Serverless on Kubernetes: WASM, Knative, or Regular Autoscaling – What Actually Works?
summary
Compares three approaches with numbers and failures: AWS Lambda scaled from 50 to 1,300 in under 2 minutes and gives predictable one-request-per-instance behavior; vanilla Kubernetes HPA is reactive and noisy—HPA reaction often takes on the order of minutes and node provisioning/image pull can add tens of seconds to minutes; Knative adds a queue and scale-to-zero but requires a service mesh and can waste concurrent requests; server-side WASM (Spin, tinyGo experiments) is promising for tiny web handlers but remains fragile and documentation-limited so it does not yet replace Lambda.
AI-written summary · about
chapters (22)
- 0:00Serverless autoscaling challenges
- 1:51AWS Lambda scaling impressed me
- 3:18Serverless flavors on Kubernetes
- 4:43Pod lifecycle delays
- 6:31Image pull times impact
- 7:33HPA metric pitfalls
- 9:22Scaling delays and latency
- 12:13Scaling up vs down complexity
- 13:18Kubernetes autoscaling reaction time
- 15:03Debugging performance issues
- 15:24Introducing Knative
- 16:30Knative scaling demo
- 17:33Knative service mesh dependency
- 17:53WebAssembly for Kubernetes
- 18:58WebAssembly past failures
- 20:23WebAssembly runtimes on K8s
- 21:27WebAssembly ecosystem gaps
- 22:53SpinKube vs WasmCloud
- 23:57WebAssembly limitations
- 24:40WebAssembly progress outlook
- 25:00Lambdas vs K8s vs Knative vs WASM
- 26:02Q&A
related talks
33mTowards Serverless As Commodity: A Case of Knative
01h 02mServerless Architecture Patterns
47mAdvanced Serverless Architectural Patterns on AWS
33mKubernetes on the edge: running low-latency workloads with WebAssembly
33mLeveling Up Wasm Support in Kubernetes
41mIs Serverless the End of Kubernetes?
21mBeyond the Basics: Production Serverless Patterns for Extreme Scale
31mBuilding Event-Driven WebAssembly on Kubernetes: Runtime, Observability, and…