Kubernetes and Checkpoint/Restore
summary
Claims that CRIU-based checkpoint/restore now works end-to-end in Kubernetes (merged for kubernetes 1.25), using kubelet APIs to call containerd/crio which invoke CRIU to produce tarball images pushed to registries. Demos show a stateful Python counter surviving reboot, fast startup by importing checkpoints and spawning multiple clones, and migration via registry restore; interactive migration still shows ~4s downtime. Implementation uses injected parasite code, respects PID namespaces, and restricts checkpoint archives to root for security.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:21Project background
- 3:12Technical and use cases
- 13:53Reboot and save demo
- 16:47Quick startup demo
- 21:13Live migration demo
- 27:28CRIU internals
- 35:07Future and conclusion
related talks
26mKubernetes and Checkpoint Restore
27mState of Checkpoint/Restore in Kubernetes – Now With GPUs
37mTransparent, Infra-Level Checkpoint and Restore for Resilient AI/ML Workloads
39mContainer Live Migration
32mForensic container checkpointing and analysis
42mForensic container checkpointing and analysis - Adrian Reber
32mEnd-to-End Encryption for Container Checkpointing in Kubernetes
34mLetting Your Containers Dream of Electric Sheep
