Kubernetes SIG/Meetings/2023-06-27
Attendees: Brian King Ben Tullis Alexandros Kosiaris Stef Dunlap Andrew Otto Ryan Kemper
Agenda:
edit- Introductions for new members (if any)
- SIG administrivia:
- Alex won’t be running these meetings in September-November.
Alex still has a TODO to make a draft policy the helm decision. Page will behttps://wikitech.wikimedia.org/wiki/Kubernetes/Upstream_Helm_charts_policy
- Misc:
- [Stef] Magic trick: KEDA HTTP autoscaler
- Immediate use case: Running out of resources in staging, silly because the deployments are barely used. Would like to scale down to nothing when not in use.
- KEDA is an k8s autoscaler
- KEDA-HTTP autoscaler not for production, experimental
- If no resources allocated, KEDA-HTTP will pause incoming HTTP request as it spawns resources
- AW Gitlab Bot from gerrit patch.
- Creates link to Environment: aw-930161.wmcloud.org, which will spin up MW using that patch.
- https://gitlab.wikimedia.org/repos/abstract-wiki/aw-ci-chart/-/blob/main/templates/wikifunctions.yaml
- AO: can we use this pattern to deploy other things from CI?
- Stef: yes, but if you don’t need scale down to 0, you might just be able to make CI do an actual temporary deployment.
- AK: how does it work?
- BK: do we have any auto scaling at WMF now?
- AK: not really, we do have one use case: video scaling. When lots of work, we just throttle.
- BK: alternate container runtimes?
- AK: looked a bit, looking at containerd as a side project. Feels very close to docker, mostly. It has namespaces? Might be making an OKR for the next quarter. Want to use gadvisor(?) for wikifunctions.
- AK: I am going on sabbatical starting sept. 8. Janis will be replacing me as meeting leader.
- [Stef] Magic trick: KEDA HTTP autoscaler
- Topic: <no specific topic was discussed>