17:59:45 #startmeeting Infrastructure (2017-05-04) 17:59:45 Meeting started Thu May 4 17:59:45 2017 UTC. The chair is smooge. Information about MeetBot at http://wiki.debian.org/MeetBot. 17:59:45 Useful Commands: #action #agreed #halp #info #idea #link #topic. 17:59:45 The meeting name has been set to 'infrastructure_(2017-05-04)' 17:59:45 #meetingname infrastructure 17:59:45 The meeting name has been set to 'infrastructure' 17:59:45 #topic aloha 17:59:45 #chair smooge relrod nirik abadger1999 dgilmore threebean pingou puiterwijk pbrobinson 17:59:45 Current chairs: abadger1999 dgilmore nirik pbrobinson pingou puiterwijk relrod smooge threebean 17:59:47 hi 17:59:59 * pingou around but likely not for long 18:00:02 * cyberworm54_ here 18:00:05 * threebean waves 18:00:15 here 18:00:22 hi 18:00:32 .hello jcline 18:00:33 jcline: jcline 'Jeremy Cline' 18:00:57 .hello bowlofeggs 18:00:58 bowlofeggs: bowlofeggs 'Randy Barlow' 18:01:15 hello everyone 18:01:16 .hello nb 18:01:17 nb: nb 'Nick Bebout' 18:01:24 morning 18:01:44 #topic New folks introductions 18:01:51 Hi do we have any new people here today? 18:02:03 smooge: I am new. 18:02:07 hi my first meeting but I joined in Feb 18:02:44 hello to you both. 18:02:54 hello!! 18:04:01 also my first ever irc meeting 18:04:23 good to meet you all. 18:04:23 welcome! 18:04:48 axk4545: :) 18:05:07 great to finally meet all of you as well 18:07:29 good to meet you also 18:07:31 next up 18:07:41 #topic announcements and information 18:07:42 #info beta freeze will start 2017-05-17 18:07:42 #info infra hackfest in RDU 2017-05-08 to 2017-05-12 - everyone 18:07:42 #info Mass update/reboot cycle completed - everyone 18:07:42 #info new nagios with templates deployed 18:07:51 yay! 18:07:53 we had a fun filled week here in infrastructure land 18:07:58 indeed. 18:08:31 with the hackfest next week a lot of folks won't be around as much as usual... so keep that in mind and file tickets or email if you can't find anyone. 18:09:01 I am rewriting /git/dns to remove built/* so everyone will have to re-clone when I am done 18:09:06 I will email the list when it finishes 18:09:29 nb++ 18:09:46 oh. nb how are the contents of built/ going to be generated? 18:10:41 threebean: they have always been generated 18:10:46 but they contain dnssec signatures. 18:11:04 And that means basically the entire file gets rewritten on every small change, which makes the size grow very hard 18:11:17 * threebean nods 18:11:44 thank you nb 18:11:46 I guess my question is.. what is responsible for generating them now? (we used to do it ourselves). is it a post-receive git hook? or a cron job? or..? 18:11:50 my disk space thanks you also 18:11:54 threebean, will be the same way as normal, I will run do-domains when the commit finishes 18:11:59 threebean: the same do-domains is responsible for it 18:12:09 threebean, nothing changes, except removing old built/* from history 18:12:15 threebean: we are just cleaning up all the historical /build builds from the git repo 18:12:16 oh!!! ok. awesome :) 18:12:20 Yeah, what nb said 18:13:02 from history but not disk then? 18:13:18 No they will be removed from disk 18:13:27 they never happened 18:13:40 history... rewritten! 18:13:41 we are never to talk about them again (until we grow to gigabytes) 18:13:46 I guess I'm a little confused then 18:14:02 pingou, there are 2 trees in the git repo 18:14:13 there is /master and /built 18:14:26 hello! 18:14:27 the /built gets generated each time with the signing keys and committed 18:14:46 so it is basically like keeping track of .o files 18:14:47 right now the repo is 3.1GB 18:15:33 currently we fix that by going through history and pruning out all those .o files regularly 18:16:08 I am not sure that is the best fix, but we have usually 100,000 things to do before 'fixing DNS to be less disk wasteful' comes near to the top 18:16:33 pingou, does that help muddy the waters any? 18:16:49 so we keep /master and drop /built 18:16:55 smooge: why not set up a thing to run a cleanup after generation? 18:17:12 axk4545, it breaks existing clones when you do a cleanup 18:17:15 because it is a force push 18:17:18 but we still rebuild /master 18:17:28 and that's no longer going into /built? 18:17:29 nb: ah 18:17:38 Sorry Im late. 18:17:44 pingou, no it is still going into /built 18:17:59 we are just making the repo smaller for a while. 18:17:59 pingou, all we are doing is basically removing 2 years of date from the history in /built 18:18:04 smooge: so we're just cleaning the history and keeping the last version 18:18:10 ok gotcha 18:18:14 cool, thanks nb 18:18:16 the repo also gets cloned to each dns server on any change 18:18:22 you can get the history from /master changes 18:18:31 so that starts getting slower and taking up more space over time 18:18:32 but anything in /built is gone 18:19:41 ok anything more on this? 18:19:59 next up then 18:20:01 #topic cleanups from mass update cycle - kevin 18:20:08 nirik, you are up 18:20:24 so yeah, this weeks mass update/reboot didn't go too great. 18:20:40 I think partly because we haven't done one in about 3 months, so there was a lot of change piled up. 18:20:51 There's a few process things we can change to help: 18:21:00 * no updates pushes allowed on days we are updating things. 18:21:25 +10 18:21:30 There's also a few bugs we still need to fix 18:21:49 I think puiterwijk already fixed/filed/whatever the fedimg and autocloud ones? They failed to start right after upgrade. 18:22:08 There's also a bug with fedmsg-relay... it starts but doesn't make a socket so nagios cannot monitor it. 18:22:09 autocloud was not issue in the end, I filed fedimg, and a patch is now live 18:22:38 There was an issue with our openvpn CRL... it was expired (I guess it never used to check it?) 18:22:52 Also, the fedmsg-datanommer package suddenly added an /etc/fedmsg.d/datanommer.py thatbroke datagrepper 18:23:26 (fedmsg-datanommer-models, that is) 18:23:36 and of course we also had hardware croak. :) as it always does when rebooted. 18:23:41 can someone update /srv/web/infra/dns? /me does not have permission to 18:23:51 puiterwijk: that one surprises me, did we do a release of some sort? 18:23:54 * nb will update the nameservers 18:24:03 pingou: I do not know of any. And it surprised me too 18:24:14 nb: wrong channel, but will do that 18:24:25 oops, sorry, i meant to talk in #fedora-noc 18:24:48 I can file the relay bug... is that part of main fedmsg? 18:25:04 I believe so 18:25:42 oh, back to openvpn CRL... the default is that it expires every month. Do we really want to have to make it every month? or should we up that to something more reasonable like a year? 18:25:56 nirik: year 18:26:26 or maybe sync it with our mass upgrade cycle 18:26:33 making this a January thing, or March thing (end of FY) 18:26:49 year is fine with me. 18:27:24 nirik, maybe it could be done like letsencrypt? 18:27:27 nirik: I have set it to a year for this generation 18:27:46 nirik: however, as I said last night, normally it gets regenerated when a new cert is generated 18:27:55 In other words: when we add a new server, we get a new CRL 18:28:02 puiterwijk: ok, well, unless you put a new one in, I put one in that still had the 30 days. 18:28:13 and changed to sha256 18:28:32 I had made those same changes, just not pushed them to the repo :). I just generated it and deployed to bastion01 18:28:32 because after the outage still a bunch of machines couldn't connect. 18:28:46 * nirik can check which one is live after the meeting and we can sync git. 18:28:59 git merge nirik 18:29:58 ok any other items? 18:30:25 I think thats it. I hope I didn't forget any... 18:30:39 we can get them next mass update after beta 18:30:50 #topic dist-git-1.0 released for epel7 - clime 18:31:02 clime, you are up 18:31:16 Okay, I would like to say a few words about this new dist-git package 18:31:57 It basically contains: the upload.cgi script, setup_git_repo, and mkbranch script, then httpd configs and selinux 18:32:17 nice. 18:32:21 there is a few more things like systemd service setup for git-smart-http but that's pretty much it 18:32:49 it does not contain gitolite or any pkgdb integration but they can be plugged in easily 18:33:00 reminds me... there's a bug in upload.cgi... puiterwijk looked into it, but I don't think he solved it.... it uploads things twice 18:33:02 ths is what runs the lookaside? 18:33:17 oh ok, then that bug is also in the upstream package 18:33:30 clime: likely, but I haven't checked for sure 18:33:49 It is very least common demoninator between Fedora Infra DistGit and any other DistGit 18:33:59 like copr-dist-git. 18:34:22 I would like to ask if we could adopt it into FedoraInfra 18:34:30 nirik: that is a client-side problem 18:35:26 clime: I'm not opposed... we will have to be carefull of course to make sure we don't break anything. 18:35:38 It has configurable paths to git repos and lookaside 18:35:42 clime: the new package? 18:35:48 axk4545: yup 18:36:01 nirik: okay cool! 18:36:21 so I would like to make a very slow transition to it 18:36:52 with lots of testing alongside 18:36:58 that is always good 18:37:26 yep 18:38:30 okay, thank you. Can I talk to anyone later about details or show a small patches for a review? 18:39:14 It's really not so much work but I would like to have an extra eye if possible. 18:39:32 I would say that an email to the list covering the changes would be a good start 18:40:01 is this something on our side or on releng to mostly help implement? 18:40:07 Yeah, there is basically no change when compared to the current state of FedoraInfra. 18:40:34 but for sure, I can write up something 18:42:33 smooge: well, it's basically on our side to implement 18:42:42 but it's just few minor changes... 18:43:02 ok thanks 18:43:09 if there is nothing more on this.. 18:43:13 #topic service account for automated rebuilds? - threebean 18:43:29 threebean, you are up 18:43:42 #link https://lists.fedoraproject.org/archives/list/infrastructure@lists.fedoraproject.org/thread/YEONLKLQ2HBTCDL4OUIBYDOYL3L55JC7/ 18:44:04 just wanted to see if anyone here had any feedback on that request. 18:44:24 the gist is that we're working on a continuous rebuild service to: 18:44:30 I'm a little worried security wise... 18:44:36 oh, yeah? 18:44:44 since basically that account will be provenpackager right? 18:44:50 as things stand, yeah. 18:44:56 but as long as we are carefull and puiterwijk doesn't yell... 18:45:01 hm. 18:45:05 threebean: I'm not sure I'm comfortable with the current way things are planned 18:45:14 Maybe we can discuss this next week? 18:45:14 here's an alternative: 18:45:23 no provenpackager for this account.. 18:45:41 but instead add some special casing to the gitolite config which gives it rights only on components in the modules/ namespace. 18:45:46 like provenpackager-lite, for modules only. 18:46:28 threebean: what initiates the changes? 18:46:28 well, either way we can discuss it next week. some of us will be together for the FAD. maybe we can discuss it there? 18:46:49 nirik: in this case, a human (kevin) commits to a specfile in dist-git. 18:47:18 the system notices that on fedmsg and looks up the list of what modules have a dependency on that specfile/rpm. 18:47:51 it then kicks off a rebuild of those modules (which necessitates doing an empty commit to those modules' yaml files in dist-git) 18:47:59 and to push changes it will need a ssh key right? 18:48:03 * threebean nods 18:48:15 yeah, that's it. shouldn't need a pw. 18:48:29 but we could lock it down to only be from the mbs machines or the like. 18:48:38 we could, yeah. 18:48:57 it would be nice to have stg dist-git be a little more open for devs to test. but certainly prod. 18:49:06 I think I like the gitolite approach 18:49:27 pagure will be more complicated but we'll see when the time comes 18:49:53 pingou: you mean using it with src.fp.o? 18:49:58 * threebean nods 18:50:18 axk4545: yup 18:50:40 yeah, I'd say lets hash this out next week if we can... you are going to be in RDU threebean ? 18:50:47 I'll be there! :) 18:51:06 OK thanks threebean 18:51:07 oh nice. cool. 18:51:12 it sounds like next week will be busy 18:51:13 going to pitch a tent in the hallway outside the conference room. :) 18:51:44 \รณ/ 18:51:51 ok. thanks all! 18:51:56 Good thinking... saves on hotel. ;) 18:52:19 I want to remind people that transportation to and from the event is provided by Smoober... where riding in the back of a 20 year old truck is style 18:52:33 ok back to the main topic 18:52:43 #topic Apprentice Office 5 minutes 18:53:33 Any questions from apprentices on tickets? 18:54:29 maybe 18:54:36 one sec. 18:54:38 Wonder if I could get some eyes on my bodhi combination template by this wekend? 18:55:29 Skeer, what is the ticket? 18:57:36 ok apprentices, we are coming up to the top of the hour. I am going to ask to move the questions over to #fedora-admin 18:57:45 Thank you all for coming to the meeting 18:57:55 sorry smooge.. boss walked in ;) 18:57:56 #endmeeting