Hung CVS server processes

[Dan Peterson]

We constantly find old CVS server processes hanging around and I'm
wondering if there's any reason they don't timeout.

For example, at the moment there's over 15 processes that have been
running (or rather sleeping) for more than 24 hours... the oldest of these
has been sleeping for almost 20 days.

Using "truss" I've determined these processes are usually in a "read"
call, sometimes in a "poll" call and on a few occasions in a "write"
call.

Using "netstat" it appears the TCP connection is ESTABLISHED.

I can't guarantee this 100%, but whenever I've checked in the past the
client associated with each connection is no longer running.

Why doesn't either the CVS server process or the TCP connection timeout?

Is this a function of the CVS server... in other words, does the server
need to set it's own timeout and terminate the connection?  If so, does
CVS do this and for some reason it isn't working on my system?  Or is this
a configurable compile time value that I haven't set?

Or is this a function of networking on Solaris?  If so, shouldn't the
socket have timed out after 20 days?  Is there some system setting which
controls this?