Achieving a Highly Available and Scalable Multi-Tenant...
summary
A reusable Eventing scheduler is implemented inside Knative Eventing that places virtual consumer replicas onto real data‑plane pods, integrates with KEDA autoscaling, and exposes consumer groups/consumers as internal CRDs. Demo runs on IBM Cloud Kubernetes (9 workers across 3 zones) with an Event Streams topic of 100 partitions; KEDA scales a KafkaSource from 12→0→1 replicas and the team manually schedules 60 replicas for 10,000 events to show recovery and delivery behavior. Kafka consumer tunings increase throughput and reduce duplicates.
AI-written summary · about
chapters (7)
- 0:00Highly available multi-tenant backends
- 1:46Event-driven architecture expectations
- 4:19Limitations and scheduler overview
- 8:54Eventing scheduler implementation
- 13:56Kafka source scaling demo
- 22:49Kafka Eventing tuning and future work
- 24:34Closing remarks
related talks
35mKEDA - Real Time And Serverless Scaling In Kubernetes
52mEvent-Driven Autoscaling for Serverless Java
36mKEDA: Event Driven and Serverless Containers in Kubernetes
37mKubernetes Event-driven Autoscaling with KEDA
01h 02mModern Event-Driven Workloads with Knative
37mKEDA: Unlocking Advanced Event-Driven Scaling for Kubernetes
56mEvent-driven autoscaling for Serverless Java
48mLet's Go Knative