Wednesday, 2026-04-22

@clarkb:matrix.orgBoth gitea and etherpad have new releases we can start preparing to deploy. Historically I've done a lot of these updates. I'm happy to walk someone else through one or both if there is interest.14:34
Note we should avoid actually upgrading etherpad until after the ptg
@tafkamax:matrix.orgThe etherpad creator is developing a new version of etherpad where the backend is in go14:34
@tafkamax:matrix.orgFYI 14:34
@tafkamax:matrix.orgHaven't tested it myself14:35
@clarkb:matrix.orgYup eventually we will need to switch I suspect. But the nodejs version just got a release so we can continue to deploy that14:35
@tafkamax:matrix.orgDo you use plugins aswell? 14:35
@tafkamax:matrix.orgI use and like to compile them in to docker instead of installing in runtime14:36
@clarkb:matrix.orgWe use one or two minor plugins. Nothing too crazy. The bigger concern I have is migrating existing pads to the new system when it comes to that14:36
@tafkamax:matrix.orgYeah14:36
@clarkb:matrix.orgAnd yes we add them directly to our docker images14:36
@tafkamax:matrix.orgNice14:36
-@gerrit:opendev.org- Clark Boylan proposed: [opendev/system-config] 985834: Update Gitea to 1.26.0 https://review.opendev.org/c/opendev/system-config/+/98583415:14
@clarkb:matrix.orgThat is a first draft on the gitea update. I figured getting the ball moving was a good idea. I'll probably do that same for etherpad. Still happy for others to jump in if interested. Chances are these changes will need updates to make thinsg work and or to update the versions as new ones may be available by the time we're ready to upgrade15:15
@mnasiadka:matrix.orgThe zp03 changes are awaiting reviews if there’s room in anybody’s todo list - https://review.opendev.org/c/opendev/system-config/+/98562015:23
@fungicide:matrix.orgi have a break in my ptg schedule so am going to take this opportunity to grab lunch, but can review when i get back15:44
-@gerrit:opendev.org- Zuul merged on behalf of yatin: [zuul/zuul-jobs] 984404: [ensure-python] Fix rpm_python_pkg_name For CentOS/RHEL https://review.opendev.org/c/zuul/zuul-jobs/+/98440415:52
@clarkb:matrix.orggithub seems to be really slow today15:53
@clarkb:matrix.orggithubstatus seems to confirm. I just need the etherpad changelog :)15:53
-@gerrit:opendev.org- Clark Boylan proposed: [opendev/system-config] 985843: Upgrade etherpad to 2.7.0 https://review.opendev.org/c/opendev/system-config/+/98584316:16
@clarkb:matrix.orghttps://review.opendev.org/c/opendev/zuul-providers/+/984866 based on the consistent post failures there I think something may be properly broken with our intermediate swift container for zuul-launcher builds16:19
@clarkb:matrix.orgI did a brief check to see if we might have the same requests exceptions issue that affected log uploads and that doesn't appear to be the problem. In this case swift appeas to return a 404 saying our container doesn't exist?16:20
@clarkb:matrix.orgIf I do a `openstack container list` against that cloud region I see the container listed16:22
@clarkb:matrix.orgif I do a container show I get a 40416:22
@clarkb:matrix.orginteresting16:22
@clarkb:matrix.orghas anyone else seen this before where container listings show a container but container show says 404? I don't think this is an auth issue because you'd expect a 403 or a failure to list the containers in the first place16:23
@clarkb:matrix.orgcorvus: ^ fyi as this affects opendev's zuul-launcher stuff16:23
@clarkb:matrix.orgpart of me wonders if we should just try using a new container but I'm not sure what is involved in setting that up16:24
@jim:acmegating.comClark: i'm about to head out for an appointment, but i would support trying a new container; i suspect there may be a bunch of old cruft in there anyway from early launcher stuff, and a restart would be good)16:25
(but -- perhaps this is just a temporary cloud issue?)
@clarkb:matrix.orgcorvus: yes I thought it may be temporary but it appears to started ~April 15 and based on my manual testing appears to be continuing. SO about a week long issue or so16:26
@clarkb:matrix.orgI can recheck the change again, but based on manual testing I don't expect a different result16:26
@clarkb:matrix.orgwithout being able to container show the container I'm not sure what sorts of acls we might need on a new container :/16:26
@jim:acmegating.comi thought we went through this during the great re-keying16:27
-@gerrit:opendev.org- Michal Nasiadka proposed: [opendev/irc-meetings] 985845: Move Magnum weekly team meeting https://review.opendev.org/c/opendev/irc-meetings/+/98584516:27
@jim:acmegating.comdid we document anything as a result of that?16:27
@clarkb:matrix.orgoh yes we did16:27
@jim:acmegating.comif not, maybe it will be in logs16:28
@clarkb:matrix.orgI don't know if we explicitly logged anything, but I can look through communication logs to see what can be found16:28
@jim:acmegating.com(or maybe in the password file)16:28
@clarkb:matrix.orgI'll work on that and if I can find it I'll go ahead and create a new container and set acls and then take it from there16:28
@jim:acmegating.comcool, sorry i can't help more right now, but will check in when i get back16:28
@clarkb:matrix.orgthanks. I don't think this is urgent so don't feel bad about that. its been broken for a week another few hours or a day won't be the end of the world16:29
@clarkb:matrix.orgok I've found where we discussed rotating the credentials. It looks like we simply created a new application credential and then updated the secret values for its uuid and token/password/secret. It does not look like we changed anything about the container itself. This implies to me that maybe we did not have any special acls on the container itself. Given then I'm inclined to create a new container and push an update to opendev/zuul-providers and see what happens16:36
@clarkb:matrix.orgI've created a new container with `openstack container create images-e86a313cdad6` and I can container show this container. I'll push up a change to use it momentarily16:42
-@gerrit:opendev.org- Clark Boylan proposed: [opendev/zuul-providers] 985847: Change the intermediate swift storage container https://review.opendev.org/c/opendev/zuul-providers/+/98584716:45
-@gerrit:opendev.org- Clark Boylan proposed: [opendev/system-config] 985834: Update Gitea to 1.26.0 https://review.opendev.org/c/opendev/system-config/+/98583417:03
@clarkb:matrix.orgsomething about how gitea is built has changed, but they super optimized their dockerfile so figuring out the delta is a bit of a pain. I don't think those optimizations make a lot of sense for us since we're not maintaining local images and building them often so if all I needed was the explicit go mod download step then maybe we're ok?17:04
@clarkb:matrix.orghowever, if things continue to not work maybe we port our dockerfile to something more closely matching theirs.17:04
@clarkb:matrix.orgoh actually maybe they removed the explicit step to build that command entirely and the thing is simply not in the codebase anymore which would explain teh compiler complaint that the name isn't in std anymore?17:06
-@gerrit:opendev.org- Zuul merged on behalf of Michal Nasiadka: [opendev/zone-opendev.org] 985619: Add zp03.opendev.org https://review.opendev.org/c/opendev/zone-opendev.org/+/98561917:11
-@gerrit:opendev.org- Clark Boylan proposed: [opendev/system-config] 985834: Update Gitea to 1.26.0 https://review.opendev.org/c/opendev/system-config/+/98583417:16
-@gerrit:opendev.org- Zuul merged on behalf of Michal Nasiadka: [opendev/system-config] 985620: Add zp03.opendev.org https://review.opendev.org/c/opendev/system-config/+/98562017:43
@mordred:waterwanders.comSo, I've been noodling a bunch with using ai agents for both coding and code-review with gerrit and zuul. It's working well so far. Like - it's almost like the tools we've built with OpenDev and Zuul are exactly the type of guardrails needed for AI sanity. :) Based on experience so far, I'd like to try something that would require a bit more buy in from OpenDev than just approving the occasional mordred repo.17:54
I'm finding that my review agent does a good enough and quick enough job of catching dumb things, that having zuul jobs not start until a change has an agent+1 feels more right. Instead of "patch upload -> zuul +1 -> human review / +A -> zuul gate" it would be "patch upload -> agent+1 -> zuul +1 -> human review / +A -> zuul gate" To do that I'd need new pipeline definitions, which in turn would need a new zuul tenant and config repo - and on the repos in question they'd want an additional gerrit review category. this feels like something that should be discussed more broadly before I start writing new config-repo patches.
Do you guys want me to write up a formal spec? Or is rambling back and forth here sufficient?
@fungicide:matrix.orgi don't have objections to creating a new zuul tenant for that, even as an experiment. alternatively you might be able to get away with just a separate pipeline in an existing tenant?17:57
@fungicide:matrix.orgobviously you need slightly different trigger conditions than the typical check pipeline, but you could set the involved projects up with no check pipeline and then have a separate validation pipeline on them17:58
@mordred:waterwanders.comyou know, I like that as a potential smaller first step for the experiment than a whole new tenant out of the gate17:59
@clarkb:matrix.orgIt also seems to be similar to what we had suggested when this came up earlier. Basically use zuul to drive the process of LLM code review. The main difference is the use of a separate pipeline which seems like it would be ok to do particularly in another tenant. I think the main drawback to that is it assumes you'll always have an LLM do pre review which is fine if tenant scoped17:59
@clarkb:matrix.orgI'm not sure how you'd do that in an existing tenant without changing the rules for all repos in that tenant18:00
@mordred:waterwanders.comyeah - I don't think it can be a global thing, definitely has to be an opt-in.18:00
@clarkb:matrix.orgSince you would need to modify the check pipeline trigger criteria18:00
@fungicide:matrix.orgwell, check could be undefined on those projects and would never trigger18:01
@clarkb:matrix.orgThough maybe you can set llm-review piepline to run noop job18:01
@mordred:waterwanders.comwell - if we do fungi's thing, we could make an "post-agent-check" pipepine (or a better name) - and then, as long as there isnt' a clean-check rule, if there weren't any jobs in check, it _should_ just work, yeah?18:01
@fungicide:matrix.orgas long as the gate pipeline in that tenant doesn't require a vote from any other pipeline as a condition, and gate is always human-triggered, i don't see why there couldn't be multiple check-like pipelines used in parallel by different projects18:02
@clarkb:matrix.orgOh ya if you invert it where check is still the first pass but opt into using LLMs or traditional jobs then ya that would work18:02
@clarkb:matrix.orgThough it might be weird to get the triggers right if using verified +1 still18:02
@clarkb:matrix.orgIt kinda seems like you want a llm-review label and llm-review pipeline then configure that pipeline to post to llm-review label and trigger check off of that18:03
@fungicide:matrix.orga custom gerrit label in those projects' acls could also be an option18:03
@mordred:waterwanders.comyeah - well, it's "have the LLM do a pass, then once they're happy, trigger Zuul" so the alternate check pipeline could have the rule that it needs agent+1 or whatever18:03
@clarkb:matrix.orgBut you can probably exercise the bulk of the workflow without that level of commitment 18:03
@fungicide:matrix.orgyeah, basically what you just said18:03
@fungicide:matrix.orgright, there's a lot of flexible bits to work with here, between project-specific gerrit and zuul configuration18:04
@clarkb:matrix.orgRight to be clear I think this is doable. I think what is awkward is to do it in a tenant with existing jobs and pipelines and expectations18:04
@clarkb:matrix.orgUsing a new tenant you side step all of that without needing to retrofit the old tenant first18:04
@mordred:waterwanders.comyah. I'll go see if I can't cook up a patch to add a new pipeline but not impact existing pipelines.18:04
@fungicide:matrix.orgi'm okay with it as long as the experiment doesn't bleed over onto other uninvolved projects/changes18:04
@mordred:waterwanders.com++ me too. I very much don't want uninvolved bleed18:05
@fungicide:matrix.orgthe only ugly bit i can think of is that there would be an additional pipeline showing up in zuul status for that tenant18:05
@fungicide:matrix.orgbut if that pipeline doesn't run jobs for or leave votes on uninvolved projects i don't have a concern18:05
@clarkb:matrix.orgI guess zuul con filter verified votes by voter so maybe that is the trick to getting the triggers right 18:06
@clarkb:matrix.org* I guess zuul can filter verified votes by voter so maybe that is the trick to getting the triggers right 18:06
@clarkb:matrix.orgWhich would allow you to overload the verified label18:06
@fungicide:matrix.organd for that matter, the default status view these days hides pipelines that aren't running any jobs at the time, so odds are people would almost never notice except when the projects participating in the experiment have changes actively running jobs18:06
@mordred:waterwanders.comyah- and since all my repos are in the opendev tenant which has a pretty low user base right now, it would be _super_ low exposure18:08
@fungicide:matrix.orgso all that's to say, i can see a few ways it could work; whether i'm okay with the idea would depend on the details of course18:08
@fungicide:matrix.orgbut seems like we can come up with something18:08
@clarkb:matrix.orgmordred: re the thesis this is the thing I've been trying to explain a few times to others. We built these tools for a couple thousand largely independent developers to go crazy making Openstack. That doesn't look too different than a bunch of agents going crazy on a shared code base18:09
@mordred:waterwanders.com++ cool. I can work on some patches to propose and we can discuss whether they are terrible or not18:09
@mordred:waterwanders.comClark: EXACTLY18:09
@mordred:waterwanders.comlike, this is not a new problem for us18:09
@mordred:waterwanders.comit's the places that have adopted the "you should build all tools assuming high trust for the devs involved" that are in a bad spot currently18:10
@fungicide:matrix.orgthat's the theme of the talk i've proposed for all things open, as well18:10
@mordred:waterwanders.comlet's be honest - our tools were built with the assumption that *I* would be writing code :) 18:10
@fungicide:matrix.orgso interested to have this as an anecdote18:10
@fungicide:matrix.orgmordred: i think some of our tools still have inline comments that say `# TODO(morderd): actually make this work`18:11
@clarkb:matrix.orgI dunno I learned pretty quickly that you just push TODOs to prompt someone else to write the actual code. Hey that isn't much different either :)18:11
@mordred:waterwanders.comI'm pondering a talk for shanghai and the lf event in slc in novemeber. haven't gotten all the way to submitting anything yet18:11
@clarkb:matrix.orgfungi: I found one today in the gitea dockerfile even18:11
@mordred:waterwanders.comhahahaha18:11
@fungicide:matrix.orgmordred: the only reason i already put in a proposal for ato is that their cfp was closing and it's the closest open source conference to me geographically. they also have an "all things ai" conference they started a few years ago, but the 2026 one was last month so won't be happening again until next march18:12
@clarkb:matrix.orgSomething to do with how files are copied not working for us because we clone the repo so we skip a step gitea does and the todo says we should fix that if we ever figure it out18:12
@mordred:waterwanders.comfungi: yeah - unfortunatley CFPs are so far ahead. I wasn't doing any of this when the CFP for the next LF event was still open :) 18:13
@mordred:waterwanders.comClark: see - now that TODO makes me want to go in and use bind-mount copy18:14
@fungicide:matrix.orgto be clear, so far the talk consists of about 5 sentences of abstract. i have faith that by the time october rolls around (if it gets accepted) i'll have something to put onto slides for it18:15
@mordred:waterwanders.comI'll definitely share whatever I come up with too18:15
@mordred:waterwanders.comI have a good amount of anecdata so far18:16
@clarkb:matrix.orgFor those that don't know one of my very first interactions with Monty was him saying the python bindings for libdrizzle were broken and someone should fix them. Guess who fixed them? This was at least a couple years before Openstack too18:17
@mordred:waterwanders.comnerdsniping is my favorite method of coding18:17
@clarkb:matrix.orgmordred: it honestly feels like a layer 8 problem more than anything else. Between the foundation policy, DCO, llm service ToS, and costs it seems like there are a bunch of hoops to jump through to do anything officially. But the tools themselves should largely plug together and support the workflow18:19
@fungicide:matrix.orgnerdsniping was vibecoding with a mechanical turk18:19
@clarkb:matrix.orgBut maybe if you unofficially nerdsnipe yourself into proving that the tools are ready then the layer 8 problems become easier to work through? "Look this could be possible"18:20
@mordred:waterwanders.comyeah. luckily many of those layer 8 problems don't exist on my personal projects. no foundation policy or dco to manage before proving. and ... we're big enough I did not really have to explain how gerrit or zuul worked, although i did make those gerrit and zuul plugins, but those were optimizations18:27
@clarkb:matrix.orgthe etherpad 2.7.0 change passes testing and the two screenshots look good. I guess the next step there is holding a node and testing it. But that isn't urgent as we shouldn't upgrade during the PTG and I'm in meetings next week too19:03
@shrews:matrix.orgwow. didn't expect to see a libdrizzle reference in matrix today19:29
@fungicide:matrix.orgnobody expects the spanish inquisition19:29
-@gerrit:opendev.org- Zuul merged on behalf of Clark Boylan: [opendev/zuul-providers] 985847: Change the intermediate swift storage container https://review.opendev.org/c/opendev/zuul-providers/+/98584719:52
@jim:acmegating.comopendev tenant has no clean check requirement (and i will continue to advocate against it!) so i think that would be a fine place to put a new pipeline if we're okay with that.19:55
@fungicide:matrix.orgyep, that's what i was thinking as well19:56
@clarkb:matrix.orgI guess we should recheck an image build change if someone hasn't already and see if the new container works?20:02
@clarkb:matrix.orgSomeone beat me to it according to zuul's status page20:04
@jim:acmegating.comwas me20:05
-@gerrit:opendev.org- Clark Boylan proposed:20:51
- [opendev/system-config] 985834: Update Gitea to 1.26.0 https://review.opendev.org/c/opendev/system-config/+/985834
- [opendev/system-config] 985877: Fix gitea selenium screenshots https://review.opendev.org/c/opendev/system-config/+/985877
@clarkb:matrix.orgI don't actually know if that name change will work as I'm not certain the selenium installation will be able to find names in /etc/hosts as I think it may run in a container?20:54
@clarkb:matrix.orgmight need to bind mount in /etc/hosts and hope that doesn't break other things? it is using network mode hosts so maybe this will just work though20:56
@clarkb:matrix.orgcorvus: looks like the zuul-providers job failures are now due to running out of disk during the compression step?21:00
@clarkb:matrix.orgso that is a different failure to the swift 404 from before. Not sure if this is before or after that step would've been previously though21:00
@fungicide:matrix.orgwe were also running out of disk on those before, so i guess the swift errors were masking that we still hadn't completely solved the disk usage?21:02
@clarkb:matrix.orgya I'm not sure which direction we may have been masking before21:02
@clarkb:matrix.organd I have to pop out for the school run in about 2 minutes so not in a good spot to debug further at the moment21:03
@jim:acmegating.comwell we were only running out of disk on the resolute build, and only barely; so i'm surprised if we're now running out of space in general21:49
@jim:acmegating.comlooks like that happened once for a resolute build, and once for a rocky build?21:50
@clarkb:matrix.orgYa I think those were the two I saw when I checked earlier21:50
@jim:acmegating.commost of the builds in the first change passed21:50
@jim:acmegating.comlike, all the others so far21:50
@clarkb:matrix.orgAh ok so maybe it's not a consistent enough issue or it's specific to those builds21:51
@jim:acmegating.comthe two that failed were on ovh21:52
@fungicide:matrix.orgwhich i think usually has a lot of available rootfs?21:54
@jim:acmegating.comone that passed was on ovh https://zuul.opendev.org/t/opendev/build/7c19f664095b486d90578d6c3e2f90ef/log/zuul-info/inventory.yaml21:55
-@gerrit:opendev.org- Monty Taylor https://matrix.to/#/@mordred:inaugust.com proposed: [opendev/project-config] 985898: Add vouched pipeline to OpenDev tenant https://review.opendev.org/c/opendev/project-config/+/98589821:55
@mordred:waterwanders.comnaming is impossible - but how does that look/sound?21:56
@fungicide:matrix.orgwfm, i'm trying to do better about sticking to my personal policy not to get into bikeshed color debates21:56
@mordred:waterwanders.comRED21:56
@fungicide:matrix.orglooking at the successful run on ovh, its rootfs is 80gb (79928754176 bytes reported by df)21:58
@jim:acmegating.comdoes it only have one fs?  (no separate /opt)21:59
@fungicide:matrix.orgyes21:59
@clarkb:matrix.orgI wonder if the number of successes is enough to just land that change as an immediate first step since we think it may make things generally more reliable 21:59
@fungicide:matrix.orgso /opt is shared with /21:59
@jim:acmegating.comClark: could -- or we could squash the other disk-saving change22:00
@jim:acmegating.comeach one gets us 10GB22:00
@fungicide:matrix.orglooking at the dstat report for the successful job, the highest i see usage go on / is 66gb22:00
@clarkb:matrix.orgcorvus: oh ya maybe that is what we should do. Combine them to maximize savings22:01
@fungicide:matrix.orgwhen zuul-info is collected at the start of the job, the rootfs has 13gb used and about 60gb free, fwiw22:01
@jim:acmegating.comyeah, i'm just stuck on thinking that we didn't think either of these was necessary before; wondering what changed22:01
@jim:acmegating.commaybe rockylinux is that much bigger?22:02
@jim:acmegating.comand maybe we were marginal before?  perhaps it's even the case that we have previously seen rocky failures on ovh nodes, but that's the only failure condition, and it's rare?22:03
@jim:acmegating.comrocky failure: /dev/vda1       78055180 69604308   4845968  94% /22:03
focal success: /dev/vda1 78055180 65657132 8793144 89% /
both on ovh
@jim:acmegating.comthat's from the "Collect disk usage info post image build" task22:04
-@gerrit:opendev.org- James E. Blair https://matrix.to/#/@jim:acmegating.com proposed:22:05
- [opendev/zuul-providers] 984866: Run vhd builds first and delete cache https://review.opendev.org/c/opendev/zuul-providers/+/984866
- [opendev/zuul-providers] 982182: Add Ubuntu resolute image build job https://review.opendev.org/c/opendev/zuul-providers/+/982182
@jim:acmegating.comokay, omnibus space saving change followed by resolute change22:05
@clarkb:matrix.orgIt wouldn't surprise me if there are differences.22:06
@jim:acmegating.com4GB is a big difference22:07
@jim:acmegating.combut it may not all be in the image, it could be in the build scaffolding22:07
@clarkb:matrix.orginfra-root https://review.opendev.org/c/opendev/system-config/+/985877 this change does appear to work and fixes the gitea screenshots. Landing that one should be safe (I stacked the 1.26.0 upgrade on top of it so they can happen separately22:38

Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!