Can I disable HPA while the argo rollout is progressing?

[sungho-rpls]

HPA scales out replicas in my environment because of high CPU usage while the argo rollout progresses. High CPU usage for startup time is acceptable for me, but too many replicas by HPA cause problems with latency, resource usage, and database connections. Is it possible to disable HPA scaling when the argo rollout is progressing?

[kostis-codefresh]

If you are doing blue/green deployment use the previewReplicaCount replica to lock down preview pods.

If you are doing canary deployments use setCanaryScale to lock down canary pods.

Would that work for you?

[florianmutter]

We are facing the same challenge. We have a lot of different workloads where some only run 2 pods and others run 20 or more. Having to come up with a good value for setCanaryScale seems hard. We would like to use the rollout feature of adjusting the pod count according to traffic split.

Right now this does not work at all because the pods do require more cpu on startup. This causes the HPA to start scaling after rollout created the first set of canary pods which causes rollout to adjust the canary replica count which causes even more cpu load.

Would it be somehow possible to make the HPA only look at the stable replicaset and ignore the canary one?
Another option could be to make rollout behave more like deployment rolling update: Only start one canary pod, wait until it is ready, shutdown a stable pod, repeat until desired number of canary pods is reached, start analysis,...

[kostis-codefresh]

Another option could be to make rollout behave more like deployment rolling update

This can be done with the dynamicStableScale Option

See example 07 here https://github.com/kostis-codefresh/rollouts-autoscaling-example