client: flush the mdlog before waiting for unsafe requests

For safe replies the clients need to wait the MDS daemons to flush
the journal logs to Rados first. The worst case it will take 5 seconds.

Just trigger to flush the journal logs before waiting.

Fixes: https://tracker.ceph.com/issues/59705
Signed-off-by: Xiubo Li <xiubli@redhat.com>
(cherry picked from commit 2b715b858e36a82b42b12129fcfcc13090f30a93)

diff --git a/src/client/Client.cc b/src/client/Client.cc
index 8613f19ed94a2ec27eca87b72e3b3e62960c9e93..04320c51606d20f947e4049c22309cb35f5c4f02 100644 (file)
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -11780,6 +11780,9 @@ int Client::_sync_fs()
   flush_caps_sync();
   ceph_tid_t flush_tid = last_flush_tid;
 
+  // flush the mdlog before waiting for unsafe requests.
+  flush_mdlog_sync();
+
   // wait for unsafe mds requests
   wait_unsafe_requests();