Re: synchronous_commit=off doesn't always return immediately

From: Merlin Moncure <mmoncure(at)gmail(dot)com>
To: tomrevam <tomer(at)fabrix(dot)tv>
Cc: pgsql-general(at)postgresql(dot)org
Subject: Re: synchronous_commit=off doesn't always return immediately
Date: 2009-08-03 14:25:03
Message-ID: b42b73150908030725s770d9b15q8703bbbca63c880c@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

On Mon, Aug 3, 2009 at 2:14 AM, tomrevam<tomer(at)fabrix(dot)tv> wrote:
>
>
>
> Tom Lane-2 wrote:
>>
>> tomrevam <tomer(at)fabrix(dot)tv> writes:
>> It might help to increase wal_buffers.
>>
>
> Hi,
>
> I increased the wal_buffers to 2 MB. Initially this improved the performance
> very much, but after 4 days of continuous operation the system is back to
> very long inserts and updates (at least as bad as it was before).
> Is there a way to know what resource the DB is running out of?

I can almost guarantee you that you are i/o bound. synchronous commit
helps, but is not magic: your single sata disk can only deliver about
100-200 iops and you are expecting more than that. I think you have
only two options: add more/faster disks (and consider separating wal
from the data volume) or disable sync completely with fsync=off (which
is of course quite dangerous).

This can be confirmed with an iostat, noting the tps on each volume
and the iowait.

merlin

In response to

Responses

Browse pgsql-general by date

  From Date Subject
Next Message Martijn van Oosterhout 2009-08-03 14:44:04 Re: Problem with Numerics multiplication in C-function
Previous Message Tom Lane 2009-08-03 14:20:36 Re: Problem with Numerics multiplication in C-function