LWG Minutes 2017-08-24

Attendance
Cray: Ben Evans, Cory Spitz

ORNL: James Simmons

SuperMicro: Abe Asraoui

Sandia: Ruth Klundt

HPE: Olaf Weber, Christopher Voltz, Stephen Champion, Robert Clark

IU: Ken Rawlings

Intel: Joe Gmitter, Peter Jones, Andreas Dilger

Actions
New Actions Captured:


 * None

Existing Open Actions:


 * None

Actions Recently Closed:


 * None

Minutes
2.10.1 Release Update

Peter


 * We are pretty close to having a release candidate. We are looking to get in the port for RHEL 7.4 along with some other minor bug fixes.
 * Has anyone been testing with it?
 * Stephen: We have seen one performance issue, but are still investigating it.  However this is for 2.10 in general and not 2.10.1 specific.
 * Will 2.10.1 contain the multi-rail dynamic discovery code just landed?
 * Peter: No, this won't be in 2.10.1 as that was never the plan.  It will be evaluated for a future release after we have had scope more time to see how it settles down on master testing.  2.10.X is a LTS release and from that perspective we have a judgement call to make if the effects of the change are worth the usability enhancements to multi-rail.
 * James: Working on two fixes, one SNMP related and one for LNet set_param.  Both of these will fix 2.10.X issues that prevent mounting Lustre on ARM systems.  These will be good to have in next 2.10.X.

2.11.0 Release Update

Peter


 * The multi-rail dynamic discovery enhancement has landed to master.
 * Is there anything else to report out on?
 * James: The tracepoint patch needs review (LU-8980).
 * Peter: It appears that Olaf and DDN have reviewed it and it can be considered for landing by the gatekeeper.

Upstream Lustre Client

James


 * Ported all the uapi changes to master upstream. We have normalized all paths.  The first batch had one bug that showed up on one platform, so it was reverted, fixed, and resubmitted.  This is a big milestone for upstream client.  Sent up tracepoint patch as well, but have not seen any feedback yet.
 * Upstream is looking pretty good. This is probably the last big push to sync up.  Looking good for 2.11 and upstream.  Last miles of the marathon.
 * Peter: How long is window open?
 * The 4.14 window will be opening soon and all these pushes should be in there.

lustre.org

Ken


 * We just held a call to work up specifics on shutting down the old VM.
 * It is Ken's understanding that, on the old wiki, the old architecture docs were generated while lustre was still at Sun? Does anyone know if this is correct?
 * Andreas: Yes, they are all from Sun timeframe.  Andreas has been trying to move new design docs into their own wiki pages for things like PFL and FLR.
 * Brief discussion of llite internals.

Other Business


 * Peter: Master still has RHEL6 patches and builds, but this has not been supported since 2.8.0.  We are looking adding back client support for the 2.10 branch, however, in terms of 2.11, 2.12 and forward, we think it would be unlikely to want to have latest Lustre and old RHEL distro.  If we leave it, the perception is that it is on an equal footing with what is supported.  We are thinking to stop it for newer releases and leave it there for 2.10.X branch.  The thought is that if someone is using production clients with bleeding edge Lustre, it is very unlikely they would not be on an old RHEL6 release?  If a production environment is using the 2.10.X LTS release, it would still be there.
 * Christopher: We have a customer wanting to use latest LTS, but they have older equipment and are stuck on older RHEL6 release.
 * Peter: This would be fine as it would remain in 2.10.X LTS releases.

Next meeting will be on 2017-09-07 at 11:00am Pacific