[LU-9952] soft lockup in osd_inode_iteration() for lustre 2.8.1

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.8.0
• Type: Bug
• Priority: Major
• Reporter: James A Simmons
• Assignee: nasf (Inactive)
• Resolution: Fixed

Environment

Lustre ldiskfs server back end running version 2.8.1 with a few additional patches. The OS is RHEL6.9

• Severity: 2

Description

One of production file systems running lustre 2.8.1 experienced a soft lock up very similar to LU-9488. I attempted to back port the patch but way to many changes have happened between 2.8.1 and lustre 2.10.0. Unsure if I would get the port right. I have attached the back trace.

Issue Links

• is related to: LU-9040 Soft lockup on CPU during lfsck
• is related to: LU-9488 soft lockup in osd_inode_iteration()

Comments

[Peter Jones]

Fan Yong

Can you please advise on this one?

Thanks

Peter

[nasf (Inactive)]

The known patches on master that are related with the OI scrub soft lockup are back ported as following:

https://review.whamcloud.com/28903
https://review.whamcloud.com/28904
https://review.whamcloud.com/28905
https://review.whamcloud.com/28906

[James A Simmons]

We are in the process of testing these patches. I attempted to recreate the problem with "lctl set_param fail_loc=0x1504" but that didn't work. What would you recommend to recreate this problem on a 2.8 system? Note we removed the offending files to make our production file system usable again.

[nasf (Inactive)]

I think that we need some new fail_loc to simulate osd_inode_iteration() trouble. For example, inject the new failure stub in the osd_iit_next() to simulate kinds of bitmap layout cases.

[James A Simmons]

Could you create a test condition before the 30th of September?

[nasf (Inactive)]

Let check whether this one https://review.whamcloud.com/#/c/29133/ works or not.

[James A Simmons]

We wouldn't be running the test framework on our production system. It looks like I just need to create a bunch of files on the file system.

lctl set_param -n osd*.MDT.force_sync=1
lctl set_param fail_val=1 fail_loc=0x190
lctl lfsck_start -M lustre-MDT0000
lctl set_param fail_val=0 fail_loc=0x198

While you check status:
lctl get_param -n osd-ldiskfs.lustre-MDT000.oi_scrub | grep status

Does this look right? What values do I use to reset it back to normal working conditions.

[James A Simmons]

We have been trying your reproducer by itself and we see it get stuck but no soft lock ups. Why is no soft lockups being reported?

[nasf (Inactive)]

"fail_loc=0x190" will slow down the OI scrub scanning, then we can have time to inject other failures before the OI scrub complete.
"fail_loc=0x198" will make the OI scrub iteration repeatedly scan the same bits for inode table. If without our former patches (28903/4/5/6), then the OI scrub will fall into soft lockup. But because we have such patches, then OI scrub can detect such dead repeat then move forward. So no soft lockup is the expected behavior.

[James A Simmons]

We have been trying patch 28903 by itself and have not seen the soft lockup BTW do we need to inject another failure?

[nasf (Inactive)]

We have been trying patch 28903 by itself and have not seen the soft lockup BTW do we need to inject another failure?

You mean you can reproduce the soft lockup every time if without any of above patch, right? And the soft lockup will be disappear if only 28903 applied, right?

[nasf (Inactive)]

Any further feedback?

Thanks!

[nasf (Inactive)]

simmonsja,
Any further feedback for this ticket?

[James A Simmons]

You can close this. With various patches applied we haven't seen problems in some time. Thanks.

[nasf (Inactive)]

The issue has bee resolved via back porting the following patches:
https://review.whamcloud.com/28903
https://review.whamcloud.com/28904
https://review.whamcloud.com/28905
https://review.whamcloud.com/28906