Comment 7 for bug 717345

I think Robert mentions it, but I think the problem could certainly be *not* that we are leaking handles, but leaking processes that aren't shutting down when they should.

I saw that a little bit in the bzr-lp-forking-service.log. When the shutdown request came, there were 138 "active" children, but not all of them shut down. I think some of them didn't even shut down when the master process signaled them (SIGINT? SIGTERM? Not sure which). I'm pretty sure the master process will even send SIGKILL if things haven't shut down cleanly by the time it decides to exit.

Though if a sysadmin force-killed the master process, then it wouldn't kill the child processes, etc.

If we can get a soft kill on some of those processes, to see what traceback they put into the log file, that could be pretty helpful. I would try SIGTERM, SIGINT, SIGABRT and maybe SIGQUIT before SIGKILL. SIGQUIT could have problems because it would want to drop into the debugger, but maybe that in itself would give us some info.