Обсуждение: Vacuum causing crashes

Hi,

We're running a Postgres 8.0.1 database and have a maintenance process
that runs vacuum on selected tables every 10 minutes. Each table takes
around 2-3 seconds to vacuum. Since we've started this process we've
seen a lot of postmaster crashes (it says it received signal 9). It
appears these crashes occur when we try to stop a vacuum by interrupting
the vacuumdb process, but sometimes they occur without us having to do this.

Any ideas? We didn't experience so many database crashes before we
started vacuuming regularly.

Regards
--
David Mitchell
Software Engineer
Telogis

David Mitchell <david.mitchell@telogis.com> writes:

> Hi,
>
> We're running a Postgres 8.0.1 database and have a maintenance process
> that runs vacuum on selected tables every 10 minutes. Each table takes
> around 2-3 seconds to vacuum. Since we've started this process we've
> seen a lot of postmaster crashes (it says it received signal 9). It
> appears these crashes occur when we try to stop a vacuum by
> interrupting the vacuumdb process, but sometimes they occur without us
> having to do this.
>
> Any ideas? We didn't experience so many database crashes before we
> started vacuuming regularly.

If you're running Linux, it's possible that your system is
overcommitted and low on memory and the kernel is picking random
processes to kill.  There should be entries in the syslogs related to
this, if that's what's happening.  There are ways to tell (some
versions of) Linux not to overcommit memory.

Otherwise, no part of Postgres sends signal 9 AFAIK, so it's either a
rogue script or a person doing the killing.

-Doug

> If you're running Linux, it's possible that your system is
> overcommitted and low on memory and the kernel is picking random
> processes to kill.  There should be entries in the syslogs related to
> this, if that's what's happening.  There are ways to tell (some
> versions of) Linux not to overcommit memory.

Cheers Doug, that's exactly what it was. That explains a lot of
mysterious process deaths recently.

Thanks heaps

--
David Mitchell
Software Engineer
Telogis

David Mitchell <david.mitchell@telogis.com> writes:

>> If you're running Linux, it's possible that your system is
>> overcommitted and low on memory and the kernel is picking random
>> processes to kill.  There should be entries in the syslogs related to
>> this, if that's what's happening.  There are ways to tell (some
>> versions of) Linux not to overcommit memory.
>
> Cheers Doug, that's exactly what it was. That explains a lot of
> mysterious process deaths recently.
>
> Thanks heaps

Glad to help!

I have /proc/sys/vm/overcommit_memory set to 2 on all my critical
servers...

-Doug