[LU-338] Lustre client reports wrong file size (zero, or too small) on first stat

• Type: Bug
• Resolution: Duplicate
• Priority: Minor
• Affects Version/s: Lustre 1.8.6
• Environment: 1.8.5.0-3chaos, RHEL5.5ish (CHAOS4.4-2)
• Severity: 3
• Assignee: Niu Yawei (Inactive)
• Reporter: Christopher Morrone (Inactive)

Description

I'm getting reports that lustre is reporting the wrong file size on the first stat (that they know of) after file creation. This seems to be relatively rare, and difficult to reproduce, but is happening at least a few times a week site wide that we know about.

The reports are a bit fuzzy so far, and I am trying to get more useful info out of folks. Often on the first stat, a file's size will be reported as zero. They claim that occasionally it will be a "power of 2", but smaller than the actual file size. I didn't get a more specific number than that yet.

The claim is that the file creation and stats happened on the same client node.

(Note that our filesystem default stripe count is 2.)

I wish that I had better info than that. I will continue to try to get more information. But I seem to remember that Oak Ridge may have seen the same issue, so I wanted to get a tracker open.

Comments

[Niu Yawei (Inactive)]

I'm afraid that if the problem is caused by another rare race in size glimpse (LU-287)? If possible, could you try to apply the patch in LU-287 to see if it can resolve your problem completely?

[Prakash Surya (Inactive)]

We have received a report of this issue happening again since applying
the LU-274 patch. Although, that patch seems to have made this issue
occur far less frequently, it appears as though it isn't completely
gone. Any suggestions as to where to go from here?

[Peter Jones]

Duplicate of LU274

[Christopher Morrone (Inactive)]

Ah, yes, that does sound like the likely culprit, thanks! You can mark this a duplicate of LU-274.

[Niu Yawei (Inactive)]

It probably be a duplicated of LU-274, but we need more information to confirm it, I think can ask customer to try out the patch in LU-274.

[Peter Jones]

Niu

Could you please confirm whether this is a duplicate of LU274?

Thanks

Peter

[Andreas Dilger]

I suspect from the description that this is a duplicate of LU-274. That issue already has a patch to fix the problem.