On 08/23/16 19:25, Jessica Otey wrote:
Our file system is mounted with the "user_xattr" option. Although it
took FOREVER to do this readlog (a few times as long as the initial
scan did), it did complete successfully. Can I assume, then, that
these errors are not some sort of misconfiguration but rather they
reflect changes in the filesystem?
This error suggest there are existing entries without "trusted.link"
attribute.
I think this may occur for still existing entries that have no longer a
path in Lustre, in particular if the file has been opened and deleted -
but is still opened. An advice from Lustre expert could help to confirm
that.
Perhaps it took so long because it was run so long after a file system
scan?
Yes of course. The longer you wait to dequeue changelogs, the more
history of changes robinhood has to process to resync its DB.
What I'm trying to understand better is the relationship between
scanning a file system and using the changelogs feature.
1) If changelogs is in fact working, does that mean that I *never*
need to do a file system scan again? Or is it still a good idea to do
occasional, full-system scans? If so, at what interval?
There may be some changes that are not reported by changelogs, in
particular in case of client or server crash or brutal shutdown.
So it is a good idea to rescan occasionally (with a interval of several
months), depending if you had many of client or servers crashes.
2) With changelogs in the picture, is there a way to indicate how
up-to-date the information issued by robinhood reports is? Does one
just assume it is real-time at that point, and thus the date of last
scan (as reported by rbh-report -a) becomes irrelevant as an indicator
of data freshness?
Robinhood regularly dump changelogs stats in its log.
The "last read record time" indicates the timestamp of the last
changelog robinhood processed.
If it is too far from the log line time, robinhood is late.
2016/08/25 *11:29:04* [945/2] STATS | last read record time =
2016/08/25 *11:28:31*.600763
Also, you can see the number of queued changelog records that robinhood
has to process on the MDS
(difference between the current index and the cl1 index).
# cat /proc/fs/lustre/mdd/fsname-MDT0000/changelog_users
current index: *1425**7232*
ID index
cl1 **1425*7022*
HTH
Thomas
Thanks,
Jessica
To troubleshoot this you can:
- check this entry path:
lfs fid2path fsname 0x93ca92:0x709d8371:0x0
- try to get this attribute from command line:
getfattr -n trusted.link <path_to_entry>
- verify that the filesystem is mounted with the "user_xattr" option.
Regards,
Thomas
------------------------------------------------------------------------------
_______________________________________________
robinhood-support mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/robinhood-support
