Branch:b_new_cmd

Disable target_stop_recovery_thread temporarily, because it
might cause dead-lock of target_recovery_thread. For example
1) in recovery, if clients are still not reconnected to the MDS.
2) this MDS failed again, try to stop_recovery_thread here.  obd_abort_recovery is set to 1.
3) MDS disconnect stale exports, obd_abort_recovery will be reset to 0. and check the next
   replay, but currently the MDS already disconnect the exports. and next replay req will
   never reached.  so the recovery-thread will hang there.

diff --git a/lustre/mdt/mdt_recovery.c b/lustre/mdt/mdt_recovery.c
index dfb7c12..4880fdd 100644 (file)
--- a/lustre/mdt/mdt_recovery.c
+++ b/lustre/mdt/mdt_recovery.c
@@ -977,7 +977,7 @@ void mdt_fs_cleanup(const struct lu_env *env, struct mdt_device *mdt)
         struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
 
         class_disconnect_exports(obd); /* cleans up client info too */
-        target_stop_recovery_thread(obd);
+        // target_stop_recovery_thread(obd);
         target_cleanup_recovery(obd);
         
         /* Remove transaction callback */