kelvinliu | hpidcock: we only sync charm files, jujud to workload pod now, right? | 02:58 |
---|---|---|
kelvinliu | in init container stage? | 02:58 |
hpidcock | charm files + ca.crt + operator.yaml | 02:58 |
hpidcock | jujud is copied from the init-container's image | 02:58 |
kelvinliu | what happens if we have a 2.6 model, we upgrade to 2.7, the workload pods does't re-init. | 02:59 |
kelvinliu | so all files are missing, we won't be able to run actions stuff unless all pods are killed and re-init | 03:00 |
kelvinliu | i think we still check if all dependencies are there or not before running an action. if no, syncs files first. | 03:03 |
kelvinliu | init container is like a performance enhancement, but we can't just rely on it to provide the deps. | 03:04 |
hpidcock | no we don't copy over anything in the exec anymore. wallyworld: safe to assume 2.6 models don't support running actions? | 03:06 |
hpidcock | probably should add a block for that | 03:06 |
hpidcock | check* | 03:06 |
wallyworld | hpidcock: kelvinliu: yeah, k8s actions not supported in 2.6 | 03:07 |
wallyworld | to test the actions upgrade issue, will need to use a iaas model | 03:08 |
hpidcock | wallyworld: should we add a check to make sure actions don't run on 2.6 models for CAAS? | 03:08 |
hpidcock | technically exec should work, you just won't be able to use juju-run | 03:09 |
kelvinliu | problem here is 2.7 models were upgraded from 2.6 won't be able to run actions as well | 03:09 |
hpidcock | or other actions | 03:09 |
wallyworld | we could add a check | 03:09 |
wallyworld | why won't upgraded 2.7 models fail? | 03:09 |
wallyworld | *work | 03:09 |
hpidcock | the 2.6->2.7 models should be able to run actions since the deployment/statefulset will change with the new operator image on the init-container | 03:13 |
kelvinliu | not sure why my mariadb wasn't re-init | 03:15 |
hpidcock | could have been timing, but that is concerning | 03:17 |
kelvinliu | EnsureService is only called if podspec has changes, right? | 03:20 |
kelvinliu | or it's not called | 03:21 |
wallyworld | that is the idea but when the agent restarts it loops through everything | 03:21 |
wallyworld | the check is done for pod-spec-set | 03:21 |
kelvinliu | good, good. i think it's just a bit delayed. ignore me, plz! 😃 | 03:27 |
hpidcock | scaring me half to death haha, I already have nightmares that the pod-init doesn't work 😨 | 03:30 |
kelvinliu | hahaha sorry for scared u... 😂 | 03:41 |
wallyworld | hpidcock: this is the fix for add-k8s https://github.com/juju/juju/pull/10842 | 04:29 |
hpidcock | wallyworld: on it | 04:29 |
kelvinliu | wallyworld: this is the fix for actionID, https://github.com/juju/juju/pull/10841 thanks! | 04:48 |
wallyworld | looking | 04:49 |
wallyworld | kelvinliu: lgtm, would be good to do a test on an iaas model | 04:55 |
kelvinliu | yep | 04:56 |
hpidcock | wallyworld: one problem with https://github.com/juju/juju/pull/10842 | 05:34 |
anastasiamac | wallyworld: kelvinliu: https://github.com/juju/juju/pull/10843 - to addres add-k8s with pipe output :D | 06:14 |
wallyworld | yay looking | 06:14 |
anastasiamac | at the end was very simple... i was petrified that i'd have to write a complex test but it was a breeze (thnx to a kind sould that did smth similar) :D | 06:15 |
wallyworld | hpidcock: ah yes, i had it as the k8s types but changed at last minute | 06:15 |
wallyworld | will fix | 06:15 |
anastasiamac | (and I mean kelvinliu of course!) | 06:15 |
anastasiamac | when i say a 'kind soul' | 06:16 |
wallyworld | lgtm ty | 06:17 |
kelvinliu | anastasiamac: lgtm as well, thanks! 😊 | 06:17 |
anastasiamac | tvym!! | 06:21 |
hpidcock | wallyworld: let me know when that change is up, I was unable to finish my testing | 06:25 |
wallyworld | hpidcock: changes up, i just tested with microk8s again. external will not work until we allow externalName to be passed in. but loadbalancer works | 06:40 |
wallyworld | which is the main bit | 06:40 |
hpidcock | LGTM | 06:45 |
wallyworld | yay ty | 06:47 |
wallyworld | i'll just do a bit more testing | 06:47 |
timClicks | wallyworld: pr 10842 will be an absolute game changer | 07:35 |
wallyworld | i guess so. the fact it was broken was because we really only tested with public k8s, microk8s, or cdk | 07:56 |
wallyworld | anything that adds new clusters to the mix is good | 07:57 |
wallyworld | be interesting to see any uptick in interest | 07:57 |
=== grumboo is now known as grumble |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!