Re: Switching timeline over streaming replication - Mailing list pgsql-hackers

From Heikki Linnakangas
Subject Re: Switching timeline over streaming replication
Date
Msg-id 50A50744.9050808@vmware.com
Whole thread Raw
In response to Re: Switching timeline over streaming replication  (Tom Lane <tgl@sss.pgh.pa.us>)
Responses Plugging fd leaks (was Re: Switching timeline over streaming replication)  (Heikki Linnakangas <hlinnakangas@vmware.com>)
List pgsql-hackers
On 15.11.2012 16:55, Tom Lane wrote:
> Heikki Linnakangas<hlinnakangas@vmware.com>  writes:
>> This is a fairly general issue, actually. Looking around, I can see at
>> least two similar cases in existing code, with BasicOpenFile, where we
>> will leak file descriptors on error:
>
> Um, don't we automatically clean those up during transaction abort?

Not the ones allocated with PathNameOpenFile or BasicOpenFile. Files 
allocated with AllocateFile() and OpenTemporaryFile() are cleaned up at 
abort.

> If we don't, we ought to think about that, not about cluttering calling
> code with certain-to-be-inadequate cleanup in error cases.

Agreed. Cleaning up at end-of-xact won't help walsender or other 
non-backend processes, though, because they don't do transactions. But a 
top-level ResourceOwner that's reset in the sigsetjmp() cleanup routine 
would work.

- Heikki



pgsql-hackers by date:

Previous
From: Andres Freund
Date:
Subject: Re: [PATCH 02/14] Add support for a generic wal reading facility dubbed XLogReader
Next
From: Alvaro Herrera
Date:
Subject: Re: [PATCH 02/14] Add support for a generic wal reading facility dubbed XLogReader