- Friday, January 5, 2018

release-team@conference.openafs.org

Friday, January 5, 2018< ^ >

Room Configuration

Room Occupants

GMT+0
[01:56:01] kaduk@jabber.openafs.org/barnowl leaves the room
[13:51:52] kaduk@jabber.openafs.org/barnowl joins the room
[13:59:20] wiesand joins the room
[13:59:35] <kaduk@jabber.openafs.org/barnowl> Yay, I have company
[13:59:42] mvita joins the room
[13:59:49] <mvita> hi
[13:59:54] <wiesand> Hi
[14:00:14] <mvita> Mike may not be joining us today, he is out of the office
[14:00:25] <wiesand> Sorry, I'll probably be distracted a bit
[14:00:38] <wiesand> (Spectre/Meltdown…)
[14:00:41] <kaduk@jabber.openafs.org/barnowl> Mike sent mail, yeah.
[14:00:57] <kaduk@jabber.openafs.org/barnowl> And Spectre/Meltdown are just making everyone's month, yes.
[14:01:06] <mvita> I may be distracted less than usual, so perhaps that makes up.
[14:01:38] <wiesand> Anyway, I guess you have read about last week's conversation regarding 1.6.22.2
[14:01:40] <kaduk@jabber.openafs.org/barnowl> Stephan, want to start us off with 1.6?
[14:01:47] <wiesand> just did ;-)
[14:02:25] <wiesand> any thoughts on the plan to release 1.6.22.2 with highsierra + Lx 4.15 + EL7.4?
[14:02:33] <wiesand> + maybe the two rh spec changes
[14:02:44] <kaduk@jabber.openafs.org/barnowl> I did read about 1.6.22.2, yes.  Was getting getcwd/EL7.4 into 1.8
timely enough for you?
[14:03:07] <wiesand> I pulled the stack up today :)
[14:03:29] <mvita> Meffie is also working on a new patchset for the solaris deadlock issue
[14:03:40] <kaduk@jabber.openafs.org/barnowl> And I also forwarded you an email where jaltman claimed another patch
needed to be pulled up to 1.6
[14:04:09] <wiesand> Yes, I saw that one
[14:04:30] <wiesand> That's what I'd pull up next, at least for discussion
[14:04:43] <kaduk@jabber.openafs.org/barnowl> sounds good
[14:05:35] <wiesand> There's a bit of code skew there. I haven't made up my mind yet whether to pull up a preceding change too or backport the one in question.
[14:05:50] <wiesand> Maybe I'll do both and we can see what looks better.
[14:05:58] <kaduk@jabber.openafs.org/barnowl> Not wrong, if you have the cycles for it.
[14:07:32] <wiesand> NB 7784
[14:08:34] <wiesand> I think we should releasw 1.6.22.2 in time for Linux 4.15. Which is probably in ~ a fortnight
[14:09:31] <kaduk@jabber.openafs.org/barnowl> no objections here
[14:09:39] <mvita> sounds fine
[14:10:53] <wiesand> Ben, do you have a desire to have 12853 included?
[14:11:15] <kaduk@jabber.openafs.org/barnowl> I don't think it's relevant
[14:11:28] <wiesand> ok
[14:11:35] <kaduk@jabber.openafs.org/barnowl> I needed 12853 as a *build* fix, since the way we use these macros has
changed for 1.8
[14:11:50] <wiesand> Mark, do you have a reference for "solaris deadlock"?
[14:12:09] <kaduk@jabber.openafs.org/barnowl> OTOH I haven't tried building 1.6 myself on freebsd recently, but
there are automated builds for ports
[14:12:09] <mvita> https://gerrit.openafs.org/#/q/topic:solaris-deadlock
[14:12:11] <wiesand> Ah, fine then.
[14:13:46] <wiesand> looks like solaris-deadlock doesn't touch any common code
[14:14:53] <mvita> I believe that's correct, but why do you mention that?
[14:15:17] <kaduk@jabber.openafs.org/barnowl> It reduces the restrictions on when it's okay to merge [without
additional prereleases]
[14:15:25] <mvita> ah
[14:15:26] <wiesand> Because it significantly lowers my bar for inclusion ;-)
[14:16:08] <wiesand> And I think a "review review review" would conclude the 1.6 part today.
[14:16:22] <mvita> aye aye
[14:16:31] <kaduk@jabber.openafs.org/barnowl> It might help if the branch:openafs-stable-1_6_x link made it into
mail
[14:16:35] <kaduk@jabber.openafs.org/barnowl> (for me, at least)
[14:17:31] <mvita> ? I didn't follow that…
[14:17:52] <kaduk@jabber.openafs.org/barnowl> If I get an email reminder to do review, with a link to the things to
review, it increases the expected amount of reviews done.
[14:18:01] <mvita> AH.
[14:18:39] <wiesand> I'll send a summary, with links to stuff to review for 1.6.22.2
[14:18:42] <kaduk@jabber.openafs.org/barnowl> Anyway, since I think that's my cue to start in on 1.8, I started
staging 1.8.0pre4 (and just pushed a web change, not that you can
verify the checksums yet, since I haven't made a volume for the
sources)
[14:18:47] <kaduk@jabber.openafs.org/barnowl> Thanks!
[14:20:31] <kaduk@jabber.openafs.org/barnowl> I also made some debian packages from it and at least that client
works fine on my laptop.  I tried to set up a freebsd server, which is
how I ran into 12853 (the kernel module doesn't build)
[14:20:56] <kaduk@jabber.openafs.org/barnowl> But I can skip the kernel module build, since the freebsd client has
never been super-functional
[14:21:29] <wiesand> pity - my interest in freebsd is growing
[14:21:29] <kaduk@jabber.openafs.org/barnowl> There's also some lingering issue with LWP stack alignment on freebsd
that I don't fully understand, but having also worked around that, I
got a functional server setup.
[14:21:35] <mvita> Mine too.
[14:21:41] <mvita> (interest in FreeBSD)
[14:22:11] <kaduk@jabber.openafs.org/barnowl> There are some known issues that "should not" be "too hard" to fix,
but have not been the best use of my (at least) time.
[14:22:42] <kaduk@jabber.openafs.org/barnowl> Anyway, I guess my question is how people feel about a pre4 that
leaves the freebsd client build broken in this way (vs. re-rolling the
tarballs/etc.)
[14:23:00] <mvita> I'm getting a new laptop soon; it should have enough disk space for a few xBSD machines that I couldn't spare space for before.
[14:23:36] <kaduk@jabber.openafs.org/barnowl> Exciting!
[14:24:20] <wiesand> re pre4: redoing it for the fbsd client is probably not worth your time
[14:24:23] <wiesand> imho
[14:24:36] <kaduk@jabber.openafs.org/barnowl> That's my thinking, but I wanted to get a second opinion.
[14:24:49] <mvita> I agree it's okay for a pre
[14:25:13] <kaduk@jabber.openafs.org/barnowl> I think we also need to allocate some more sysnames for freebsd
releases, too, which I've been remiss about.
[14:25:48] <kaduk@jabber.openafs.org/barnowl> But likewise, that doesn't affect other OSes and has a lower bar for
when it can be merged.
[14:26:49] <kaduk@jabber.openafs.org/barnowl> It is perhaps premature to ask, but the current list of things we know
we want to put in 1.8 after pre4 includes:
the freebsd fix+sysnames
the configure/etc. refactor stuff
Am I missing anything?  Do we think that just those would require a
pre5?
[14:27:18] <kaduk@jabber.openafs.org/barnowl> I was a little bit considering calling pre4 "release candidate" in the
announcement email, instead of "beta".
[14:27:54] <kaduk@jabber.openafs.org/barnowl> Does that inspire a reaction from anyone?
[14:28:31] <mvita> I would prefer not to call it an rc if we know FreeBSD is broken
[14:29:03] <kaduk@jabber.openafs.org/barnowl> I had some thoughts along those lines, yeah.
[14:29:12] <kaduk@jabber.openafs.org/barnowl> I will be patient and not try to jump the gun, then :)
[14:29:15] <mvita> because you don't plan to release it that way.
[14:29:59] <mvita> To me, release candidate means I can make this GA unchanged if nothing else breaks.
[14:30:11] <kaduk@jabber.openafs.org/barnowl> *nods*
[14:30:59] <wiesand> "final beta" ?
[14:31:12] <kaduk@jabber.openafs.org/barnowl> I don't think I have other 1.8 topics; just trying to get pre4 out the
door
[14:31:21] <mvita> "gamma"
[14:31:33] <mvita> ;-)
[14:31:39] <kaduk@jabber.openafs.org/barnowl> final beta is nice, as tempting as gamma is
[14:32:14] <mvita> yes, final beta is definitely better
[14:33:14] <kaduk@jabber.openafs.org/barnowl> Shall we move on to master, then?
[14:33:22] <mvita> sure
[14:33:28] <wiesand> ok
[14:33:57] <kaduk@jabber.openafs.org/barnowl> Not too much news there, though.
[14:33:58] <wiesand> yay rxgk
[14:34:19] <kaduk@jabber.openafs.org/barnowl> I started applying Andrew's review comments for rxgk on the plane back
from christmas
[14:34:54] <kaduk@jabber.openafs.org/barnowl> But the current base point for that branch is on a master too old to
be compatible with the 4.14 kernel on my laptop.  So I may need to
rebase it in order to properly test things.
[14:35:47] <kaduk@jabber.openafs.org/barnowl> I see Mike sent in this vol/VGC stack, though I don't fully understand
the scope of it yet.
[14:36:53] <kaduk@jabber.openafs.org/barnowl> Oh, and back to rxgk: I think for at least most of the things Andrew
reviewed, I'm okay with taking his comments as input, doing a new
version, and then just doing a self-review before merging.  But if he
was planning to to re-reviews of updates, I'm happy to wait for them.
[14:37:40] <mvita> In general he would be glad to re-review - but I can't speak for his time or priorities.
[14:38:03] <mvita> I can ask him if you like…
[14:38:44] <kaduk@jabber.openafs.org/barnowl> I wouldn't worry about it too much.
[14:38:50] <mvita> Okay
[14:40:44] <kaduk@jabber.openafs.org/barnowl> is akimpersonate-on-master on our radar or still in "needs work"?
[14:41:08] <mvita> I think it's on Mike's radar.
[14:41:27] <kaduk@jabber.openafs.org/barnowl> Okay.
[14:41:44] <kaduk@jabber.openafs.org/barnowl> (I'm just looking over what's in gerrit for master to see if there are
other topics we should talk about)
[14:41:54] <kaduk@jabber.openafs.org/barnowl> Oh, I should probably merge the glibc 2.26 compatibility stuff
[14:42:37] <kaduk@jabber.openafs.org/barnowl> And Andrew has the "solaris-nonembed-vnode" topic from October that is
probably worth a look.
[14:43:56] <kaduk@jabber.openafs.org/barnowl> And that goes back to May already, so probably plenty for this
meeting.
[14:44:23] <mvita> Finishing review of the Solaris vnode stack is on my list of tasks for this month.
[14:44:51] <kaduk@jabber.openafs.org/barnowl> So I'll plan to go ahead and release 1.8.0pre4 today and try to get
some of this other stuff in.
Any other topics for today?
[14:45:36] <mvita> What, if anything, should we be doing for OpenAFS re: Meltdown/Spectre?
[14:45:42] <kaduk@jabber.openafs.org/barnowl> You could also rebase the ubik_perf topic, I suppose
[14:46:06] <mvita> maybe we should wait for questions on openafs-info
[14:46:14] <kaduk@jabber.openafs.org/barnowl> We don't currently ourselves have much of a privilege boundary, so I
think it will be limited to coping with what linux does
[14:46:37] <mvita> I think the main concern will be performance impact
[14:47:00] <kaduk@jabber.openafs.org/barnowl> Anders had tried out a kernel with those patches and got an oops from
kernel_read or kernel_write, IIRC.  But that's very preliminary and I
don't think anyone's actually looked yet.
[14:47:00] <mvita> "How much slower will this make OpenAFS"?, things along those lines
[14:47:57] <kaduk@jabber.openafs.org/barnowl> I don't think it should be a whole lot worse than any other
filesystem; we do have some downcalls from the kernel into the
userspace helper for things like DNS lookup, but I don't remember any
that are really in the hot path.
[14:48:06] <wiesand> Luckily, openafs is so bad at efficiently using modern hardware… I doubt there'll be noticeable impact…
[14:50:02] <wiesand> And it looks like we don't have to do anything to cope with what linux does - yet
[14:51:14] <kaduk@jabber.openafs.org/barnowl> (web change is 12861, btw)
[14:56:25] <wiesand> it lacks the symlink
[14:59:50] <wiesand> Anything else to discuss today?
[15:00:38] <mvita> nothing from me
[15:01:47] <wiesand> Let's adjourn then, back to the spectre/meltdown mines…
[15:01:49] <kaduk@jabber.openafs.org/barnowl> nor from me
[15:02:01] <wiesand> Thanks a lot!
[15:02:02] <kaduk@jabber.openafs.org/barnowl> thanks everyone!
[15:02:18] <mvita> Bye!
[15:02:49] wiesand leaves the room
[18:36:33] meffie joins the room
[22:44:45] meffie leaves the room