On Fri, Mar 14, 2014 at 10:31 AM, thorkill <
trackid-ohcee4ve at grid.einherjar.de> wrote:

> I my case it's "master" on FreeBSD 9.2 and a "slave" on FreeBSD 9.1. It
> seems that in some time around net.inet.tcp.keepidle (sysctl -a to
> check) when keepalive from OS kicks in the transsmission will be droped
> depending on your keepalive configuration. You should also see error
> messages in postgresql logs like:
>
> getsockopt(TCP_KEEPCNT) failed: Protocol not available
>
> > Any ideas? This is so confusing because it is such an odd time interval
> > before it locks up. What's magical about 58 minutes?
>
> My wild guess is:
>
> http://lists.freebsd.org/pipermail/freebsd-stable/2013-November/075781.html
>
>
>
>
>
Wow. These symptoms line up *exactly*.

I notice the getsockopt error is logged a handful of times when slon starts
and connects.

I applied the patch from the above message thread and rebooted the kernel.
When slon started, no getsockopt errors were reported. I'll let you know in
an hour what happens.

By the way, my net.inet.tcp.keepidle is 7200000, which should be 2 hours.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.slony.info/pipermail/slony1-general/attachments/20140314/ff4ca79b/attachment.htm