Lately we have found a creeping problem on our Linux based CVS server. Over the course of a few days many cvs processes are piling up. Eventually when about 40+ of them are open, the service stops responding to clients. Then I manually kill the cvs processes and re-enable the CVS service. Everything appear to work fine from the client side after that. At least until too many processes are created.
The extra cvs processes are listed as "sleeping" (ie. Stat == 'S'). They all list xinetd as the parent, as they should. We are using pserver protocol[1] to access the service. I know there are many more details involved here. But I am not sure where to begin looking for traces of what is going on. If anyone has seen such a thing before, perhaps you could share some hints about where to look for clues. BTW, this is CVS version 1.11.17 running on a Fedora Core 2 system on an internal network. Alan [1] I know pserver is not secure and may be bad in other ways. This thread is not intended to open a discussion about pserver vs. other CVS connection protocols unless that is an obvious solution to this symptom. I'd not mind a discussion about that if someone wants to start another thread.
