Custom Metrics

Custom Metrics

Custom and External Metrics for Scaling

CPU is a proxy; an API may scale better on requests per second, a worker on queue depth. Besides Resource, an autoscaling/v2 HPA accepts ContainerResource, Pods (a per-Pod custom metric), Object (a value from another object) and External (a value from outside the cluster). The last three need an adapter serving custom.metrics.k8s.io or external.metrics.k8s.io, and this cluster has none:

Which metrics APIs this cluster servesShell

kubectl get apiservices | grep -e NAME -e metrics

Output

NAME                                SERVICE                      AVAILABLE   AGE
v1beta1.metrics.k8s.io              kube-system/metrics-server   True        33s

Prometheus Adapter 2,094 (github.com/kubernetes-sigs/prometheus-adapter (https://github.com/kubernetes-sigs/prometheus-adapter 2,094 ), v0.12.0) turns PromQL queries into custom metrics. KEDA 434,961 (github.com/kedacore/keda (https://github.com/kedacore/keda 10,549 ), a CNCF graduated project, v2.21.0) brings 70+ scalers (Kafka 129 , RabbitMQ 28,807 , SQS, Prometheus 16,091 , cron) and can scale a Deployment to zero, which a plain HPA cannot. With several metrics, the HPA takes the largest replica count.

MENU