On Tue, Dec 21, 2010 at 9:26 PM, Fujii Masao <masao.fujii@gmail.com> wrote:
> On Fri, Dec 17, 2010 at 7:27 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Tatsuhito Kasahara <kasahara.tatsuhito@oss.ntt.co.jp> writes:
>>> When I tested simple query as following, backend process used much memo=
ry
>>> and not freed until the backend was finished.
>>> # This is reproduced on PostgreSQL8.3 (PostgreSQL8.3.6 - PostgreSQL8.3.=
12)
>>
>> Hmm ... this test case doesn't appear to produce any significant memory
>> leakage in 8.4 and up. =A0What is happening in 8.3 is that the
>> CurTransactionContext of each subtransaction becomes nonempty, so it
>> eats 8K or so even though the snapshot gets released shortly later.
>> There are plenty of other ways to cause that to happen, though, so I'm
>> not particularly excited about fixing this one ... especially not in a
>> stable branch that's not getting a lot of developer testing anymore.
>> I'm inclined to leave this alone --- I think the risks of patching only
>> an old branch will outweigh the benefits.
>
> The proposed patch looks very simple. I don't think that applying that
> patch will cause serious risk.
>
> Unless the bug is fixed, the users who encountered the memory-leak
> cannot update their postgres to the latest version of 8.3. This would
> cause more serious situation. And, since psqlODBC frequently issues
> SAVEPONT and RELEASE SAVEPOINT, this memory-leak is not rare case,
> I think.
Are you saying that this problem does not exist in 8.3.0 but does
exist in later 8.3.x revs?
--=20
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
