[12:40:43] I'm about to merge https://gerrit.wikimedia.org/r/c/operations/puppet/+/934309/ which makes a small change to the webproxy Squid config. it was reviewed by multiple people, but let me know if that breaks anything :) [13:03:23] <_joe_> dhinus: uhm the webproxy, not url-downloader right? [13:07:06] <_joe_> yep, answering myself, that is the case [14:49:45] jbond: I am going to separate the timesyncd and resolv.conf patches, since the common change is a separate commit anyway, just as an FYI as I am going to abandon and submit again [14:51:55] or rather let's merge timesyncd one first, running PCC [14:51:57] sukhe: ack sgtm [14:52:19] jbond: this affects all hosts that calle C:systemd::timesyncd, 2059 in total. want me to run on all :) [14:52:41] sukhe: the $::ntp_peers change is a very simple one to make [14:52:49] im not sure it needs s epertae change [14:53:03] oh wait ignore me let me take a second look [14:53:03] yeah I already did that [14:53:07] https://puppet-compiler.wmflabs.org/output/933473/42228/ [14:53:14] see this and the latest patchest [14:54:08] sukhe: as for pcc id do a full run just to be on the safe side [14:54:26] yeah! [14:54:31] let's do that just to be extra safe [14:54:35] sgtm [14:55:31] sukhe: fyi if you are doing a $randome set of hosts like in https://puppet-compiler.wmflabs.org/output/933473/42228/ also add the word 'basic' [14:55:44] this means it will also get tested on at least one wmcs host as well [14:56:22] cool thanks, for now I am doing cumin:P{C:systemd::timesyncd} [14:56:25] basic expads to the pcc-worker host running the test and one of the sretest hosts its a good one to use for easing out comon issue [14:56:33] sukhe: [14:56:35] no [14:56:37] oh? [14:56:49] that will run on everything pcc knows about [14:57:02] so like 3000 hosts [14:57:05] I thought that's what you meant by a "full run" above and that's what I had in mind [14:57:08] well 2059 :) [14:57:09] but yeah! [14:57:18] better to just `Hosts: C:systemd::timesyncd` [14:57:44] withthat instead of running on every mw server for instance it will just pick one joberunner, one api etc [14:57:53] or just one dnsbox instead of all of them [14:58:20] the dnsbox don't pick these up anyway since timesyncd is set to false there [14:58:26] pcc is better at storing things but still 2059 (and thats before we have started n the cloud nodes) is still a lot [14:58:30] (talking about the NTP peers/timesyncd change) [14:59:34] sukhe: yes but i mean it just picks one host per role. there is not a lot of value in running pcc on every mw serveres as there are only three different roles [14:59:50] so we only need to run pcc 3 times not 340 [14:59:55] ah in that way right [15:00:14] using `Hosts: C:systemd::timesyncd` [15:00:14] and for the resolv.conf change, that one is easy, it's just limited to the dns hosts [15:00:18] achives that filtering [15:00:29] its not perfect but for this change its more then enough [15:01:04] got it, I misread what you are saying above and confused it with the DNS change [15:01:12] ack no probs [15:02:21] ftr i very rarley use the cumin syntax, theses days i tenmd to use auto unless that touches everything in which case i use basic while developing then do a run with autop once i have got most issues [15:03:35] yeah usually auto is fine but since it restricts itself to one host? [15:04:28] auto works the same as described above i.e. one host per role [15:04:43] in this case auto should work exactly the same as setting C:systemd::timesyncd [15:05:05] one host per role, right [15:05:42] also i agree the dns case is a good case where auto is not the best option as you really do want to check more all hosts have the correct config pulled in from hiera or dynamically set [15:05:58] I agree that sometimes because of hiera settings hosts within the same role are different [15:05:59] yeah, I think this is a tricky change [15:06:06] it's simple but it changes a lot of things [15:06:10] maybe we could add an 'all' option that selects all hosts imatching [15:06:10] so better safe than sorry :) [15:06:18] I will also be super careful with the rollout [15:06:31] disable puppet, run manually on each host (for the DNS ones) [15:06:35] volans: i think for that use the cumin syntax, i want to avoid people defaulting to all [15:06:55] that works toot [15:07:06] e.e Hosts: cumin:A:all [15:07:15] jbond: I don't mean all hosts [15:07:20] I mean all matching the change [15:07:27] instead of one per role [15:07:28] yes i know buyt e.g. cumin A:dnsbox [15:07:31] sure [18:06:54] godog: if you're availalbe today, could use a merge on https://gerrit.wikimedia.org/r/c/operations/puppet/+/934710. Otherwise, I'll pull submodule one time by hand on the server.