LWG Minutes 2015-07-01

Agenda

 * Lustre 2.8 development status
 * Client in upstream kernel status
 * Discuss OpenSFS's role in Lustre development

Attendance

 * Chris Morrone (LLNL)
 * Sebastien Buisson (Atos)
 * James Simmons (ORNL)
 * John Suykerbuyk (Seagate)
 * Patrick Farrell (Cray)
 * Ben Evans (Cray)
 * Justin Miller(Cray)
 * Chris Hanna (IU)
 * Brad Settlemyer (Sandia)
 * Cory Spitz (Cray)
 * Vitaly Fertman (Seagate)

Upstream kernel status
Survived through the Lustre release deadline. Upstream has seen enough effort going into the client to allow work to continue to get Lustre out of Linux's staging area.

It is important to keep the momentum going. We can't afford to fall behind again.

Goals: Probably next week start working on topic branch to merge upstream fixes back into Lustre master. That will allow libcfs and lnet changes to sync up. Client changes still pretty different.

Focus on user-space tools soon. Patch already started to support sysfs.

Will memory management changes make it back into master? Yes, they are working on that.

Lustre 2.8 Dev
One last patch for DNE2, didn't land before feature freeze. No Intel folks available to comment. Chris to look into what will happen with remaining DNE2 work.

Question about landing about ptlrpc smp scaling patch LU-6325.

SLES 12, ldiskfs support...

LU-5319 - landed just under the wire. Some sort of MDT-MDT work still outstanding, but not essential to the base feature. Support for multiple MD rpcs between MDTs.

Is DNE2 really done? What about all of the work on LU-3534? Also patchs from LU-3540?

Vitaly discussed issues with LU-5560. Andrew Perepechko made some comments in gerrit, and Vitaly thinks they were ignored. Thinks patch needs reverting. Sebastien explained that it doesn't add any new security issues. Just does not address all SELinux issues. Does not agree that the patch needs reverting. Chris looked at the gerrit page in real time, and noted that Andrew failed to give the patch a negative review. Chris explained that the comments were not ignored, but that a lack of negative review is a signal that the reviewer is not objecting to landing the patch. So now what do we do after the fact? Chris recommended opening a new Jira ticket explaining the issues with the patch, and making the case for a revert.

Most kerberos revival patches have landed. Kerberos is back and useable again. Some bug fixes need landing, but should be possible before code freeze.

Patrick, lock-ahead. Was code complete three weeks ago, but no code reviews. Chris to check why reviews didn't happen.

LU-6433 needs review. Chris to check.

Discuss OpenSFS's role in Lustre development
Discussion ensued, but time ran short. We plan to continue the conversation at the next meeting.