From: Patrick Donnelly <pdonnell@redhat.com>
Date: Sat, 22 Oct 2016 22:47:03 +0000 (-0400)
Subject: mds: check if down mds is known
X-Git-Tag: v10.2.6~36^2~10^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=54d6c1dd1c5a7ea2d34f0c0986d9b422a67b6f2d;p=ceph.git

mds: check if down mds is known

This avoids an assertion failure where an MDS receives an mdsmap that
causes it to enter up:replay and also see another MDS go down.

Fixes: http://tracker.ceph.com/issues/17670

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit 99a21133cd3133088e3977d46aab7120524eb350)
---

diff --git a/src/mds/MDSRank.cc b/src/mds/MDSRank.cc
index 19a584a65ccc..76b4662f8a64 100644
--- a/src/mds/MDSRank.cc
+++ b/src/mds/MDSRank.cc
@@ -1556,7 +1556,7 @@ void MDSRankDispatcher::handle_mds_map(
     oldmap->get_down_mds_set(&olddown);
     mdsmap->get_down_mds_set(&down);
     for (set<mds_rank_t>::iterator p = down.begin(); p != down.end(); ++p) {
-      if (olddown.count(*p) == 0) {
+      if (oldmap->have_inst(*p) && olddown.count(*p) == 0) {
         messenger->mark_down(oldmap->get_inst(*p).addr);
         handle_mds_failure(*p);
       }