From FedoraProject

Jump to: navigation, search


Fedora Release Engineering Meeting :: Monday 2000-04-21

Preview Release

  • PR was made available via torrents on Friday
  • Attempting to sync release trees to master mirrors so that we can offer direct (and jigdo) downloads
  • upload is mostly done--waiting for snapmirror to sync up the various RH netapps
  • after sync is complete will flip a bit and let the mirrors get it

CVS Branching

  • 36+ hours of outage for branching
  • cannot do this again--must to figure out a way to do this without an outage
  • the second batch of cvs branches is about 50% of the way through which started 11~ hours ago
  • part of the delay is the pkgdb updates--a different SCM wouldn't necessarily fix the problem

Release Candidate

1. They're real RCs, as in "we think we fixed all we're going to take, so now we're going to do the composes that, if they turn out right, are what we ship"

  • not "hey that RC passed, now go compose the final"

1. RC will not be a public release

  • lead time to get bits moved around just takes too long for it to be really worth it

1. Everyone can do composes leading up to RC for testing various fixes 1. Using pungi people outside of Red Hat can create their own RC using rawhide

Tagging SOP

IRC Transcript


-!- f13 changed the topic of #fedora-meeting to: Fedora Release Engineering Meeting - Roll Call13:11
* poelcat here :)13:11
f13ping: jwb jeremy spot warren wwoods poelcat rdieter notting lmacken13:11
* f13 wonders if he forgot anybody13:12
nottingnot really here13:12
jwb1/2 here13:12
* warren here13:12
* rdieter here13:12
* lmacken is kind of here13:12
-!- f13 changed the topic of #fedora-meeting to: releng - Preview Release13:15
f13Sorry for not being good with agenda creation, I suck.13:15
f13PR was made available via torrents on Friday, and a number of people have them.13:15
f13I've also been trying to get the release trees synced to the master mirrors so that we can offer direct (and jigdo) downloads13:15
f13that upload is mostly done, now we're just waiting for snapmirror to sync up the various RH netapps13:16
f13then I'll flip a bit and let the mirrors get it13:16
f13then PR will be out there, and everybody will be happy13:19
-!- f13 changed the topic of #fedora-meeting to: releng - CVS branching13:19
f13this is a disaster.13:20
f13looking like 36+ hours of outage for branching13:21
f13cannot do that again.  We /have/ to figure out a way to do this without an outage.13:21
f13and I think all we do an outage for is to turn off mail, and short circuit some slow things.  If we just make our mail senders smart enough to opt-out given certain options, we can live with teh other slowness because there won't be an outage.13:21
f13I haven't filed tickets on it, but toshio and I have talked a bit13:22
f13as it stands, the second batch of cvs branches is about 50% of the way through, and I started it 11~ hours ago13:25
poelcatf13: w/o intending to start a flamewar on source control13:26
poelcatwould git or something else make this easier?13:26
f13poelcat: tis the season.13:26
f13poelcat: that's really ahrd to say, as we can't just drop in replace our system with git13:26
jwbpoelcat, it would from the mass-<doing something> perspective.  dunno about the rest of the VCS uses13:26
f13poelcat: theoretically it would get better, but a good chunk of the delay was doing all the branch requests in pkgdb13:27
f13we have to tell pkgdb that all these packages now have an F-9 branch, and to clone the permissions from devel/ over to the F-9 version13:27
f13the good news is that we had very few failures thus far, and it's gone smoother (although longer) than last time13:28
jwbf13, abadger1999 said that took 15s/package13:30
jwbwhereas CVS took 3min?13:30
f13jwb: 3 min seems high.13:30
jwbthose were the numbers i saw yesterday13:31
abadger1999jwb: 3pkgs /minute13:31
jwboh, sorry13:31
jwbi suck13:31
abadger1999jwb: pkgdb-client 4 pkgs / minute; pkgdb2branch 3 pkgs/minute13:32
f13plus or minus the spinup time of the task13:33
f13the cvs branching can be done somewhat parallel, I wasn't sure if the pkgdb-client stuff could or not13:34
f13moving on.13:37
-!- f13 changed the topic of #fedora-meeting to: releng - Release Candidates13:37
abadger1999It should be fine to do in parallel.  There's three app servers so using three parallel processes ought to be reasonable.13:37
f13abadger1999: ok, we'll have to make note of doing that in the future13:37
f13Couple things with RC.  1) they're real RCs, as in "we think we fixed all we're going to take, so now we're going to do the composes that, if they turn out right, are what we ship"13:38
f13not "hey that RC passed, now go compose the final"13:38
f132) they likely won't be very public.  Lead time to get bits moved around just takes too long for it to be really worth it.13:39
f133) we can do all the composes we want leading up to then for testing various fixes, and with pungi other folks can too, plus rawhide.13:39
jwbso if the RCs aren't public...13:39
jwbor could you define "very"?13:39
f134) I'm somewhat not available for a week starting this coming Thursday.13:39
f13jwb: until we get more infrastructure in place, the release composes happen on my workstation. They're as public as my workstation is.13:40
f13which sucks, but so does the tubes.13:40
poelcatf13: could they be put on download.fedora.r.c but not sync'd to the mirrors?13:40
jwbpoelcat, so people will swamp d.f.r.c?13:40
f13poelcat: given that it takes about a day to sync stuff out there, they'd be quite quickly obsoleted, and cause a lot of churn to hte mirrors.13:41
rdieteror torrents? (or that probably means even more work)13:41
f13and yes, swamp d.f.r.c which right now == PHX13:41
poelcatit seems funny that everyone can participate in the whole process until the end :)13:41
f13poelcat: it's far better than other places where you can't participate at all13:41
f13poelcat: the turn around time we have with RCs just doesn't work with waiting for N folks around the world to download it and give their approval13:41
poelcatbut we say 'all our bits are free' ;-) jk13:42
f13unless we insert week delays for each RC and change absolutely nothing between it and the next RC unless a blocker comes up.13:42
poelcatit's not a big deal, it just struck me13:42
jwbwe can still direct people to do network installs of rawhide13:43
f13sure, we won't even wait for other "RH" folks in other offices to sign off on it.13:43
f13really, the folks around where the compose are happening will be doing most hte smoke testing, but everything should be known fixed or passed on at that point, so really we're just sanity checking the compose results itself.13:43
f13poelcat: for lolz, read the FreeBSD release process sometime.  It's cute (:13:44
jwbf13, i will note that with 2ndary arches we might have to think harder13:44
jwbor care less13:44
wwoodsthe difference between the RC composes and rawhide will usually be one or two packages, and frequently nil13:45
wwoodsgiven that all the bits are public, and the code to produce the RC from the bits is also public13:45
wwoodsthe fact that we don't invite people to totally frag our tubes just to get already-public bits13:46
wwoodsis Not A Big Deal13:46
poelcatwwoods: yep, i said that above :)13:46
wwoodsesp. given the average time between RC composes versus turnaround time on community testing13:46
wwoodsjust reaffirmign13:46
rdieterI wantz my fedoraz, now now now13:46
rdietermmm... tasty13:46
jwbwwoods, but... making shiny discs is hard13:47
wwoodswe aren't *hiding* anything from anyone, this *isn't* a loss of transparency13:47
f13anywho, the week I'm gone should only need tag requesting and signing done as far as F9 is concerned13:49
f13many people can tag, bill can sign.13:49
f13I'll probably poke online once a day or so, and make sure F7/8 updates keep flowing13:49
poelcatf13: i've been watching all the mail go by on releng list about tagging... do we have an SOP page that explains how it all works?13:50
poelcati'd consider helping, but don't know the process or what to do :)13:51
poelcator maybe I'd make things worse and that is by design ;-)13:51
f13poelcat: we have the FinalFreeze policy page, but that's mostly a developer POV.  I don't think I have anything written up from the releng pov13:51
jwbpoelcat, you just wait for 2 +1s13:51
rdieternod, the only thing missing there is how to actually do the tagging.13:52
poelcatjwb: and then what?13:52
f13Could probably add something to
jwbthen you run: koji tag-pkg --force f9-final <package>13:52
f13poelcat: but to run --force, you have to have super cow releng powers13:52
-!- f13 changed the topic of #fedora-meeting to: releng - open floor13:54
rdieter(selfishly) this blocker scares the bejezus out of me: , no idea what's going on, only started happening recently. ??13:57
buggbotBug 443320: low, low, ---, X/OpenGL Maintenance List, NEW , Cannot enter keyboard input into kdm after logout13:57
poelcatrdieter: i saw that issue after an initial install trying to log in for the first time13:59
poelcatbut couldn't reproduce it13:59
rdieternot that I expect miracles here/now, just wanted to raise awareness.13:59
rdieterI can (semi?) reliably get my laptop into a state to reproduce on every logout.  (suspend->resume seems to be the trigger for me) ??14:00
f13well, if there is anything else...14:00
rdieternothin here14:01
jwbf13, random question14:02
jwbf13, RHEL5 ppc box (aka quad) for developers to muck with... after F9 goes out?14:02
f13yes, I need to mail whatshisname to ask about apple store details.14:03
f13although we still need a place to host it I think.  OLPC may be hostile14:03
f13dgilmore was saying they didn't want any non-olpc machines in the racks anymore14:04
f13rdieter: I have no clue.14:05

Generated by 2.3 by Marius Gedminas - find it at!