- Wednesday, October 22, 2014

release-team@conference.openafs.org

Wednesday, October 22, 2014< ^ >

Room Configuration

Room Occupants

GMT+0
[05:04:23] kaduk leaves the room
[05:15:39] Jeffrey Altman leaves the room
[05:21:58] Jeffrey Altman joins the room
[05:22:23] Jeffrey Altman leaves the room
[05:32:51] Jeffrey Altman joins the room
[06:31:41] Jeffrey Altman leaves the room
[06:41:36] Jeffrey Altman joins the room
[11:47:55] wiesand joins the room
[13:23:54] kaduk joins the room
[14:00:02] <kaduk> $timeofday
[14:02:27] <wiesand> Hello
[14:03:24] <wiesand> I prepared the 1.6.10 release.
[14:03:33] <kaduk> Thank you!
[14:03:39] <wiesand> Uploaded all available binaries.
[14:04:02] <wiesand> The web change was updated. Would you care to look at 11515?
[14:06:03] <wiesand> Daria, Jeffrey, are you actually here?
[14:06:52] <kaduk> Jeffrey is IRC-ing...
[14:07:59] <wiesand> Anyway, it looks like it's just the two of us today.
[14:08:16] <Jeffrey Altman> I am here for about 30 minutes
[14:08:19] deason joins the room
[14:08:56] <wiesand> We should probably make use of those 30 minutes to discuss the status w.r.t. 1.8 .
[14:09:32] <kaduk> I might say that the suspected regression would be linux-only.
[14:10:02] <kaduk> These web changes are not super-easy to review in gerrit, I think.
[14:10:07] <wiesand> You're right. Working on PS3...
[14:10:18] <Jeffrey Altman> For those that are not aware, Yosemite released last Thursday and Yosemite has no DES support at all.   A large number of orgs did not upgrade their Kerberos realms or update various user or realm principals to be non-DES.  There is a large support burden on support providers as a result.   Not much time at the moment for OpenAFS.
[14:10:35] <kaduk> So much for the writing on the wall...
[14:11:02] <kaduk> Anyway, I did push some updates last night/this morning for the tmay zeroing thing.
[14:12:42] <Jeffrey Altman> thanks for that.   I have not had time yet to look
[14:12:51] <kaduk> I am thinking about what should be the top priority for documentation updates for 1.8.
[14:12:55] <wiesand> New version of 11515 pushed.
[14:13:17] <deason> that regression is linux-only, and is only a regression from 1.6.9, and is fixing another regression
[14:14:08] meffie joins the room
[14:14:10] <wiesand> That's reworded in PS3. Still not ok?
[14:14:53] <deason> that's probably fine, but it's up to you
[14:14:56] <kaduk> I would be okay with PS3.
[14:15:15] <kaduk> The only thing Andrew said that's not particularly covered is that it's a regression from 1.6.9 in particular.
[14:15:47] <deason> I don't really know if I'd <strong> that; I don't think it's a completely new (or extremely serious) issue; it seems like that'd make it look scary
[14:16:04] <deason> (not that I have a preview of what it actually looks like :)
[14:16:06] <deason> but I don't object to it
[14:17:09] <deason> er, and sorry, not a regression from 1.6.9, that would be bad; regression from 1.6.8 or so
[14:17:17] <deason> whenever the getcwd thing was last 'fixed'
[14:17:52] <wiesand> Files to preview should be available from http://www.zeuthen.desy.de/~wiesand/oaw.tar.bz2
[14:18:23] <wiesand> It was "fixed" in 1.6.8 .
[14:18:32] <wiesand> And not broken in 1.6.9.
[14:19:15] <deason> I might mention it only affects scenarios involving fakestat mounts; if that's too tricky to word clearly then don't worry about it
[14:19:47] <wiesand> I wasn't certain about that...
[14:20:03] <kaduk> Do we know of other 1.8-related things to do than the tmay issue and documentation?
[14:21:44] <Jeffrey Altman> How about (without the  "strong"):
Note: On Linux, the 1.6.9 client can suppress ENOENT errors when file or directory contents are altered by another AFS client and that file or directory is in-use on the local machine.  The 1.6.10 release contains a fix for that suppression but it is believed to generate spurious ENOENT errors when evaluating mount points and volume root directories in conjunction with dynamic root functionality.
[14:23:50] <Jeffrey Altman> Ben, I haven't gone through the open Gerrit issues recently to pull out things that should be addressed on master.  The volume update time issues are one that should probably be addressed.   Last time I was in Pittsburgh Jeff Hutzelman said that he wanted to review that in detail but has yet to do so.
[14:24:40] <kaduk> I was wondering if anyone had gone through the changes...it's been a while since I tried, too.
[14:24:57] wiesand leaves the room
[14:25:14] wiesand joins the room
[14:25:38] <wiesand> Did we all get kicked out for a moment?
[14:25:45] <kaduk> Just you.
[14:25:55] <wiesand> Good ;-)
[14:26:45] <deason> I'd just maybe say:
"There is a known issue affecting Linux clients, causing spurious ENOENT errors on fakestat mounts Please find the detauls..."
[14:26:57] <deason> er, without the typos
[14:27:37] <deason> jeff's text is mostly accurate (fakestat mounts, not dynroot), but I don't think that much detail is needed just for a website note
[14:27:58] <deason> linux client, fakestat, and enoent should be enough landmarks for people to recognize
[14:28:01] <deason> and more details can be in the ticket
[14:31:23] <deason> bit wiesand, it's whatever you think is best :)
[14:31:30] <deason> er, "but wiesand"
[14:31:33] <Jeffrey Altman> I tend to err on the too much info side
[14:31:58] <deason> yeah, and I don't mind that either
[14:32:39] <deason> sorry for bikeshedding, just trying to provide input
[14:33:08] <wiesand> no that's fine.
[14:33:55] <Jeffrey Altman> Stephen, its up to you.
[14:36:50] <wiesand> Note: There is a known issue affecting Linux clients, causing spurious ENOENT errors on fakestat mounts, often showing up as getcwd() failures. This problem had been fixed in the 1.6.8 release, but seems to be back in 1.6.10 due to a change fixing another issue. Please find the details...   ?
[14:37:15] <deason> sure
[14:37:18] <kaduk> Sounds fine here
[14:38:20] <wiesand> Ok, I'll push that in a moment. Feel free to go on with 1.8 topics.
[14:38:43] <Jeffrey Altman> great
[14:40:05] <kaduk> Do you think we need to go through outstanding gerrit changes for master prior to branching, or can that be done after?
[14:42:44] <Jeffrey Altman> it would be better to go through now to ensure that things are not forgotten.  get fixed on master later but failed to get pulled up
[14:42:55] <Jeffrey Altman> but there isn't a strict requirement to do so
[14:43:01] <kaduk> Okay.
[14:43:43] <kaduk> In terms of documentation, we add a KeyFileExt file which will actually be used for rxkad-k5 (which is also undocumented).
[14:44:36] <kaduk> How much rein to we have to go about removing outdated stuff from the xml docs?
(I just noticed that they still refer to the package utility, removed in 2006.)
[14:44:57] <Jeffrey Altman> take an axe
[14:45:21] <Jeffrey Altman> the only reason the IBM docs are still used is because we don't have anything better
[14:45:53] <kaduk> Okay.
"But where to start"
[14:46:07] <deason> if you ever find anything wrt obsolete things, just get rid of them; if people need it it's in older versions
[14:46:11] <deason> and if you make a mistake it's not lost
[14:50:03] <kaduk> So at the moment, the only blocker we know about for branching 1.8 is tmay, but we would like to get doc updates and go through changes sitting in gerrit?
[14:52:35] <Jeffrey Altman> yes.   I'm going to need to go.  The volume header updates issue raised a related question of the volume update time not be updated by administrative operations that alter the volume metadata such a quota and other fields.   As a result these values can be lost.
[14:53:26] <kaduk> Okay.  Thanks.
[14:54:09] <Jeffrey Altman> Another thing that comes to mind is the vos release -stayup functionality that was committed and pulled up to 1.6.x but doesn't work.   It was mentioned in IRC multiple times by Mark Vitale that patches to fix it were forthcoming but they haven't arrived.  Perhaps reverting that change would be appropriate on master before 1.8.
[14:55:13] <Jeffrey Altman> A migration guide from 1.6 to 1.8 keying.   Perhaps as a miscellaneous man page
[14:56:23] <Jeffrey Altman> I can't think of anything else at the moment.   I will look at the revised patches you pushed later.   Wasn't able to do so this morning.
[14:57:02] <wiesand> Updated 1.6.10 draft announcement in http://www.zeuthen.desy.de/~wiesand/1610draft.txt
[14:59:02] <kaduk> I will try to update the 1.8 todo list on the wiki.
[14:59:28] Jeffrey Altman leaves the room
[14:59:49] <wiesand> Good.
[15:00:42] <kaduk> I expect that means we're done for the day, then.
[15:00:56] <deason> there was some other change I was nervous about... the "lockless path through d_revalidate" or something, which I think is on master
[15:01:00] <meffie> is upserver really going away (according to that wiki page) ?
[15:01:02] <wiesand> Well, we haven't talked about 1.6.11.
[15:01:05] <deason> which I have an alternative fix for in progress
[15:02:25] <kaduk> Mike: I would like for upserver to go away.  No one has spoken against doing so (yet). :)
[15:02:43] <kaduk> Oh right, 1.6.11, that was on the agenda, too.
[15:02:48] <meffie> i'm pretty sure some sites are still using it, for config files.
[15:02:56] <kaduk> I think the aklog 524_conv_principal thing is fine for 1.6.11
[15:04:07] <wiesand> Yes, I think so too. I'm mostly wondering about 11538
[15:04:27] <kaduk> Do you think those sites would be averse to switching to something like puppet for those files?
[15:05:12] <kaduk> Andrew thinks he'll have a proper fix for the 11538 bits "soon", as I understand it.
[15:05:15] <deason> yes, their upserver usage is not going away without a fight
[15:05:18] <wiesand> As a data point, my site won't miss upserver.
[15:05:27] <meffie> well. i'm all for scrubbing obsolete stuff. but it's one more thing to make upgrading harder :)
[15:05:53] <deason> at least one site in particular uses somewhat real config management, but runs it out of afs
[15:05:55] <wiesand> Will we get Anders to test it before we tag 1.6.11?
[15:06:00] <kaduk> "Can I ask when they switched from 1.4 to 1.6?"
[15:06:07] <deason> so for afs itself they just use upserver
[15:06:24] wiesand leaves the room
[15:06:35] wiesand joins the room
[15:06:57] <kaduk> I expect we can get Anders/Scripts to test candidates for that issue, yes.
[15:08:16] <deason> 1.6 switch would have been 1 or 2 years ago at my guess
[15:08:37] <wiesand> Fine. If we expect to fix it in 1.6.11, no need to talk about our intention to revert 11358 in the 1.6.10 announcement.
[15:08:44] <deason> if another major release removes a feature they were using without an equivalent that takes zero setup, another migration might take 10 years
[15:09:26] <deason> or instead they might say "if it takes this much effort to migrate to 1.8, maybe we'll instead use that effort to migrate off of afs"
[15:09:56] <deason> or more realistically, we (I) would have to maintain a patch that keeps upserver around for a while
[15:11:09] <meffie> yeah
[15:11:19] <deason> well, I think I meantioned 11538 could be fixed soon
[15:11:32] <kaduk> Well, this would be for the stable release after 1.8, not 1.8, but I guess they're probably not going to change much without prompting.
[15:11:36] <deason> it's not a priority at the moment, but if I'm given a timeline or something it could be
[15:11:54] <deason> we could talk to them about it
[15:12:02] <deason> I wasn't aware upserver was causing much grief
[15:12:30] <kaduk> It's still LWP.
[15:12:57] <kaduk> I don't have much interest in converting it to pthreads; it's easier to just rip it out.
Now, if someone else does the work, I don't mind keeping it around...
[15:13:09] Jeffrey Altman joins the room
[15:13:30] <kaduk> But one of the stated goals for post-1.8 is to get rid of LWP entirely.
[15:14:17] <meffie> ah. it's lwp.
[15:14:44] <meffie> that's the motivation. ok, that makes sense.
[15:15:14] <wiesand> Andrew: would it be feasible to have the proposed 11538 fix in gerrit next wednesday?
[15:17:01] <deason> wiesand: sure
[15:21:08] <wiesand> Thanks.
[15:21:18] <wiesand> Anything else to discuss today?
[15:22:20] <wiesand> I guess that's a "no".
[15:24:05] <kaduk> It does look like one, yes.
[15:24:38] <wiesand> I should get out minutes today. Unleashing 1.6.10 may not happen before Friday though :-(
[15:25:16] <wiesand> I have to run. Thanks a lot everyone!
[15:25:50] wiesand leaves the room
[15:27:18] kaduk leaves the room
[15:32:49] meffie leaves the room
[21:48:37] Jeffrey Altman leaves the room
[21:56:19] Jeffrey Altman joins the room
[22:24:21] deason leaves the room