[Lustre-discuss] lustre 2.5.58 MDT 100% full
Dilger, Andreas
andreas.dilger at intel.com
Tue Jun 10 19:21:01 PDT 2014
The underlying MDT filesystem is actually ZFS and not ldiskfs, but it should be equally possible to expand the MDT storage capacity by adding another mirrored VDEV to the MDT pool.
It is also important to ask if there are any snapshots of the pool? That would also hold space in the pool even if the files are deleted.
That said, there wasn't enough information to determine if there is actually a problem with the filesystem or if it is just full. Just because the MDT did not fill up previously, is no indication that the MDT will not fill up if the average file size is smaller this time around.
What does "lfs df -i" and "lfs df" report for filesystem? That would allow figuring out how much space is used on the MDT and how many files in the filesystem, and the average space used per file on the MDT. It should be in the 2-3KB or less range.
Cheers, Andreas
On Jun 10, 2014, at 6:14, "E.S. Rosenberg" <esr+lustre at mail.hebrew.edu<mailto:esr+lustre at mail.hebrew.edu>> wrote:
Did anyone answer this?
>From my way too basic understanding should it be possible to expand an MDT? after all the underlying fs is ext4?
On Mon, Jun 2, 2014 at 5:12 PM, Anjana Kar <kar at psc.edu<mailto:kar at psc.edu>> wrote:
The MGS/MDT pool and filesystem were created using these commands:
zpool create -f -o ashift=12 -O canmount=off lustre-mgs-mdt
mirror /dev/disk/by-path/pci-0000:00:1f.2-scsi-0:0:0:0 /dev/disk/by-path/pci-0000:
00:1f.2-scsi-1:0:0:0
mkfs.lustre --mgs --mdt --fsname=iconfs --backfstype=zfs --dev
ice-size 131072 --index=0 lustre-mgs-mdt/mgsmdt0
However, now the MDT is full, and 8 OSTs are 86% full.
lustre-mgs-mdt/mgsmdt0
73G 73G 0 100% /mnt/lustre/local/mgsmdt
lustre-ost0/ost0 13T 11T 1.8T 86% /mnt/lustre/local/ost0
Is there a way to clear space, or will the filesystem/MDT need rebuilt?
Previously, the same disks were used for creating a mirrored MDT w/o
-ashift=12, and did not fill up even when the OSTs were at 89%.
Thanks for any advice.
-Anjana Kar
Pittsburgh Supercomputing Center
kar at psc.edu<mailto:kar at psc.edu>
_______________________________________________
Lustre-discuss mailing list
Lustre-discuss at lists.lustre.org<mailto:Lustre-discuss at lists.lustre.org>
http://lists.lustre.org/mailman/listinfo/lustre-discuss
_______________________________________________
Lustre-discuss mailing list
Lustre-discuss at lists.lustre.org<mailto:Lustre-discuss at lists.lustre.org>
http://lists.lustre.org/mailman/listinfo/lustre-discuss
More information about the lustre-discuss
mailing list