[LU-4522] ldlm_cli_enqueue and ll_inode_revalidate_fini LustreError messages on 2.4.1 clients

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.6.0
• Affects Version/s: Lustre 2.4.1
• Severity: 3
• Assignee: Andreas Dilger
• Reporter: Oz Rentas (Inactive)

Description

From time to time, KIT are seeing ldlm_cli_enqueue and ll_inode_revalidate_fini LustreError messages. They mainly appear on login nodes of our clusters.

Here is an example of the error messages:
Dec 17 10:24:28 ic2n988 kernel: [512992.139174] LustreError: 11865:0 (mdc_locks.c:840:mdc_enqueue()) ldlm_cli_enqueue: -13
Dec 17 10:24:28 ic2n988 kernel: [512992.139183] LustreError: 11865:0:(mdc_locks.c:840:mdc_enqueue()) Skipped 2 previous similar messages
Dec 17 10:24:28 ic2n988 kernel: [512992.139202] LustreError: 11865:0:(file.c:2716:ll_inode_revalidate_fini()) pfs2wor1: revalidate FID [0x1d080001:0x86b7421d:0x0] error: rc = -13
Dec 17 10:24:28 ic2n988 kernel: [512992.139208] LustreError: 11865:0:(file.c:2716:ll_inode_revalidate_fini()) Skipped 2 previous similar messages
Dec 17 10:24:29 ic2n988 kernel: [512993.347645] LustreError: 13000:0:(file.c:2716:ll_inode_revalidate_fini()) pfs2wor1: revalidate FID [0x1d080001:0x86b7421d:0x0] error: rc = -13

The Lustre client in this case is at version 2.4.1 plus patch for LU-3645. The servers are also at version 2.4.1.

What do the messages mean, and how can we get rid of these error messages?

Issue Links

• is related to: LU-4705 LustreError: 89827:0:(mdc_locks.c:916:mdc_enqueue()) ldlm_cli_enqueue: -2

Comments

[Andreas Dilger]

Patch 8828 was landed to master for 2.6.0.

[Oz Rentas (Inactive)]

We can go ahead and close this one. Thanks.

From the customer:
The update from Andreas Dilger at 24/Jan/14 5:30 AM provided what we expected from this case: Improving the error message or get rid of unnecessary LustreError messages. Therefore, you can close this case.

[Andreas Dilger]

In Lustre 1.8 we used to return -EIDRM (Identifier Removed) so that it was clear this was an issue with missing users in the MDS user database, rather than some other kind of permission problem.

We should also avoid printing these messages on the console, since they are just a distraction. I pushed http://review.whamcloud.com/8988 to quiet the client error messages for the common -EACCES error message.

[Oleg Drokin]

Could this be that login nodes have some users that are not known to MDS, similar to LU-4084 for example?