LWG Minutes 2015-10-21

From OpenSFS
Jump to: navigation, search

Attendance:

Argonne: Kevin Harms
Cray: Justin Miller, Ben Evans, Chris Horn, Cory Spitz
Fermilab: Alex Kulyavtsev
Indiana: Steve Simms
Intel: Paul Sathis, Andreas Dilger, Richard Henwood, John Hammond
ORNL: James Simmons, Sarp Oral
Sandia: Ruth Klundt
Seagate: Kalpak Shah

Apologies if I [Cory] missed anyone


Lustre 2.8 update:

Peter Jones is traveling to Japan LUG, unable to provide his usual 2.8 update.

Andreas notes that there are 7 blockers remaining at this time. Most have patches in-progress. 11 critical issues remain. Won’t hold the release for major/minor. No release ETA at this time, but it is likely several weeks away. Peter hopes to get it done before SC’15 though. Most of the bugs are related to DNE2, but there are other nasty bugs, not necessarily regressions.

James @ORNL reported on some large scale test. Ran into LU-6271 (should be fixed, client) and one other one that should land soon. MDS oops being investigated under LU-7173. Not running with multiple MDTs. 2.8 client to 2.5 test server (for now).


In-tree client development:

James reported lots of cleanup. It’s been quite active. There has been more server code than he expected and that’s being pruned. No date yet for getting out of staging. Goal is to have libcfs and LNET in sync with master by next LUG.


Scope of the leadership role:

Peter previously proposed a reasonable list of roles/responsibilities to the LWG list. It would be nice to get this down on the wiki. That is left as an action for the next leader to more formally define and document.


Selecting a new leader:

Cory shared leadership rules to the list last week. Only individuals from Promotor or Adopter level organizations can serve. The board will decide. They would like some recommendation from the LWG. Steve S. addressed what the board needs. They would like a recommendation, but no time frame is set.

LWG would like to take a vote to form a recommendation. Cory agreed to send mail to the LWG and collect email votes. We’ll try to wrap up voting by 10/30.

Three nominees at this time: Peter Jones, Sarp Oral, Kalpak Shah. Nominees should take an opportunity to mail the list with their “stump speech” prior to 10/30.

Kalpak has officially withdrawn his name. He was nominated by Peter B. because there was some concern about a conflict of interest with OpenSFS contracts.

Sarp is the contract administrator for OpenSFS, but that doesn’t disqualify either him or Peter. Sarp might have to step down from that role anyway as ORNL is stepping down as an OpenSFS Promotor. Steve will check with the OpenSFS lawyer.

Sarp would like to speak to Peter J. first as he is thinking that he would also withdraw his nomination. Sarp expressed his concern about 2020-and-beyond for extreme computing. He wants to make sure that Lustre is on-track for that and that we have a release at least once a year. Sarp is open to helping as a co-lead. Sarp will follow up with the LWG after his talk with Peter.

Cory will follow up to the list with instructions if a vote is necessary. We’ll still aim to make a recommendation to the board on 10/31.


Update on feature branches:

Andreas reported that developers can’t create a branch on their own. A Gerrit sysadmin needs to do it. We should reserve this for fairly significant features. The LWG should help establish the qualifications needed to create a feature branch. Andreas agreed to take documenting the process as an action. He has been moving some of the developer notes from the HPDD wiki to the lustre.org wiki [thank you!] and he’ll easily be able to add this work. It was agreed that the LWG should play some role in publishing what’s being done in feature branches to avoid stepping on one another’s toes. Developers should communicate development on lustre-devel and on the Projects wiki page at http://wiki.lustre.org/Projects. Test plan execution prior to master integration? What about the gatekeepers role? Whose responsibility is it to run the test plan? What about large scale testing? Note: Hyperion is supposed to be retired next year. Cory took an action to put a straw man for a more formalized feature branch strategy on the wiki. Richard was concerned about additional steps for developers. The intention isn’t to create extra work but to ensure integration is ready. Cray and others doing testing may be asked to step up to help.


Face to face LWG meeting at SC?

No, not enough critical mass. Room/phone logistics would be a concern. Cory will send out an invitation for the next call scheduled for 11/4.