SYSTEM Cited by 2 sources
Keda (Kubernetes Event-Driven Autoscaler)¶
Keda is an open-source (CNCF graduated) event-driven Kubernetes pod auto-scaler — a controller on top of the Horizontal Pod Autoscaler (HPA) that adds scaling triggers beyond CPU / memory. Canonical Keda triggers: queue length (systems/aws-sqs, Kafka topic lag, RabbitMQ depth), custom metrics from Datadog / Prometheus / CloudWatch, scheduled cron triggers, external event systems.
Minimum viable page — expand on future Keda-internals sources.
Why the CNCF triggers matter¶
HPA alone on CPU/memory is coarse — a queue-consumer pod can be idle on CPU while a large backlog piles up on SQS. Keda scales pods against the metric that actually expresses demand (queue length) and then scales them to zero when the queue drains, pairing naturally with systems/karpenter node-level scaling.
Seen in¶
- sources/2024-08-08-figma-migrated-onto-k8s-in-less-than-12-months — Figma lists Keda as one of the CNCF primitives that motivated the ECS→EKS migration. Explicitly called out triggers: CPU utilization, AWS SQS queue length, custom Datadog metrics. Keda (pod-level autoscaling) was deferred from the initial migration to a fast-follow so that it wouldn't add complexity to the already- large ECS→EKS cutover.
- sources/2026-04-06-aws-unlock-efficient-model-deployment-simplified-inference-operator-setup-on-amazon-sagemaker-hyperpod
— Keda bundled as the autoscaler in AWS's
SageMaker HyperPod
Inference Operator EKS add-on (2026-04-06). The add-on
provisions a dedicated KEDA Operator IAM role via IRSA
(
keda.auth.aws.irsa.roleArnin the add-on's configuration blob). Notable shape: Keda bundled as part of a managed inference-platform add-on (lifecycle managed by AWS) rather than as a customer CNCF-primitive choice — a new delivery-shape for the event-driven-autoscaler-on-HPA pattern.