hi there,
On May 10, 9:11am, marchuk@xxxxxxxxxxxxxxxxx wrote:
> Subject: Re: XFS 1.0/Quota
> I've recompiled kernel, this time making sure XFS quota is selected. When
> the machine starts up it does say "XFS doing quotacheck.. Please wait." I
Well, at least we know for sure quota are enabled now. ;-)
> wait 5 minutes or so and then the machine freezes (by freezing I mean no
> more hard drive activity and machine does not respond).
>
> I booted into a single user mode. did a mount -a, got the "doing
> quotacheck... Please wait." After awhile I get a funny error "Run out of
> memory.. Cancelling processs (sh)" The error is not in exact words but
> something like that.
>
Can you capture some output & send to me (offlist, it'll
probably amount to alot of data):
- cat /proc/meminfo
- "quot -ug" on the filesystem (mounted without quota
option(s) obviously);
- xqmstats, before & during the quotacheck if possible;
- cat /proc/slabinfo, also before & during if possible;
Also, are you enabling both user and group quota on this
filesystem? That would likely increase the dquot memory
pressure - still shouldn't fail, but could try just one
or the other & see if that helps (or one at a time might
work as a temporary workaround, to reduce the number of
incore dquots required during quotacheck).
> The machine has 256MB memory with 658MB disk cache. The xfs filesystem is
> a hardware raid with 8 disks totalling 350MB. Does anyone have experience
That doesn't sound right - did you mean 350GB? During a
quotacheck its shouldn't be the size of the filesystem
that matters, but the number of unique uids/gids in use,
as each of these must map to a unique incore dquot (hmm -
which has just tickled a distant memory & I may know what
the problem is... I need to think about it a bit though,
and I'll need to see your data from above to confirm).
> running xfs on large partitions?
>
This is largely based on the IRIX/XFS quota implementation,
so I'd imagine something has been lost in the translation to
Linux (my bad). IRIX quota works with enormous filesystems,
so large should be easy. :^)
thanks.
--
Nathan
|