); npages++; } diff --git a/src/backend/access/transam/xact.c b/src/backend/access/transam/xact.c index 0181976964..92bda87804 100644 --- a/src/backend/access/transam/xact.c +++ b/src/backend/access/transam/xact.c @@ -48,6 +48,7 @@ #include "replication/walsender.h" #include "storage/condition_variable.h" #include "storage/fd.h" +#include "storage/freespace.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" @@ -2493,6 +2494,12 @@ AbortTransaction(void) pgstat_report_wait_end(); pgstat_progress_end_command(); + /* + * In case we aborted during RelationGetBufferForTuple(), clear the local + * map of heap pages. + */ + FSMClearLocalMap(); + /* Clean up buffer I/O and buffer context locks, too */ AbortBufferIO(); UnlockBuffers(); @@ -4714,6 +4721,13 @@ AbortSubTransaction(void) pgstat_report_wait_end(); pgstat_progress_end_command(); + + /* + * In case we aborted during RelationGetBufferForTuple(), clear the local + * map of heap pages. + */ + FSMClearLocalMap(); + AbortBufferIO(); UnlockBuffers(); diff --git a/src/backend/storage/freespace/README b/src/backend/storage/freespace/README index e7ff23b76f..0d3cd29772 100644 --- a/src/backend/storage/freespace/README +++ b/src/backend/storage/freespace/README @@ -8,7 +8,41 @@ free space to hold a tuple to be stored; or to determine that no such page exists and the relation must be extended by one page. As of PostgreSQL 8.4 each relation has its own, extensible free space map stored in a separate "fork" of its relation. This eliminates the disadvantages of the former -fixed-size FSM. +fixed-size FSM. There are two exceptions: + +1. Hash indexes never have a FSM. +2. For very small tables, a 3-page relation fork would be relatively large +and wasteful, so to save space we refrain from creating the FSM if the +heap has HEAP_FSM_CREATION_THRESHOLD pages or fewer. + +To locate free space in the latter case, we simply try pages directly without +knowing ahead of time how much free space they have. To maintain good +performance, we create a local in-memory map of pages to try, and only mark +every other page as available. For example, in a 3-page heap, the local map +would look like: + +ANAN +0123 + +Pages 0 and 2 are marked "available", and page 1 as "not available". +Page 3 is beyond the end of the relation, so is likewise marked "not +available". First we try page 2, and if that doesn't have sufficient free +space we try page 0 before giving up and extending the relation. There may +be some wasted free space on block 1, but if the relation extends to 4 pages: + +NANA +0123 + +We not only have the new page 3 at our disposal, we can now check page 1 +for free space as well. + +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. + +FSM data structure +------------------ It is important to keep the map small so that it can be searched rapidly. Therefore, we don't attempt to record the exact free space on a page. @@ -192,5 +226,3 @@ TODO ---- - fastroot to avoid traversing upper nodes with just 1 child -- use a different system for tables that fit into one FSM page, with a - mechanism to switch to the real thing as it grows. diff --git a/src/backend/storage/freespace/freespace.c b/src/backend/storage/freespace/freespace.c index eee8286057..d3f207b854 100644 --- a/src/backend/storage/freespace/freespace.c +++ b/src/backend/storage/freespace/freespace.c @@ -76,6 +76,14 @@ #define FSM_ROOT_LEVEL (FSM_TREE_DEPTH - 1) #define FSM_BOTTOM_LEVEL 0 +/* Status codes for the local map. */ + +/* Either already tried, or beyond the end of the relation */ +#define FSM_LOCAL_NOT_AVAIL 0x00 + +/* Available to try */ +#define FSM_LOCAL_AVAIL 0x01 + /* * The internal FSM routines work on a logical addressing scheme. Each * level of the tree can be thought of as a separately addressable file. @@ -89,6 +97,23 @@ typedef struct /* Address of the root page. */ static const FSMAddress FSM_ROOT_ADDRESS = {FSM_ROOT_LEVEL, 0}; +/* Local map of block numbers for small heaps with no FSM. */ +typedef struct +{ + BlockNumber nblocks; + uint8 map[HEAP_FSM_CREATION_THRESHOLD]; +} FSMLocalMap; + +static FSMLocalMap fsm_local_map = +{ + 0, + { + FSM_LOCAL_NOT_AVAIL + } +}; + +#define FSM_LOCAL_MAP_EXISTS (fsm_local_map.nblocks > 0) + /* functions to navigate the tree */ static FSMAddress fsm_get_child(FSMAddress parent, uint16 slot); static FSMAddress fsm_get_parent(FSMAddress child, uint16 *slot); @@ -107,10 +132,14 @@ static Size fsm_space_cat_to_avail(uint8 cat); /* workhorse functions for various operations */ static int fsm_set_and_search(Relation rel, FSMAddress addr, uint16 slot, uint8 newValue, uint8 minValue); +static void fsm_local_set(Relation rel, BlockNumber cur_nblocks); static BlockNumber fsm_search(Relation rel, uint8 min_cat); +static BlockNumber fsm_local_search(void); static uint8 fsm_vacuum_page(Relation rel, FSMAddress addr, BlockNumber start, BlockNumber end, bool *eof); +static bool fsm_allow_writes(Relation rel, BlockNumber heapblk, + BlockNumber nblocks, BlockNumber *get_nblocks); /******** Public API ********/ @@ -127,13 +156,46 @@ static uint8 fsm_vacuum_page(Relation rel, FSMAddress addr, * amount of free space available on that page and then try again (see * RecordAndGetPageWithFreeSpace). If InvalidBlockNumber is returned, * extend the relation. + * + * For very small heap relations that don't have a FSM, we try every other + * page before extending the relation. To keep track of which pages have + * been tried, initialize a local in-memory map of pages. */ BlockNumber -GetPageWithFreeSpace(Relation rel, Size spaceNeeded) +GetPageWithFreeSpace(Relation rel, Size spaceNeeded, bool check_fsm_only) { uint8 min_cat = fsm_space_needed_to_cat(spaceNeeded); + BlockNumber target_block, + nblocks; - return fsm_search(rel, min_cat); + /* First try the FSM, if it exists. */ + target_block = fsm_search(rel, min_cat); + + if (target_block == InvalidBlockNumber && + (rel->rd_rel->relkind == RELKIND_RELATION || + rel->rd_rel->relkind == RELKIND_TOASTVALUE) && + !check_fsm_only) + { + nblocks = RelationGetNumberOfBlocks(rel); + + if (nblocks > HEAP_FSM_CREATION_THRESHOLD) + { + /* + * 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. + */ + target_block = nblocks - 1; + } + else if (nblocks > 0) + { + /* Create or update local map and get first candidate block. */ + fsm_local_set(rel, nblocks); + target_block = fsm_local_search(); + } + } + + return target_block; } /* @@ -144,16 +206,47 @@ GetPageWithFreeSpace(Relation rel, Size spaceNeeded) * also some effort to return a page close to the old page; if there's a * page with enough free space on the same FSM page where the old one page * is located, it is preferred. + * + * For very small heap relations that don't have a FSM, we update the local + * map to indicate we have tried a page, and return the next page to try. */ BlockNumber RecordAndGetPageWithFreeSpace(Relation rel, BlockNumber oldPage, Size oldSpaceAvail, Size spaceNeeded) { - int old_cat = fsm_space_avail_to_cat(oldSpaceAvail); - int search_cat = fsm_space_needed_to_cat(spaceNeeded); + int old_cat; + int search_cat; FSMAddress addr; uint16 slot; int search_slot; + BlockNumber nblocks = InvalidBlockNumber; + + /* First try the local map, if it exists. */ + if (FSM_LOCAL_MAP_EXISTS) + { + Assert((rel->rd_rel->relkind == RELKIND_RELATION || + rel->rd_rel->relkind == RELKIND_TOASTVALUE) && + fsm_local_map.map[oldPage] == FSM_LOCAL_AVAIL); + + fsm_local_map.map[oldPage] = FSM_LOCAL_NOT_AVAIL; + return fsm_local_search(); + } + + if (!fsm_allow_writes(rel, oldPage, InvalidBlockNumber, &nblocks)) + { + /* + * If we have neither a local map nor a FSM, we probably just tried + * the target block in the smgr relation entry and failed, so we'll + * need to create the local map. + */ + fsm_local_set(rel, nblocks); + return fsm_local_search(); + } + + /* Normal FSM logic follows */ + + old_cat = fsm_space_avail_to_cat(oldSpaceAvail); + search_cat = fsm_space_needed_to_cat(spaceNeeded); /* Get the location of the FSM byte representing the heap block */ addr = fsm_get_location(oldPage, &slot); @@ -176,20 +269,44 @@ RecordAndGetPageWithFreeSpace(Relation rel, BlockNumber oldPage, * Note that if the new spaceAvail value is higher than the old value stored * in the FSM, the space might not become visible to searchers until the next * FreeSpaceMapVacuum call, which updates the upper level pages. + * + * Callers have no need for a local map. */ void -RecordPageWithFreeSpace(Relation rel, BlockNumber heapBlk, Size spaceAvail) +RecordPageWithFreeSpace(Relation rel, BlockNumber heapBlk, + Size spaceAvail, BlockNumber nblocks) { - int new_cat = fsm_space_avail_to_cat(spaceAvail); + int new_cat; FSMAddress addr; uint16 slot; + BlockNumber dummy; + + if (!fsm_allow_writes(rel, heapBlk, nblocks, &dummy)) + /* No FSM to update and no local map either */ + return; /* Get the location of the FSM byte representing the heap block */ addr = fsm_get_location(heapBlk, &slot); + new_cat = fsm_space_avail_to_cat(spaceAvail); fsm_set_and_search(rel, addr, slot, new_cat, 0); } +/* + * Clear the local map. We must call this when we have found a block with + * enough free space, when we extend the relation, or on transaction abort. + */ +void +FSMClearLocalMap(void) +{ + if (FSM_LOCAL_MAP_EXISTS) + { + fsm_local_map.nblocks = 0; + memset(&fsm_local_map.map, FSM_LOCAL_NOT_AVAIL, + sizeof(fsm_local_map.map)); + } +} + /* * XLogRecordPageWithFreeSpace - like RecordPageWithFreeSpace, for use in * WAL replay @@ -204,6 +321,31 @@ XLogRecordPageWithFreeSpace(RelFileNode rnode, BlockNumber heapBlk, BlockNumber blkno; Buffer buf; Page page; + bool write_to_fsm; + + /* This is meant to mirror the logic in fsm_allow_writes() */ + if (heapBlk >= HEAP_FSM_CREATION_THRESHOLD) + write_to_fsm = true; + else + { + /* Open the relation at smgr level */ + SMgrRelation smgr = smgropen(rnode, InvalidBackendId); + + if (smgrexists(smgr, FSM_FORKNUM)) + write_to_fsm = true; + else + { + BlockNumber heap_nblocks = smgrnblocks(smgr, MAIN_FORKNUM); + + if (heap_nblocks > HEAP_FSM_CREATION_THRESHOLD) + write_to_fsm = true; + else + write_to_fsm = false; + } + } + + if (!write_to_fsm) + return; /* Get the location of the FSM byte representing the heap block */ addr = fsm_get_location(heapBlk, &slot); @@ -904,3 +1046,134 @@ fsm_vacuum_page(Relation rel, FSMAddress addr, return max_avail; } + +/* + * For heaps, we prevent creation of the FSM unless the number of pages + * exceeds HEAP_FSM_CREATION_THRESHOLD. For tables that don't already have + * a FSM, this will save an inode and a few kB of space. + * + * XXX The API is a little awkward -- if the caller passes a valid nblocks + * value, it can avoid invoking a system call. If the caller passes + * InvalidBlockNumber and receives a false return value, it can get an + * up-to-date relation size from get_nblocks. This saves a few cycles in + * the caller, which would otherwise need to get the relation size by itself. + */ +static bool +fsm_allow_writes(Relation rel, BlockNumber heapblk, + BlockNumber nblocks, BlockNumber *get_nblocks) +{ + bool skip_get_nblocks; + + if (heapblk >= HEAP_FSM_CREATION_THRESHOLD) + return true; + + /* Non-heap rels can always create a FSM. */ + if (rel->rd_rel->relkind != RELKIND_RELATION && + rel->rd_rel->relkind != RELKIND_TOASTVALUE) + return true; + + /* + * If the caller knows nblocks, we can avoid a system call later. If it + * doesn't, maybe we have relpages from a previous VACUUM. Since the table + * may have extended since then, we still have to count the pages later if + * we can't return now. + */ + if (nblocks != InvalidBlockNumber) + { + if (nblocks > HEAP_FSM_CREATION_THRESHOLD) + return true; + else + skip_get_nblocks = true; + } + else + { + if (rel->rd_rel->relpages != InvalidBlockNumber && + rel->rd_rel->relpages > HEAP_FSM_CREATION_THRESHOLD) + return true; + else + skip_get_nblocks = false; + } + + RelationOpenSmgr(rel); + if (smgrexists(rel->rd_smgr, FSM_FORKNUM)) + return true; + + if (skip_get_nblocks) + return false; + + /* last resort */ + *get_nblocks = RelationGetNumberOfBlocks(rel); + if (*get_nblocks > HEAP_FSM_CREATION_THRESHOLD) + return true; + else + return false; +} + +/* + * Initialize or update the local map of blocks to try, for when there is + * no FSM. + * + * When we initialize the map, the whole heap is potentially available to + * try. Testing revealed that trying every block can cause a small + * performance dip compared to when we use a FSM, so we try every other + * block instead. + */ +static void +fsm_local_set(Relation rel, BlockNumber cur_nblocks) +{ + BlockNumber blkno, + cached_target_block; + + /* The local map must not be set already. */ + Assert(!FSM_LOCAL_MAP_EXISTS); + + /* + * Starting at the current last block in the relation and working + * backwards, mark alternating blocks as available. + */ + blkno = cur_nblocks - 1; + while (true) + { + fsm_local_map.map[blkno] = FSM_LOCAL_AVAIL; + if (blkno >= 2) + blkno -= 2; + else + break; + } + + /* Cache the number of blocks. */ + fsm_local_map.nblocks = cur_nblocks; + + /* Set the status of the cached target block to 'unavailable'. */ + cached_target_block = RelationGetTargetBlock(rel); + if (cached_target_block != InvalidBlockNumber && + cached_target_block < cur_nblocks) + fsm_local_map.map[cached_target_block] = FSM_LOCAL_NOT_AVAIL; +} + +/* + * Search the local map for an available block to try, in descending order. + * As such, there is no heuristic available to decide which order will be + * better to try, but the probability of having space in the last block in the + * map is higher because that is the most recent block added to the heap. + * + * This function is used when there is no FSM. + */ +static BlockNumber +fsm_local_search(void) +{ + BlockNumber target_block; + + /* Local map must be set by now. */ + Assert(FSM_LOCAL_MAP_EXISTS); + + target_block = fsm_local_map.nblocks; + do + { + target_block--; + if (fsm_local_map.map[target_block] == FSM_LOCAL_AVAIL) + return target_block; + } while (target_block > 0); + + return InvalidBlockNumber; +} diff --git a/src/backend/storage/freespace/indexfsm.c b/src/backend/storage/freespace/indexfsm.c index 58cedeaa9f..9d8f43d373 100644 --- a/src/backend/storage/freespace/indexfsm.c +++ b/src/backend/storage/freespace/indexfsm.c @@ -37,7 +37,7 @@ BlockNumber GetFreeIndexPage(Relation rel) { - BlockNumber blkno = GetPageWithFreeSpace(rel, BLCKSZ / 2); + BlockNumber blkno = GetPageWithFreeSpace(rel, BLCKSZ / 2, true); if (blkno != InvalidBlockNumber) RecordUsedIndexPage(rel, blkno); @@ -51,7 +51,7 @@ GetFreeIndexPage(Relation rel) void RecordFreeIndexPage(Relation rel, BlockNumber freeBlock) { - RecordPageWithFreeSpace(rel, freeBlock, BLCKSZ - 1); + RecordPageWithFreeSpace(rel, freeBlock, BLCKSZ - 1, InvalidBlockNumber); } @@ -61,7 +61,7 @@ RecordFreeIndexPage(Relation rel, BlockNumber freeBlock) void RecordUsedIndexPage(Relation rel, BlockNumber usedBlock) { - RecordPageWithFreeSpace(rel, usedBlock, 0); + RecordPageWithFreeSpace(rel, usedBlock, 0, InvalidBlockNumber); } /* diff --git a/src/include/storage/freespace.h b/src/include/storage/freespace.h index 8b00033438..dbaae651c5 100644 --- a/src/include/storage/freespace.h +++ b/src/include/storage/freespace.h @@ -18,15 +18,20 @@ #include "storage/relfilenode.h" #include "utils/relcache.h" +/* Only create the FSM if the heap has greater than this many blocks */ +#define HEAP_FSM_CREATION_THRESHOLD 4 + /* prototypes for public functions in freespace.c */ extern Size GetRecordedFreeSpace(Relation rel, BlockNumber heapBlk); -extern BlockNumber GetPageWithFreeSpace(Relation rel, Size spaceNeeded); +extern BlockNumber GetPageWithFreeSpace(Relation rel, Size spaceNeeded, + bool check_fsm_only); extern BlockNumber RecordAndGetPageWithFreeSpace(Relation rel, BlockNumber oldPage, Size oldSpaceAvail, Size spaceNeeded); extern void RecordPageWithFreeSpace(Relation rel, BlockNumber heapBlk, - Size spaceAvail); + Size spaceAvail, BlockNumber nblocks); +extern void FSMClearLocalMap(void); extern void XLogRecordPageWithFreeSpace(RelFileNode rnode, BlockNumber heapBlk, Size spaceAvail); diff --git a/src/test/regress/expected/fsm.out b/src/test/regress/expected/fsm.out new file mode 100644 index 0000000000..b02993188c --- /dev/null +++ b/src/test/regress/expected/fsm.out @@ -0,0 +1,48 @@ +-- +-- Free Space Map test +-- +CREATE TABLE fsm_check_size (num int, str text); +-- With one block, there should be no FSM +INSERT INTO fsm_check_size VALUES(1, 'a'); +VACUUM fsm_check_size; +SELECT pg_relation_size('fsm_check_size', 'main') AS heap_size, +pg_relation_size('fsm_check_size', 'fsm') AS fsm_size; + heap_size | fsm_size +-----------+---------- + 8192 | 0 +(1 row) + +-- Extend table with enough blocks to exceed the FSM threshold +DO $$ +DECLARE curtid tid; +num int; +BEGIN +num = 11; + LOOP + INSERT INTO fsm_check_size VALUES (num, 'b') RETURNING ctid INTO curtid; + EXIT WHEN curtid >= tid '(4, 0)'; + num = num + 1; + END LOOP; +END; +$$; +VACUUM fsm_check_size; +SELECT pg_relation_size('fsm_check_size', 'fsm') AS fsm_size; + fsm_size +---------- + 24576 +(1 row) + +-- Add long random string to extend TOAST table to 1 block +INSERT INTO fsm_check_size +VALUES(0, (SELECT string_agg(md5(chr(i)), '') + FROM generate_series(1,100) i)); +VACUUM fsm_check_size; +SELECT pg_relation_size(reltoastrelid, 'main') AS toast_size, +pg_relation_size(reltoastrelid, 'fsm') AS toast_fsm_size +FROM pg_class WHERE relname = 'fsm_check_size'; + toast_size | toast_fsm_size +------------+---------------- + 8192 | 0 +(1 row) + +DROP TABLE fsm_check_size; diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule index cc0bbf5db9..4051a4ad4e 100644 --- a/src/test/regress/parallel_schedule +++ b/src/test/regress/parallel_schedule @@ -68,6 +68,12 @@ test: create_aggregate create_function_3 create_cast constraints triggers inheri # ---------- test: sanity_check +# ---------- +# fsm does a delete followed by vacuum, and running it in parallel can prevent +# removal of rows. +# ---------- +test: fsm + # ---------- # Believe it or not, select creates a table, subsequent # tests need. diff --git a/src/test/regress/serial_schedule b/src/test/regress/serial_schedule index 0c10c7100c..ac1ea622d6 100644 --- a/src/test/regress/serial_schedule +++ b/src/test/regress/serial_schedule @@ -80,6 +80,7 @@ test: roleattributes test: create_am test: hash_func test: sanity_check +test: fsm test: errors test: select test: select_into diff --git a/src/test/regress/sql/fsm.sql b/src/test/regress/sql/fsm.sql new file mode 100644 index 0000000000..332c3e2b2d --- /dev/null +++ b/src/test/regress/sql/fsm.sql @@ -0,0 +1,41 @@ +-- +-- Free Space Map test +-- + +CREATE TABLE fsm_check_size (num int, str text); + +-- With one block, there should be no FSM +INSERT INTO fsm_check_size VALUES(1, 'a'); + +VACUUM fsm_check_size; +SELECT pg_relation_size('fsm_check_size', 'main') AS heap_size, +pg_relation_size('fsm_check_size', 'fsm') AS fsm_size; + +-- Extend table with enough blocks to exceed the FSM threshold +DO $$ +DECLARE curtid tid; +num int; +BEGIN +num = 11; + LOOP + INSERT INTO fsm_check_size VALUES (num, 'b') RETURNING ctid INTO curtid; + EXIT WHEN curtid >= tid '(4, 0)'; + num = num + 1; + END LOOP; +END; +$$; + +VACUUM fsm_check_size; +SELECT pg_relation_size('fsm_check_size', 'fsm') AS fsm_size; + +-- Add long random string to extend TOAST table to 1 block +INSERT INTO fsm_check_size +VALUES(0, (SELECT string_agg(md5(chr(i)), '') + FROM generate_series(1,100) i)); + +VACUUM fsm_check_size; +SELECT pg_relation_size(reltoastrelid, 'main') AS toast_size, +pg_relation_size(reltoastrelid, 'fsm') AS toast_fsm_size +FROM pg_class WHERE relname = 'fsm_check_size'; + +DROP TABLE fsm_check_size;