Auto-scaling in Openstack without Telemetry
summary
Explains why OpenStack Heat plus Telemetry fails in production: Gnocchi/Aodh development stalled since Rocky, Ceilometer only collects OpenStack metrics, adding custom metrics requires rebuilding three services, and RabbitMQ overload kills listeners. Replaces Telemetry with Prometheus + Faythe bridge that evaluates PromQL rules and calls Heat APIs via OpenStack service discovery. Demonstrates autoscaling actions (scale out/scale in) and a UI, including scaling to 12 instances using configurable intervals, retries, and Faythe’s clustering for reliability.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 0:55Auto scaling in OpenStack
- 2:39Telemetry limitations
- 4:39New architecture approach
- 6:40How the system works
- 9:35System demo and workflow
- 15:22Conclusion
related talks
28mHow and Why We Rebuilt Auto-scaling in OpenFaaS with Prometheus
23mObservability in OpenStack Learnings and building blocks from the SIG Monitoring at SCS
45mDecentralizing SaaS Monitoring Using Open-Source Tools and Standards
28mImprove Monitoring and Observability for Kubernetes with OSS Tools
36mPutting the PaaS in OpenStack
30mOpenStack Works, so now what?
03mRally: Benchmarking as a service for OpenStack
12mOpenStack for Computational workloads