On Tue, Mar 25, 2014 at 5:50 AM, Shinji Matsumoto <shinji.matsumoto@xxxxxxxxxxx> wrote: > Hello all, > > I have a question about Ceph notification mechanism. > http://ceph.com/docs/master/architecture/#object-watch-notify > > Scenario: > > (1) 3 clients (client1, client2, client3) have interests on a Ceph > object (object A). Each client calls rados_watch() to register > notification callback on the object.. > > (2) Client1 calls rados_notify() on object A. On client1 and client2, > the notification is delivered and processed successfully within a > timeout period. > > (3) However on client3, the notification is delivered, but it doesn't > complete within a timeout period for a temporary reason, such as > temporarily busy process for instance. > > (4) On client1, rados_notify() returns a status of timeout. > > (5) On client3, the temporary reason goes away and the notification is > processed. > > (6) Then, client3 attempts to read from object A by a call to rados_read(). > > Question: > > Will the call to rados_read() attempted at step (6) always succeed, > always fail or is the behavior unpredictable, which is, it's affected > by the factors that are not defined in this scenario? The success of the read is not impacted by the notify timeout -- you can order reads using watch-notify, but issues in the watch-notify framework won't impact reads. Watch-notify is a voluntary thing. IIRC the OSD involved might kill client3's connection, but it'll just establish a new one transparently. -Greg Software Engineer #42 @ http://inktank.com | http://ceph.com -- 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
