Hi, Andy,
Impact of other VMs is reflected as %steal time. And it's only this
specific UPDATE query on this specific table that ever exhibits the problem,
as measured over many days, a query that must represent something like <2%
of all queries run against the DB over the same period and <5% of all other
UPDATE queries running on the system across other tables in the same
tablespace over, again, the same period.
V.
----- Original Message -----
From: "Andy Colson" <andy@squeakycode.net>
To: "Vlad Romascanu" <vromascanu@monaqasat.com>;
<pgsql-performance@postgresql.org>
Sent: Thursday, September 17, 2009 9:45 AM
Subject: Re: [PERFORM] Possible causes of sometimes slow single-row UPDATE
with trivial indexed condition?
> Vlad Romascanu wrote:
>> Problem occurs when running (in production) Postgres 8.3.7 64-bit (from
>> RPM) on Ubuntu 8.04.2, on an Amazon EC2 (xen) "Large" instance (8GB RAM),
>> with the DB on a 50GB EC2 block device.
>> Problem does not occur when running (in staging/pre-production) Postgres
>> 8.3.5 32-bit (from RPM) on Ubuntu 8.04.1, on a less beefy Amazon EC2
>> (xen) "Small" instance, with the DB on a 5GB EC2 block device.
>>
>> I am running with slow query logging on, and somewhat inexplicably I've
>> been getting the following slow UPDATE query several times in the past
>> weeks (I'm also including some context lines above and below):
>>
>
> I'm not sure how Amazon vm's work, but are there multiple vm's on one box?
> Just because your vm has zero cpu/disk does not mean the host isn't pegged
> out of its mind.
>
> Does Amazon give any sort of host stats?
>
> -Andy
>
