]> granicus.if.org Git - postgresql/commit
Make heap TID a tiebreaker nbtree index column.
authorPeter Geoghegan <pg@bowt.ie>
Wed, 20 Mar 2019 17:04:01 +0000 (10:04 -0700)
committerPeter Geoghegan <pg@bowt.ie>
Wed, 20 Mar 2019 17:04:01 +0000 (10:04 -0700)
commitdd299df8189bd00fbe54b72c64f43b6af2ffeccd
tree931ef720687d61cf5e75464fa0b1c1d75fb3f9d3
parente5adcb789d80ba565ccacb1ed4341a7c29085238
Make heap TID a tiebreaker nbtree index column.

Make nbtree treat all index tuples as having a heap TID attribute.
Index searches can distinguish duplicates by heap TID, since heap TID is
always guaranteed to be unique.  This general approach has numerous
benefits for performance, and is prerequisite to teaching VACUUM to
perform "retail index tuple deletion".

Naively adding a new attribute to every pivot tuple has unacceptable
overhead (it bloats internal pages), so suffix truncation of pivot
tuples is added.  This will usually truncate away the "extra" heap TID
attribute from pivot tuples during a leaf page split, and may also
truncate away additional user attributes.  This can increase fan-out,
especially in a multi-column index.  Truncation can only occur at the
attribute granularity, which isn't particularly effective, but works
well enough for now.  A future patch may add support for truncating
"within" text attributes by generating truncated key values using new
opclass infrastructure.

Only new indexes (BTREE_VERSION 4 indexes) will have insertions that
treat heap TID as a tiebreaker attribute, or will have pivot tuples
undergo suffix truncation during a leaf page split (on-disk
compatibility with versions 2 and 3 is preserved).  Upgrades to version
4 cannot be performed on-the-fly, unlike upgrades from version 2 to
version 3.  contrib/amcheck continues to work with version 2 and 3
indexes, while also enforcing stricter invariants when verifying version
4 indexes.  These stricter invariants are the same invariants described
by "3.1.12 Sequencing" from the Lehman and Yao paper.

A later patch will enhance the logic used by nbtree to pick a split
point.  This patch is likely to negatively impact performance without
smarter choices around the precise point to split leaf pages at.  Making
these two mostly-distinct sets of enhancements into distinct commits
seems like it might clarify their design, even though neither commit is
particularly useful on its own.

The maximum allowed size of new tuples is reduced by an amount equal to
the space required to store an extra MAXALIGN()'d TID in a new high key
during leaf page splits.  The user-facing definition of the "1/3 of a
page" restriction is already imprecise, and so does not need to be
revised.  However, there should be a compatibility note in the v12
release notes.

Author: Peter Geoghegan
Reviewed-By: Heikki Linnakangas, Alexander Korotkov
Discussion: https://postgr.es/m/CAH2-WzkVb0Kom=R+88fDFb=JSxZMFvbHVC6Mn9LJ2n=X=kS-Uw@mail.gmail.com
29 files changed:
contrib/amcheck/expected/check_btree.out
contrib/amcheck/sql/check_btree.sql
contrib/amcheck/verify_nbtree.c
contrib/pageinspect/btreefuncs.c
contrib/pageinspect/expected/btree.out
contrib/pgstattuple/expected/pgstattuple.out
doc/src/sgml/indices.sgml
src/backend/access/common/indextuple.c
src/backend/access/nbtree/README
src/backend/access/nbtree/nbtinsert.c
src/backend/access/nbtree/nbtpage.c
src/backend/access/nbtree/nbtree.c
src/backend/access/nbtree/nbtsearch.c
src/backend/access/nbtree/nbtsort.c
src/backend/access/nbtree/nbtutils.c
src/backend/access/nbtree/nbtxlog.c
src/backend/access/rmgrdesc/nbtdesc.c
src/backend/utils/sort/tuplesort.c
src/include/access/nbtree.h
src/include/access/nbtxlog.h
src/test/regress/expected/btree_index.out
src/test/regress/expected/create_index.out
src/test/regress/expected/dependency.out
src/test/regress/expected/event_trigger.out
src/test/regress/expected/foreign_data.out
src/test/regress/expected/rowsecurity.out
src/test/regress/sql/btree_index.sql
src/test/regress/sql/create_index.sql
src/test/regress/sql/foreign_data.sql