serving What is the mechanism of panic mode?

Ask your question here:

What is the mechanism of panic mode?Why does it trigger when the target concurrency is reached or exceeded by a factor of two?Is there any rationable behind setting it to twice the target concurrency?Can this value be modified?Is there a data flow diagram for painc mode?I only found the diagram for stable mode at this URL:https://github.com/knative/serving/blob/main/docs/scaling/SYSTEM.md

Oct 18 '23 03:10 baosiling

I found a way to set the panic mode threshold. https://knative.dev/docs/serving/autoscaling/kpa-specific/

Oct 18 '23 07:10 baosiling

See https://github.com/knative/serving/blob/main/docs/scaling/README.md#autoscaler. I don't think there is more to it.

If too many requests appear in the short time, the Autoscaler panics, which means it decides the required scale based on a shorter window. In normal scenarios, the Autoscaler decides on a trailing average of the past 60 seconds, but in panic mode it's on the last 6 seconds only. This makes the decisions more sensitive to bursty traffic.

Dec 07 '23 10:12 ReToCode