[LU-13438] Rhel8.1 / lustre-client 2.12.4-1

• Type: Question/Request
• Resolution: Unresolved
• Priority: Major
• Affects Version/s: Lustre 2.10.4
• Environment: Rhel 8.1 clients, 7.7 servers

Epic/Theme

client

• Assignee: Peter Jones
• Reporter: Jon Symon (Inactive)

Description

I took your advice from LU-13382 and went back to a 2.12.4-1 client with IB support, released 11th of feb this year.  What we are finding now is that the (rhel8.1) compute nodes randomly reboot frequently, leaving us with a rather unstable cluster and some rather unhappy users.

We have done some testing and found that the issue appears to be caused by the change from radix_tree_exceptional_entry to xa_is_value as described in LU-13136.  We were able to patch the 2.12.4-1 client source and create a patched client that appears to solve the problem.

However, we did have some issues getting this to compile and had to make some semi intelligent guesses to get it to work.  That leaves us less than confident that we haven't introduced other bugs that will be appear if we were to use our patched client version in production.

Do you have a version of the 2.12.4 client with that patch applied and IB support that you can release to us?  Or a version even a version 2.12.5 client with IB support? Something that would give us a little more confidence than our current patched version.

Thanks

jon

Issue Links

• is related to: LU-13136 (layout.c:2121:__req_capsule_get()) @@@ Wrong buffer for field 'niobuf_inline' (7 of 7) in format 'LDLM_INTENT_OPEN', 0 vs. 0 (server)
• is related to: LU-13382 Wrong buffer for field 'niobuf_inline' (7 of 7) in format 'LDLM_INTENT_OPEN'