On Fri, Feb 18, 2011 at 2:50 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Fri, Feb 18, 2011 at 2:35 PM, Andrew Dunstan <andrew@dunslane.net> wrote:
>>> It's not running HS, so there's no query to wait on.
>
>> That seems to imply that recovery has leaked a buffer pin.
>
> No, because then the sanity check in LockBufferForCleanup would have
> fired:
>
> /* There should be exactly one local pin */
> if (PrivateRefCount[buffer - 1] != 1)
> elog(ERROR, "incorrect local pin count: %d",
> PrivateRefCount[buffer - 1]);
Hmm, yeah.
> Some sort of deadly embrace with the bgwriter, maybe?
Maybe.
I think it'd be useful to know what the buffer header thinks the
refcount on that buffer is, and what the startup process and the
bgwriter each have for PrivateRefCount[buffer].
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company