You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
At present when an AnalysisRun is in progress the HPA can continue to scale. While this behavior makes sense as a default in order to ensure the needs of an application are met it can also lead to additional complexity. The ask is to add a feature that will pause HPA scaling during the analysis. This will avoid confusion when deploying and will simplify the analysis which may need to account for the auto-scaling if it is comparing traffic on the stable against the canary traffic.
The feature would also reduce the time a rollout takes if the HPA scaled up and caused the cluster autoscaler to add nodes during the analysis.
Use Cases
This would be an option that could be specified as part of a Rollout spec and would be used whenever a Rollout and the HPA are used in conjunction with each other.
Message from the maintainers:
Impacted by this bug? Give it a 👍. We prioritize the issues with the most 👍.
The text was updated successfully, but these errors were encountered:
. Isn't this dangerous? What happens if your load goes up during the Analysis run and you would have needed HPA to scale up? You can't predict these things...
I definitely don't see this as the default behavior. For workloads with predictable behavior in terms of traffic load I see this as simplifying the rollout. What we've seen in production is that the HPA will decide to scale up by one or two pods as the canary is being promoted. This particular application just runs one pod per node so the CA also scales up one or two nodes accordingly. Now the canary also has to scale up before it completes the promotion. With the additional time needed to add nodes to the cluster it has caused a fair amount of confusion where individuals running deploys assume their Rollout has failed in some manner.
We run with enough headroom that if the pods had not been placed we would still be able to process all traffic.
I could be convinced that by virtue of addressing #1029 this feature would not be needed.
Summary
At present when an AnalysisRun is in progress the HPA can continue to scale. While this behavior makes sense as a default in order to ensure the needs of an application are met it can also lead to additional complexity. The ask is to add a feature that will pause HPA scaling during the analysis. This will avoid confusion when deploying and will simplify the analysis which may need to account for the auto-scaling if it is comparing traffic on the stable against the canary traffic.
The feature would also reduce the time a rollout takes if the HPA scaled up and caused the cluster autoscaler to add nodes during the analysis.
Use Cases
This would be an option that could be specified as part of a Rollout spec and would be used whenever a Rollout and the HPA are used in conjunction with each other.
Message from the maintainers:
Impacted by this bug? Give it a 👍. We prioritize the issues with the most 👍.
The text was updated successfully, but these errors were encountered: