Prevent termination of pods with long-running functions using KEDA

[conreaux]

What is the general guidance for preventing termination of pods with long-running functions while using KEDA?
In testing, I've seen that during a scale down, the HPA can terminate a pod that is still running an orchestration/activity function.
This behavior is documented here: https://keda.sh/docs/1.4/concepts/scaling-deployments/#long-running-executions
What's the recommended method to prevent this from occurring?

[conreaux]

Bumping topic. Also, how does this compare with MSSQL provider scale-in on the Elastic Premium and Consumption plans? Does it behave as described here where "functions that are currently executing are given time to finish executing." Does this need to be explicitly handled by the user with KEDA?

[cgillum]

In the case of Azure Functions hosting, the Azure Functions runtime is able to detect when a function is still running and will try to not tear down the container until execution has finished.

It looks like there is more work that we'd need to do to support something similar in Kubernetes. The suggestion to respond to lifecycle hooks seems promising. For Azure Functions, however, the runtime would probably need to be augmented to support this. I don't think there's anything this MSSQL extension could do to prevent the host from shutting down, other than potentially exposing some signal indicating whether there is still work pending (which could be leveraged by some 3rd party component that responds to container lifecycle events).