From: Yan, Zheng <zyan@redhat.com>
Date: Thu, 2 Aug 2018 02:05:12 +0000 (+0800)
Subject: mds: update MDSRank::cluster_degraded before handling mds failure
X-Git-Tag: v13.2.2~74^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=refs%2Fpull%2F23180%2Fhead;p=ceph.git

mds: update MDSRank::cluster_degraded before handling mds failure

Migrator checks MDSRank::is_cluster_degraded() to decide if it needs to
send message to other mds. MDSRank::cluster_degraded should get updated
before calling Migrator::handle_mds_failure_or_stop().

Introduced by commit 7de9da4a "mds: handle discontinuous mdsmap"

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 93458c7306c3822175ed3c048f0937af8f02c341)
---

diff --git a/src/mds/MDSRank.cc b/src/mds/MDSRank.cc
index 4d473a7d6801..f44ec1d126e7 100644
--- a/src/mds/MDSRank.cc
+++ b/src/mds/MDSRank.cc
@@ -1738,6 +1738,8 @@ void MDSRankDispatcher::handle_mds_map(
   if (g_conf->mds_dump_cache_on_map)
     mdcache->dump_cache();
 
+  cluster_degraded = mdsmap->is_degraded();
+
   // mdsmap and oldmap can be discontinuous. failover might happen in the missing mdsmap.
   // the 'restart' set tracks ranks that have restarted since the old mdsmap
   set<mds_rank_t> restart;
@@ -1896,7 +1898,6 @@ void MDSRankDispatcher::handle_mds_map(
     }
   }
 
-  cluster_degraded = mdsmap->is_degraded();
   if (oldmap->is_degraded() && !cluster_degraded && state >= MDSMap::STATE_ACTIVE) {
     dout(1) << "cluster recovered." << dendl;
     auto it = waiting_for_active_peer.find(MDS_RANK_NONE);