Average of metric value over a time window

[prasad-pillala-by]

Hi,

My requirement is to autoscale based on a average of kafka lag for last 10 minutes. For example, if the average of incoming kafka messages stays above 1000 for last 10 minutes, then I want to add more pods using HPA.

Means, I want to avoid acting on a sudden spike of 2000 messages 1 time, which can be handled in 5 minutes by current pod. If in-flow continues for 10 minutes and the average is more than 1000; then I want to add more pods to handle the load.

Can you please help me on how to achieve this?

Thanks,
Prasad

[tomkerkhove]

Looks like you are looking for pollingInterval:

This is the interval to check each trigger on. By default KEDA will check each trigger source on every ScaledObject every 30 seconds.

Example: in a queue scenario, KEDA will check the queueLength every pollingInterval, and scale the deployment up or down accordingly.

[prasad-pillala-by]

I think pollingInterval check the instant value at this moment. Does it support average of last 10 minutes values i.e. 20 values average?

[tomkerkhove]

Oh that way, I don't think that's supported at the moment, right @zroubalik

[prasad-pillala-by]

If this is not available today. Can you consider it as an enhancement?
Thanks

[zroubalik]

Yeah, it is not available. @prasad-by let's open an issue with feature request. So we can discuss it there.