| *** jgilaber__ is now known as jgilaber | 08:16 | |
| dtantsur | Hi! Quick q on node labels: when I request ubuntu-noble-16GB, does it also match ubuntu-noble-32GB or is it specific? And what's the default for ubuntu-noble? | 12:19 |
|---|---|---|
| dtantsur | I'm checking some NODE_FAILURES on Ironic and researching the possibility of reducing the requirement on certain jobs. | 12:20 |
| frickler | the default should be -8GB. and the labels are specific, mapping to flavors that should match the respective ram sizes | 12:29 |
| fungi | dtantsur: it's specific, but they may still map to the same flavor in some clouds. you can find their definitions and that mapping in https://opendev.org/opendev/zuul-providers/src/branch/master/zuul.d | 12:30 |
| fungi | for example, the nested-virt-8GB label in vexxhost-base uses their v3-standard-8 flavor which has 32gb ram | 12:35 |
| fungi | er, the nested-virt-8GB zuul flavor i should say, mapped to the v3-standard-8 nova flavor in vexxhost clouds | 12:37 |
| fungi | anyway, that providers.yaml file has most of what you'd need to identify cases where you may get a node with significantly more ram than the flavor you requested would imply | 12:38 |
| fungi | you can also browse a bit of that at https://zuul.opendev.org/t/openstack/providers though the details and descriptions aren't plumbed through yet, i expect that'll be expanded further in time | 12:43 |
| dtantsur | aha, thanks! | 12:50 |
| dtantsur | My takeaway is that 16G nodes are much more common, so it does make sense to switch the jobs that don't need 32G | 12:57 |
| fungi | yes, the 32gb nodes i think are currently only available in a single provider region, so if the node request to that provider fails you'll end up with a node_error result for the job | 12:58 |
| fungi | (e.g. if the provider tries to boot it but then nova returns an error state on the resulting server instance) | 12:59 |
| dtantsur | I see, I see. Yeah, we have these on Ironic even right now. | 13:06 |
| fungi | if there's simply insufficient quota, the requests will queue up until they get a turn, but if say there's no capacity in a host aggregate on the cloud or something then that can result in openstack server create errors | 13:26 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!