| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-operator] 894504: DNM checking CI for Publish container images to quay.io https://review.opendev.org/c/zuul/zuul-operator/+/894504 | 06:13 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-operator] 894504: DNM checking CI for Publish container images to quay.io https://review.opendev.org/c/zuul/zuul-operator/+/894504 | 07:05 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-operator] 894504: DNM checking CI for Publish container images to quay.io https://review.opendev.org/c/zuul/zuul-operator/+/894504 | 08:10 | |
| -@gerrit:opendev.org- Bernhard Berg proposed wip on behalf of Lukas Kranz: [zuul/zuul-jobs] 887917: prepare-workspace-git: Add ability to define synced pojects https://review.opendev.org/c/zuul/zuul-jobs/+/887917 | 08:47 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-operator] 894504: DNM checking CI for Publish container images to quay.io https://review.opendev.org/c/zuul/zuul-operator/+/894504 | 10:09 | |
| -@gerrit:opendev.org- Benjamin Schanzel proposed: [zuul/nodepool] 894636: Fix UnboundLocalError in error handling of runStateMachine https://review.opendev.org/c/zuul/nodepool/+/894636 | 11:00 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-operator] 894504: DNM checking CI for Publish container images to quay.io https://review.opendev.org/c/zuul/zuul-operator/+/894504 | 12:51 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-operator] 894504: DNM checking CI for Publish container images to quay.io https://review.opendev.org/c/zuul/zuul-operator/+/894504 | 12:53 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-jobs] 894755: Add feature to set --vm-driver name for minikube https://review.opendev.org/c/zuul/zuul-jobs/+/894755 | 13:35 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-jobs] 894755: Add feature to set --vm-driver name for minikube https://review.opendev.org/c/zuul/zuul-jobs/+/894755 | 13:38 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-operator] 894504: DNM checking CI for Publish container images to quay.io https://review.opendev.org/c/zuul/zuul-operator/+/894504 | 13:40 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-jobs] 894755: Add feature to set --vm-driver name for minikube https://review.opendev.org/c/zuul/zuul-jobs/+/894755 | 14:18 | |
| -@gerrit:opendev.org- daniel.pawlik https://matrix.to/#/@dpawlik:matrix.org proposed: [zuul/zuul-jobs] 894755: Add feature to set --vm-driver name for minikube https://review.opendev.org/c/zuul/zuul-jobs/+/894755 | 14:23 | |
| @jpew:matrix.org | Would someone mind a look at https://review.opendev.org/c/zuul/nodepool/+/891635 please? | 14:23 |
|---|---|---|
| @clarkb:matrix.org | element updated and all the colors changed | 17:22 |
| -@gerrit:opendev.org- Bernhard Berg proposed wip on behalf of Lukas Kranz: [zuul/zuul-jobs] 887917: prepare-workspace-git: Add ability to define synced pojects https://review.opendev.org/c/zuul/zuul-jobs/+/887917 | 17:27 | |
| @clarkb:matrix.org | corvus: tobiash ianw I'm skimming https://review.opendev.org/c/zuul/zuul/+/794688 in order to possibly resurrect it. One thing I'm confused about is that we drop the fallback of anonymous access compeltely. Should we keep that as a fallback to our fallback particularly since that means we are backward compatible? | 17:27 |
| @clarkb:matrix.org | I think we can try the app token, if that fails, check for a a user token and if not present then go anyonymous? | 17:28 |
| @jim:acmegating.com | Clark: yes, and i interpret tobiash 's comment about not adding it to all the test fixtures as supporting that position too. | 17:33 |
| @clarkb:matrix.org | ack ok I think I understand the change well enough as well as the comments and the future direction to push an update eventually. | 17:40 |
| @jpew:matrix.org | R.E. https://review.opendev.org/c/zuul/nodepool/+/891635 : Aside from our use special use case where our quota changes, it seems generally useful if someone would rather nodepool wait for (a fixed amount of) quota to become available instead of declining the request and failing a build (in particular, maybe they have way more requests than can be satisfied at once but are happy to just wait?) | 19:54 |
| @jim:acmegating.com | that's just it, nodepool doesn't decline requests when it's at quota, it pauses providers and fulfills them | 19:54 |
| @jpew:matrix.org | ``` | 19:56 |
| elif not self.hasProviderQuota(self.request.node_types): | ||
| declined_reasons.append('it would exceed quota') | ||
| ``` | ||
| I think I must be missing something | ||
| @jim:acmegating.com | a declined request is not a failed request | 19:58 |
| @jim:acmegating.com | jpew: this is ancient, but still mostly relevant to understand the protocol: https://docs.opendev.org/opendev/infra-specs/latest/specs/zuulv3.html#id1 | 20:00 |
| @jpew:matrix.org | "If all of the currently on-line launchers are represented in declined_by the request should be marked failed" which matches the code implementation..... but also means the request will fail if declined? | 20:02 |
| @jim:acmegating.com | if all providers decline, it fails. but check the docstring for hasProviderQuota; it doesn't take current nodes into account. that method returns false if you ask for 100 nodes and your limit is 10. doesn't matter how many nodes there are currently, it can never satisfy a request for 100 nodes. so it declines and if no provider has a limit of 100 it fails. | 20:05 |
| @jpew:matrix.org | mmm, right that makes sense | 20:06 |
| @jpew:matrix.org | Well.... that is kinda disappointing..... I _depise_ forking :( | 20:08 |
| @jim:acmegating.com | yeah, i don't disagree. but also, the custom behavior for openstack quota handling is a contributing factor here. i think that just reinforces your point about forking. :) | 20:10 |
| @jpew:matrix.org | Ya, I blame Jenkins. The only reason we have the weird dynamic quota is because Zuul is sharing limited openstack resources with 20 other Jenkins instances. | 20:11 |
| @jpew:matrix.org | Anyway, thanks for explaining it | 20:12 |
| @jim:acmegating.com | jpew: out of curiosity, why not just let them all fight it out? | 20:12 |
| @jpew:matrix.org | I don't run openstack, so I don't really know | 20:13 |
| @jpew:matrix.org | We have multiple segements and I believe the quota balancer tries to fairly spread out "the pain" when the cluster is overloaded | 20:13 |
| @jpew:matrix.org | (so probably a mostly political problem, actually) | 20:14 |
| @jim:acmegating.com | gotcha. well, once you get all those jenkinseses converted over to zuul tenants, know that zuul will try to balance node requests across tenants. :) | 20:14 |
| @jpew:matrix.org | Right | 20:14 |
| @jpew:matrix.org | Someday.... | 20:14 |
| @jim:acmegating.com | jpew: if the problem is that you're hitting a node request > openstack limit problem, any way you can establish a minimum size for the quota balancer? so at least your quota would never be < your node request size? then if you tell nodepool not to ignore quota, it should maybe mostly work as designed | 20:17 |
| @jim:acmegating.com | (it will be surprised when the quota changes, but it should correct itself and continue) | 20:17 |
| @jpew:matrix.org | Ah, good idea | 20:18 |
| @jim:acmegating.com | you might run into edge cases more than other folks, but at least we'd be back into bugfix territory :) | 20:19 |
| @jpew:matrix.org | Right | 20:19 |
| @fungicide:matrix.org | > <@clarkb:matrix.org> element updated and all the colors changed | 20:33 |
| oww, my eyes! almost everyone is pastel now in dark mode | ||
| @jpew:matrix.org | corvus: Got the minimum quota bumped up to be the largest single node we can allocate; hopefully that's our fix. Thanks! | 22:02 |
| -@gerrit:opendev.org- James E. Blair https://matrix.to/#/@jim:acmegating.com proposed: [zuul/zuul] 893592: WIP: Fix bundle updates https://review.opendev.org/c/zuul/zuul/+/893592 | 22:04 | |
| @jim:acmegating.com | jpew: a clarification, just in case it's relevant: nodepool's going to check that the entire request could potentially be satisfied; so if you only ever request one node for a job, that's fine; but if you say, request two nodes with 2 cpus each, then the min quota would need to be 4 cpus. | 22:07 |
| @jpew:matrix.org | Ah, right | 22:07 |
| @jpew:matrix.org | I always forget that's a thing since we do it so rarely | 22:08 |
| -@gerrit:opendev.org- James E. Blair https://matrix.to/#/@jim:acmegating.com proposed: [zuul/zuul] 894839: Use a seeded PRNG when setting timer triggers https://review.opendev.org/c/zuul/zuul/+/894839 | 23:21 | |
| @iwienand:matrix.org | as long as https://opendev.org/zuul/zuul/src/branch/master/zuul/driver/github/githubconnection.py#L1759 is making a graphql call, anonymous access will not really work though? i.e. you'll still have the problem that you can't depends-on: a project you're not installed in? | 23:51 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!