Running Wasm Inside Your Storage Cluster With CSI and Gateway API
summary
A Kubernetes-native storage gateway runs user-defined WebAssembly modules next to NVMe devices, using CSI and Gateway API for volume lifecycle and HTTP routing, Helm/OCI delivery for modules, and Rust/Wasmtime-style runtimes. A 29.61 GB log input reduces to 6.64 MB returned (29.61 GB -> 6.64 MB (99.978% reduction)), and end-to-end latency falls from 46.98s (client filter) to 25.62s (gateway Wasm) while native offload hits 15.45s (25.62s wasm vs 46.98s client vs 15.45s native). Multi-tenant isolation and full host-acceleration remain future work.
AI-written summary · about
chapters (12)
- 0:00Wasm in storage clusters
- 1:29Moving computation to data
- 4:00Wasm and Kubernetes architecture
- 7:25Operator workflow and integration
- 11:47Client contract and registry
- 14:22Security boundaries and policies
- 18:00Limitations and multi-tenancy
- 19:26Experimental setup and topology
- 21:10Performance results and trade-offs
- 22:16Hardware offload with DPUs
- 24:25Key conclusions and future work
- 25:30Q&A
related talks
33mKubernetes on the edge: running low-latency workloads with WebAssembly
30mWASM in the Wild West: A Practical Application Tale
28mWasm + Kubernetes: Beyond Containers
29mSupercharge Your Kubernetes AI Deployments with Wasm
50mWebAssembly: A Recovering Kubernetes Engineer's View of the Future
39mMind the Gap Between the Future and the Present
31mBuilding Event-Driven WebAssembly on Kubernetes: Runtime, Observability, and…
42mExotic Runtime Targets: Ruby and Wasm on Kubernetes and GitOps Delivery Pipelines