ReleaseEngineering/Meetings/2007-jun-11

From FedoraProject

Jump to: navigation, search

Contents

Fedora Release Engineering Meeting :: Monday 11-Jun-2007

Buildroot contents for updates

  • The buildroot used for fedora 7 updates building is not self updating. It only contains things from f7-gold, and stable released updates. This means that one update candidate cannot be built against another update candidate without rel-eng interaction. We haven't been very vocal about this yet, and it isn't documented anywhere.
  • Do we want to adjust things or leave them be?
  • See IRC log for discussion points

Decision: Talk to lmacken about modifying Bodhi to do a sanity check on published buildroot contents before allowing an update be pushed. After that we will make the dist-fc7-build buildroot auto-update with -candidate builds and see what happens.

Expand Standard Operating Procedures (SOPs)

  • f13
  • rel-eng has more and more people helping out so we want to make sure that things are well documented
  • Infrastructure team started doing SOP/ pages (Standard Operating Procedure) and they are proving very useful. I'd like to start doing them for Release Engineering too
  • thinking of ReleaseEngineering/SOP/<task> as the layout, the SOP page itself could be a list of pages and info on how to add an SOP
  • Proposal: As a release engineer, as you do tasks for Fedora, check to see if there is an SOP/<task> page. If there isn't, create one and poke rel-eng folks for review.
  • This is more of a mandate than a vote item.

Early Torrent Release - Rahul Sundaram

  • explore possibility of doing early torrent releases.
  • See IRC log for discussion of pros and cons and affect on mirrors

Decision: Investigate with Infrastructure team ways of getting more mirrors to participate in seeding the torrent (early?). Do not release torrent to general public before the agreed upon coordinated release date/time.

Upgrade path enforcement - Rahul Sundaram

  • Discussion surrounding policy that upgrade path should not break from a particular point on, for example, after Test1 or Test2--enforced by Release Engineering.
  • See IRC log for discussion details

Decision: Recommend to FESCo policy that upgrade path never be compromised either by removing builds or NEVR regressions. File RFE in Koji to enforce rule at build time

IRC Transcript

Idea.png f13 changed the topic of #fedora-meeting to: RELENG-Meeting - Buildroot contents for updates - JesseKeating<a href="#t12:57" class="time">12:57</a>
f13Meeting time coming up.<a href="#t12:57" class="time">12:57</a>
f13Release Engineering meeting starting now.<a href="#t12:59" class="time">12:59</a>
f13jwb_gone: notting: jeremy: rdieter: warren: wwoods: ping<a href="#t13:00" class="time">13:00</a>
nottingyes?<a href="#t13:00" class="time">13:00</a>
rdieterhere<a href="#t13:00" class="time">13:00</a>
* nirik sits in the bleachers with his coffee. <a href="#t13:00" class="time">13:00</a>
f13notting: just role call.<a href="#t13:00" class="time">13:00</a>
wwoodsptchu<a href="#t13:00" class="time">13:00</a>
wwoodscurmudgeonly grammar teacher living in my frontal lobe requires me to say: it's roll call.<a href="#t13:01" class="time">13:01</a>
* jeremy is here<a href="#t13:01" class="time">13:01</a>
wwoodslike honor roll.<a href="#t13:01" class="time">13:01</a>
nottingwwoods: cinnamon roll?<a href="#t13:01" class="time">13:01</a>
nottingpork roll?<a href="#t13:01" class="time">13:01</a>
jeremymy clock is just slow :)<a href="#t13:01" class="time">13:01</a>
f13wwoods: but what if I'm asking to see who's here to fill what role?<a href="#t13:01" class="time">13:01</a>
f13role roll call?<a href="#t13:02" class="time">13:02</a>
* mmcgrath pongs<a href="#t13:02" class="time">13:02</a>
wwoodsf13: totally<a href="#t13:02" class="time">13:02</a>
wwoodsand if you're playing back a filmstrip of this event, you can say: roll role roll call<a href="#t13:02" class="time">13:02</a>
f13ok, first topic is kind of fun.<a href="#t13:02" class="time">13:02</a>
warrenfun<a href="#t13:02" class="time">13:02</a>
f13As it currently stands, the buildroot used for fedora 7 updates building is not self updating.  It only contains things from f7-gold, and stable released updates.<a href="#t13:03" class="time">13:03</a>
f13this means that one update candidate cannot be built against another update candidate without rel-eng interaction.<a href="#t13:03" class="time">13:03</a>
f13We haven't been very vocal about this yet, and it isn't documented anywhere.  I have to ask if we want to adjust things or leave them be.<a href="#t13:03" class="time">13:03</a>
mmcgrathf13: was this by design or an oversight?<a href="#t13:04" class="time">13:04</a>
f13mmcgrath: design<a href="#t13:04" class="time">13:04</a>
wwoodsI think we want to adjust that. People have been complaining about how huge a PITA it is to get several interdependent packages built<a href="#t13:04" class="time">13:04</a>
f13it's how things work internally since before I got here.<a href="#t13:04" class="time">13:04</a>
rdietermight be worth trying, anyone against self-updated buildroots?<a href="#t13:04" class="time">13:04</a>
* mmcgrath just wonders what we might break if we change it.<a href="#t13:04" class="time">13:04</a>
wwoodsoh. then how did updated stuff get put into the buildroot for interdependent builds?<a href="#t13:04" class="time">13:04</a>
nottingnothing would break until someone puts something broke in updates-candidate and someone builds against it<a href="#t13:05" class="time">13:05</a>
f13mmcgrath: we run the risk of broken buildroots, things built against things that don't go out as updates, etc...<a href="#t13:05" class="time">13:05</a>
warrenHere's one possible solution, similar to the fedora-cvs work queue.  Create a fedora-rel-eng work queue, where people can ask for that kind of request in a controlled manner.  We can use that until we have a more automated process in packagedb.<a href="#t13:05" class="time">13:05</a>
f13wwoods: we have a 'dist-fc7-override' tag that we can tag builds with to make them available.<a href="#t13:05" class="time">13:05</a>
nirikhow many requests has releng had to add something to the buildroots?<a href="#t13:05" class="time">13:05</a>
jeremywarren: you more want to eventually get to where people can specify "use these unreleased updates" for my build...  not something in the packagedb<a href="#t13:05" class="time">13:05</a>
f13I'm bringing pjones in, I know he's had some opinions on this in the past.<a href="#t13:05" class="time">13:05</a>
wwoodsf13: right, which requires rel-eng intervention (and extra action by maintainers --> OMGWTFREGRESSION)<a href="#t13:06" class="time">13:06</a>
warrenf13, oops, I shouldn't have added a package to dist-fc7-build I'm guessing.  I'll have to undo that.<a href="#t13:06" class="time">13:06</a>
pjoneshey, so where are we?<a href="#t13:06" class="time">13:06</a>
wwoodshow did this work for Core / Extras?<a href="#t13:06" class="time">13:06</a>
rdieterwwoods: Extras self-updated, Core, dunno (not?)<a href="#t13:06" class="time">13:06</a>
nirikextras always got everything after builing.<a href="#t13:06" class="time">13:06</a>
nottingwwoods: extras always built against latest, even unreleased<a href="#t13:06" class="time">13:06</a>
nottingwwoods: core required intervention<a href="#t13:06" class="time">13:06</a>
f13pjones: basically I've stated the current situation (not updating) and gathering thoughts/opinions on either keeping it like it is, or changing it and risking the breakage.<a href="#t13:06" class="time">13:06</a>
f13mschwendnt (hwoever the hell you spell that) and dgilmore had some stats for how often the buildroot got fubar in Extras<a href="#t13:07" class="time">13:07</a>
nirikI know it will be fun for me on the next Xfce update... there are lots of dependency chains there... 3 inital packages, then everythign needs them and each one in the row after that...<a href="#t13:07" class="time">13:07</a>
wwoodsIt would seem like we need some kind of a hybrid where people can create private branched buildroots for building their stuff, or something?<a href="#t13:07" class="time">13:07</a>
pjonesf13: at the very least, if we leave the default like it is, we need the ability to make it pull from updates and such via BuildRequires:<a href="#t13:07" class="time">13:07</a>
f13pjones: well, the buildroot currently is populated with f7-gold and stable updates<a href="#t13:07" class="time">13:07</a>
wwoodsI admit not having deep knowledge of how much overhead that would require. All I know is that a bunch of maintainers are unhappy about it. Which is not news to anyone.<a href="#t13:08" class="time">13:08</a>
f13pjones: it just doesn't have any updates-testing or -candidate builds.<a href="#t13:08" class="time">13:08</a>
pjonesso if, for example, I know a new anaconda needs a newer kudzu which I've just built, then I can BR: it and have that work.<a href="#t13:08" class="time">13:08</a>
nirikhow about a bodhi tag/box? "add to buildroot for more builds"?<a href="#t13:08" class="time">13:08</a>
mdomschpjones, BR with version?<a href="#t13:08" class="time">13:08</a>
f13pjones: in the current situation, unless kudzu has been pushed out as stable, you'd have to request rel-eng interaction to tag the kudzu build.<a href="#t13:08" class="time">13:08</a>
pjonesmdomsch: yes<a href="#t13:09" class="time">13:09</a>
mdomschpjones, even then, you need to have the repo available for mock to use it<a href="#t13:09" class="time">13:09</a>
pjonesmdomsch: agreed<a href="#t13:09" class="time">13:09</a>
f13mdomsch: if the buildroot was self updating, chain-build would work<a href="#t13:09" class="time">13:09</a>
f13almost.<a href="#t13:09" class="time">13:09</a>
nottingf13: that's a big win for peopel<a href="#t13:09" class="time">13:09</a>
rdieterI'm of the KISS mind, auto-update buildroot, rel-eng can step in and fix any breakage (ie, untag bustedness) if need be.<a href="#t13:09" class="time">13:09</a>
pjonesf13: chain-build has other problems though.  not the least of which being the syntax is weird<a href="#t13:09" class="time">13:09</a>
f13I don't think chainbuild has the concept of 'this was already built, but wait unti it's available to build me'<a href="#t13:09" class="time">13:09</a>
f13pjones: yeah, syntax is interesting.  NOt sure how to make it better.  Open to suggestions<a href="#t13:10" class="time">13:10</a>
rdieterf13: chainbuild worked for me that way, when I tried it.<a href="#t13:10" class="time">13:10</a>
pjonesrdieter: *nod*.  We should be able to tell "oh, kudzu is fucked, we need to roll it out as well as ${dependent packages that got built}"<a href="#t13:10" class="time">13:10</a>
f13rdieter: oh cool, maybe it is smart enough (:<a href="#t13:10" class="time">13:10</a>
f13pjones: my worry is that $(dependent packages that got buitl) may have made it out to stable updates<a href="#t13:10" class="time">13:10</a>
f13and then it's not so easy/forgiving to just make it go away<a href="#t13:11" class="time">13:11</a>
pjonesrdieter: I think we have problems where an update isn't available to build with more often than we have problems where an update was bad.<a href="#t13:11" class="time">13:11</a>
pjonesf13: that seems like something we should be able to preclude from happening<a href="#t13:11" class="time">13:11</a>
rdieterpjones: agreed.<a href="#t13:11" class="time">13:11</a>
f13pjones: with some work sure, bodhi could query to see if any component in the buildroot for an update is as of yet unreleased.<a href="#t13:11" class="time">13:11</a>
pjonesf13: nod.<a href="#t13:11" class="time">13:11</a>
warrenf13, good idea.<a href="#t13:11" class="time">13:11</a>
* f13 signs lmacken up for more work.<a href="#t13:12" class="time">13:12</a>
rdieterA compromise maybe would be to enable updates-testing in the buildroot?<a href="#t13:12" class="time">13:12</a>
f13rdieter: that's not good enough I don't thik.<a href="#t13:12" class="time">13:12</a>
f13think.<a href="#t13:12" class="time">13:12</a>
jeremyf13: with that, I'd be okay with allowing updates-candidates in buildroots.  but it's going to cause problems<a href="#t13:12" class="time">13:12</a>
f13rdieter: you may want to issue a big stack as one update.<a href="#t13:12" class="time">13:12</a>
jeremyeg, jakub builds new glibc, it's in updates-testing and he wants it to be there a week<a href="#t13:12" class="time">13:12</a>
rdieterf13: ok, nevermind.<a href="#t13:12" class="time">13:12</a>
jeremypeople that want to build and immediately push (or security updates, etc) will get flagged<a href="#t13:12" class="time">13:12</a>
f13jeremy: well, jakub builds into dist-fc7-candidate anyway, something completely not imported for his own personal tests, then tags it for updates-cnadidate by hand when he knows it's good<a href="#t13:13" class="time">13:13</a>
f13but yes, I can see that causing problems, so perhaps an override option.<a href="#t13:13" class="time">13:13</a>
f13unless that will break deps :/<a href="#t13:13" class="time">13:13</a>
f13So I'm willing to experiment with Fedora 7 updates to see how it goes, once we get some bodhi love.<a href="#t13:14" class="time">13:14</a>
warrenf13, how could bodhi know if something that was in the buildroot is not yet pushed?  Does mock output list *all* packages that were installed?<a href="#t13:14" class="time">13:14</a>
pjonesf13: perhaps allow package builders to specifically tag their updates-testing things for building?  you'd probably still need the "this can't be released until that is" infrastructure, but it wouldn't have the problem jeremy mentioned<a href="#t13:14" class="time">13:14</a>
f13warren: koji tracks buildroot components for a build.<a href="#t13:15" class="time">13:15</a>
warrenf13, ah.<a href="#t13:15" class="time">13:15</a>
wwoodsI thought we were going to start allowing updates to have a set of packages?<a href="#t13:16" class="time">13:16</a>
nirikyeah, I think it would make sense to leave default as it is now, and allow a checkbox for maintainers in bodhi to add their update-testing update to the buildroot?<a href="#t13:16" class="time">13:16</a>
f13pjones: that's a possiblity.  Would have to think about how to construct that workflow<a href="#t13:16" class="time">13:16</a>
nottingnirik: that really doesn't help with things like chain-build<a href="#t13:16" class="time">13:16</a>
warrennirik, oh, so a third unpublished layer?<a href="#t13:16" class="time">13:16</a>
f13ah right.<a href="#t13:16" class="time">13:16</a>
f13that doesn't help chain-build.<a href="#t13:16" class="time">13:16</a>
niriktrue.<a href="#t13:16" class="time">13:16</a>
f13so...<a href="#t13:16" class="time">13:16</a>
warrenbodhi could push to THREE layers instead of the current two.<a href="#t13:16" class="time">13:16</a>
warrenunpushed -> buildrootonly -> updates-testing -> updates<a href="#t13:17" class="time">13:17</a>
nirikwell, it already has pending/updates-testing/updates<a href="#t13:17" class="time">13:17</a>
f13Proposal:  We talk to lmacken about modifying bodhi to do a sanity check on published buildroot contents before allowing an update be pushed.  We then make the buildroot self updating and see what happens.<a href="#t13:17" class="time">13:17</a>
warrenah<a href="#t13:17" class="time">13:17</a>
rdieterf13: +1<a href="#t13:17" class="time">13:17</a>
f13warren: you'd still introduce a break in the workflow where you can't rely upon chain-build.<a href="#t13:17" class="time">13:17</a>
warrennirik, what does pending mean?<a href="#t13:17" class="time">13:17</a>
warrenf13, +1<a href="#t13:17" class="time">13:17</a>
f13warren: it means it hasn't gone out in updates-testing yet<a href="#t13:17" class="time">13:17</a>
nirikwarren: someone added that they want it as an update, but hasn't pushed it to updates-testing yet<a href="#t13:18" class="time">13:18</a>
wwoodsf13: update pushed live, or pushed to -testing?<a href="#t13:18" class="time">13:18</a>
pjonesf13: as an aside, the problem with chain-build is that it doesn't really act on a checked out tree<a href="#t13:19" class="time">13:19</a>
pjonesf13: +1 on your proposal above, btw.<a href="#t13:19" class="time">13:19</a>
f13pjones: lets talk about chain-build over in #fedora-devel after this with mbonnet ok?<a href="#t13:20" class="time">13:20</a>
pjonesok<a href="#t13:20" class="time">13:20</a>
f13wwoods: I don't understand the question?<a href="#t13:20" class="time">13:20</a>
abadger1999f13: +1<a href="#t13:20" class="time">13:20</a>
f13wwoods: oh I see.  Well, it shouldn't allow something to go into -testing unless what was used to produce it is on the way to -testing, ditto final.<a href="#t13:21" class="time">13:21</a>
wwoodsf13: I'm trying to clarify "pushed" in your proposal - pushed where?<a href="#t13:21" class="time">13:21</a>
f13although we could be more relaxed with -testing.<a href="#t13:21" class="time">13:21</a>
wwoodsah, gotcha<a href="#t13:21" class="time">13:21</a>
wwoods+1 either way, but I wanted to be clear<a href="#t13:21" class="time">13:21</a>
f13k.  I'll work with lmacken to see which is more feasable.<a href="#t13:22" class="time">13:22</a>
f13This may take a while to implement so we'll still need documented workflow for people to ask rel-eng to tag things for them.<a href="#t13:22" class="time">13:22</a>
f13jeremy: notting: your votes?<a href="#t13:23" class="time">13:23</a>
nottingf13: open the floodgates!<a href="#t13:23" class="time">13:23</a>
jeremylet's see what we can do...  I'm open to trying things, though still a little concerned for the obvious reasons<a href="#t13:23" class="time">13:23</a>
f13jeremy: ditto (:<a href="#t13:23" class="time">13:23</a>
f13Ok, so that passed.<a href="#t13:23" class="time">13:23</a>
f13poelcat: Decision: We talk to lmacken about modifying bodhi to do a sanity check on published buildroot contents before allowing an update be pushed.  We then make the dist-fc7-build buildroot auto-update with -candidate builds and see what happens.<a href="#t13:24" class="time">13:24</a>
Idea.png f13 changed the topic of #fedora-meeting to: RELENG-Meeting - Flesh out SOPs for rel-eng tasks - JesseKeating<a href="#t13:24" class="time">13:24</a>
f13so rel-eng has more and more people doing things for us, we'll want to make sure that things are well documented.<a href="#t13:25" class="time">13:25</a>
f13The Infrastructure team started doing SOP/ pages (Standard Operating Proceedure) and they are proving very useful.  I'd like to start doing them for Release Engineering too<a href="#t13:25" class="time">13:25</a>
* mclasen mumbles about silly acronyms<a href="#t13:26" class="time">13:26</a>
f13I'm thinking of ReleaseEngineering/SOP/<task>  as the layout, the SOP page itself could be a list of pages and info on how to add an SOP<a href="#t13:26" class="time">13:26</a>
f13mclasen: suggestions?<a href="#t13:26" class="time">13:26</a>
warrenf13, sounds good to me.<a href="#t13:26" class="time">13:26</a>
mclasensince you spell out release engineering there, why not spell out procedure as well<a href="#t13:27" class="time">13:27</a>
f13*shrug* I'm lazy, and there is already precidence for SOP/ in the Infrastructure pages<a href="#t13:27" class="time">13:27</a>
* mclasen relurks<a href="#t13:27" class="time">13:27</a>
f13I don't really think it matters if the page is linked to from ReleaseEngineering.<a href="#t13:27" class="time">13:27</a>
f13Proposal: As a release engineer, as you do tasks for Fedora, check to see if there is an SOP/<task> page.  If there isn't, create one and poke rel-eng folks for review.<a href="#t13:28" class="time">13:28</a>
f13I guess this is more of a mandate than a vote item.<a href="#t13:28" class="time">13:28</a>
f13Ok, moving on<a href="#t13:29" class="time">13:29</a>
Idea.png f13 changed the topic of #fedora-meeting to: RELENG-Meeting - Early Torrent Release - RahulSundaram<a href="#t13:29" class="time">13:29</a>
f13mether wanted us to talk about doing early torrent releases.<a href="#t13:29" class="time">13:29</a>
mmcgrathf13: I found out at linux tag that OpenSuSE actually has some of their mirrors setup as torrent seeds.  We could also do that.<a href="#t13:30" class="time">13:30</a>
f13I'm personally not a huge fan of this, last time the mirror admins caught wind of something like this they got kinda pissy.<a href="#t13:30" class="time">13:30</a>
wwoodslike.. torrents open up before the mirrors finish synching?<a href="#t13:30" class="time">13:30</a>
jeremymmcgrath: some of our mirrors act as torrent seeds<a href="#t13:30" class="time">13:30</a>
jeremymmcgrath: eg, I know kernel.org does<a href="#t13:30" class="time">13:30</a>
* mmcgrath didn't know that.<a href="#t13:30" class="time">13:30</a>
f13wwoods: yeah, I don't recall from the email how soon he wanted it open.<a href="#t13:30" class="time">13:30</a>
jeremyf13: I'm not a fan, so -1 from me<a href="#t13:30" class="time">13:30</a>
mmcgrathhow do we get the seed to them?<a href="#t13:30" class="time">13:30</a>
f13mmcgrath: it's done after release though, not before release.<a href="#t13:30" class="time">13:30</a>
mmcgrathahh<a href="#t13:30" class="time">13:30</a>
f13mmcgrath: they probably just copy the mirrored isos into place then fire up bittorrent<a href="#t13:31" class="time">13:31</a>
wwoodsI think I kind of fail to see the point<a href="#t13:31" class="time">13:31</a>
jeremyyeah, believe so<a href="#t13:31" class="time">13:31</a>
* f13 tries to find the posting<a href="#t13:31" class="time">13:31</a>
warrenA mirror with lots of bandwidth doesn't need the data to be useful to a torrent cloud.<a href="#t13:31" class="time">13:31</a>
mmcgrathwwoods: I think the point is to encourage torrent usage.<a href="#t13:31" class="time">13:31</a>
warrenThey will rapidly obtain it and become a seed within 10-20 minutes.<a href="#t13:31" class="time">13:31</a>
mmcgrathdid anyone try the torrent right after release this time around?  I wonder what the speeds were.<a href="#t13:31" class="time">13:31</a>
nottingf13: right, we don't actually publish the layout for torrent, though<a href="#t13:31" class="time">13:31</a>
f13Torrent only release<a href="#t13:32" class="time">13:32</a>
f13--------------------<a href="#t13:32" class="time">13:32</a>
f13I remember this was discussed before but when a new release is made and<a href="#t13:32" class="time">13:32</a>
f13we are waiting for the mirrors to sync why not do a torrent only release<a href="#t13:32" class="time">13:32</a>
f13immediately?<a href="#t13:32" class="time">13:32</a>
rdietermmcgrath: I did, speed was good (not great).<a href="#t13:32" class="time">13:32</a>
f13The advantage is that we won't suffer as much from infrastructure and<a href="#t13:32" class="time">13:32</a>
f13mirror problems with the mad rush in everybody trying to get it at the<a href="#t13:32" class="time">13:32</a>
f13same time. Not sure if this has any disadvantages.<a href="#t13:32" class="time">13:32</a>
mmcgrathrdieter: but you didn't think "gosh this is so terrible I'll just go to download.fedora.redhat.com and get it from there" ? :)<a href="#t13:32" class="time">13:32</a>
poelcatmmcgrath: speed was not that great for at least 24 hours, maybe longer<a href="#t13:32" class="time">13:32</a>
warrenwhy would  it be terrible?<a href="#t13:32" class="time">13:32</a>
warrenThe more people that download a torrent, the faster it goes.  A mirror server's bandwidth only accelerates it.<a href="#t13:32" class="time">13:32</a>
rdietermmcgrath: nope, I think it took ~2 hrs for me.<a href="#t13:33" class="time">13:33</a>
f13Well, bittorrent is pretty damn dumb in how it works<a href="#t13:33" class="time">13:33</a>
poelcatwarren: but not if everyone is starting at zero to begin with<a href="#t13:33" class="time">13:33</a>
mmcgrathwarren: not if no one has it and everyone is hitting torrent.fp.o to get it in the first many hours.<a href="#t13:33" class="time">13:33</a>
f13so if there are a lot of seeders in the other side of hte world from you, and a couple local, torrent will try all of them anyway and you'll be stuck getting bits slowly<a href="#t13:33" class="time">13:33</a>
nottingf13: so, disadvantages: 1) annoy mirrors 2) at most, you're getting a day or two extra release<a href="#t13:33" class="time">13:33</a>
f13notting: that was my thought too.<a href="#t13:34" class="time">13:34</a>
mmcgrathreally we should ask ourselves "Is increasting torrent speed over the first 24 hours something we need to fix"<a href="#t13:34" class="time">13:34</a>
f13plus confusion in the masses as to what the release date really is<a href="#t13:34" class="time">13:34</a>
f13mmcgrath: We should always consider making things faster.<a href="#t13:34" class="time">13:34</a>
mmcgrathbut at what cost!  Oh the humanity!<a href="#t13:34" class="time">13:34</a>
f13I think we could publish torrent layouts early or ask for mirrors officailly to help torrent.<a href="#t13:34" class="time">13:34</a>
f13so I'm interested in looking at improving that.  I don't thin kjust releasing the torrent to the general public is the way to achieve this though.<a href="#t13:35" class="time">13:35</a>
wwoodsI think getting some mirrors to seed the torrents might be the smarter thing to do<a href="#t13:35" class="time">13:35</a>
mmcgrath<nod><a href="#t13:35" class="time">13:35</a>
wwoodsI don't think breaking the release dates is a good idea<a href="#t13:35" class="time">13:35</a>
mmcgrathwe could limit early release to people with a fedoraproject.org account and ask them to start seeding early.<a href="#t13:35" class="time">13:35</a>
warrenmmcgrath, how do you limit that?<a href="#t13:35" class="time">13:35</a>
f13Proposal:  Investigate with Infrastructure team ways of getting more mirrors to participate in seeding the torrent (early?).  Do not release torrent to general public before the agreed upon coordinated release date/time.<a href="#t13:36" class="time">13:36</a>
warrenmmcgrath, torrent has no login<a href="#t13:36" class="time">13:36</a>
poelcathow much impact would there be if 20-30 individ people had adv access to GA IOS and started their own seed at the minute release went live? not before<a href="#t13:36" class="time">13:36</a>
mmcgrathwarren: by restricting access to the .torrent file<a href="#t13:36" class="time">13:36</a>
warrenmmcgrath, so it could be leaked easily<a href="#t13:36" class="time">13:36</a>
mmcgrathwarren: your right, lets do nothing.<a href="#t13:37" class="time">13:37</a>
warrenthat'll work though<a href="#t13:37" class="time">13:37</a>
nottingpoelcat: such as.. mirrors?<a href="#t13:37" class="time">13:37</a>
warrenmmcgrath, I'm not suggesting that we do nothing.<a href="#t13:37" class="time">13:37</a>
nirikwith f7 it was leaked and appeared on some of the torrent sites... torrentspy, etc.<a href="#t13:37" class="time">13:37</a>
poelcatnotting: no idividual people like you and me... running from home<a href="#t13:37" class="time">13:37</a>
nirik(before release)<a href="#t13:37" class="time">13:37</a>
rdieterf13: +1<a href="#t13:37" class="time">13:37</a>
warrenLeaks are OK IMHO, but we better make sure the ACTUAL release is leaked.<a href="#t13:38" class="time">13:38</a>
warren(People don't receive a trojaned copy, or not quite final.)<a href="#t13:38" class="time">13:38</a>
nottingi think mirrors would do better<a href="#t13:38" class="time">13:38</a>
f13I do too<a href="#t13:38" class="time">13:38</a>
mmcgrathf13: sounds good to me.<a href="#t13:38" class="time">13:38</a>
f13mirrors can make use of symlinks/hardlinks whatever to participate, mirrormanager could potentially help<a href="#t13:38" class="time">13:38</a>
warrenWhen I ran a mirror with DS-3 bandwidth (45mbit) and I joined the user swarm, I was downloading FC1 ISO's at like 7MB/sec.<a href="#t13:39" class="time">13:39</a>
warrenhmm... that doesn't add up.  we might have had an OC-3 by then.<a href="#t13:39" class="time">13:39</a>
rdieterafk 15 mins...<a href="#t13:40" class="time">13:40</a>
* f13 pushes for more votes.<a href="#t13:40" class="time">13:40</a>
warrenf13, +1<a href="#t13:41" class="time">13:41</a>
f13warren, notting, jeremy, wwoods ?<a href="#t13:41" class="time">13:41</a>
jeremysure!<a href="#t13:41" class="time">13:41</a>
wwoods+1<a href="#t13:41" class="time">13:41</a>
f13poelcat: Decision: Investigate with Infrastructure team ways of getting more mirrors to participate in seeding the torrent (early?).  Do not release torrent to general public before the agreed upon coordinated release date/time.<a href="#t13:42" class="time">13:42</a>
Idea.png f13 changed the topic of #fedora-meeting to: RELENG-Meeting - Upgrade path enforcement - RahulSundaram<a href="#t13:42" class="time">13:42</a>
f13Another mether topic<a href="#t13:42" class="time">13:42</a>
f13Supported live upgrade from last test to general release<a href="#t13:42" class="time">13:42</a>
f13---------------------------------------------------------<a href="#t13:42" class="time">13:42</a>
f13This was agreed upon for Fedora 7 test 4 in QA meetings but we didn't<a href="#t13:42" class="time">13:42</a>
f13actually announce it. It would encourage us to be more careful in<a href="#t13:42" class="time">13:42</a>
f13pushing out any updates after the last test release and it would<a href="#t13:42" class="time">13:42</a>
f13encourage more people to try it out the development branch and provide<a href="#t13:42" class="time">13:42</a>
f13us feedback.<a href="#t13:42" class="time">13:42</a>
f13I think all he's looking for here is that we set policy that upgrade path shouldn't break from that point on, and we enforce that.<a href="#t13:43" class="time">13:43</a>
warrenThis means don't pull packages from rawhide in "oops" cases?<a href="#t13:43" class="time">13:43</a>
f13THere has been some discussion around making it so that upgrade path works from day one of a release cycle though, pre-test1 even.<a href="#t13:43" class="time">13:43</a>
f13warren: or revert with epoch<a href="#t13:43" class="time">13:43</a>
wwoodsat the very least, we're saying that official support for (Last Test Release) -> (Final Release) upgrades should be considered<a href="#t13:44" class="time">13:44</a>
warrenI'm all for disallowing pulling of packages from rawhide *always*.<a href="#t13:44" class="time">13:44</a>
f13I'm somewhat lazy and and would like to set policy only from say test2 or test3 on.<a href="#t13:44" class="time">13:44</a>
warrenIf you need to undo a mistake, use epoch.   People are really overly emotional about epoch.<a href="#t13:44" class="time">13:44</a>
f13but I can be convinced otherwise.<a href="#t13:44" class="time">13:44</a>
wwoodsin theory there would only be critical fixes in that time period anyway<a href="#t13:44" class="time">13:44</a>
f13wwoods: "this time" we'll actually be good about that!<a href="#t13:44" class="time">13:44</a>
warrenWe lose testers whenever we require users to undo a mistake manually.<a href="#t13:44" class="time">13:44</a>
f13</straight_face><a href="#t13:44" class="time">13:44</a>
wwoodsright, we're saying test3 (where test3 is final) - I'm sure mether would be happy to have it at test2 or earlier<a href="#t13:44" class="time">13:44</a>
nottingf13: an addict will never change until *they* decide to change<a href="#t13:45" class="time">13:45</a>
wwoodsI'll leave it to you all to determine whether that's feasible or not<a href="#t13:45" class="time">13:45</a>
wwoodstest2 would be even better because in theory that's the feature freeze<a href="#t13:45" class="time">13:45</a>
warrenI would be OK with a compromise of test2.<a href="#t13:45" class="time">13:45</a>
wwoodsso guaranteeing that users will be able to easily upgrade from test2 to final<a href="#t13:45" class="time">13:45</a>
wwoodswill mean getting lots more testers by test2. or so we'd hope.<a href="#t13:45" class="time">13:45</a>
f13wwoods: that's not what we're saying.<a href="#t13:45" class="time">13:45</a>
warrenf13, part of this is a rule, "After test2, rawhide yum update should not require manual intervention."?<a href="#t13:46" class="time">13:46</a>
f13wwoods: all I'm talking about right now is enforceing upgrade paths.<a href="#t13:46" class="time">13:46</a>
f13warren: I don't know if we can really ensure that either<a href="#t13:46" class="time">13:46</a>
nottingf13: how is that different? why wouldn't upgrades work?<a href="#t13:46" class="time">13:46</a>
* lmacken strolls in<a href="#t13:46" class="time">13:46</a>
f13notting: %post fuckups<a href="#t13:46" class="time">13:46</a>
wwoodsyeah, why wouldn't that work? assuming that you've already installed test2 by a supported method..<a href="#t13:46" class="time">13:46</a>
f13other such things.<a href="#t13:46" class="time">13:46</a>
f13beyond the "fixing" of here's a new package<a href="#t13:46" class="time">13:46</a>
f13I don't want to put us on the line for some things that may be out of our control.  I'm happy to enforce upgrade paths, which goes a long long way to this, but I'm not going to say we can prevent manual intervention completely<a href="#t13:47" class="time">13:47</a>
warrenf13, i'm unclear on what exactly this proposal would mean.<a href="#t13:47" class="time">13:47</a>
f13warren: from say test2 on, any new package build has to have a clean upgrade path from the previous released package<a href="#t13:48" class="time">13:48</a>
warrenf13, will we stop the insanity of undoing a %{version} upgrade by removing it from rawhide and expecting users to manually downgrade?<a href="#t13:48" class="time">13:48</a>
f13no making builds dissappear, no downgrades without epoch<a href="#t13:48" class="time">13:48</a>
abadger1999Right. things like broken deps could also need manual intervention.<a href="#t13:48" class="time">13:48</a>
warrenf13, is that the entire scope of this proposal?<a href="#t13:48" class="time">13:48</a>
f13warren: are you looking for something else?<a href="#t13:49" class="time">13:49</a>
warrenno<a href="#t13:49" class="time">13:49</a>
warrenI'd be happy with that.<a href="#t13:49" class="time">13:49</a>
warren+1 for after test2<a href="#t13:49" class="time">13:49</a>
f13Can we come up with a reasonable cutoff point?  forever?  test1/2/3?<a href="#t13:49" class="time">13:49</a>
warrenI would prefer forever, but test2 is a good compromise.<a href="#t13:49" class="time">13:49</a>
warrenor even test1<a href="#t13:49" class="time">13:49</a>
wwoodstest2 seems like a reasonable compromise for the first run for this<a href="#t13:50" class="time">13:50</a>
wwoodstest1 is a bit early - it's pre-feature freeze so we can't guarantee there won't be big changes between test1 and test2<a href="#t13:50" class="time">13:50</a>
f13perhaps rel-eng can suggest test2 to FESCo, and they'll ratify/adjust as necessary.<a href="#t13:50" class="time">13:50</a>
f13wwoods: good point, like packages going away<a href="#t13:50" class="time">13:50</a>
warrenWho opposes "forever"?<a href="#t13:50" class="time">13:50</a>
wwoodssince the goal of this change is to make upgrades from test->final much easier, we can't necessarily hold those promises for anything pre-feature-freeze<a href="#t13:51" class="time">13:51</a>
f13warren: I do.<a href="#t13:51" class="time">13:51</a>
warrenf13, do you similarly oppose test1?<a href="#t13:51" class="time">13:51</a>
f13yeah, pretty much the same reasons.<a href="#t13:51" class="time">13:51</a>
f13up to test2 things are going to be somewhat fast and loose.<a href="#t13:51" class="time">13:51</a>
warrenbut you're OK with the test2 compromise?<a href="#t13:51" class="time">13:51</a>
f13yeah, I'm fine with test2 on<a href="#t13:52" class="time">13:52</a>
warrenOK, let's do it.<a href="#t13:52" class="time">13:52</a>
* f13 is interested in notting and jeremy's opinion.<a href="#t13:52" class="time">13:52</a>
warrenjeremy said "sure!"<a href="#t13:53" class="time">13:53</a>
nottingi'd sort of like earlier, but there are occasionally going to be things that have to 'go away'<a href="#t13:53" class="time">13:53</a>
jeremyas a general thing, it's fine... but as notting says, there are always going to be exceptions<a href="#t13:53" class="time">13:53</a>
mclasenof course, it doesn't mean that we shouldn't strive to have it at all times<a href="#t13:54" class="time">13:54</a>
nottingexactly<a href="#t13:54" class="time">13:54</a>
nottingwe need to make things easier for users, not harder<a href="#t13:55" class="time">13:55</a>
jeremyyep, I'm 110% agreed.  just want to make sure we're realistic about the fact that we're not going to hit 100%<a href="#t13:55" class="time">13:55</a>
f13well, if that<a href="#t13:55" class="time">13:55</a>
* warren tries to understand the logic of 110% not reaching 100%.<a href="#t13:56" class="time">13:56</a>
f13's the desire, should we just set policy to have upgrade paths from day one, and only make exceptions when necessary?<a href="#t13:56" class="time">13:56</a>
jeremyf13: yep<a href="#t13:56" class="time">13:56</a>
warrenf13, +1<a href="#t13:57" class="time">13:57</a>
warrenf13, does this mean builds in F7 should fail if it isn't already superceded by a build in devel now?<a href="#t13:57" class="time">13:57</a>
jwb_gonewhat do you do with packages that don't make it<a href="#t13:58" class="time">13:58</a>
f13jwb_gone: untag them!<a href="#t13:58" class="time">13:58</a>
f13oh wait...<a href="#t13:58" class="time">13:58</a>
warrenjwb_gone, concrete example?<a href="#t13:58" class="time">13:58</a>
jwb_gonehow does that fix the repo?<a href="#t13:58" class="time">13:58</a>
nottingwarren: huh?<a href="#t13:58" class="time">13:58</a>
jwb_gonefoo-1.1.2.fc6 > foo-1.1.1.fc7<a href="#t13:58" class="time">13:58</a>
jwb_gonewhat do you do in that case?<a href="#t13:58" class="time">13:58</a>
f13we're talking about devel, not f7<a href="#t13:58" class="time">13:58</a>
f13warren: ^^<a href="#t13:59" class="time">13:59</a>
warrenjwb_gone, epoch<a href="#t13:59" class="time">13:59</a>
jwb_gonewas an example.  s/fc6/fc7 s/fc7/fc8<a href="#t13:59" class="time">13:59</a>
warrenPeople really need to get over their fear of epoch.  It is better to use epoch than to break automated upgrades and to lose testing.<a href="#t13:59" class="time">13:59</a>
jwb_gonemether's option was to pull that package from the repo in that case.  i think that is wrong<a href="#t13:59" class="time">13:59</a>
warrenYes, mether is wrong in that case.<a href="#t13:59" class="time">13:59</a>
f13who does the epoch and build though?  Should we just file bugs and expect people to fix it, or will rel-eng have to go on a warpath at some point and fix all these things ourselves?<a href="#t14:00" class="time">14:00</a>
f13policy is one thing, but how does one enforce policy?<a href="#t14:00" class="time">14:00</a>
warrenf13, before brew, beehive would reject builds if N-V-R went backwards.<a href="#t14:01" class="time">14:01</a>
warrenWe could do something like that.<a href="#t14:01" class="time">14:01</a>
wwoodsat some point we should have an rpm-diff tool running for proposed updates<a href="#t14:02" class="time">14:02</a>
wwoodsand that's something it should test for<a href="#t14:02" class="time">14:02</a>
warrenf13, I don't see anything wrong with enforcing N-V-R progression in the buildsystem itself.<a href="#t14:02" class="time">14:02</a>
wwoodsbut yeah, putting that into the buildsys works<a href="#t14:03" class="time">14:03</a>
mclasenwarren: you mean E-V-R ?<a href="#t14:03" class="time">14:03</a>
f13ok, so we set policy now, and ask for koji enhancements to block a build of nevr goes backwards, for some value of backwards.<a href="#t14:04" class="time">14:04</a>
warren"N-V-R ADVANCEMENT ERROR: foo-1.2.3-2.fc8 is older than foo-1.2.4-1.fc8, please consider releasing a new version upstream or use an Epoch."<a href="#t14:04" class="time">14:04</a>
warrenmclasen, yeah<a href="#t14:04" class="time">14:04</a>
f13Proposal:  Recommend to FESCo policy that upgrade path never be compromised either by removing builds or NEVR regressions.  File RFE in Koji to enforce rule at build time<a href="#t14:05" class="time">14:05</a>
warren"E-V-R ADVANCEMENT ERROR: foo-1.2.5-1.fc7 is newer than foo-1.2.4-1.fc8, please upgrade the version in F8 before F7."<a href="#t14:06" class="time">14:06</a>
nottingsorry. gotta head out<a href="#t14:06" class="time">14:06</a>
mclasenf13: does that affect our flexibility wrt to building updates in different tags, if they inherit from each other ?<a href="#t14:07" class="time">14:07</a>
f13mclasen: example?<a href="#t14:07" class="time">14:07</a>
warrenmclasen, you mean FC6, F7 and devel for example?<a href="#t14:07" class="time">14:07</a>
mclasenyeah<a href="#t14:07" class="time">14:07</a>
f13warren: I'm not quite sure I want to do forward blocking yet.<a href="#t14:07" class="time">14:07</a>
warrenmclasen, version-release-disttag should be sufficient to prevent problems<a href="#t14:08" class="time">14:08</a>
f13warren: because it's more important to get a security update out for a releaesd product than it is for rawhide.<a href="#t14:08" class="time">14:08</a>
mclasenI was thinking about a situation where I build something for f7 first, and then my next devel build has to be newer than that<a href="#t14:08" class="time">14:08</a>
mclasenprobably not a problem in practise<a href="#t14:08" class="time">14:08</a>
warrenf13, if both F7 and F8 are out as stable distros that might become a good idea.<a href="#t14:08" class="time">14:08</a>
f13warren: that's more enforcement at update push time, rather than build time<a href="#t14:08" class="time">14:08</a>
warrenmclasen, I'm thinking the buildsys only enforces forward E-V-R progression for a "stable" distro build if both distros are stable.<a href="#t14:08" class="time">14:08</a>
f13warren: lets not dictate what order a maintainer builds things in.<a href="#t14:09" class="time">14:09</a>
warrenf13, I don't see a problem with doing it between two stable releases, when F7 and F8 are both released.<a href="#t14:09" class="time">14:09</a>
f13warren: and really, upgrading from a releaed f7 to a released f8 is going to have issues anyway<a href="#t14:09" class="time">14:09</a>
warrensure, but there shouldn't be any reason why a package in F8 should be < than F7<a href="#t14:09" class="time">14:09</a>
f13becuase currently your upgrade to f8 via supported methods is anaconda, and you can't add update repos then.<a href="#t14:10" class="time">14:10</a>
f13warren: at release of hte udpate time sure.  At build time, no.<a href="#t14:10" class="time">14:10</a>
warrenf13, how is this a problem really?<a href="#t14:10" class="time">14:10</a>
f13warren: what right do we have to dictate what order people do their builds in?<a href="#t14:10" class="time">14:10</a>
warrenf13, most cases where N+1 < N in extras were accidents.<a href="#t14:10" class="time">14:10</a>
warrenf13, OK, I guess we can consider this later.<a href="#t14:11" class="time">14:11</a>
f13so long as the update requests land at the right time I'm happy.  I could give a shit what order they built the updates in.<a href="#t14:11" class="time">14:11</a>
jwb_gonef13, there's nothing to prevent fc-6 builds from going higher<a href="#t14:11" class="time">14:11</a>
jwb_goneit doesn't use koji<a href="#t14:11" class="time">14:11</a>
f13jwb_gone: not forever<a href="#t14:11" class="time">14:11</a>
f13so I didn't see any votes on the last proposal<a href="#t14:12" class="time">14:12</a>
f13<f13> Proposal:  Recommend to FESCo policy that upgrade path never be compromised either by removing builds or NEVR regressions.  File RFE in Koji to enforce rule at build time<a href="#t14:12" class="time">14:12</a>
* bpepple gives a +1 from the peanut gallery.<a href="#t14:13" class="time">14:13</a>
warrenf13, what happened to the test2 part?<a href="#t14:13" class="time">14:13</a>
warrenf13, or at least the mention the "get an exception"?<a href="#t14:13" class="time">14:13</a>
warrenwho grants exceptions?  who makes that choice?<a href="#t14:13" class="time">14:13</a>
f13warren: I adjusted it to be forever.<a href="#t14:13" class="time">14:13</a>
wwoodswe were convinced that "forever" is workable and useful<a href="#t14:14" class="time">14:14</a>
f13warren: rel-eng grants exception, dealt with on case by case basis.<a href="#t14:14" class="time">14:14</a>
warrenf13, ok.<a href="#t14:14" class="time">14:14</a>
rdieterf13: +1 proposal<a href="#t14:14" class="time">14:14</a>
warrenf13, +1<a href="#t14:14" class="time">14:14</a>
wwoods+1<a href="#t14:15" class="time">14:15</a>
f13jeremy: ?<a href="#t14:15" class="time">14:15</a>
f13well, he was pretty happy with this before.<a href="#t14:15" class="time">14:15</a>
f13poelcat: Decision: Recommend to FESCo policy that upgrade path never be compromised either by removing builds or NEVR regressions.  File RFE in Koji to enforce rule at build time<a href="#t14:16" class="time">14:16</a>
Idea.png f13 changed the topic of #fedora-meeting to: RELENG-Meeting - Open Discussion<a href="#t14:16" class="time">14:16</a>
f13that's the end of the agenda.  We're over time.  ANy pressing things?<a href="#t14:16" class="time">14:16</a>
wwoodsdo we have an official (or draft) policy for updates-testing stuff?<a href="#t14:17" class="time">14:17</a>
warrenf13, until we're able to enable more rel-eng people to do pushing, could we have pushing done a little more often?<a href="#t14:18" class="time">14:18</a>
f13wwoods: all we decided on last week wwas to allow maintainers to skip updates-testing if they wanted to.<a href="#t14:18" class="time">14:18</a>
warrenf13, sometimes pushing has waited on 2-3 days for example.  That has to be frustrating to our contributors.<a href="#t14:18" class="time">14:18</a>
f13warren: yeah, I'm trying to do the best we can here.  Once a day isn't totally unreasonable.<a href="#t14:18" class="time">14:18</a>
f13um, since release of f7 I don't believe I've gone for more than 2 days without a push<a href="#t14:19" class="time">14:19</a>
warrenAs long as we don't go beyond 1 day I think we're OK.<a href="#t14:19" class="time">14:19</a>
Idea.png f13 changed the topic of #fedora-meeting to: <a href="http://fedoraproject.org/wiki/Communicate/FedoraMeetingChannel">http://fedoraproject.org/wiki/Communicate/FedoraMeetingChannel</a> -- Meetings often get logged -- see schedule in the wiki for next meeting<a href="#t14:20" class="time">14:20</a>
f13thanks all.<a href="#t14:20" class="time">14:20</a>

Generated by irclog2html.py 2.3 by Marius Gedminas - find it at mg.pov.lt!