TODO: write developer/operator facing documentation.
There are cases when Operators need to set lower and upper bounds on the number of pods serving their apps (e.g. avoiding cold-start, control compute costs, etc).
The following annotations can be used on spec.template.metadata.annotations
(propagated to PodAutoscaler objects) to do exactly that:
# +optional
# When not specified, the revision can scale down to 0 pods
autoscaling.knative.dev/minScale: "2"
# +optional
# When not specified, there's no upper scale bound
autoscaling.knative.dev/maxScale: "10"You can also use these annotations directly on PodAutoscaler objects.
NOTE: These annotations apply for the full lifetime of a revision. Even
when a revision is not referenced by any route, the minimal pod count
specified by autoscaling.knative.dev/minScale will still be provided. Keep in
mind that non-routeable revisions may be garbage collected, which enables
Knative to reclaim the resources. These annotations are specific to Autoscaler
implementations but NOT subject to Conformance.