Re: BUG #14714: long running sessions from remote instance seems to hang some times

Hi,

On 2017-06-19 14:49:07 +0000, josef(dot)machytka(at)gmail(dot)com wrote:
> The following bug has been logged on the website:
>
> Bug reference: 14714
> Logged by: Josef Machytka
> Email address: josef(dot)machytka(at)gmail(dot)com
> PostgreSQL version: 9.6.3
> Operating system: Debian jessie
> Description:
>
> On pg 9.6.* we have sometimes strange problem with some sessions. We
> experienced it from bash scripts, golang program and node.js applications.
> Environment are Google Compute Engine instances with Debian 8.
>
> Sometimes some session goes into some kind of "drowsy" state and runs
> incredibly slowly. It causes some very low disk IO and runs like 20x slower
> then usual. This happens mainly when database is under heavy load and it
> happens randomly to different tasks. It happens randomly to both connections
> from local instance (cronjob running on instance with PG) and remote
> connections from other instances.
>
> When we restart task it runs normally. I cannot see any changes in process
> priority in OS, process switches between sleeping and running state and I do
> not see any locks or waits in pg_stat_activity or pg_locks. Usually all
> other processes finish OK and this "drowsy" one runs "for ever..."

Which version of the os & kernel is this? This kind of sounds like the
symptoms that we've observed more frequently a while ago with
the kernel's transparent_hugepages management.

Greetings,

Andres Freund