From: | Mark Kirkwood <mark(dot)kirkwood(at)catalyst(dot)net(dot)nz> |
---|---|
To: | Cédric Villemain <cedric(dot)villemain(dot)debian(at)gmail(dot)com> |
Cc: | PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: patch review : Add ability to constrain backend temporary file space |
Date: | 2011-06-03 01:12:04 |
Message-ID: | 4DE834E4.4020100@catalyst.net.nz |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On 03/06/11 12:33, Cédric Villemain wrote:
> 2011/6/2 Mark Kirkwood<mark(dot)kirkwood(at)catalyst(dot)net(dot)nz>:
>> On 01/06/11 09:24, Cédric Villemain wrote:
>>> * I am not sure it is better to add a fileSize like you did or use
>>> relationgetnumberofblock() when file is about to be truncated or
>>> unlinked, this way the seekPos should be enough to increase the global
>>> counter.
>>>
>> The temp files are not relations so I'd have to call stat I guess. Now
>> truncate/unlink can happen quite a lot (e.g hash with many files) and I
>> wanted to avoid adding too many library calls to this code for performance
>> reasons, so on balance I'm thinking it is gonna be more efficient to
>> remember the size in the Vfd.
> I am not sure temporary relation are truncated. I have not checked
> right now, but IIRC, we don't need to truncate it. And I believe it
> would defeat the log_temp feature because log_temp is done on
> FileClose() only.
>
> If we are to add a field in struct vfd to keep the filesize then some
> optimization may happens for logging too....
We pretty much need to store the file size I think, due to needing to
work out if FileWrite is re-writing inside the file or extending it. So
maybe we could avoid a stat on unlink too (mind you it is a nice
validity check that the ongoing size accounting is correct). I was/am
keen to avoid changing too much here (heh - maybe I'm being too timid...).
Cheers
Mark
From | Date | Subject | |
---|---|---|---|
Next Message | Jun Ishiduka | 2011-06-03 01:58:21 | Re: Online base backup from the hot-standby |
Previous Message | Jeff Davis | 2011-06-03 00:48:53 | Re: storing TZ along timestamps |