mds: heartbeat_reset in dispatch

Previously we only heartbeated in tick.  However, our locking is
not guaranteed to be fair, so on a super-busy dispatch queue it may be
possible for the heartbeat to time out while the tick() function
is waiting for mds_lock.

Fixes: http://tracker.ceph.com/issues/19118
Signed-off-by: John Spray <john.spray@redhat.com>
(cherry picked from commit 819394549af10532419d88742fae3a69d2ea487d)

diff --git a/src/mds/MDSRank.cc b/src/mds/MDSRank.cc
index e1f28f6c0218f62a0856f2f8457d8a17229a5221..4a487cd50ea801c8131a5f2fffe6bc6b9c93946b 100644 (file)
--- a/src/mds/MDSRank.cc
+++ b/src/mds/MDSRank.cc
@@ -438,6 +438,8 @@ bool MDSRank::_dispatch(Message *m, bool new_msg)
       dout(0) << "unrecognized message " << *m << dendl;
       return false;
     }
+
+    heartbeat_reset();
   }
 
   if (dispatch_depth > 1)