Обсуждение: BUG #5675: out of memory at request size 32Mb

The following bug has been logged online:

Bug reference:      5675
Logged by:          Andreas Pflug
Email address:      pgadmin@pse-consulting.de
PostgreSQL version: 8.4.4
Operating system:   Linux 2.6.26 i686
Description:        out of memory at request size 32Mb
Details:

A slony 2.0 cluster running for a year, no config changes lately. Slony has
to replicate a table with a bytea column; maximum so far was a little over
32MB.

Lately, the slave's slon process fails to replicate. The server reports "out
of memory", "Failed on request of size 33554432.". The slon log shows an
insert was the failing statement, the server log spits out its memory
contexts but not the failing statement.

One log line that catched my attention was
MessageContext: 485902696 total in 6 blocks; 32160 free (16 chunks);
485870536 used
which is well beyond the configured memory limits (280MB shared_buffers,
temp 32MB, workmem 64MB)

On Fri, Sep 24, 2010 at 8:42 AM, Andreas Pflug
<pgadmin@pse-consulting.de> wrote:
>
> The following bug has been logged online:
>
> Bug reference: =A0 =A0 =A05675
> Logged by: =A0 =A0 =A0 =A0 =A0Andreas Pflug
> Email address: =A0 =A0 =A0pgadmin@pse-consulting.de
> PostgreSQL version: 8.4.4
> Operating system: =A0 Linux 2.6.26 i686
> Description: =A0 =A0 =A0 =A0out of memory at request size 32Mb
> Details:
>
> A slony 2.0 cluster running for a year, no config changes lately. Slony h=
as
> to replicate a table with a bytea column; maximum so far was a little over
> 32MB.
>
> Lately, the slave's slon process fails to replicate. The server reports "=
out
> of memory", "Failed on request of size 33554432.". The slon log shows an
> insert was the failing statement, the server log spits out its memory
> contexts but not the failing statement.
>
> One log line that catched my attention was
> MessageContext: 485902696 total in 6 blocks; 32160 free (16 chunks);
> 485870536 used
> which is well beyond the configured memory limits (280MB shared_buffers,
> temp 32MB, workmem 64MB)

Are you sure you meant to report this here?  Seems likely to be a
Slony issue, although of course I'm not sure of that.

--=20
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise Postgres Company

Am 24.09.10 14:54, schrieb Robert Haas:
> On Fri, Sep 24, 2010 at 8:42 AM, Andreas Pflug
> <pgadmin@pse-consulting.de>  wrote:
>> The following bug has been logged online:
>>
>> Bug reference:      5675
>> Logged by:          Andreas Pflug
>> Email address:      pgadmin@pse-consulting.de
>> PostgreSQL version: 8.4.4
>> Operating system:   Linux 2.6.26 i686
>> Description:        out of memory at request size 32Mb
>> Details:
>>
>> A slony 2.0 cluster running for a year, no config changes lately. Slony has
>> to replicate a table with a bytea column; maximum so far was a little over
>> 32MB.
>>
>> Lately, the slave's slon process fails to replicate. The server reports "out
>> of memory", "Failed on request of size 33554432.". The slon log shows an
>> insert was the failing statement, the server log spits out its memory
>> contexts but not the failing statement.
>>
>> One log line that catched my attention was
>> MessageContext: 485902696 total in 6 blocks; 32160 free (16 chunks);
>> 485870536 used
>> which is well beyond the configured memory limits (280MB shared_buffers,
>> temp 32MB, workmem 64MB)
> Are you sure you meant to report this here?  Seems likely to be a
> Slony issue, although of course I'm not sure of that.
>
It's the postgres server that reports the oom situation on the insert query.

On further invesigation, the slon process (running on the same machine)
consumed all memory on the 144MB-Insert (for a 29MB column) making pgsql
bail out.

Regards,
Andreas