From: | Josh Berkus <josh(at)agliodbs(dot)com> |
---|---|
To: | pgsql-hackers(at)postgresql(dot)org |
Cc: | Bruce Momjian <bruce(at)momjian(dot)us>, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
Subject: | Re: What happens at BIND time? (pg_upgrade issue) |
Date: | 2013-08-30 20:23:38 |
Message-ID: | 5220FF4A.9090809@agliodbs.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On 08/28/2013 11:44 AM, Josh Berkus wrote:
> Tom,
>
>> Does the backend's memory usage climb, or hold steady? If the former,
>> I'd bet on client failure to release resources, eg not closing the
>> portals when done with them. A memory map from MemoryContextStats
>> would help determine exactly what's leaking.
>
> FS cache usage increases through the test run, as you'd expect, but the
> amount of pinned memory actually remains pretty much constant -- and has
> the same usage in both 8.4 (where the BIND issue doesn't happen) and
> 9.3b2 (where it does).
So, this just got a lot stranger.
What we've been testing here is upgrading from 8.4 to 9.X via
pg_upgrade, because that's what they have to do in production for time
reasons. We recently confirmed that this issue affects 9.0 also.
So I finally got results back from the test where we do dump/restore (to
9.3b2) instead of pg_upgrade ... and the BIND regression does not occur.
So this now has something to do with pg_upgrade, not just BIND and Java.
cc'd Bruce for that reason.
We'll be rerunning these tests next week, comparing a 9.3 with the issue
to a 9.3 without it under oprofile etc. Suggestions on what we should
look for are welcome.
--
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com
From | Date | Subject | |
---|---|---|---|
Next Message | Stephen Frost | 2013-08-30 20:24:06 | Re: [v9.4] row level security |
Previous Message | Josh Berkus | 2013-08-30 19:57:28 | Re: [v9.4] row level security |