Re: Service outage: each postgres process use the exact amount of the configured work_mem

[Moreno Andreo <moreno.andreo@xxxxxxxxxx>]

Sorry,
 my mistake (I'm a bit nervous...)

that's not work_mem, but shared_buffers

Thanks

Il 14/04/2017 19:33, Melvin Davidson ha scritto:

On Fri, Apr 14, 2017 at 1:12 PM, Moreno Andreo <moreno.andreo@xxxxxxxxxx> wrote:

Hi all,
About 2 hours and half ago, suddenly (and on the late afternoon of the Easter Friday), customers reported failing connections to our server, or even very slow.
After a bit of checking (that also involved server reboot) I noticed (using top) that every process regarding postgres is using exactly the amout I configured as work_mem (3 GB). And AFAIK it's not good.

30085 postgres 20 0 3370048 156656 153876 S 6.7 0.3 0:00.20 postgres
29833 postgres 20 0 3370000 65260 62416 S 1.7 0.1 0:00.17 postgres
29632 postgres 20 0 3372468 11712 6028 S 0.7 0.0 0:00.60 postgres

What can be happened?
Nothing has been touched....
postgresql 9.5.6 on debian 8 just apt-get upgrade'd

Any help would be appreciated.
Moreno.

>using exactly the amout I configured as work_mem (3 GB).

You are right, that is bad, but that is your own fault. 3GB of work_mem is very bad, Try lowing in to something more reasonable, like 20GB.

https://www.postgresql.org/docs/9.5/static/runtime-config-resource.html#RUNTIME-CONFIG-RESOURCE-MEMORY
"several running sessions could be doing such operations concurrently. Therefore, the total memory used could be many times the value of work_mem; it is necessary to keep this fact in mind when choosing the value."

--

Melvin Davidson
I reserve the right to fantasize.  Whether or not you
wish to share my fantasy is entirely up to you.