Re: Hung CVS server processes

[Dan Peterson]

On Thu, 19 Sep 2002 address@hidden wrote:

> Dan Peterson writes:
> >
> > Why doesn't either the CVS server process or the TCP connection timeout?
>
> Good question.  The CVS server uses the KEEPALIVE socket option, if
> available, to detect broken connections.  KEEPALIVE tells the system to
> periodically ping the other side of an idle connection and break the
> connection if it fails to get a response in a reasonable amount of time.
> The fact that your connections are still in the ESTABLISHED state
> implies that this isn't working on your system -- many systems take a
> number of hours to detect a broken connection, but not 20 days.  As of
> CVS 1.11.1, CVS will syslog any errors it receives setting KEEPALIVE, so
> check your syslog.

Well, I do see several errors (84 over a 3.5 month period) related to
KEEPALIVE:

  ... error setting KEEPALIVE: Invalid argument

But the interesting thing is, none of the current list of 15 procs older
than 24 hours logged one of these KEEPALIVE message.  At least I assume I
should see the same process id in /var/adm/messages as the current "hung"
process?

What does "Invalid argument" in the context of setsockopt mean anyway?
The Solaris man page for setsockopt doesn't even show that EINVAL is a
valid error response.

> If there aren't any errors, that implies that your system is accepting
> the option but ignoring it, which is extremely unfriendly, to say the
> least.  You may want to check with some Solaris experts to see if
> there's some kind of option to enable it.