MDS: damage reporting by ino number is useless
We had two damaged directories on the long-running cluster, but examining the directories in question (other than through raw RADOS) is pretty difficult.
When we generate a damage report a little more info is sent to clog, but we can't find any entries and think it was so long ago the files have been logrotated into nothingness.
In a reasonably healthy cluster one could also look at the raw RADOS objects, but assuming I've done the name translation correctly, we actually have lost these objects!
#1 Updated by John Spray about 1 year ago
The log message reporting the path is still there:
/var/log/ceph/ceph.log.3.gz:2017-01-10 04:29:24.415773 mds.0 172.21.5.114:6816/1992312860 363 : cluster [ERR] dir 100047040d6 object missing on disk; some files may be lost (/teuthology-archive/teuthology-2014-10-18_19:33:01-upgrade:dumpling-firefly-x:stress-split-giant-distro-basic-multi)
Not that it wouldn't still be reasonable to add an informational field to the damage entries, shall we just make it a path string or can we think of other advisory metadata that might be useful?