Re: BUG #7670: BUG #7545: Unresponsive server with error log reporting: "poll() failed: Invalid argument"

On 2012-11-18 17:36:40 +0000, Peter Geoghegan wrote:
> On 18 November 2012 17:12, Andres Freund <andres(at)2ndquadrant(dot)com> wrote:
> > Ok, this is already helpful. Do you have anything the log that shows
> > what process pid 95528 is? Could you grep for it?
>
> I haven't looked at this in detail, but are you sure that this isn't
> the bug that was fixed by commit
> e81e8f9342b037246b284bad15e42e21b1929301 ? The "invalid argument"
> poll() error is too generic to be sure what is really at fault, but
> this doesn't need to be a FATAL error, it seems.

I looked at that commit at first as well, but I don't really see how it
could have fixed that issue.
Btw, its not a FATAL but a plain ERROR.

> Is this a 32-bit platform? It's probably just that there is a large
> number of sockets, which Google seems to think can cause this error
> with poll(), so this seems like a slight variant of the problem
> reported by Sean Chittenden (kernel resource exhaustion causes pipe
> creation to fail, with unnecessarily bad consequences for Postgres).

We never have more than 3 fds in the poll. And a failure to create the
pipe should have caused problems at other locations before this...

Greetings,

Andres Freund

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services