From: | Teodor Sigaev <teodor(at)sigaev(dot)ru> |
---|---|
To: | Pgsql Hackers <pgsql-hackers(at)postgresql(dot)org> |
Cc: | Martin Schäfer <Martin(dot)Schaefer(at)cadcorp(dot)com> |
Subject: | Incorrect cursor behaviour with gist index |
Date: | 2008-10-17 18:33:20 |
Message-ID: | 48F8DA70.7010804@sigaev.ru |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
I'm back, sorry for a long absence.
About this bug: http://archives.postgresql.org/pgsql-bugs/2008-09/msg00086.php
Unfortunately, GiST index doesn't work with change direction of scan. I.e. it
can't move forward then move backward and this behaviour was from the beginning.
I think it's fixable, although GiST doesn't store on page left links (only right
links) and doesn't store parent page pointer. That's needed because matched
entries in GiST isn't stored consecutively unlike to BTree. So, price for it
will be storing in memory or stack all numbers of visited pages in index even
they will not be used. In worst case it's 2^32 of GISTSearchStack. Right now it
occupies 24-32 bytes depending on sizeof(void*), and it will be needed to add
one more pointer to make double-linked stack. So, about 32Mb. Is it acceptable?
Nevertheless, it doesn't solve problem with concurrent page splitting what can
cause different order between forward and backward scan in one cursor.
--
Teodor Sigaev E-mail: teodor(at)sigaev(dot)ru
WWW: http://www.sigaev.ru/
From | Date | Subject | |
---|---|---|---|
Next Message | Simon Riggs | 2008-10-17 19:13:19 | Re: Block-level CRC checks |
Previous Message | Teodor Sigaev | 2008-10-17 17:44:25 | Re: small bug in hlCover |