LWG Minutes 2015-06-17
Chris was on travel, so the notes are little spotty this week. No attendance nodes.
Client Upstreaming Effort Update
The upstreaming effort appears to be going pretty well. Some upstream patches have landed.
Ben Evans starting to get involved. James pointed him to work.
The upstream kernel is at the RC7 tag, so the next kernel release is not far away. The recent lustre work seems to have mollified the kernel powers that been. Andreas is hopefully that lustre has earned a reprieve.
James has a topic branch full of patches for the upstream kernel client that need to be backported into Lustre's master branch. This will be our first example of landing a topic branch in lustre. It is planned to be merged after the feature freeze.
Andreas explained that several way of dealing with topic branches were tested. It was decided that the one that works best is to make a new named branch in gerrit for each topic branch. That requires an administrative action, but it appears to be the most sane way to get gerrit to allow the branch to be reasonably merged into master and otherwise working in ways that we are familiar with.
Andreas moved the lustre coding style guide from HPDD wiki to the lustre.org Wiki.
The wiki markup is not the same between those two wikis, so some conversion was needed. Andreas found that OpenOffice has a mediawiki exporter plugin. He used Confluence's ability to save a page as a .doc file, opened that in OpenOffice, and then exported the file in mediawiki format. It wasn't a perfect process, but it saved him some time manually editing it on that complicated page.
For simpler pages, it would probably be quicker to just do it by hand.
Chris to unlock main wiki.lustre.org page for more editors. (Done.)
Feature freeze is soon. Any red flags?
Andreas - no feature yet landed completely, but some are close and should be done on time.
DNE - only 1 patch for feature completeness, and a few known bugs
shared key crypto not reads
Tree is 70% converted to tabs.
It was asked if Channel Bonding looks like it can land for the feature freeze. Chris did not think it was close enough to finished to land in time. James seemed to agree.
CLIO cleanup - three patches left to land. LU-5823, which has dependency on LU-6377
sysfs conversion work - probably not ready in time for 2.8. 2.9 timeframe probably more likely.
There was some discussion about whether sysfs and proc are going to coexist in the code base. They may just for users who are used to proc. But Lustre's user space tools need to move to sysfs as soon as possible, because the upstream kernel client will not have user space tools that currently will work with it until that move is made. lctl and lfs are the most important to get updated to sysfs.
LU-5319 (multiple slots per client in last_rcvd file)
Sebastien was concerned that LU-5319 would not make the feature freeze.
Andreas said that he was staying on top of the reviews, and was still hopeful that the work would be completed in time.