Обсуждение: BUG #6229: Postgresql crashes after: LOG: statistics buffer is full

Поиск
Список
Период
Сортировка

BUG #6229: Postgresql crashes after: LOG: statistics buffer is full

От
"Max Kunz"
Дата:
The following bug has been logged online:

Bug reference:      6229
Logged by:          Max Kunz
Email address:      admin@m4ximus.de
PostgreSQL version: 8.1.3
Operating system:   Suse
Description:        Postgresql crashes after: LOG: statistics buffer is full
Details:

Hello,

i have detected a problem after postgres logs "statistics buffer is full":
The database achieves farther on connections from the clients but no one
could process a statement and neither of the connection was closed.
It was also not possible to shutdown postgresql regularly.

Is there a dependence on statistics buffer, while processing statements?
I've read a lot about similar problems on the internet, but could not finde
a satisfying solution.

Is that a known problem?

Here is the statistic collector configuration:

# - Query/Index Statistics Collector -

stats_start_collector = on
stats_command_string = on
#stats_block_level = off
stats_row_level = on
#stats_reset_on_server_start = off

I need the collector to perform autovacuum. Is it helpful to disable the
command_string?

I am very greatful for any idea to avoid this problem in future.

best regards,
Max

Re: BUG #6229: Postgresql crashes after: LOG: statistics buffer is full

От
Alvaro Herrera
Дата:
Excerpts from Max Kunz's message of mar sep 27 13:50:57 -0300 2011:
>
> The following bug has been logged online:
>
> Bug reference:      6229
> Logged by:          Max Kunz
> Email address:      admin@m4ximus.de
> PostgreSQL version: 8.1.3
> Operating system:   Suse
> Description:        Postgresql crashes after: LOG: statistics buffer is full
> Details:
>
> Hello,
>
> i have detected a problem after postgres logs "statistics buffer is full":
> The database achieves farther on connections from the clients but no one
> could process a statement and neither of the connection was closed.
> It was also not possible to shutdown postgresql regularly.

The 8.1 branch is no longer supported.  You need to upgrade, at least to
8.2; but since that's going to lose support just past December 2011, you
should look into something more recent than that too.

If you really can't do anything about it, you should be using 8.1.21
which has more than four years of bugfixes on top of 8.1.3.

--
Álvaro Herrera <alvherre@commandprompt.com>
The PostgreSQL Company - Command Prompt, Inc.
PostgreSQL Replication, Consulting, Custom Development, 24x7 support

Re: BUG #6229: Postgresql crashes after: LOG: statistics buffer is full

От
Tom Lane
Дата:
Alvaro Herrera <alvherre@commandprompt.com> writes:
> Excerpts from Max Kunz's message of mar sep 27 13:50:57 -0300 2011:
>> i have detected a problem after postgres logs "statistics buffer is full":
>> The database achieves farther on connections from the clients but no one
>> could process a statement and neither of the connection was closed.
>> It was also not possible to shutdown postgresql regularly.

> The 8.1 branch is no longer supported.  You need to upgrade, at least to
> 8.2; but since that's going to lose support just past December 2011, you
> should look into something more recent than that too.

FWIW, I suspect the problem is at a much deeper level and "statistics
buffer is full" is just a minor symptom.  If the stats buffer process is
stuck, that should really have no effect on the system other than
failure to deliver statistics updates.  Probably the reason why it's
reporting this is that the stats collector process is stuck, and perhaps
the underlying reason for that is also affecting regular backends.  But
there's not nearly enough info here to diagnose the true cause.

In any case I agree with Alvaro that 8.1.3 is horribly obsolete,
and you should at least update to the end of the 8.1 release branch
if you can't easily migrate to a supported branch.

            regards, tom lane