| eolivare | hi there, we are facing an issue with two multinode jobs that require that their nodes run on a common cloud provider. This issue started occurring when openmetal was disabled. Please look at for more details https://bugs.launchpad.net/neutron/+bug/2170727 | 07:43 |
|---|---|---|
| eolivare | We have found a workaround and it seems it fixed the issue: https://review.opendev.org/c/openstack/neutron/+/1010218 | 07:44 |
| eolivare | it looks to me more like a workaround than a solution because it doesn't force using nodes from a common cloud provider. What I wanted to ask you is: is there a way to specify that the nodes these jobs use have to be obtained from a common provider? | 07:46 |
| *** bauzas4 is now known as bauzas | 09:41 | |
| fungi | eolivare: zuul tries to always allocate all nodes in a job from the same provider, unless the labels requested are not all available from any one provider | 12:53 |
| fungi | the build links in your bug report are malformed, missing the tenant identifier | 12:54 |
| fungi | i think you meant something like https://zuul.opendev.org/t/openstack/build/3c1736e633324eabaf04cb8fa9182928 | 12:55 |
| eolivare | fungi, yes, those links are wrong, sorry | 12:57 |
| fungi | it looks like that build wanted 2x nested-virt-8gb nodes and 3x 4gb nodes, it's possible that combination can't be satisfied out of the same provider | 12:58 |
| fungi | (also 5 nodes is a lot for a job) | 12:59 |
| eolivare | fungi, it was possible until openmetal was disabled https://review.opendev.org/c/opendev/zuul-providers/+/1008144 | 12:59 |
| fungi | i meant *today* | 13:00 |
| eolivare | ok, I think the patch I proposed will fix the problem. Yesterday ot worked fine. It took all nodes from vexxxhost | 13:02 |
| fungi | yeah, satisfying 5 nodes in a request is already a bit of a challenge for zuul, especially when providers are all running near quota. requesting multiple different node labels adds complexity that may or may not all be satisfiable from the same provider | 13:10 |
| fungi | if you can get away with requesting 5 identical nodes (all using the same label) that may improve your chances | 13:11 |
| fungi | also be aware that requesting atypical flavors (larger or smaller than 8gb ram, support for nested virt) limits the providers where your job can get nodes to only those where they're available | 13:14 |
| fungi | for example, recently rackspace flex dropped the processor flags that signal nested virt support, so we had to drop the labels in that provider's regions | 13:15 |
| clarkb | right these resources have never been static and we've done our best for many years to paper over the changes but occasionally that isn't possible | 15:30 |
| cardoe | https://review.opendev.org/c/openstack/openstack-helm/+/1009681 ready to submit! but how do I submit? | 23:11 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!