From: Xiubo Li <xiubli@redhat.com>
Date: Wed, 3 Nov 2021 06:27:02 +0000 (+0800)
Subject: mds: just respawn mds daemon when osd op requests timeout
X-Git-Tag: v15.2.16~37^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=101ae53a6c333058cd8f6f4d169d29457133b446;p=ceph.git

mds: just respawn mds daemon when osd op requests timeout

Fixes: https://tracker.ceph.com/issues/51280
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---

diff --git a/src/mds/MDSContext.cc b/src/mds/MDSContext.cc
index 94726a2cb4f..f96fc54202a 100644
--- a/src/mds/MDSContext.cc
+++ b/src/mds/MDSContext.cc
@@ -107,8 +107,11 @@ void MDSIOContextBase::complete(int r) {
     return;
   }
 
-  if (r == -EBLACKLISTED) {
-    derr << "MDSIOContextBase: blacklisted!  Restarting..." << dendl;
+  // It's possible that the osd op requests will be stuck and then times out
+  // after "rados_osd_op_timeout", the mds won't know what we should it, just
+  // respawn it.
+  if (r == -EBLACKLISTED || r == -ETIMEDOUT) {
+    derr << "MDSIOContextBase: failed with " << r << ", restarting..." << dendl;
     mds->respawn();
   } else {
     MDSContext::complete(r);