[LU-15724] MDT failover hang

Related — Wed, 12 Apr 2023 06:45:50 +0000

With a ~~LU-8367~~ deadlock between osp_precreate_reserve() and

osp_precreate_cleanup_orphans(), I've found a problem with MDT failover.

00000020:02000400:31.0:1644539398.776433:0:454249:0:(obd_config.c:854:class_cleanup()) Failing over kjcf05-MDT0001
...
00010000:02020000:20.0:1644539461.204784:0:454249:0:(ldlm_resource.c:1188:__ldlm_namespace_free()) 0-0: Forced cleanup waiting for mdt-kjcf05-MDT0001_UUID namespace with 46 resources in use, (rc=-110)
00010000:02020000:8.0:1644539699.332763:0:454249:0:(ldlm_resource.c:1188:__ldlm_namespace_free()) 0-0: Forced cleanup waiting for mdt-kjcf05-MDT0001_UUID namespace with 46 resources in use, (rc=-110)

So the situation is - MDT failover does not produce disconnect event, so osp_precreate_cleanup_orphans() cannot be awakened. Also it does not cleanup opd_pre_recovering and osp_precreate_reserve() wait skips wakeup signal. This hang would be ended after ~obd_timeout.

Whamcloud Community JIRA

[LU-15724] MDT failover hang