On Wed, Feb 15, 2017 at 01:04:51PM +0100, Karsten Hilbert wrote:
> > Nope, that pops too. The query runs for a long time at a somewhat
> > normal rate of ram consumption, using ~1G of RSS then suddenly spikes
> > to about 6G, at which point the OOM killer pops it. Box has 8G of ram
> > and 4G of swap.
>
> By any chance:
>
> - when it happens has the kernel considered using swap ?
>
> - which kernel are you running ?
>
> There's been (for some workloads) massive problems with RAM
> exhaustion / swapping / OOM killer going wild with
> 4.7/4.8/some 4.9 kernels.
I guess what I'm trying to say is that it may actually not be
PostgreSQL's fault but rather the kernel invoking the OOM
killer way prematurely.
Karsten
--
GPG key ID E4071346 @ eu.pool.sks-keyservers.net
E167 67FD A291 2BEA 73BD 4537 78B9 A9F9 E407 1346