[lustre-discuss] Incorrect quota summary when file system has inactive (deactivated) OSTs

[Steve Crusan]

Hi all,

We are having a strange issue where it seems like the quota summary for a
particular group is wrong, which seems directly related to having inactive
OSTs (they were fully drained, then active=0 on the MGS, for both the osc
and osp services).

For instance:
"""
$ lfs quota -g groupname /h4
Disk quotas for grp groupname (gid XXXX):
     Filesystem  kbytes   quota   limit   grace   files   quota   limit
grace
            /fs4 [5109145]       0       0       -  537293       0       0
      -
Some errors happened when getting quota info. Some devices may be not
working or deactivated. The data in "[]" is inaccurate.
"""

If you run with the -v flag with lfs quota, all of the columns for each
active OST add up to the expected quota. We do get these extra messages
when running with -v:

"""
quotactl ost19 failed.
quotactl ost20 failed.
quotactl ost21 failed.
"""

The current client I'm testing with is 2.10.5 (also tested 2.10.1, 2.12.3),
servers are 2.10.1 (yes, I know old). But even upgrading a client to 2.12.3
doesn't resolve the issue. We've tried a variety of things, but the only
systems that haven't unmounted the file system in awhile. It gets more
bizarre that other client machines running the same exact software stack
(including Lustre version) show different results.

The machines that do show the correct output on the summary line about do
not have any OSTs in "IN" state (via lctl dl), not sure what else is
different. This means that "working" systems have an import state of FULL,
whereas non-working systems have a state of disconnected. For the
particular OSTs that we've removed from this file system, both working and
non-working quota clients show the osc.${ost}.active as '0'.

For now we are summing the verbose output columns, but does the issue jump
out to anyone?


-- 

*Steve Crusan*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20200121/f74274b3/attachment.html>