osd/PG: discard msgs from down peers

if a repop is replied after a replica goes down in a new osdmap, and
before the pg advances to this new osdmap, the repop replies before this
repop can be discarded by that replica OSD, because the primary resets the
connection to it when handling the new osdmap marking it down, and also
resets the messenger sesssion when the replica reconnects. to avoid the
out-of-order replies, the messages from that replica should be discarded.

Fixes: http://tracker.ceph.com/issues/19605
Signed-off-by: Kefu Chai <kchai@redhat.com>

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index 42e3883a1a80de31ee4ac5cec082e8b410075068..2ea5610911ad4241ddad19cfc26f6549b107f524 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -5696,12 +5696,21 @@ bool PG::can_discard_replica_op(OpRequestRef& op)
   const T *m = static_cast<const T *>(op->get_req());
   assert(m->get_type() == MSGTYPE);
 
+  int from = m->get_source().num();
+
+  // if a repop is replied after a replica goes down in a new osdmap, and
+  // before the pg advances to this new osdmap, the repop replies before this
+  // repop can be discarded by that replica OSD, because the primary resets the
+  // connection to it when handling the new osdmap marking it down, and also
+  // resets the messenger sesssion when the replica reconnects. to avoid the
+  // out-of-order replies, the messages from that replica should be discarded.
+  if (osd->get_osdmap()->is_down(from))
+    return true;
   /* Mostly, this overlaps with the old_peering_msg
    * condition.  An important exception is pushes
    * sent by replicas not in the acting set, since
    * if such a replica goes down it does not cause
    * a new interval. */
-  int from = m->get_source().num();
   if (get_osdmap()->get_down_at(from) >= m->map_epoch)
     return true;