Adventures in stateful applications on Kubernetes
summary
Shows that Kubernetes StatefulSets handle persistent-data apps but not in-memory, long-lived computations, so a small framework called acbar prototype is built to assign data sources as partitions to running pods and expose a partition-to-worker ratio metric via a custom metrics adapter to drive HPA. Demo uses a Redis-backed config store, the HPA scales up correctly when a fourth camera is added, but autoscaler overreacts briefly and scale-down can lose the desired pod→partition binding.
AI-written summary · about
chapters (6)
- 0:00Stateful apps on Kubernetes
- 4:15Understanding stateful sets in Kubernetes
- 11:59StatefulSets limitations
- 17:57Introducing Acbar
- 23:00Acbar demo
- 27:00Q&A on Acbar
related talks
25mStatefulsets on Kubernetes: The Guardians of Your Data Galaxy
08mGetting Creative with HPA Scaling in Kubernetes
30mHorizontal Pod Autoscaler Reloaded - Scale on Custom Metrics
42mImproving Availability for Stateful Applications in Kubernetes
35mKubernetes Persistent Data Challenges – AZ, Region and Multi-Clou...
35mStateful Apps On Kubernetes - Bring Them On!
29mStateful Workloads in Kubernetes: A Deep Dive
31mAn Adventure in Multi Cloud with Kubernetes and Distributed Data