[LU-2358] LBUG reading /proc/fs/lustre/mgs/MGS/fstype

• Type: Story
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.4.0
• Affects Version/s: Lustre 2.4.0
• Labels: patch procfs
• Environment: CentOS 6.3
• Assignee: Emoly Liu
• Reporter: John Hammond

Description

An MGS has a NULL obd_fsops.

cat /proc/fs/lustre/mgs/MGS/fstype

LustreError: 4219:0:(lprocfs_status.c:604:lprocfs_rd_fstype()) ASSERTION( obd->obd_fsops != ((void *)0) ) failed: 
LustreError: 4219:0:(lprocfs_status.c:604:lprocfs_rd_fstype()) LBUG

Comments

[Peter Jones]

Landed for 2.4

[John Hammond]

An issue for the sanity check has been created; see LU-2376.

[Emoly Liu]

Should we add a sanity.sh test to check all the readable lustre proc entries to avoid such error?

[Mikhail Pershin]

I worry that we will have many links in lustre procfs tree making that difficult to understand where each entry belongs to. I am just not sure how e2fsck uses that, but if all data is placed already in OSD procfs directory, maybe we need to check it there? For example, proposed patch makes two things - link proper OSD proc entry to the particular MGS and then link fstype and mntdev right from there to the MGS too. I wonder just if we have already proper OSD link, can't we check those fields there?

[Andreas Dilger]

Mike, if they symlink to the proper OSD device, then it should be ok? One reason for these files is that the mnt dev is used by e2fsck to detect that the device is mounted by Lustre. It is also useful for the admin on occasions to see this.

[Emoly Liu]

Andreas, do you have any idea about these entries?

[Mikhail Pershin]

Sorry, I had wrong impression just after reading comments in patch. I see now that it will link properly. But I wonder still do we really need these entries?

[Mikhail Pershin]

I am afraid this will not work if MGS is set up separately from MDT and/or if it is based on osd-zfs. Can we just remove fstype and mntdev entries from MGS/ completely?

[Emoly Liu]

I will review this patch.