LWG Minutes 2015-04-22

From OpenSFS
Revision as of 16:06, 23 April 2015 by Morrone (talk | contribs) (Created page with "== Agenda == * Automated testing integration of CPPCheck * Build system improvements for 2.8 * Begin monitoring tools effort == Attendance == * Chris Morrone (LLNL) * Rich...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Agenda

  • Automated testing integration of CPPCheck
  • Build system improvements for 2.8
  • Begin monitoring tools effort

Attendance

  • Chris Morrone (LLNL)
  • Richard Henwood (Intel)
  • Paul Sathis (Intel)
  • Roman Grigoryev (Seagate)
  • Colin Faber (Seagate)
  • James Simmons (ORNL)
  • Cory Spitz (Cray)
  • Justin Miller (Cray)
  • John Hammond (Intel)
  • Alex Kulyavtsev (FNAL)
  • Brad Settlemyer (LANL)

Minutes

Automated testing integration of CPPCheck

LU-6372, created job for high level observation of master branch status, started gerrit integration.

For gerrit integration need to use jenkins. Need special permission to observe events in gerrit. Asked for permission from Intel about a month ago, but no progress yet.

Hammond - Still queued up with gerrit admin to take a look at. No idea when they will get to it.

It was pointed out that HPDDCheckPatch is in the contrib directory of the lustre source tree. That patch uses regular polling rather than gerrit events, and the CPPCheck scripting could perhaps borrow code from that temporarily.

In the mean time, the LWG will continue to apply pressure to Intel IT to evaluate the request.

Build system improvements for 2.8

Chris proposed that we take some of the work that Brian Behlendorf did recently for b2_5 to improve the RPM packing or lustre and apply it to master in time for 2.8. He also suggested that the Intel build farm be improved to build using the new method. Some benefits of Brian's recent work are:

  • kmod-style packages (pretty much standard in the RPM world for packaging linux kernel modules)
  • some better separation between kernel modules and user space tool packages
  • DKMS packages
  • User space tools can be built independantly and all kernel components disabled (necessary to build user space tools to use with DKMS packages)
  • Ability to build once for both client and server side. Subsets of built packages chosen for different node types. (Currently we build much of Lustre twice to accomodate client installs.)

Chris suggested that we make this part of the planned work for the 2.8.0 release, and make it part of the work under Contract SFS-MAINT-001.

James added that it would be nice to package lnet by itself. ORNL routers are a light weight OS environment, and they would rather not include all of the rest of lustre just to get lnet installed. Chris agreed that it was a good idea.

James generally agreed with the proposal.

There was no opposition to the proposal, but also no additional support.

It was requested that Chris send out an email detailing the proposal to give others a chance to consider it more fully.

Begin monitoring tools effort

At LUG, there seemed to be broad interest in collaborating on monitoring tools, since there are many different lustre monitoring efforts under way that seem to have a great deal of overlap in scope.

Chris attempted to get a project started to kick off that collaboration. He asked for volunteers to participate, and a volunteer to coordinate the project.

One person asked if this isn't an area where vendors want to specialize? Would this effort drive vendors away?

Cray expressed that they did not feel alienated, but would prefer to put their effort towards something else.

No one wanted to lead or participate in an OpenSFS monitoring tools effort.

It was decided to table the topic until such time as someone is willing to apply effort to the topic. All were encouraged to let Chris Morrone know when that happens so that the topic can be revived.

Other Discussion

Action Items

  • Chris to send email explaining proposed build system changes