]> git.ipfire.org Git - thirdparty/postgresql.git/commit
This patch makes some improvements to the rtree index implementation:
authorNeil Conway <neilc@samurai.com>
Tue, 18 Jan 2005 23:25:55 +0000 (23:25 +0000)
committerNeil Conway <neilc@samurai.com>
Tue, 18 Jan 2005 23:25:55 +0000 (23:25 +0000)
commitb4297c177c466d2cc0268df6b16261ab4f3cb777
treebc2f25fcf8ed589ba0fb82d37fc5c76bd9dd14ac
parent1f5299bc3fde2e738e7080af72d26a08da53cabd
This patch makes some improvements to the rtree index implementation:

(1) Keep a pin on the scan's current buffer and mark buffer. This
avoids the need to do a ReadBuffer() for each tuple produced by the
scan. Since ReadBuffer() is expensive, this is a significant win.

(2) Convert a ReleaseBuffer(); ReadBuffer() pair into
ReleaseAndReadBuffer(). Surely not a huge win, but it saves a lock
acquire/release...

(3) Remove a bunch of duplicated code in rtget.c; make rtnext() handle
both the "initial result" and "subsequent result" cases.

(4) Add support for index tuple killing

(5) Remove rtscancache(): it is dead code, for the same reason that
gistscancache() is dead code (an index scan ought not be invoked with
NoMovementScanDirection).

The end result is about a 10% improvement in rtree index scan perf,
according to contrib/rtree_gist/bench.
src/backend/access/rtree/rtget.c
src/backend/access/rtree/rtree.c
src/backend/access/rtree/rtscan.c
src/include/access/rtree.h