CRI: The Second Boom of Container Runtimes - Harry Zhang & Xu Wang, HyperHQ
summary
CRI standardizes kubelet-to-runtime calls so Kubernetes can swap in containerd, cri-o, frakti or other shims without changing kubelet, and the talk shows how those shims actually implement pod sandboxes, image/storage plugins and CNI/CSI hooks. It explains Kata Containers: VM isolation with ~100 ms startup and two integration approaches — CLI shim vs library integration — plus agent comms via vsock or serial, and demonstrates writing a CRI shim from zero and the practical tradeoffs that determine which shim design is simpler or faster.
AI-written summary · about
chapters (18)
- 0:00Intro
- 1:24A Brief History of CRI
- 3:11The Birth of CRI
- 5:36The CRI Interface
- 7:31Implement a CRI Runtime (2)
- 9:56The Existing CRI Runtimes
- 10:33Containerd (wl cri plugin) in brief
- 12:21CRI-O in brief
- 14:28Frakti 1.0 in brief
- 15:36History of Kata
- 16:22Technical Vision of Kata
- 18:18How Kata Works
- 20:39CRI for Kata Contaienrs
- 22:50containerd/CRI-O + Kata
- 26:06Frakti+ Kata
- 27:26Comparison
- 28:22What's the Next of Frakti
- 30:02Related Sessions in KubeCon EU
related talks
42mEnhance Workload QoS with Pluggable and Customizable Smarter Runtimes
36mMitigating Noisy Neighbours: Advanced Container Resource Management
30mBeyond Orchestration: The Cloud Native Runtimes Ecosystem for Performance and Security
20mFrom Cold Start To Warp Speed: Triton Kernel Caching With OCI Container Images
29mThere is no container
40mUnpacking the Container: A Deep Dive into Virtualized Container Technology
27mKubeletosaurus Wrecks: When Memory Requests Meet Limits!
01h 02mOptimizing Containerized #Java Apps with Jib, #GraalVM, CDS, & #CRaC