The party started at 8PM. Your Kafka topic started accumulating lag, your job queue started backing up, your users started waiting. And your pods? Still getting dressed.
HPA isn't broken - it's just fashionably late. It waits for CPU and memory to spike before reacting, and by the time new pods are ready, the damage is done.
KEDA (Kubernetes Event-Driven Autoscaler, a CNCF graduated project): gets your pods to the party on time.
Scale on Kafka consumer lag, GitHub Actions job queues, Prometheus business metrics, and 60+ external triggers. Scale to zero when it's over. Back up the moment the next one starts.
In this talk we'll cover how KEDA works under the hood, walk through three production-grade scalers, and end with a live demo — zero to autoscaling in one YAML file.

Key Takeaways
- Why HPA is always late and what that costs you in production
- How KEDA extends Kubernetes autoscaling with 60+ external scalers
- Production patterns for Kafka, GitHub Actions, and Prometheus scalers
- How scale-to-zero changes your cost model overnight
- Honest tradeoffs: when KEDA is the right tool and when it isn't

Ref blog: https://medium.com/the-tech-matter/your-pods-are-always-late-to-the-party-fix-it-with-keda-ba8bb1d77688