I'd say use both. The goal of the Predictive Autoscaler is to increase the number of pods before the load hits. This gives k8s ample time to go ahead and schedule new pods. Even if the replica count predicted by our autoscaler is off, its fine cause the backup scaler can cover the difference once the load hits.
Hi, great video!
Wanted to know how is this more reliable or resilient than using vanilla Prometheus query in keda, especially in case of spikes?
I'd say use both. The goal of the Predictive Autoscaler is to increase the number of pods before the load hits. This gives k8s ample time to go ahead and schedule new pods.
Even if the replica count predicted by our autoscaler is off, its fine cause the backup scaler can cover the difference once the load hits.
Amazing use case
Glad you liked it
Nice
Glad you liked it