[LU-8909] urgently need RH7.3 client for Endeavour

• Status: Resolved
• Project: Lustre
• Type: Task
• Priority: Blocker
• Reporter: Michael Hebensteit (Inactive)
• Assignee: Dmitry Eremin (Inactive)
• Resolution: Fixed

Environment

RH7.3

Description

currently working with commit 5763c175c9a1a330b664a01bc08f329b2bee54f7
this works on BDW, but not on KNL (because of https://jira.hpdd.intel.com/browse/LU-8703 )

can you integrate the KNL fix into your RH7.3, and then give me a new commit?

Comments

[Peter Jones]

Michael

I'd like to understand more fully what it is that you need. So, you are using an interim master build, and you want to use it with RHEL 7.3 but there is something necessary for KNL that is missing from the default RH-supplied kernel? Could you please elaborate as to what that is?

Thanks

Peter

[Michael Hebensteit (Inactive)]

The Lustre client I compiled from the interim master build contains an error exposed only on KNL. You already have that error (and the fix) in LU-8703. I need an interim master build WITH the fix included

 

[Peter Jones]

Michael

So by "that error" you mean the unnecessary error messages that are generated? Having discussed with engineering, there is no functional change for this patch. And do you specifically want just that change on that specific commit or would landing the fix to the current tip of master (or based on the 2.9 release) meet your needs?

Peter

[Andreas Dilger]

Michael, there is a new build underway with all of the LU-8703 patches against master (essentially 2.9.0) that will be available at:

https://build.hpdd.intel.com/job/lustre-reviews/43020/

in a couple of hours, as our build system is currently backlogged.

[Michael Hebensteit (Inactive)]

could you please point me to the GIT entry?

[Andreas Dilger]

The last patch in the series is https://review.whamcloud.com/24304 or https://git.hpdd.intel.com/?p=fs%2Flustre-release.git;a=commit;h=5a928f213bfb24a6c7e5ae353bc7872528e32a1d though it should be noted that last patch in the series has not yet been reviewed.

[Dmitry Eremin (Inactive)]

The issue is resolved by specific configuration in lustre.conf

[Andreas Dilger]

Dmitry, is Michael aware of the specific lustre.conf configuration?  Otherwise, please pass that information on in this ticket.

[Dmitry Eremin (Inactive)]

From: Hebenstreit, Michael
Sent: Tuesday, December 13, 2016 1:52 AM
To: Eremin, Dmitry <[e-mail]>
Subject: RE: problems with KNL and Lustre

Thanks, that seems to solve my problem for now

You can close 8909

From: Eremin, Dmitry
Sent: Friday, December 9, 2016 1:52 AM
To: Hebenstreit, Michael <[e-mail]>
Subject: RE: problems with KNL and Lustre

Hi Michael,

I’m reworking CPU partition code now. This work is not complete yet. It’s not implemented in optimal way for KNL so far. Therefore I would recommend you to use the following settings in lustre.conf for KNL.

options libcfs cpu_pattern="" cpu_npartitions=34

options ptlrpc ptlrpcd_partner_group_size=4

Dmitry.

[Michael Hebensteit (Inactive)]

yes, that works

thanks for help