On Tue, Jan 27, 2009 at 01:52:46PM +0100, Daniel Schreiber wrote: > Iustin Pop schrieb: >> Thanks. This is starting to look like a different problem. >> >> The errno 9 we had before, but it should be solved. The strace shows >> that the ganeti-noded is not actually listening, and the call_version >> talks to a different host - it completes a connect(), while the node >> daemon doesn't get any traffic. > > For some reason the listening socket is not included in the select list. > So it never accepts incoming connections. > See lines 5861, 5863, 5920 in the strace. > >> I'll try to understand why you get the errno 9 error, but in the >> meantime also a "lsof -p $pid_of_node_daemon" and an strace of the node >> daemon startup would be helpful to understand in what state a node >> daemon is. > > Twisted included a closed file descriptor in the select list.
Ok, so it seems our cleaning up/closing filedescriptors is conflicting with twisted opening some pipes for itself. Why it only does in some cases and not always it's a mistery to me. In any case, could you try and run the node daemon in foreground (ganeti-noded -f)? In the non-daemon case, we don't cleanup the file descriptors, and this will confirm or infirm my theory. If this is the problem, we could backport a patch from trunk that closes the file descriptors very very early in the startup sequence, and not after forking. regards, iustin -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
