Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2026-06-12
15:54:26 sean-k-mooney so restoring fork seams to work
15:55:07 sean-k-mooney with threaded mode it fixed 1 of the these but the other got the token expired thing you were seeing on ubuntu 24.04
15:55:46 sean-k-mooney so for now i think doing multiprocessing.set_start_method('fork', force=True) uncondtionaly regarless fo eventlet or threaidng is the right move
15:56:05 sean-k-mooney until we fiture out how to make novnc without that in general
16:00:21 melwitt sean-k-mooney: had to change approach on ^ due to it still failing bc the scoping of the mock did not actually help bc they apparently all resolve to the same place in the end
16:00:40 sean-k-mooney yes they do
16:00:41 melwitt I'll be rechecking it some times
16:00:49 sean-k-mooney did you swap to useign mock_open
16:01:43 melwitt no, used list.count() to assert the specified fd was called close with once. a claude idea that _seems_ ok afaict
16:01:49 sean-k-mooney yep just saw
16:01:55 melwitt to just ignore the rest. yeah
16:02:08 sean-k-mooney ill pull that patch locally and run it in a loop on one of my systems for like 30 mins
16:02:53 melwitt the shame deepens
16:03:05 melwitt I'll run a loop
16:03:13 sean-k-mooney --until-failure
16:03:43 melwitt although I guess I need to run all tests that's key to the issue
16:03:47 sean-k-mooney ill let it run on my mac mini for a while that really fast at this
16:04:16 sean-k-mooney yep but i have like 3 systems taht can do that pretty quickly
16:04:53 melwitt :) sean-k-mooney's data center
16:05:23 melwitt I'm lucky it failed in the gate. this sucks haha
16:05:30 sean-k-mooney my home sever may be 10+ years old but if you trho 48 cores at nova unit tests it still flies
16:06:28 melwitt 48 o_O
16:06:31 sean-k-mooney my m4 base spec mac mini almost but not quite catchs my server
16:06:31 opendevreview Merged openstack/nova master: Add irrelevant-files for the graceful shutdown jobs https://review.opendev.org/c/openstack/nova/+/993063
16:06:49 melwitt my 10+ year old home server only has 12 and it finally got too slow
16:07:15 sean-k-mooney techinaly it 2 14 core cpus so 24 cores and 48 threads
16:08:37 sean-k-mooney https://paste.opendev.org/show/bzdwOUsGPvkQPl8HU6Rz/
16:09:10 melwitt wow
16:09:59 sean-k-mooney oh i was wrong it 2 12 cores
16:10:18 sean-k-mooney but that still 24 cors and 48 threads im just bad at math
16:28:50 opendevreview melanie witt proposed openstack/nova master: Fix cover env HTML testr results missing unit tests https://review.opendev.org/c/openstack/nova/+/993151
16:28:56 sean-k-mooney hum so thinkign abou tthis a bit
16:29:28 sean-k-mooney your sill mockign all calls to open clsoe and unlink
16:30:34 sean-k-mooney which we need to do
16:31:08 sean-k-mooney but i one der if we shoudl make the call to open and unlink more robust as well
16:32:04 melwitt what do you have in mind?
16:32:05 sean-k-mooney on the ohter hand i have doen over 20 complet runs locally without any issue
16:32:37 sean-k-mooney removing the once assert_called_once_with
16:32:51 sean-k-mooney and juse using assert_called_with
16:33:07 melwitt yeah. I think the specific asserts are pretty much avoiding issues but maybe the mock_open.assert_called_once_with('/swap', os.O_CREAT | os.O_EXCL | os.O_WRONLY) could technically run into trouble somehow
16:33:49 sean-k-mooney well all 3 could use assert_called_with without really losing much in terms of coverage
16:34:22 melwitt true
16:35:13 sean-k-mooney i know normally we like to assert the call count as well but in this case i dont know if it get us much
16:35:30 melwitt yeah, I hear ya
16:35:36 sean-k-mooney if ci say no again we can try that
16:35:56 sean-k-mooney if it has not failed locally by the tim it report back i think im ok to leave it as is and proceed
16:36:37 melwitt yeah, same. I have a loop going too and if it's also still going then extra good haha
16:36:55 melwitt --until-failure is nice, I forgot about that
16:38:27 sean-k-mooney this is basiclly all i use if for but ya i think gibi first told me about it many moons ago
16:41:17 opendevreview Merged openstack/os-vif master: Make ``PyPyroute2._lookup_interface`` private https://review.opendev.org/c/openstack/os-vif/+/991292
16:46:25 opendevreview Balazs Gibizer proposed openstack/nova master: Avoid stacktrace at CLI exit https://review.opendev.org/c/openstack/nova/+/993153
16:52:50 gmaan dansmith: so you want to inherit it from ServersUpdateSampleJsonTest instead of ServersSampleBase, same way other update serve rmicroversion tests did https://review.opendev.org/c/openstack/nova/+/990682/comment/7a022078_05092941/
16:53:03 gmaan I also missed to check those update test
16:53:28 gmaan that way, you do not need to create the GET server samples in new microvesion dir
17:02:08 dansmith gmaan: ah meant to
17:02:53 dansmith I thought that still required the get samples though which is why I left them in there,
17:03:24 dansmith but perhaps I missed an editor save or something, saw it still requiring the get samples, put those back, and then never committed the save
17:05:07 gmaan ohk, this file is getting huge and difficult to read. I think I should split it.
17:06:14 dansmith yeah, it is for sure
17:09:28 opendevreview Dan Smith proposed openstack/nova master: Add unpin AZ functionality https://review.opendev.org/c/openstack/nova/+/990682
17:09:28 opendevreview Dan Smith proposed openstack/nova master: Doc changes for unpin-az https://review.opendev.org/c/openstack/nova/+/992469
17:09:46 dansmith okay sorry about that.. I had a successful run in my terminal buffer from when I removed everything but the update samples
17:10:36 dansmith then when I was rebasing to apply the change to the first commit I must have lost the inheritance part and then running made me add the get/list things back in so I thought I had stale samples on disk last time or something
17:11:11 gmaan k
17:11:41 dansmith trying to get claude to "help" with this stuff is maybe not the best plan past initial generation.. like having a pair programmer messing with things on disk while you're trying to clean up :)
17:12:25 gmaan yeah, i think you should change the tag also now as you updated it manually https://review.opendev.org/c/openstack/nova/+/990682/comment/9a9abe22_4f0d2b93/
17:13:00 dansmith gmaan: meh
17:13:19 dansmith it's still mostly generated and I think the tag is still okay
17:13:31 dansmith Generated-By: claude, Fixed-By: human meat bag
17:14:08 melwitt claude has been betraying me a lot lately
17:14:56 dansmith same
17:15:32 dansmith there's betrayal and there's also the aspect of just being a third thing competing for context... you have on-disk state, you have in-git state, you have the current state of your editor, and then whatever claude is doing
17:15:36 melwitt in my commit message tag I also include the claude code version bc I think it's interesting and it's been updated/upgraded at least twice since I stated this morning
17:15:45 melwitt *started
17:16:28 melwitt so I'm like 🙏 they are fixing things
17:16:41 gmaan it is difficult to say if Generated-By is still ok when we modified the generated code but does not matter much anyways
17:16:58 melwitt Clauded-By
17:18:21 sean-k-mooney well i nomrally update to assisted-by once i have reworked some
17:18:32 sean-k-mooney but generate by does not have to mean entrilly llm output
17:18:52 melwitt I never use Generated-By, problem solved
17:19:06 melwitt #efficiency
17:19:28 sean-k-mooney ya for many things is v 2 or 3 before i push for review the first time anyway
17:22:31 sean-k-mooney as an aside how ould folk feel about add a funtional py 3.14 job, myign the unit test job voting and maybe moving the coverage job to 3.14 as well
17:23:13 sean-k-mooney the reason i ask is openstack-tox-py314 has been passing stablty for a good few weeks
17:23:28 sean-k-mooney and if the funciotnal job also passes we coudl mvoe coverge to 3.14
17:23:51 sean-k-mooney and look into the alternitve backend that stephen mentioned that speed it up
17:24:59 melwitt on coverage, I asked claude about it again today and it said tl;dr that you can only get the speedup without eventlet
17:25:18 melwitt which I guess we can't do bc coverage runs functional now too
17:25:59 melwitt "There it is — concurrency = ["eventlet"] in pyproject.toml forces coverage.py to use sys.settrace instead of sys.monitoring, even on Python 3.12+. That's likely the biggest performance hit."
17:26:26 sean-k-mooney hum ok well we are getting closer on functionla coverage in threading mode
17:26:35 melwitt *we can't switch coverage to native threading bc we don't have the func tests sorted out yet
17:26:37 sean-k-mooney so maybe we can just run it in threading mode in a bit
17:26:45 melwitt yeah
17:28:09 sean-k-mooney i might put up a patch thetn jsut for makeing the py314 unit test voting and addign a non voting functionl 3.14 job to see how that goes
17:28:16 melwitt I had tried a patch to help with that and then lots of unrelated stuff happened and I dropped the ball
17:47:44 melwitt gmaan: fyi I noticed you got some questions on the graceful shutdown func tests https://review.opendev.org/c/openstack/nova/+/991743 so I cc'ed myself when you have got a chance to reply
17:49:46 melwitt sean-k-mooney: I just approved https://review.opendev.org/c/openstack/nova/+/993109 but then I wondered, shouldn't that change go in whitebox not nova?
17:50:13 sean-k-mooney whitebox is not branched
17:50:31 sean-k-mooney so normally in whitebox we do the opisce of tempet
17:50:44 sean-k-mooney we disabel new tests and make them opt in

Earlier   Later