KEDA Interview Q&A
Q: What is KEDA and why use it?
Kubernetes Event-Driven Autoscaler. Scales workloads based on external metrics (Kafka lag, queue depth, Prometheus metrics, cron schedule). Unlike HPA (CPU/memory only), KEDA can scale to zero and from zero. Critical for cost optimization in event-driven architectures.
Q: KEDA vs HPA?
HPA: built-in, CPU/memory based, min replicas >= 1. KEDA: event-driven, 50+ scalers, can scale to zero. KEDA creates HPA under the hood but with custom metrics. They can work together.
Q: What is a ScaledObject?
KEDA CRD that defines scaling behavior: which deployment to scale, what triggers (scalers), min/max replicas, polling interval. KEDA watches the trigger source and adjusts HPA target replicas.
Q: What happens when KEDA scales to zero?
When trigger metric drops to zero (e.g., Kafka topic has no messages), KEDA sets HPA target replicas to 0. Kubernetes terminates all pods. When a new message arrives, KEDA detects it via the scaler, sets replicas to 1+ (minReplicaCount in the spec), and pods spin up. Cold start latency depends on image pull and application startup time.
Q: How does KEDA differ from cluster autoscaler?
KEDA: scales individual Deployments/StatefulSets based on custom metrics (Kafka lag, queue depth, cron). Cluster Autoscaler: scales the number of nodes in the cluster based on pending pods and node utilization. They work together: KEDA scales pods, Cluster Autoscaler adds nodes when pods can't be scheduled.
Q: What are KEDA scalers? Name five common ones.
Scalers are connectors to external metric sources: (1) Kafka — consumer group lag, (2) RabbitMQ — queue length, (3) Azure Service Bus — message count, (4) AWS SQS — queue depth, (5) Prometheus — any custom metric query, (6) Cron — scale based on time schedule, (7) Redis — list length, (8) NATS JetStream — consumer lag.
Q: What is KEDA's TriggerAuthentication?
CRD for managing credentials used by scalers. Stores secrets (Kafka SASL credentials, Azure connection strings) separately from ScaledObject. Can reference: Kubernetes secrets, pod identity (Azure MSI, AWS IRSA), HashiCorp Vault. Avoids embedding credentials in ScaledObject definition.