Custom Metrics AutoScaler, Red Hat’s KEDA build, makes Kubernetes autoscaling more flexible, event-driven, and production-ready. With support for diverse scalers, fallback behavior, lifecycle events, and observability signals, CMA helps platform teams scale workloads based on the metrics that matter most to their applications.
A key reliability improvement is KEDA’s fallback behavior. When an external metrics source becomes unavailable, CMA can preserve the last-known replica count instead of making unsafe scaling decisions. This behavior is validated for standard Deployments and also for Argo Rollouts, helping teams maintain stability during temporary metrics outages while supporting progressive delivery patterns. You should use the GitOps Operator to use with IBM Power and OpenShift.
KEDA also improves event-driven visibility through the CloudEventSource custom resource. Scaling lifecycle events can be emitted as structured CloudEvents, including meaningful source, subject, and type fields. This makes it easier to integrate KEDA activity with event routers, audit systems, and operational workflows.
The scaler ecosystem remains broad and practical. The cron scaler supports scheduled scale-up and scale-down windows, making it useful for predictable traffic patterns. The kubernetes-workload scaler enables one workload to scale based on the pod count of another workload, with activationValue thresholds helping avoid unnecessary scaling from low-signal activity.
Resource-based autoscaling is also covered through CPU and memory scalers, which scale deployments when container utilization crosses configured percentage thresholds. These provide familiar autoscaling behavior while remaining part of KEDA’s unified scaling model.
For event streaming workloads, the Kafka scaler supports scenarios such as consumer group lag, offset commit policies, and partition distribution strategies. This makes KEDA a strong fit for streaming systems where scale should reflect backlog and consumption pressure.
For metrics-driven platforms, the Prometheus scaler allows deployments to scale from the result of a configurable PromQL query. This gives teams the freedom to scale from application, infrastructure, or business metrics already exposed through Prometheus.
KEDA also strengthens operational confidence through observability. Tests validate that KEDA exports correct OpenTelemetry metrics to collectors and exposes well-formed Prometheus metrics from operator and adapter endpoints. These signals help teams monitor scaler health, adapter behavior, reconciliation activity, and autoscaling outcomes.
Finally, Custom Metrics AutoScaler v2.19 adds support for IBM Power, expanding deployment options for organizations running Kubernetes on Power-based infrastructure.
Together, these features make CMA a powerful Custom Metrics AutoScaler for modern Kubernetes environments: flexible in what it scales from, resilient when dependencies fail, observable by default, and increasingly portable across infrastructure platforms.
Go forth and build with IBM Power and CMA. https://community.ibm.com/community/user/blogs/paul-bastide/2026/06/17/custom-metrics-autoscaler-operator-v2191-supports
Leave a Reply