| jhesketh | jeblair: ah cool. Will look shortly :-) | 00:03 |
|---|---|---|
| *** saneax is now known as saneax-_-|AFK | 00:04 | |
| *** harlowja has quit IRC | 00:04 | |
| openstackgerrit | James E. Blair proposed openstack-infra/zuul: Add infrastructure for pre/post playbooks https://review.openstack.org/425989 | 00:10 |
| jeblair | jhesketh: thanks... and you can tell me if that ^ fits in with how you saw it being used | 00:11 |
| jlk | With OpenStack's implementation of gerrit/zuul, Im looking at the layout, trying to determine if the system blocks on -1 reviews or -2 reviews, or if it's just human convention to do so. It looks like there are no rejects on the "gate" pipeline, there's only a requirement on verified 1 or 2 from jenkins, and a "workflow: 1". Is it just human convention to not give a workflow +1 if there are negative reviews, or does gerrit prevent that somehow? | 00:21 |
| jlk | (asking to read responses later, going afk) | 00:21 |
| jeblair | jlk: convention | 00:21 |
| jeblair | jlk: no -2 code reviews is a requirement. i believe it's not encoded in the zuul config because zuul can detect that from gerrit. | 00:24 |
| jeblair | jlk: (so to clarify: no -1 code-reviews is a convention (i mean, it's not even a convention stated like that, it's just input for a core reviewer to consider). no -2 code-reviews is a system requirement) | 00:25 |
| *** bhavik1 has joined #zuul | 04:26 | |
| *** saneax-_-|AFK is now known as saneax | 05:40 | |
| *** bhavik1 has quit IRC | 05:56 | |
| *** yolanda has quit IRC | 07:12 | |
| *** yolanda has joined #zuul | 07:20 | |
| *** hashar has joined #zuul | 09:01 | |
| *** isaacb has joined #zuul | 10:50 | |
| *** isaacb has quit IRC | 11:01 | |
| *** saneax is now known as saneax-_-|AFK | 11:24 | |
| SpamapS | jeblair: the talk went well btw. There were no less than _3_ Zuul users in the audience.. which surprised me. :-D | 14:26 |
| *** saneax-_-|AFK is now known as saneax | 14:29 | |
| *** saneax is now known as saneax-_-|AFK | 14:40 | |
| openstackgerrit | James E. Blair proposed openstack-infra/zuul: Add infrastructure for pre/post playbooks https://review.openstack.org/425989 | 15:28 |
| openstackgerrit | Clint 'SpamapS' Byrum proposed openstack-infra/zuul: Re-enable test_dependent_behind_dequeue https://review.openstack.org/413768 | 15:49 |
| *** saneax-_-|AFK is now known as saneax | 15:49 | |
| *** yolanda has quit IRC | 15:51 | |
| *** yolanda has joined #zuul | 15:52 | |
| openstackgerrit | Clint 'SpamapS' Byrum proposed openstack-infra/zuul: Re-enable test_dependent_behind_dequeue https://review.openstack.org/413768 | 16:19 |
| *** saneax is now known as saneax-_-|AFK | 16:21 | |
| jeblair | SpamapS: hooray! :) | 16:30 |
| SpamapS | jeblair: I notice 413768 conflicts with your stack-of-destiny ... should I rebase on top? | 16:33 |
| jeblair | SpamapS: an experimental cherry-pick has it applying clean, so i wouldn't worry about it | 16:34 |
| jeblair | i think gerrit is conservative in that calculation | 16:35 |
| SpamapS | jeblair: oh good | 16:44 |
| SpamapS | time to go find some dumplings and fried cheeze #CzechHealthFood | 16:59 |
| openstackgerrit | James E. Blair proposed openstack-infra/zuul: Move job running into own class https://review.openstack.org/385964 | 16:59 |
| openstackgerrit | James E. Blair proposed openstack-infra/zuul: Add infrastructure for pre/post playbooks https://review.openstack.org/425989 | 16:59 |
| openstackgerrit | James E. Blair proposed openstack-infra/zuul: Implement job aborting https://review.openstack.org/426306 | 16:59 |
| *** hashar has quit IRC | 17:09 | |
| openstackgerrit | James E. Blair proposed openstack-infra/zuul: Implement job aborting https://review.openstack.org/426306 | 17:13 |
| jlk | jeblair: where is the "no -2" codified? I don't see it in project-config/zuul/layout.yaml | 17:56 |
| clarkb | jlk: in gerrit | 17:56 |
| clarkb | zuul does a "can this thing merge" query against gerrit, and gerrit has rules that include no merges if you have a max negative vote | 17:57 |
| jlk | OIC | 17:58 |
| jlk | Github has a "mergable" query, but I think its limited in what it decides against | 17:59 |
| clarkb | worth noting that same rule applies to verified and workflow columns for us too | 18:01 |
| clarkb | so its more generic than just -2 code review | 18:01 |
| *** harlowja has joined #zuul | 18:02 | |
| jlk | yeah, it's just gerrit that does it, not zuul. | 18:02 |
| jlk | Do you know how gerrit does it? Is it strictly /any/ -1 will block, or does it have to be more -1s than +1s? or..? | 18:02 |
| jeblair | jlk: theoretically -- anything. it's user-customizable prolog. :| | 18:03 |
| jlk | oof. | 18:03 |
| jlk | alright. | 18:03 |
| jeblair | jlk: but the standard is... | 18:03 |
| clarkb | jlk: Gerrit's default "function" (I think that is the term they use) is at least one maximum positive votes in all columns and no maximum negative votes in any columns | 18:03 |
| jeblair | that ^ | 18:03 |
| clarkb | they ship several functions that you can apply but you can also write custom prolog to provdie the yeses as appropriate for you | 18:04 |
| jlk | yeah, just trying to replicate some of it over in zuul+github land. Trying to define the rules, at least for our implementation of it | 18:05 |
| jeblair | i don't consider the zuul-gerrit negotiation over this to be very central. zuul can duplicate most of the basic functionality. i'm somewhat inclined to remove the (somewhat clunky) part of zuul that does query gerrit for this, but it could be useful. | 18:05 |
| jlk | was thinking of trying to map "people with write access" == 2s (+/-) and people without would be 1s (+/-), and figuring out which should block. | 18:06 |
| jeblair | jlk: well, zuul should just expose what's available in github and let operators configure that in the layout. | 18:06 |
| jeblair | don't translate to gerrit, just expose github | 18:06 |
| jlk | yeah, I'm trying to assign some special meaning to reviews by people with write access though | 18:06 |
| jlk | because those are what can block (or allow) a merger | 18:07 |
| jlk | whereas people without write access cannot block | 18:07 |
| jlk | I want to expose that difference to zuul, so that layouts can decide how to treat them | 18:07 |
| jlk | (instead of maintaining a list of special people in zuul to match against) | 18:07 |
| jeblair | jlk: that makes sense -- i would just look for a way of expressing that in those terms. "require: {score: positive; group:owner}" or something. | 18:08 |
| jlk | that would make sense for v3, where there is more flexibility. For v2.5 I'm trying to shove it into the same construct as gerrit approvals | 18:08 |
| jlk | so that I don't have to touch nearly as much core zuul code | 18:08 |
| jeblair | jlk: v3 doesn't change that | 18:09 |
| jlk | Well, I'm hoping it will, that's why I filed a storyboard on it. | 18:10 |
| jlk | mostly it's just in zuul/model.py for 2.5 | 18:11 |
| jlk | in ChangeishFilter. | 18:11 |
| jeblair | the *intent* is that provides a generic base for supporting multiple systems. you're working on the first alternative system and we know some things will need to change. | 18:12 |
| jlk | yup. Doing this on 2.5 now is helping us to find places where it may be awkward, so that we can inform v3(+) direction. | 18:13 |
| jeblair | so i guess the thing i'm trying to convey is that there is no specific design that has gone into v3 to make this easier | 18:13 |
| jeblair | we just need to do it. :) | 18:13 |
| jlk | agreed | 18:13 |
| jeblair | at any rate, i look forward to working on this once we have a minimally viable v3 running | 18:14 |
| jlk | me too | 18:14 |
| jeblair | jlk: until then, i'm going to continue to try to pull you into v3 things. :) | 18:15 |
| jlk | :) I've been doing some reviews... and thinking real hard about it, I swear! | 18:15 |
| jeblair | jlk: thanks! i think there will be some more interesting ansible stuff next week. | 18:17 |
| clarkb | speaking of, rcarrillocruz any chance you will be able to address the comment on https://review.openstack.org/#/c/404243/13 so that we can close out that portion of the devstack-gate work? | 18:19 |
| jeblair | jamielennox: 406699 has some comments (and could use an update) | 18:21 |
| jeblair | i should probably send that again in a couple of days :) | 18:22 |
| jlk | heh yeah | 18:22 |
| *** bhavik1 has joined #zuul | 19:15 | |
| *** nigelb has quit IRC | 19:17 | |
| *** bhavik1 has quit IRC | 19:38 | |
| openstackgerrit | Merged openstack-infra/zuul: Merger: fully update repo on repo update https://review.openstack.org/389850 | 20:46 |
| openstackgerrit | James E. Blair proposed openstack-infra/zuul: Cloner: use zuul_url always when project set https://review.openstack.org/389851 | 21:02 |
| Shrews | sometimes I *really* hate git. Especially when I back myself into a corner that I cannot find a way out of, like right now | 21:40 |
| openstackgerrit | David Shrewsbury proposed openstack-infra/nodepool: WIP: Assign node set to node requests https://review.openstack.org/426406 | 21:41 |
| Shrews | jeblair: When you get some time, could a take a look at 426406 ^^^. Really, only the _getNodeSet() part is what I'd like your input on, and how it's using the new NodeLaunchManager to do the thread-per-launch handling we discussed yesterday | 21:45 |
| Shrews | jeblair: i tried to come up with an easily understandable way to do that thread handling, and that's what my brain finally came up with (using the manager to hide the complexity) | 21:46 |
| clarkb | Shrews: the trick I have found is thinking about what you want to do to the git dag, unfortunately some times this maens you have to learn far more about git than should be necessary | 22:06 |
| Shrews | clarkb: In this instance, I got myself so lost that the only recourse I could think of was to reset the dag. | 22:15 |
| Shrews | i can't even explain what i did to get so lost | 22:16 |
| *** yolanda has quit IRC | 22:20 | |
| *** yolanda has joined #zuul | 22:31 | |
| jeblair | Shrews: that lgtm -- the process makes a lot of sense. related: i think that we should consider having noderequestworker not be a thread (in the simple case of a 1-node request we will have 2 threads, compared to 1 today, and we're already running into thread contention problems. distributed provider managers alleviate that somewhat, but i still don't want to lose the ground we've gained. the requestworker is a pretty good candidate to not be ... | 22:38 |
| jeblair | ... a thread since it's fairly straightforward -- do some zk operations, then have the nodelaunchmanager spawn launch threads if needed, then wait for them. that could become a state machine pretty easily.) having said all that, this is very easy to reason about right now, so if you want to proceed with this and visit de-threadifying it later, that works for me. | 22:38 |
| jeblair | Shrews: (the main thing about this patch that would need to change is the launch_manager.wait() would become a polling method instead of a blocking method. not too hard.) | 22:39 |
| Shrews | jeblair: *nod* and thanks. let me mull on that bit over the weekend | 22:40 |
| *** yolanda has quit IRC | 22:44 | |
| *** yolanda has joined #zuul | 22:49 | |
| *** yolanda has quit IRC | 22:54 | |
| *** saneax-_-|AFK is now known as saneax | 22:59 | |
| *** yolanda has joined #zuul | 23:00 | |
| *** greghaynes has quit IRC | 23:10 | |
| *** yolanda has quit IRC | 23:16 | |
| *** yolanda has joined #zuul | 23:34 | |
| *** yolanda has quit IRC | 23:36 | |
| *** greghaynes has joined #zuul | 23:37 | |
| *** yolanda has joined #zuul | 23:40 | |
| *** yolanda has quit IRC | 23:48 | |
| *** greghaynes has quit IRC | 23:52 | |
| openstackgerrit | James E. Blair proposed openstack-infra/zuul: WIP: Reload tenant configuration on change merged https://review.openstack.org/426433 | 23:55 |
| jeblair | friday afternoon brain twister ^ | 23:56 |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!