Update autoscaling/v1 to autoscaling/v2beta2

[pmalek-sumo]

Description

Update autoscaling/v1 to autoscaling/v2beta2 in order to enable custom metrics autoscaling and autoscaling based on memory utilization.

This PR adds targetMemoryUtilizationPercentage key to enable autoscaling based on memory utilization.

ref: https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale-walkthrough/#autoscaling-on-multiple-metrics-and-custom-metrics

Testing performed

• ci/build.sh
• Redeploy fluentd and fluentd-events pods
• Confirm events, logs, and metrics are coming in

[sumo-drosiek]

@frankreno @perk-sumo
Do we want to support more factors than CPU? IMHO, we should avoid to copy k8s templates to values.yaml if it's possible

[frankreno]

I agree with @sumo-drosiek. I feel like we should be able to handle this w/o a breaking change as well. In the values.yaml we can have the same key we do today (targetCPUUtilizationPercentage). We can modify the HPA with new format in the template as opposed to the values. If customers want to add more metrucs, we can add an extraHPAMetrics property that the template could render no?

[pmalek-sumo]

I agree with @sumo-drosiek. I feel like we should be able to handle this w/o a breaking change as well. In the values.yaml we can have the same key we do today (targetCPUUtilizationPercentage). We can modify the HPA with new format in the template as opposed to the values. If customers want to add more metrucs, we can add an extraHPAMetrics property that the template could render no?

@frankreno

The reason I'm hesitant on this approach is that

• we would be "overriding" the "scope" (?) of targetCPUUtilizationPercentage (v1 vs v2beta) - this in itself wouldn't be that big of a deal
• we already had a thread on slack about a customer who'd like to configure hpa based on memory, this will require either extraHPAMetrics as suggested or something like targetMemoryUtilizationPercentage which doesn't exist neither in v1 nor in v2beta2 which might confuse users

Not my call, just chippin' in :)

[perk-sumo]

There is one problem with doing a breaking change and fully migrating - it's still a beta version.
I'd rather have some "scope" inconsistence than changing our configuration back and forth in case the hpa api changes.

[pmalek-sumo]

There is one problem with doing a breaking change and fully migrating - it's still a beta version.
I'd rather have some "scope" inconsistence than changing our configuration back and forth in case the hpa api changes.

Ok, sure. Let's do that then.
One thing though: do we set the default and if so then what value do we want for targetMemoryUtilizationPercentage?

[perk-sumo]

Let's set it at 50% but not use it by default.

[perk-sumo]

Let's not change the values.yaml keys around autoscaling just yet.

[pmalek-sumo]

Removing breaking-change label as we decided to use targetMemoryUtilizationPercentage key for autoscaling based on memory utilization

[pmalek-sumo]

@perk-sumo Please take a look one more time

[perk-sumo]

👍