Re: Nasty VACUUM/bgwriter/segmentation bug - Mailing list pgsql-hackers

From Heikki Linnakangas
Subject Re: Nasty VACUUM/bgwriter/segmentation bug
Date
Msg-id 4560BCEB.6050206@enterprisedb.com
Whole thread Raw
In response to Re: Nasty VACUUM/bgwriter/segmentation bug  ("Florian G. Pflug" <fgp@phlo.org>)
Responses Re: Nasty VACUUM/bgwriter/segmentation bug  ("Florian G. Pflug" <fgp@phlo.org>)
List pgsql-hackers
Florian G. Pflug wrote:
> Joshua D. Drake wrote:
>> On Sun, 2006-11-19 at 11:28 -0800, Josh Berkus wrote:
>>> Tom,
>>>> Let's go with the easy fix. With regular 1GB segment size, having a few
>>>> empty files in the data directory isn't going to hurt anyone.
>>> No, but it will confuse DBAs ("What the heck are all these 0B 
>>> files?").  Maybe we should add code to VACUUM to look for these empty 
>>> file segments and unlink them if they haven't been touched in a while 
>>> (say, a day?).
>>
>> I don't buy this argument. A smart DBA isn't going to do any such thing.
>> If you go around deleting unknown files you deserve everything you get.
>> Especially if you are in the middle of a PostgreSQL cluster tree.
> 
> Couldn't some application trigger this problem regularly (Say, once a 
> day while
> importing new data), and therefore create an ever increasing number of 
> empty
> files that you can't get rid of without backup,reinit and restore?

No, the old empty files would get reused if the relation grows again. 
And they would get deleted if you truncate or drop the relation.

> Maybe postgres could delete them during recovery?

Yes, though it would have to run not only when recovering from WAL, but 
on normal startup as well. It would require sweeping through the data 
directory, looking for the empty files, which would increase the startup 
time a bit. We discussed a solution like that to find orphaned relation 
files some time ago, which IMHO is a real problem, but that idea got 
forgotten.

--   Heikki Linnakangas  EnterpriseDB   http://www.enterprisedb.com


pgsql-hackers by date:

Previous
From: "Florian G. Pflug"
Date:
Subject: Re: Nasty VACUUM/bgwriter/segmentation bug
Next
From: "Florian G. Pflug"
Date:
Subject: Re: Nasty VACUUM/bgwriter/segmentation bug