postgresql/src/backend/access/nbtree/nbtdedup.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1106 lines
37 KiB
C
Raw Normal View History

Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
/*-------------------------------------------------------------------------
*
* nbtdedup.c
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* Deduplicate or bottom-up delete items in Postgres btrees.
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
*
* Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
* Portions Copyright (c) 1994, Regents of the University of California
*
*
* IDENTIFICATION
* src/backend/access/nbtree/nbtdedup.c
*
*-------------------------------------------------------------------------
*/
#include "postgres.h"
#include "access/nbtree.h"
#include "access/nbtxlog.h"
#include "access/xloginsert.h"
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
#include "miscadmin.h"
#include "utils/rel.h"
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
static void _bt_bottomupdel_finish_pending(Page page, BTDedupState state,
TM_IndexDeleteOp *delstate);
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
static bool _bt_do_singleval(Relation rel, Page page, BTDedupState state,
OffsetNumber minoff, IndexTuple newitem);
static void _bt_singleval_fillfactor(Page page, BTDedupState state,
Size newitemsz);
#ifdef USE_ASSERT_CHECKING
static bool _bt_posting_valid(IndexTuple posting);
#endif
/*
Deprecate nbtree's BTP_HAS_GARBAGE flag. Streamline handling of the various strategies that we have to avoid a page split in nbtinsert.c. When it looks like a leaf page is about to overflow, we now perform deleting LP_DEAD items and deduplication in one central place. This greatly simplifies _bt_findinsertloc(). This has an independently useful consequence: nbtree no longer relies on the BTP_HAS_GARBAGE page level flag/hint for anything important. We still set and unset the flag in the same way as before, but it's no longer treated as a gating condition when considering if we should check for already-set LP_DEAD bits. This happens at the point where the page looks like it might have to be split anyway, so simply checking the LP_DEAD bits in passing is practically free. This avoids missing LP_DEAD bits just because the page-level hint is unset, which is probably reasonably common (e.g. it happens when VACUUM unsets the page-level flag without actually removing index tuples whose LP_DEAD-bit was set recently, after the VACUUM operation began but before it reached the leaf page in question). Note that this isn't a big behavioral change compared to PostgreSQL 13. We were already checking for set LP_DEAD bits regardless of whether the BTP_HAS_GARBAGE page level flag was set before we considered doing a deduplication pass. This commit only goes slightly further by doing the same check for all indexes, even indexes where deduplication won't be performed. We don't completely remove the BTP_HAS_GARBAGE flag. We still rely on it as a gating condition with pg_upgrade'd indexes from before B-tree version 4/PostgreSQL 12. That makes sense because we sometimes have to make a choice among pages full of duplicates when inserting a tuple with pre version 4 indexes. It probably still pays to avoid accessing the line pointer array of a page there, since it won't yet be clear whether we'll insert on to the page in question at all, let alone split it as a result. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wz%3DYpc1PDdk8OVJDChGJBjT06%3DA0Mbv9HyTLCsOknGcUFg%40mail.gmail.com
2020-11-17 18:45:56 +01:00
* Perform a deduplication pass.
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
*
* The general approach taken here is to perform as much deduplication as
* possible to free as much space as possible. Note, however, that "single
* value" strategy is used for !bottomupdedup callers when the page is full of
* tuples of a single value. Deduplication passes that apply the strategy
* will leave behind a few untouched tuples at the end of the page, preparing
* the page for an anticipated page split that uses nbtsplitloc.c's own single
* value strategy. Our high level goal is to delay merging the untouched
* tuples until after the page splits.
*
* When a call to _bt_bottomupdel_pass() just took place (and failed), our
* high level goal is to prevent a page split entirely by buying more time.
* We still hope that a page split can be avoided altogether. That's why
* single value strategy is not even considered for bottomupdedup callers.
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
*
Deprecate nbtree's BTP_HAS_GARBAGE flag. Streamline handling of the various strategies that we have to avoid a page split in nbtinsert.c. When it looks like a leaf page is about to overflow, we now perform deleting LP_DEAD items and deduplication in one central place. This greatly simplifies _bt_findinsertloc(). This has an independently useful consequence: nbtree no longer relies on the BTP_HAS_GARBAGE page level flag/hint for anything important. We still set and unset the flag in the same way as before, but it's no longer treated as a gating condition when considering if we should check for already-set LP_DEAD bits. This happens at the point where the page looks like it might have to be split anyway, so simply checking the LP_DEAD bits in passing is practically free. This avoids missing LP_DEAD bits just because the page-level hint is unset, which is probably reasonably common (e.g. it happens when VACUUM unsets the page-level flag without actually removing index tuples whose LP_DEAD-bit was set recently, after the VACUUM operation began but before it reached the leaf page in question). Note that this isn't a big behavioral change compared to PostgreSQL 13. We were already checking for set LP_DEAD bits regardless of whether the BTP_HAS_GARBAGE page level flag was set before we considered doing a deduplication pass. This commit only goes slightly further by doing the same check for all indexes, even indexes where deduplication won't be performed. We don't completely remove the BTP_HAS_GARBAGE flag. We still rely on it as a gating condition with pg_upgrade'd indexes from before B-tree version 4/PostgreSQL 12. That makes sense because we sometimes have to make a choice among pages full of duplicates when inserting a tuple with pre version 4 indexes. It probably still pays to avoid accessing the line pointer array of a page there, since it won't yet be clear whether we'll insert on to the page in question at all, let alone split it as a result. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wz%3DYpc1PDdk8OVJDChGJBjT06%3DA0Mbv9HyTLCsOknGcUFg%40mail.gmail.com
2020-11-17 18:45:56 +01:00
* The page will have to be split if we cannot successfully free at least
* newitemsz (we also need space for newitem's line pointer, which isn't
* included in caller's newitemsz).
*
* Note: Caller should have already deleted all existing items with their
* LP_DEAD bits set.
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
*/
void
_bt_dedup_pass(Relation rel, Buffer buf, IndexTuple newitem, Size newitemsz,
bool bottomupdedup)
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
{
OffsetNumber offnum,
minoff,
maxoff;
Page page = BufferGetPage(buf);
BTPageOpaque opaque = BTPageGetOpaque(page);
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
Page newpage;
BTDedupState state;
Size pagesaving PG_USED_FOR_ASSERTS_ONLY = 0;
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
bool singlevalstrat = false;
int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
/* Passed-in newitemsz is MAXALIGNED but does not include line pointer */
newitemsz += sizeof(ItemIdData);
/*
* Initialize deduplication state.
*
* It would be possible for maxpostingsize (limit on posting list tuple
* size) to be set to one third of the page. However, it seems like a
* good idea to limit the size of posting lists to one sixth of a page.
* That ought to leave us with a good split point when pages full of
* duplicates can be split several times.
*/
state = (BTDedupState) palloc(sizeof(BTDedupStateData));
state->deduplicate = true;
state->nmaxitems = 0;
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
state->maxpostingsize = Min(BTMaxItemSize(page) / 2, INDEX_SIZE_MASK);
/* Metadata about base tuple of current pending posting list */
state->base = NULL;
state->baseoff = InvalidOffsetNumber;
state->basetupsize = 0;
/* Metadata about current pending posting list TIDs */
state->htids = palloc(state->maxpostingsize);
state->nhtids = 0;
state->nitems = 0;
/* Size of all physical tuples to be replaced by pending posting list */
state->phystupsize = 0;
/* nintervals should be initialized to zero */
state->nintervals = 0;
Deprecate nbtree's BTP_HAS_GARBAGE flag. Streamline handling of the various strategies that we have to avoid a page split in nbtinsert.c. When it looks like a leaf page is about to overflow, we now perform deleting LP_DEAD items and deduplication in one central place. This greatly simplifies _bt_findinsertloc(). This has an independently useful consequence: nbtree no longer relies on the BTP_HAS_GARBAGE page level flag/hint for anything important. We still set and unset the flag in the same way as before, but it's no longer treated as a gating condition when considering if we should check for already-set LP_DEAD bits. This happens at the point where the page looks like it might have to be split anyway, so simply checking the LP_DEAD bits in passing is practically free. This avoids missing LP_DEAD bits just because the page-level hint is unset, which is probably reasonably common (e.g. it happens when VACUUM unsets the page-level flag without actually removing index tuples whose LP_DEAD-bit was set recently, after the VACUUM operation began but before it reached the leaf page in question). Note that this isn't a big behavioral change compared to PostgreSQL 13. We were already checking for set LP_DEAD bits regardless of whether the BTP_HAS_GARBAGE page level flag was set before we considered doing a deduplication pass. This commit only goes slightly further by doing the same check for all indexes, even indexes where deduplication won't be performed. We don't completely remove the BTP_HAS_GARBAGE flag. We still rely on it as a gating condition with pg_upgrade'd indexes from before B-tree version 4/PostgreSQL 12. That makes sense because we sometimes have to make a choice among pages full of duplicates when inserting a tuple with pre version 4 indexes. It probably still pays to avoid accessing the line pointer array of a page there, since it won't yet be clear whether we'll insert on to the page in question at all, let alone split it as a result. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wz%3DYpc1PDdk8OVJDChGJBjT06%3DA0Mbv9HyTLCsOknGcUFg%40mail.gmail.com
2020-11-17 18:45:56 +01:00
minoff = P_FIRSTDATAKEY(opaque);
maxoff = PageGetMaxOffsetNumber(page);
/*
* Consider applying "single value" strategy, though only if the page
* seems likely to be split in the near future
*/
if (!bottomupdedup)
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
singlevalstrat = _bt_do_singleval(rel, page, state, minoff, newitem);
/*
* Deduplicate items from page, and write them to newpage.
*
* Copy the original page's LSN into newpage copy. This will become the
* updated version of the page. We need this because XLogInsert will
* examine the LSN and possibly dump it in a page image.
*/
newpage = PageGetTempPageCopySpecial(page);
PageSetLSN(newpage, PageGetLSN(page));
/* Copy high key, if any */
if (!P_RIGHTMOST(opaque))
{
ItemId hitemid = PageGetItemId(page, P_HIKEY);
Size hitemsz = ItemIdGetLength(hitemid);
IndexTuple hitem = (IndexTuple) PageGetItem(page, hitemid);
if (PageAddItem(newpage, (Item) hitem, hitemsz, P_HIKEY,
false, false) == InvalidOffsetNumber)
elog(ERROR, "deduplication failed to add highkey");
}
for (offnum = minoff;
offnum <= maxoff;
offnum = OffsetNumberNext(offnum))
{
ItemId itemid = PageGetItemId(page, offnum);
IndexTuple itup = (IndexTuple) PageGetItem(page, itemid);
Assert(!ItemIdIsDead(itemid));
if (offnum == minoff)
{
/*
* No previous/base tuple for the data item -- use the data item
* as base tuple of pending posting list
*/
_bt_dedup_start_pending(state, itup, offnum);
}
else if (state->deduplicate &&
_bt_keep_natts_fast(rel, state->base, itup) > nkeyatts &&
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
_bt_dedup_save_htid(state, itup))
{
/*
* Tuple is equal to base tuple of pending posting list. Heap
* TID(s) for itup have been saved in state.
*/
}
else
{
/*
* Tuple is not equal to pending posting list tuple, or
* _bt_dedup_save_htid() opted to not merge current item into
* pending posting list for some other reason (e.g., adding more
* TIDs would have caused posting list to exceed current
* maxpostingsize).
*
* If state contains pending posting list with more than one item,
* form new posting tuple and add it to our temp page (newpage).
* Else add pending interval's base tuple to the temp page as-is.
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
*/
pagesaving += _bt_dedup_finish_pending(newpage, state);
if (singlevalstrat)
{
/*
* Single value strategy's extra steps.
*
* Lower maxpostingsize for sixth and final large posting list
* tuple at the point where 5 maxpostingsize-capped tuples
* have either been formed or observed.
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
*
* When a sixth maxpostingsize-capped item is formed/observed,
* stop merging together tuples altogether. The few tuples
* that remain at the end of the page won't be merged together
* at all (at least not until after a future page split takes
* place, when this page's newly allocated right sibling page
* gets its first deduplication pass).
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
*/
if (state->nmaxitems == 5)
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
_bt_singleval_fillfactor(page, state, newitemsz);
else if (state->nmaxitems == 6)
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
{
state->deduplicate = false;
singlevalstrat = false; /* won't be back here */
}
}
/* itup starts new pending posting list */
_bt_dedup_start_pending(state, itup, offnum);
}
}
/* Handle the last item */
pagesaving += _bt_dedup_finish_pending(newpage, state);
/*
* If no items suitable for deduplication were found, newpage must be
* exactly the same as the original page, so just return from function.
*
* We could determine whether or not to proceed on the basis the space
* savings being sufficient to avoid an immediate page split instead. We
* don't do that because there is some small value in nbtsplitloc.c always
* operating against a page that is fully deduplicated (apart from
* newitem). Besides, most of the cost has already been paid.
*/
if (state->nintervals == 0)
{
/* cannot leak memory here */
pfree(newpage);
pfree(state->htids);
pfree(state);
return;
}
/*
* By here, it's clear that deduplication will definitely go ahead.
*
Deprecate nbtree's BTP_HAS_GARBAGE flag. Streamline handling of the various strategies that we have to avoid a page split in nbtinsert.c. When it looks like a leaf page is about to overflow, we now perform deleting LP_DEAD items and deduplication in one central place. This greatly simplifies _bt_findinsertloc(). This has an independently useful consequence: nbtree no longer relies on the BTP_HAS_GARBAGE page level flag/hint for anything important. We still set and unset the flag in the same way as before, but it's no longer treated as a gating condition when considering if we should check for already-set LP_DEAD bits. This happens at the point where the page looks like it might have to be split anyway, so simply checking the LP_DEAD bits in passing is practically free. This avoids missing LP_DEAD bits just because the page-level hint is unset, which is probably reasonably common (e.g. it happens when VACUUM unsets the page-level flag without actually removing index tuples whose LP_DEAD-bit was set recently, after the VACUUM operation began but before it reached the leaf page in question). Note that this isn't a big behavioral change compared to PostgreSQL 13. We were already checking for set LP_DEAD bits regardless of whether the BTP_HAS_GARBAGE page level flag was set before we considered doing a deduplication pass. This commit only goes slightly further by doing the same check for all indexes, even indexes where deduplication won't be performed. We don't completely remove the BTP_HAS_GARBAGE flag. We still rely on it as a gating condition with pg_upgrade'd indexes from before B-tree version 4/PostgreSQL 12. That makes sense because we sometimes have to make a choice among pages full of duplicates when inserting a tuple with pre version 4 indexes. It probably still pays to avoid accessing the line pointer array of a page there, since it won't yet be clear whether we'll insert on to the page in question at all, let alone split it as a result. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wz%3DYpc1PDdk8OVJDChGJBjT06%3DA0Mbv9HyTLCsOknGcUFg%40mail.gmail.com
2020-11-17 18:45:56 +01:00
* Clear the BTP_HAS_GARBAGE page flag. The index must be a heapkeyspace
* index, and as such we'll never pay attention to BTP_HAS_GARBAGE anyway.
* But keep things tidy.
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
*/
if (P_HAS_GARBAGE(opaque))
{
BTPageOpaque nopaque = BTPageGetOpaque(newpage);
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
nopaque->btpo_flags &= ~BTP_HAS_GARBAGE;
}
START_CRIT_SECTION();
PageRestoreTempPage(newpage, page);
MarkBufferDirty(buf);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
XLogRecPtr recptr;
xl_btree_dedup xlrec_dedup;
xlrec_dedup.nintervals = state->nintervals;
XLogBeginInsert();
XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
XLogRegisterData((char *) &xlrec_dedup, SizeOfBtreeDedup);
/*
* The intervals array is not in the buffer, but pretend that it is.
* When XLogInsert stores the whole buffer, the array need not be
* stored too.
*/
XLogRegisterBufData(0, (char *) state->intervals,
state->nintervals * sizeof(BTDedupInterval));
recptr = XLogInsert(RM_BTREE_ID, XLOG_BTREE_DEDUP);
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
/* Local space accounting should agree with page accounting */
Assert(pagesaving < newitemsz || PageGetExactFreeSpace(page) >= newitemsz);
/* cannot leak memory here */
pfree(state->htids);
pfree(state);
}
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
/*
* Perform bottom-up index deletion pass.
*
* See if duplicate index tuples (plus certain nearby tuples) are eligible to
* be deleted via bottom-up index deletion. The high level goal here is to
* entirely prevent "unnecessary" page splits caused by MVCC version churn
* from UPDATEs (when the UPDATEs don't logically modify any of the columns
* covered by the 'rel' index). This is qualitative, not quantitative -- we
* do not particularly care about once-off opportunities to delete many index
* tuples together.
*
* See nbtree/README for details on the design of nbtree bottom-up deletion.
* See access/tableam.h for a description of how we're expected to cooperate
* with the tableam.
*
* Returns true on success, in which case caller can assume page split will be
* avoided for a reasonable amount of time. Returns false when caller should
* deduplicate the page (if possible at all).
*
* Note: Occasionally we return true despite failing to delete enough items to
* avoid a split. This makes caller skip deduplication and go split the page
* right away. Our return value is always just advisory information.
*
* Note: Caller should have already deleted all existing items with their
* LP_DEAD bits set.
*/
bool
_bt_bottomupdel_pass(Relation rel, Buffer buf, Relation heapRel,
Size newitemsz)
{
OffsetNumber offnum,
minoff,
maxoff;
Page page = BufferGetPage(buf);
BTPageOpaque opaque = BTPageGetOpaque(page);
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
BTDedupState state;
TM_IndexDeleteOp delstate;
bool neverdedup;
int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
/* Passed-in newitemsz is MAXALIGNED but does not include line pointer */
newitemsz += sizeof(ItemIdData);
/* Initialize deduplication state */
state = (BTDedupState) palloc(sizeof(BTDedupStateData));
state->deduplicate = true;
state->nmaxitems = 0;
state->maxpostingsize = BLCKSZ; /* We're not really deduplicating */
state->base = NULL;
state->baseoff = InvalidOffsetNumber;
state->basetupsize = 0;
state->htids = palloc(state->maxpostingsize);
state->nhtids = 0;
state->nitems = 0;
state->phystupsize = 0;
state->nintervals = 0;
/*
* Initialize tableam state that describes bottom-up index deletion
* operation.
*
* We'll go on to ask the tableam to search for TIDs whose index tuples we
* can safely delete. The tableam will search until our leaf page space
* target is satisfied, or until the cost of continuing with the tableam
* operation seems too high. It focuses its efforts on TIDs associated
* with duplicate index tuples that we mark "promising".
*
* This space target is a little arbitrary. The tableam must be able to
* keep the costs and benefits in balance. We provide the tableam with
* exhaustive information about what might work, without directly
* concerning ourselves with avoiding work during the tableam call. Our
* role in costing the bottom-up deletion process is strictly advisory.
*/
delstate.irel = rel;
delstate.iblknum = BufferGetBlockNumber(buf);
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
delstate.bottomup = true;
delstate.bottomupfreespace = Max(BLCKSZ / 16, newitemsz);
delstate.ndeltids = 0;
delstate.deltids = palloc(MaxTIDsPerBTreePage * sizeof(TM_IndexDelete));
delstate.status = palloc(MaxTIDsPerBTreePage * sizeof(TM_IndexStatus));
minoff = P_FIRSTDATAKEY(opaque);
maxoff = PageGetMaxOffsetNumber(page);
for (offnum = minoff;
offnum <= maxoff;
offnum = OffsetNumberNext(offnum))
{
ItemId itemid = PageGetItemId(page, offnum);
IndexTuple itup = (IndexTuple) PageGetItem(page, itemid);
Assert(!ItemIdIsDead(itemid));
if (offnum == minoff)
{
/* itup starts first pending interval */
_bt_dedup_start_pending(state, itup, offnum);
}
else if (_bt_keep_natts_fast(rel, state->base, itup) > nkeyatts &&
_bt_dedup_save_htid(state, itup))
{
/* Tuple is equal; just added its TIDs to pending interval */
}
else
{
/* Finalize interval -- move its TIDs to delete state */
_bt_bottomupdel_finish_pending(page, state, &delstate);
/* itup starts new pending interval */
_bt_dedup_start_pending(state, itup, offnum);
}
}
/* Finalize final interval -- move its TIDs to delete state */
_bt_bottomupdel_finish_pending(page, state, &delstate);
/*
* We don't give up now in the event of having few (or even zero)
* promising tuples for the tableam because it's not up to us as the index
* AM to manage costs (note that the tableam might have heuristics of its
* own that work out what to do). We should at least avoid having our
* caller do a useless deduplication pass after we return in the event of
* zero promising tuples, though.
*/
neverdedup = false;
if (state->nintervals == 0)
neverdedup = true;
pfree(state->htids);
pfree(state);
/* Ask tableam which TIDs are deletable, then physically delete them */
_bt_delitems_delete_check(rel, buf, heapRel, &delstate);
pfree(delstate.deltids);
pfree(delstate.status);
/* Report "success" to caller unconditionally to avoid deduplication */
if (neverdedup)
return true;
/* Don't dedup when we won't end up back here any time soon anyway */
return PageGetExactFreeSpace(page) >= Max(BLCKSZ / 24, newitemsz);
}
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
/*
* Create a new pending posting list tuple based on caller's base tuple.
*
* Every tuple processed by deduplication either becomes the base tuple for a
* posting list, or gets its heap TID(s) accepted into a pending posting list.
* A tuple that starts out as the base tuple for a posting list will only
* actually be rewritten within _bt_dedup_finish_pending() when it turns out
* that there are duplicates that can be merged into the base tuple.
*/
void
_bt_dedup_start_pending(BTDedupState state, IndexTuple base,
OffsetNumber baseoff)
{
Assert(state->nhtids == 0);
Assert(state->nitems == 0);
Assert(!BTreeTupleIsPivot(base));
/*
* Copy heap TID(s) from new base tuple for new candidate posting list
* into working state's array
*/
if (!BTreeTupleIsPosting(base))
{
memcpy(state->htids, &base->t_tid, sizeof(ItemPointerData));
state->nhtids = 1;
state->basetupsize = IndexTupleSize(base);
}
else
{
int nposting;
nposting = BTreeTupleGetNPosting(base);
memcpy(state->htids, BTreeTupleGetPosting(base),
sizeof(ItemPointerData) * nposting);
state->nhtids = nposting;
/* basetupsize should not include existing posting list */
state->basetupsize = BTreeTupleGetPostingOffset(base);
}
/*
* Save new base tuple itself -- it'll be needed if we actually create a
* new posting list from new pending posting list.
*
* Must maintain physical size of all existing tuples (including line
* pointer overhead) so that we can calculate space savings on page.
*/
state->nitems = 1;
state->base = base;
state->baseoff = baseoff;
state->phystupsize = MAXALIGN(IndexTupleSize(base)) + sizeof(ItemIdData);
/* Also save baseoff in pending state for interval */
state->intervals[state->nintervals].baseoff = state->baseoff;
}
/*
* Save itup heap TID(s) into pending posting list where possible.
*
* Returns bool indicating if the pending posting list managed by state now
* includes itup's heap TID(s).
*/
bool
_bt_dedup_save_htid(BTDedupState state, IndexTuple itup)
{
int nhtids;
ItemPointer htids;
Size mergedtupsz;
Assert(!BTreeTupleIsPivot(itup));
if (!BTreeTupleIsPosting(itup))
{
nhtids = 1;
htids = &itup->t_tid;
}
else
{
nhtids = BTreeTupleGetNPosting(itup);
htids = BTreeTupleGetPosting(itup);
}
/*
* Don't append (have caller finish pending posting list as-is) if
* appending heap TID(s) from itup would put us over maxpostingsize limit.
*
* This calculation needs to match the code used within _bt_form_posting()
* for new posting list tuples.
*/
mergedtupsz = MAXALIGN(state->basetupsize +
(state->nhtids + nhtids) * sizeof(ItemPointerData));
if (mergedtupsz > state->maxpostingsize)
{
/*
* Count this as an oversized item for single value strategy, though
* only when there are 50 TIDs in the final posting list tuple. This
* limit (which is fairly arbitrary) avoids confusion about how many
* 1/6 of a page tuples have been encountered/created by the current
* deduplication pass.
*
* Note: We deliberately don't consider which deduplication pass
* merged together tuples to create this item (could be a previous
* deduplication pass, or current pass). See _bt_do_singleval()
* comments.
*/
if (state->nhtids > 50)
state->nmaxitems++;
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
return false;
}
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
/*
* Save heap TIDs to pending posting list tuple -- itup can be merged into
* pending posting list
*/
state->nitems++;
memcpy(state->htids + state->nhtids, htids,
sizeof(ItemPointerData) * nhtids);
state->nhtids += nhtids;
state->phystupsize += MAXALIGN(IndexTupleSize(itup)) + sizeof(ItemIdData);
return true;
}
/*
* Finalize pending posting list tuple, and add it to the page. Final tuple
* is based on saved base tuple, and saved list of heap TIDs.
*
* Returns space saving from deduplicating to make a new posting list tuple.
* Note that this includes line pointer overhead. This is zero in the case
* where no deduplication was possible.
*/
Size
_bt_dedup_finish_pending(Page newpage, BTDedupState state)
{
OffsetNumber tupoff;
Size tuplesz;
Size spacesaving;
Assert(state->nitems > 0);
Assert(state->nitems <= state->nhtids);
Assert(state->intervals[state->nintervals].baseoff == state->baseoff);
tupoff = OffsetNumberNext(PageGetMaxOffsetNumber(newpage));
if (state->nitems == 1)
{
/* Use original, unchanged base tuple */
tuplesz = IndexTupleSize(state->base);
Assert(tuplesz == MAXALIGN(IndexTupleSize(state->base)));
Assert(tuplesz <= BTMaxItemSize(newpage));
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
if (PageAddItem(newpage, (Item) state->base, tuplesz, tupoff,
false, false) == InvalidOffsetNumber)
elog(ERROR, "deduplication failed to add tuple to page");
spacesaving = 0;
}
else
{
IndexTuple final;
/* Form a tuple with a posting list */
final = _bt_form_posting(state->base, state->htids, state->nhtids);
tuplesz = IndexTupleSize(final);
Assert(tuplesz <= state->maxpostingsize);
/* Save final number of items for posting list */
state->intervals[state->nintervals].nitems = state->nitems;
Assert(tuplesz == MAXALIGN(IndexTupleSize(final)));
Assert(tuplesz <= BTMaxItemSize(newpage));
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
if (PageAddItem(newpage, (Item) final, tuplesz, tupoff, false,
false) == InvalidOffsetNumber)
elog(ERROR, "deduplication failed to add tuple to page");
pfree(final);
spacesaving = state->phystupsize - (tuplesz + sizeof(ItemIdData));
/* Increment nintervals, since we wrote a new posting list tuple */
state->nintervals++;
Assert(spacesaving > 0 && spacesaving < BLCKSZ);
}
/* Reset state for next pending posting list */
state->nhtids = 0;
state->nitems = 0;
state->phystupsize = 0;
return spacesaving;
}
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
/*
* Finalize interval during bottom-up index deletion.
*
* During a bottom-up pass we expect that TIDs will be recorded in dedup state
* first, and then get moved over to delstate (in variable-sized batches) by
* calling here. Call here happens when the number of TIDs in a dedup
* interval is known, and interval gets finalized (i.e. when caller sees next
* tuple on the page is not a duplicate, or when caller runs out of tuples to
* process from leaf page).
*
* This is where bottom-up deletion determines and remembers which entries are
* duplicates. This will be important information to the tableam delete
* infrastructure later on. Plain index tuple duplicates are marked
* "promising" here, per tableam contract.
*
* Our approach to marking entries whose TIDs come from posting lists is more
* complicated. Posting lists can only be formed by a deduplication pass (or
* during an index build), so recent version churn affecting the pointed-to
* logical rows is not particularly likely. We may still give a weak signal
* about posting list tuples' entries (by marking just one of its TIDs/entries
* promising), though this is only a possibility in the event of further
* duplicate index tuples in final interval that covers posting list tuple (as
* in the plain tuple case). A weak signal/hint will be useful to the tableam
* when it has no stronger signal to go with for the deletion operation as a
* whole.
*
* The heuristics we use work well in practice because we only need to give
* the tableam the right _general_ idea about where to look. Garbage tends to
* naturally get concentrated in relatively few table blocks with workloads
* that bottom-up deletion targets. The tableam cannot possibly rank all
* available table blocks sensibly based on the hints we provide, but that's
* okay -- only the extremes matter. The tableam just needs to be able to
* predict which few table blocks will have the most tuples that are safe to
* delete for each deletion operation, with low variance across related
* deletion operations.
*/
static void
_bt_bottomupdel_finish_pending(Page page, BTDedupState state,
TM_IndexDeleteOp *delstate)
{
bool dupinterval = (state->nitems > 1);
Assert(state->nitems > 0);
Assert(state->nitems <= state->nhtids);
Assert(state->intervals[state->nintervals].baseoff == state->baseoff);
for (int i = 0; i < state->nitems; i++)
{
OffsetNumber offnum = state->baseoff + i;
ItemId itemid = PageGetItemId(page, offnum);
IndexTuple itup = (IndexTuple) PageGetItem(page, itemid);
TM_IndexDelete *ideltid = &delstate->deltids[delstate->ndeltids];
TM_IndexStatus *istatus = &delstate->status[delstate->ndeltids];
if (!BTreeTupleIsPosting(itup))
{
/* Simple case: A plain non-pivot tuple */
ideltid->tid = itup->t_tid;
ideltid->id = delstate->ndeltids;
istatus->idxoffnum = offnum;
istatus->knowndeletable = false; /* for now */
istatus->promising = dupinterval; /* simple rule */
istatus->freespace = ItemIdGetLength(itemid) + sizeof(ItemIdData);
delstate->ndeltids++;
}
else
{
/*
* Complicated case: A posting list tuple.
*
* We make the conservative assumption that there can only be at
* most one affected logical row per posting list tuple. There
* will be at most one promising entry in deltids to represent
* this presumed lone logical row. Note that this isn't even
* considered unless the posting list tuple is also in an interval
* of duplicates -- this complicated rule is just a variant of the
* simple rule used to decide if plain index tuples are promising.
*/
int nitem = BTreeTupleGetNPosting(itup);
bool firstpromising = false;
bool lastpromising = false;
Assert(_bt_posting_valid(itup));
if (dupinterval)
{
/*
* Complicated rule: either the first or last TID in the
* posting list gets marked promising (if any at all)
*/
BlockNumber minblocklist,
midblocklist,
maxblocklist;
ItemPointer mintid,
midtid,
maxtid;
mintid = BTreeTupleGetHeapTID(itup);
midtid = BTreeTupleGetPostingN(itup, nitem / 2);
maxtid = BTreeTupleGetMaxHeapTID(itup);
minblocklist = ItemPointerGetBlockNumber(mintid);
midblocklist = ItemPointerGetBlockNumber(midtid);
maxblocklist = ItemPointerGetBlockNumber(maxtid);
/* Only entry with predominant table block can be promising */
firstpromising = (minblocklist == midblocklist);
lastpromising = (!firstpromising &&
midblocklist == maxblocklist);
}
for (int p = 0; p < nitem; p++)
{
ItemPointer htid = BTreeTupleGetPostingN(itup, p);
ideltid->tid = *htid;
ideltid->id = delstate->ndeltids;
istatus->idxoffnum = offnum;
istatus->knowndeletable = false; /* for now */
istatus->promising = false;
if ((firstpromising && p == 0) ||
(lastpromising && p == nitem - 1))
istatus->promising = true;
istatus->freespace = sizeof(ItemPointerData); /* at worst */
ideltid++;
istatus++;
delstate->ndeltids++;
}
}
}
if (dupinterval)
{
state->intervals[state->nintervals].nitems = state->nitems;
state->nintervals++;
}
/* Reset state for next interval */
state->nhtids = 0;
state->nitems = 0;
state->phystupsize = 0;
}
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
/*
* Determine if page non-pivot tuples (data items) are all duplicates of the
* same value -- if they are, deduplication's "single value" strategy should
* be applied. The general goal of this strategy is to ensure that
* nbtsplitloc.c (which uses its own single value strategy) will find a useful
* split point as further duplicates are inserted, and successive rightmost
* page splits occur among pages that store the same duplicate value. When
* the page finally splits, it should end up BTREE_SINGLEVAL_FILLFACTOR% full,
* just like it would if deduplication were disabled.
*
* We expect that affected workloads will require _several_ single value
* strategy deduplication passes (over a page that only stores duplicates)
* before the page is finally split. The first deduplication pass should only
* find regular non-pivot tuples. Later deduplication passes will find
* existing maxpostingsize-capped posting list tuples, which must be skipped
* over. The penultimate pass is generally the first pass that actually
* reaches _bt_singleval_fillfactor(), and so will deliberately leave behind a
* few untouched non-pivot tuples. The final deduplication pass won't free
* any space -- it will skip over everything without merging anything (it
* retraces the steps of the penultimate pass).
*
* Fortunately, having several passes isn't too expensive. Each pass (after
* the first pass) won't spend many cycles on the large posting list tuples
* left by previous passes. Each pass will find a large contiguous group of
* smaller duplicate tuples to merge together at the end of the page.
*/
static bool
_bt_do_singleval(Relation rel, Page page, BTDedupState state,
OffsetNumber minoff, IndexTuple newitem)
{
int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
ItemId itemid;
IndexTuple itup;
itemid = PageGetItemId(page, minoff);
itup = (IndexTuple) PageGetItem(page, itemid);
if (_bt_keep_natts_fast(rel, newitem, itup) > nkeyatts)
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
{
itemid = PageGetItemId(page, PageGetMaxOffsetNumber(page));
itup = (IndexTuple) PageGetItem(page, itemid);
if (_bt_keep_natts_fast(rel, newitem, itup) > nkeyatts)
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
return true;
}
return false;
}
/*
* Lower maxpostingsize when using "single value" strategy, to avoid a sixth
* and final maxpostingsize-capped tuple. The sixth and final posting list
* tuple will end up somewhat smaller than the first five. (Note: The first
* five tuples could actually just be very large duplicate tuples that
* couldn't be merged together at all. Deduplication will simply not modify
* the page when that happens.)
*
* When there are six posting lists on the page (after current deduplication
* pass goes on to create/observe a sixth very large tuple), caller should end
* its deduplication pass. It isn't useful to try to deduplicate items that
* are supposed to end up on the new right sibling page following the
* anticipated page split. A future deduplication pass of future right
* sibling page might take care of it. (This is why the first single value
* strategy deduplication pass for a given leaf page will generally find only
* plain non-pivot tuples -- see _bt_do_singleval() comments.)
*/
static void
_bt_singleval_fillfactor(Page page, BTDedupState state, Size newitemsz)
{
Size leftfree;
int reduction;
/* This calculation needs to match nbtsplitloc.c */
leftfree = PageGetPageSize(page) - SizeOfPageHeaderData -
MAXALIGN(sizeof(BTPageOpaqueData));
/* Subtract size of new high key (includes pivot heap TID space) */
leftfree -= newitemsz + MAXALIGN(sizeof(ItemPointerData));
/*
* Reduce maxpostingsize by an amount equal to target free space on left
* half of page
*/
reduction = leftfree * ((100 - BTREE_SINGLEVAL_FILLFACTOR) / 100.0);
if (state->maxpostingsize > reduction)
state->maxpostingsize -= reduction;
else
state->maxpostingsize = 0;
}
/*
* Build a posting list tuple based on caller's "base" index tuple and list of
* heap TIDs. When nhtids == 1, builds a standard non-pivot tuple without a
* posting list. (Posting list tuples can never have a single heap TID, partly
* because that ensures that deduplication always reduces final MAXALIGN()'d
* size of entire tuple.)
*
* Convention is that posting list starts at a MAXALIGN()'d offset (rather
* than a SHORTALIGN()'d offset), in line with the approach taken when
* appending a heap TID to new pivot tuple/high key during suffix truncation.
* This sometimes wastes a little space that was only needed as alignment
* padding in the original tuple. Following this convention simplifies the
* space accounting used when deduplicating a page (the same convention
* simplifies the accounting for choosing a point to split a page at).
*
* Note: Caller's "htids" array must be unique and already in ascending TID
* order. Any existing heap TIDs from "base" won't automatically appear in
* returned posting list tuple (they must be included in htids array.)
*/
IndexTuple
_bt_form_posting(IndexTuple base, ItemPointer htids, int nhtids)
{
uint32 keysize,
newsize;
IndexTuple itup;
if (BTreeTupleIsPosting(base))
keysize = BTreeTupleGetPostingOffset(base);
else
keysize = IndexTupleSize(base);
Assert(!BTreeTupleIsPivot(base));
Assert(nhtids > 0 && nhtids <= PG_UINT16_MAX);
Assert(keysize == MAXALIGN(keysize));
/* Determine final size of new tuple */
if (nhtids > 1)
newsize = MAXALIGN(keysize +
nhtids * sizeof(ItemPointerData));
else
newsize = keysize;
Assert(newsize <= INDEX_SIZE_MASK);
Assert(newsize == MAXALIGN(newsize));
/* Allocate memory using palloc0() (matches index_form_tuple()) */
itup = palloc0(newsize);
memcpy(itup, base, keysize);
itup->t_info &= ~INDEX_SIZE_MASK;
itup->t_info |= newsize;
if (nhtids > 1)
{
/* Form posting list tuple */
BTreeTupleSetPosting(itup, nhtids, keysize);
memcpy(BTreeTupleGetPosting(itup), htids,
sizeof(ItemPointerData) * nhtids);
Assert(_bt_posting_valid(itup));
}
else
{
/* Form standard non-pivot tuple */
itup->t_info &= ~INDEX_ALT_TID_MASK;
ItemPointerCopy(htids, &itup->t_tid);
Assert(ItemPointerIsValid(&itup->t_tid));
}
return itup;
}
/*
* Generate a replacement tuple by "updating" a posting list tuple so that it
* no longer has TIDs that need to be deleted.
*
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* Used by both VACUUM and index deletion. Caller's vacposting argument
* points to the existing posting list tuple to be updated.
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
*
* On return, caller's vacposting argument will point to final "updated"
* tuple, which will be palloc()'d in caller's memory context.
*/
void
_bt_update_posting(BTVacuumPosting vacposting)
{
IndexTuple origtuple = vacposting->itup;
uint32 keysize,
newsize;
IndexTuple itup;
int nhtids;
int ui,
d;
ItemPointer htids;
nhtids = BTreeTupleGetNPosting(origtuple) - vacposting->ndeletedtids;
Assert(_bt_posting_valid(origtuple));
Assert(nhtids > 0 && nhtids < BTreeTupleGetNPosting(origtuple));
/*
* Determine final size of new tuple.
*
* This calculation needs to match the code used within _bt_form_posting()
* for new posting list tuples. We avoid calling _bt_form_posting() here
* to save ourselves a second memory allocation for a htids workspace.
*/
keysize = BTreeTupleGetPostingOffset(origtuple);
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
if (nhtids > 1)
newsize = MAXALIGN(keysize +
nhtids * sizeof(ItemPointerData));
else
newsize = keysize;
Assert(newsize <= INDEX_SIZE_MASK);
Assert(newsize == MAXALIGN(newsize));
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
/* Allocate memory using palloc0() (matches index_form_tuple()) */
itup = palloc0(newsize);
memcpy(itup, origtuple, keysize);
itup->t_info &= ~INDEX_SIZE_MASK;
itup->t_info |= newsize;
if (nhtids > 1)
{
/* Form posting list tuple */
BTreeTupleSetPosting(itup, nhtids, keysize);
htids = BTreeTupleGetPosting(itup);
}
else
{
/* Form standard non-pivot tuple */
itup->t_info &= ~INDEX_ALT_TID_MASK;
htids = &itup->t_tid;
}
ui = 0;
d = 0;
for (int i = 0; i < BTreeTupleGetNPosting(origtuple); i++)
{
if (d < vacposting->ndeletedtids && vacposting->deletetids[d] == i)
{
d++;
continue;
}
htids[ui++] = *BTreeTupleGetPostingN(origtuple, i);
}
Assert(ui == nhtids);
Assert(d == vacposting->ndeletedtids);
Assert(nhtids == 1 || _bt_posting_valid(itup));
Assert(nhtids > 1 || ItemPointerIsValid(&itup->t_tid));
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
/* vacposting arg's itup will now point to updated version */
vacposting->itup = itup;
}
/*
* Prepare for a posting list split by swapping heap TID in newitem with heap
* TID from original posting list (the 'oposting' heap TID located at offset
* 'postingoff'). Modifies newitem, so caller should pass their own private
* copy that can safely be modified.
*
* Returns new posting list tuple, which is palloc()'d in caller's context.
* This is guaranteed to be the same size as 'oposting'. Modified newitem is
* what caller actually inserts. (This happens inside the same critical
* section that performs an in-place update of old posting list using new
* posting list returned here.)
*
* While the keys from newitem and oposting must be opclass equal, and must
* generate identical output when run through the underlying type's output
* function, it doesn't follow that their representations match exactly.
* Caller must avoid assuming that there can't be representational differences
* that make datums from oposting bigger or smaller than the corresponding
* datums from newitem. For example, differences in TOAST input state might
* break a faulty assumption about tuple size (the executor is entitled to
* apply TOAST compression based on its own criteria). It also seems possible
* that further representational variation will be introduced in the future,
* in order to support nbtree features like page-level prefix compression.
*
* See nbtree/README for details on the design of posting list splits.
*/
IndexTuple
_bt_swap_posting(IndexTuple newitem, IndexTuple oposting, int postingoff)
{
int nhtids;
char *replacepos;
char *replaceposright;
Size nmovebytes;
IndexTuple nposting;
nhtids = BTreeTupleGetNPosting(oposting);
Assert(_bt_posting_valid(oposting));
/*
* The postingoff argument originated as a _bt_binsrch_posting() return
* value. It will be 0 in the event of corruption that makes a leaf page
* contain a non-pivot tuple that's somehow identical to newitem (no two
* non-pivot tuples should ever have the same TID). This has been known
* to happen in the field from time to time.
*
* Perform a basic sanity check to catch this case now.
*/
if (!(postingoff > 0 && postingoff < nhtids))
elog(ERROR, "posting list tuple with %d items cannot be split at offset %d",
nhtids, postingoff);
Add deduplication to nbtree. Deduplication reduces the storage overhead of duplicates in indexes that use the standard nbtree index access method. The deduplication process is applied lazily, after the point where opportunistic deletion of LP_DEAD-marked index tuples occurs. Deduplication is only applied at the point where a leaf page split would otherwise be required. New posting list tuples are formed by merging together existing duplicate tuples. The physical representation of the items on an nbtree leaf page is made more space efficient by deduplication, but the logical contents of the page are not changed. Even unique indexes make use of deduplication as a way of controlling bloat from duplicates whose TIDs point to different versions of the same logical table row. The lazy approach taken by nbtree has significant advantages over a GIN style eager approach. Most individual inserts of index tuples have exactly the same overhead as before. The extra overhead of deduplication is amortized across insertions, just like the overhead of page splits. The key space of indexes works in the same way as it has since commit dd299df8 (the commit that made heap TID a tiebreaker column). Testing has shown that nbtree deduplication can generally make indexes with about 10 or 15 tuples for each distinct key value about 2.5X - 4X smaller, even with single column integer indexes (e.g., an index on a referencing column that accompanies a foreign key). The final size of single column nbtree indexes comes close to the final size of a similar contrib/btree_gin index, at least in cases where GIN's posting list compression isn't very effective. This can significantly improve transaction throughput, and significantly reduce the cost of vacuuming indexes. A new index storage parameter (deduplicate_items) controls the use of deduplication. The default setting is 'on', so all new B-Tree indexes automatically use deduplication where possible. This decision will be reviewed at the end of the Postgres 13 beta period. There is a regression of approximately 2% of transaction throughput with synthetic workloads that consist of append-only inserts into a table with several non-unique indexes, where all indexes have few or no repeated values. The underlying issue is that cycles are wasted on unsuccessful attempts at deduplicating items in non-unique indexes. There doesn't seem to be a way around it short of disabling deduplication entirely. Note that deduplication of items in unique indexes is fairly well targeted in general, which avoids the problem there (we can use a special heuristic to trigger deduplication passes in unique indexes, since we're specifically targeting "version bloat"). Bump XLOG_PAGE_MAGIC because xl_btree_vacuum changed. No bump in BTREE_VERSION, since the representation of posting list tuples works in a way that's backwards compatible with version 4 indexes (i.e. indexes built on PostgreSQL 12). However, users must still REINDEX a pg_upgrade'd index to use deduplication, regardless of the Postgres version they've upgraded from. This is the only way to set the new nbtree metapage flag indicating that deduplication is generally safe. Author: Anastasia Lubennikova, Peter Geoghegan Reviewed-By: Peter Geoghegan, Heikki Linnakangas Discussion: https://postgr.es/m/55E4051B.7020209@postgrespro.ru https://postgr.es/m/4ab6e2db-bcee-f4cf-0916-3a06e6ccbb55@postgrespro.ru
2020-02-26 22:05:30 +01:00
/*
* Move item pointers in posting list to make a gap for the new item's
* heap TID. We shift TIDs one place to the right, losing original
* rightmost TID. (nmovebytes must not include TIDs to the left of
* postingoff, nor the existing rightmost/max TID that gets overwritten.)
*/
nposting = CopyIndexTuple(oposting);
replacepos = (char *) BTreeTupleGetPostingN(nposting, postingoff);
replaceposright = (char *) BTreeTupleGetPostingN(nposting, postingoff + 1);
nmovebytes = (nhtids - postingoff - 1) * sizeof(ItemPointerData);
memmove(replaceposright, replacepos, nmovebytes);
/* Fill the gap at postingoff with TID of new item (original new TID) */
Assert(!BTreeTupleIsPivot(newitem) && !BTreeTupleIsPosting(newitem));
ItemPointerCopy(&newitem->t_tid, (ItemPointer) replacepos);
/* Now copy oposting's rightmost/max TID into new item (final new TID) */
ItemPointerCopy(BTreeTupleGetMaxHeapTID(oposting), &newitem->t_tid);
Assert(ItemPointerCompare(BTreeTupleGetMaxHeapTID(nposting),
BTreeTupleGetHeapTID(newitem)) < 0);
Assert(_bt_posting_valid(nposting));
return nposting;
}
/*
* Verify posting list invariants for "posting", which must be a posting list
* tuple. Used within assertions.
*/
#ifdef USE_ASSERT_CHECKING
static bool
_bt_posting_valid(IndexTuple posting)
{
ItemPointerData last;
ItemPointer htid;
if (!BTreeTupleIsPosting(posting) || BTreeTupleGetNPosting(posting) < 2)
return false;
/* Remember first heap TID for loop */
ItemPointerCopy(BTreeTupleGetHeapTID(posting), &last);
if (!ItemPointerIsValid(&last))
return false;
/* Iterate, starting from second TID */
for (int i = 1; i < BTreeTupleGetNPosting(posting); i++)
{
htid = BTreeTupleGetPostingN(posting, i);
if (!ItemPointerIsValid(htid))
return false;
if (ItemPointerCompare(htid, &last) <= 0)
return false;
ItemPointerCopy(htid, &last);
}
return true;
}
#endif