LWG Minutes 2019-01-10

Attendance
BP: Shawn Hall

Cray: Cory Spitz

ORNL: Dustin Leverman, James Simmons

HPE: Olaf Weber, Christopher Voltz

Sandia: Ruth Klundt

Supermicro: Abe Asraoui

Whamcloud: Joe Gmitter, Peter Jones, Andreas Dilger, James Nunez

Actions
New Actions Captured:

Existing Open Actions:
 * None

Actions Recently Closed:
 * None


 * None

Minutes
2.13.0 Release Update

Peter


 * Discussion about whether RHEL 8.0 and/or ZFS 0.8 would be available during this release cycle. We will aim to keep 2.13 available so it could be used with 0.8 but likely keep 0.7.x as the default shipping version for this release. We'll see how the timing of the RHEL 8.0 release ties in with the 2.12 cycle but we'll aim to be ready by getting vanilla 4.18 kernel support ready.

Upstream Lustre Client

James


 * We are not far off from being up to a 2.11 level. The next thing to take on is DoM to get to 2.11.
 * Peter: Who would have to make the call to get it released?
 * James: Neil is going to approach Al Viro and Linus to discuss.  The next step would be for it to get directly into Linus' tree.
 * Peter: Is there any time frame being discussed?
 * James: The goal is to have it in Linus' tree by LUG this year.
 * Andreas: It may be worthwhile to consider stopping at the latest 2.10 LTS release and do all the final cleanups.
 * James is already past that point with other 2.11 content.
 * Cory: Can we shoot to have it all the way up to the 2.12 LTS for the LUG milestone?
 * Andreas: There is only a few week period every 3-4 months.  That’s puts a lot of pressure to get in done.
 * Peter: Suggests to wait until Neil has that initial conversation and go from there.  This will give us a better idea of how far off we are and what the window will look like.

lustre.org

Peter


 * Ken was not able to make the call today. He did drop a note to say that there was no substantive progress since last call.
 * Andreas: If you find issues with the wiki site, feel free to take directly to ken or bring up on lustre-discuss.  Ken has been doing a great job with fixing any website issues as they are found.

Other Business


 * Shawn: We enabled PFL on a new filesystem and things went well.  We are in the process of enabling PFL on the rest of the filesystems.  When that is complete we will be a full PFL site across all filesystems.  The only thing we have run into is that we discovered that there is a 143 limit on number of object indexes in the metadata.
 * Andreas: This was just being discussed elsewhere as well.  The limit is 143.  If you enable the xaddr inode feature on the MD, that will let you have the larger ldiskfs xaddr and you can then have basically as much as your heart desires.
 * Andreas: This would be useful to document in the manual so that everyone is aware.  Also, we should also enable it by default in 2.13 development window.
 * Cory: We are also recommending it as part of our best practices going forward for our customers.

Next meeting will be on 2019-01-24 at 11:00am Pacific