| opendevreview | Merged openstack/oslo.middleware master: Make ConfigInvalid generic https://review.opendev.org/c/openstack/oslo.middleware/+/923458 | 05:22 |
|---|---|---|
| opendevreview | liuyulong proposed openstack/oslo.messaging master: Add RPC incoming and reply log https://review.opendev.org/c/openstack/oslo.messaging/+/923337 | 06:57 |
| frickler | this looks to me like there might be an issue with the recent oslo.log release https://storage.gra.cloud.ovh.net/v1/AUTH_dcaab5e32b234d56b626f72581e3644c/zuul_opendev_logs_732/923578/3/check/neutron-tempest-plugin-designate-scenario/7321903/controller/logs/screen-designate-producer.txt | 08:05 |
| frickler | ralonsoh found https://review.opendev.org/c/openstack/oslo.log/+/918426 as possible trigger | 08:05 |
| frickler | damani2: sean-k-mooney: hberaud: tkajinam: ^^ | 08:06 |
| frickler | cc JayF because it seems eventlet is involved | 08:06 |
| sean-k-mooney | frickler: in what way. if neutron was impacted by not setting the debug feature then nueton was already broken as the debug feature was enebaling unsafe file descriptor aceess | 08:07 |
| sean-k-mooney | ah this is designate not neutron | 08:08 |
| sean-k-mooney | i see so that means that designate has a latent bug | 08:09 |
| sean-k-mooney | i would suggest that the correct way to fix that in the short term is for designate alone to enable the non production debug feature in there own code when monky patching until they can remove the dependcy | 08:09 |
| sean-k-mooney | frickler: note that the debug feature is not supproted in teh aio_hub so designate wont be able to use that to transtion off eventlet until they stop using it. | 08:15 |
| frickler | hmm, ok, I'll take this to the designate team, then, wasn't clear to me whether the change would affect other projects, too | 08:18 |
| sean-k-mooney | it may but its a bug in any that are affected. the enablment of the debug flag was done unintentionally when the code was ported form swift to oslo 12-18 months ago | 08:19 |
| sean-k-mooney | and projects were not aware of it | 08:19 |
| sean-k-mooney | so they may have acidentlly merged code that used it without knowing | 08:19 |
| sean-k-mooney | this is unlikely but may happen if they are spawning threads directly adn tring to log form those threads | 08:20 |
| sean-k-mooney | looking at the backtrace there was one call to logging via a task module an 2 form perodic tasks vai looping call | 08:21 |
| sean-k-mooney | haleyb: by the way with the current direction we are heading in we should be able to supprot the asycio NATS driver and with awaitlet that could be used by non asyncio code too | 09:20 |
| sean-k-mooney | im not sure that nova will adopt asyncio in the next 1-2 years although i hope we remove eventlet usage in that time but i would like to at least have the option to use the NATS driver if it progresses so that could be an interesting test case for awaitlet | 09:22 |
| sean-k-mooney | zzzeek: haleyb: by the way just listenign to the tc call i dont think 4-5 years for the eventlet transtion is fiesible. i think if its not done in 1-2 we are screwed. | 09:27 |
| sean-k-mooney | we ideally need to be able to run without eventlet to soem degree in 2025.1 and need to be able to default to not using eventlet by 2026.1 and perhaps even start droping support for eventlet in that time frame | 09:28 |
| sean-k-mooney | we effecctly have a hard deadline of completeding this work before python 3.12 goes end of life given we dont expect eventlet to support 3.13+ | 09:29 |
| sean-k-mooney | granted 3.12 EOL is 2028-10 | 09:30 |
| sean-k-mooney | so technially yes we have 4 years but practially we cant expect distro like debian or ubunut to continue to package openstack if its stuck in 3.12 for that long | 09:31 |
| hberaud | frickler: I agree with sean-k-mooney, fixes can be done directly on the impacted deliverables, leaving oslo.log free from this multiple reader non sense, and thus, limiting side effects into other deliverables | 11:42 |
| hberaud | sean-k-mooney: as far as possible, I'd love to see the discontinuation of the support of new cpython versions, but I still prefer to offer a minimal window for Py3.13 (see https://github.com/eventlet/eventlet/pull/966 and https://github.com/eventlet/eventlet/pull/965) | 11:47 |
| sean-k-mooney | hberaud: if it works and can be supproted with minimal addtional work the cool | 11:48 |
| hberaud | our problem now is more that greenlet is stuck... https://github.com/python-greenlet/greenlet/pull/410 same scenario repeat again... | 11:48 |
| sean-k-mooney | when you say greenlet is stuch you mean there ci is broken or the grenlet lib has no maintianer or the greenlet lib is broken? | 11:49 |
| sean-k-mooney | hopefully not all 3 | 11:49 |
| hberaud | victor submitted patches to greenlet to fix the compat, but nobody review them since months | 11:49 |
| hberaud | the ci could be fixed by victor's patches | 11:49 |
| hberaud | the py313 support too | 11:50 |
| hberaud | now we need reviews | 11:50 |
| hberaud | and if greenlet do not support py313, eventlet will be impacted too | 11:51 |
| sean-k-mooney | greenlet last had a release Dec 21, 2023 which isnt terribel since i belvie the api they have is resonable stale at this point | 11:51 |
| hberaud | even if "technically" eventlet is ok with py313 | 11:51 |
| sean-k-mooney | ya no commits for 7 months is kind of worrying however | 11:52 |
| hberaud | let observe that a couple of additional weeks | 11:52 |
| hberaud | py313 is still "far" | 11:53 |
| hberaud | and we are now in the summer period | 11:54 |
| hberaud | if these patches are still opened by the beginning of september I think we can really start to worry about | 11:54 |
| opendevreview | Daniel Bengtsson proposed openstack/oslo-specs master: Use cotyledon and futurist. https://review.opendev.org/c/openstack/oslo-specs/+/922597 | 14:31 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!