[LU-5555] Inode usage quota are corrupted when running e2fsck

• Type: Bug
• Resolution: Fixed
• Priority: Major
• Affects Version/s: Lustre 2.4.3
• Environment: RHEL6 w/ patched kernel for Lustre; Lustre 2.4.3 + bull patches; e2fsprogs-1.42.9.wc1; ; Details about the FS:; - 1 MDT; - 32 OSSs; - 480 OSTs; - 150 000 000+ inodes; - 3400+ users (uid/gid range: 2312-65534)

Epic/Theme

Quota

• Severity: 3
• Assignee: Niu Yawei (Inactive)
• Reporter: Bruno Travouillon (Inactive)

Description

We have an issue with the inode usage quota on a Lustre 2.4 filesystem.

After a run of e2fsk on the MDT, the inode usage quota and block usage quota are 0 for each user on the MDT. We observe that new files are accounted. After a new run of e2fsck, all usage quota are reset to 0.

As a workaround, we have to disable/enable the quota feature on the device after each e2fsck run. In this case, the quota usage is correct.

It seems that e2fsck corrupts the quota inode.

We did not observe this issue on OSTs, but no correction has been made by e2fsck on these targets.

I believe there is a fix for this issue in e2fsprogs 1.42.10, merged in the master-lustre branch of git://git.whamcloud.com/tools/e2fsprogs.git (see commit ac7ea7b081c119a843ed6105d37461d84baaf544 )

Can you confirm?