KubeElasti: True Scale-to-Zero for Kubernetes Services
Shriira Press
True scale-to-zero for Kubernetes HTTP services — idle workloads drop to zero replicas, and not a single request is lost when traffic returns.
Welcome to KubeElasti: True Scale-to-Zero for Kubernetes Services.
KubeElasti is a CNCF Sandbox project that closes a long-standing gap in Kubernetes autoscaling. The Horizontal Pod Autoscaler and KEDA are excellent at moving a workload between one and many replicas, but they will not take it all the way to zero, because a service with no pods has nowhere to send the next request. KubeElasti solves this with a small operator and a lightweight resolver: when a service goes quiet it is scaled to zero, and when a request arrives the resolver catches it, queues it, triggers a scale-up, and replays the request once pods are ready — so a cold start costs latency but never a dropped connection. This book starts with the idle-workload problem and the economics behind it, then works through the ElastiService custom resource, the operator-and-resolver architecture, the request path that distinguishes proxy mode from serve mode, the way KubeElasti cooperates with HPA, KEDA, and Prometheus rather than replacing them, and finally how to install, operate, and reason about it in production alongside your existing ingress and service mesh.
This title is part of the ShriIra library and is free to read in full, right here — our small contribution to making world-class knowledge easy to reach.
A note on reading it: open the Contents menu at the top of the reader to jump between chapters, use the Aa menu to set a comfortable text size, theme (light, sepia, or night), and single- or two-page layout. Your place is saved automatically, so you can always pick up where you left off.
We hope it serves you well.
— Shriira Press