Hi Henry, > 3. http://github.com/tcloud/ceph/commit/01e3bd78fe7ae26b394f5acda5c0ea875726a387 > > We found that every execution of 'ceph -s' causes a thread leaking on > the monitor. > This patch sort of fixed the problem: the thread will be terminated > when the subscription > expires. However, I wonder if we should provide an "unsubscribe" mechanism to > allow the client to explicitly unsubscribe it. There is an ms_handle_reset callback that triggers when the socket closes--that's where the cleanup should happen. Looking into this, though, it turns out the problem was somewhat unrelated. The monitor has a "lossy server" policy for non-monitor peers, which means that things connect to it, but it shouldn't initiate any outgoing connections. That piece of logic was missing from SimpleMessenger::submit_message(). Should be fixed by http://ceph.newdream.net/git/?p=ceph.git;a=commitdiff;h=01b58f38cc94ae7f209c215b782b1c7dec2c1b65 It will need a bit of testing. (I caught this was setting debug ms = 10 for [mon], and noticing that there were two different Pipe instances for the same peer address.) We're looking at the other two issues now. Thanks! sage -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
