From: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
---|---|
To: | mark <markir(at)i4free(dot)co(dot)nz> |
Cc: | pgsql-general(at)hub(dot)org |
Subject: | Re: Hash Indexes Causing Deadlock Notices |
Date: | 2000-10-19 01:49:56 |
Message-ID: | 7051.971920196@sss.pgh.pa.us |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-general |
mark <markir(at)i4free(dot)co(dot)nz> writes:
> I noticed that a large number of log notice messages announcing a
> deadlock being detected together with transactions being aborted
> The tables impacted in the benchmark had HASH indexes. ( only = scans
> were being performed, it seemed a good idea at the time ..). Recreating
> these indexes as BTREE made all the messages go away. ( and produced a
> better request/s result ! )
IIRC, hash indexes use per-page locking that is not guaranteed deadlock
free. It's also bad for concurrency even when you don't get a deadlock.
The btree index type is by far the best choice we offer for
high-concurrency applications. Perhaps at some point someone will
be motivated to improve the other index types, but right now they
are definitely poor stepchildren ...
regards, tom lane
From | Date | Subject | |
---|---|---|---|
Next Message | Tom Lane | 2000-10-19 01:55:29 | Re: plpgsql - cache lookup error 18977 |
Previous Message | Larry Rosenman | 2000-10-19 00:07:03 | Re: [HACKERS] pg_connect error |