NRI: Extending Containerd And CRI-O With Common Plugins
summary
NRI turns runtime extension into a formal protobuf protocol with ttrpc services, letting plugins modify a controlled subset of the OCI spec (annotations, env, mounts, devices, CPU/memory pinning, RDT and block‑IO classes) during create/update/stop events. Experimental integrations exist for containerd and CRI‑O; a template plugin and examples (CDI/device injection, OCI hook injection that reproduces CRI‑O behavior, event logger, IRQ‑migration high‑perf hook) are available. PR is near merge; security ACLs are not yet implemented and plugin ordering uses an index.
AI-written summary · about
chapters (7)
- 0:00Start
- 5:55What Is The Benefit of NRI...
- 7:21How Does NRI Work...
- 13:11NRI Execution Model: Plugin Events
- 15:21NRI Data Model: Plugin Input And Output
- 21:22How To Write An NRI Plugin
- 26:29Examples And Use Cases
related talks
28mCNI, CRI, and OCI - Oh My! [I]
24mEnhancing CRI-O With CDI: Streamlining Device Integration in Kubernetes
38mIntroduction to CRI-O
31mCRI-O: The Runtime Control Room
23mOCI, CRI, ??: Making Sense of the Container Runtime Landscape in Kubernetes - Phil Estes, IBM
42mEnhance Workload QoS with Pluggable and Customizable Smarter Runtimes
04mCRI-O: Container Runtime Intensifies
33mCRI-O Still Loves Kubernetes