Welcome to CRIU, a project to implement checkpoint/restore functionality for Linux in userspace.
Checkpoint/Restore In Userspace, or CRIU (pronounced kree-oo, IPA: /krɪʊ/, Russian: криу), is a software tool for Linux operating system. Using this tool, you can freeze a running application (or part of it) and checkpoint it to a hard drive as a collection of files. You can then use the files to restore and run the application from the point it was frozen at. The distinctive feature of the CRIU project is that it is mainly implemented in user space.
Using
- Installation
- What to do to have CRIU on your system
- CLI and RPC
- Two ways to start using the C/R functionality
- Usage scenarios
- Ideas how criu can be used (some are crazy indeed)
- What software is supported
- Describes TODO list in higher level terms
- Category:HOWTO
- Collection of real world examples of how to use CRIU. Some are complex, some are not. HOW TO dump a simple loop might be the best one to start with.
- What can change after C/R
- CRIU cannot (yet) save and restore every single bit of tasks' state. This page describes what bits visible through standard kernel API are such.
- What cannot be checkpointed
- This article describes what application can do to make CRIU refuse to dump it. It can be used as some sort of troubleshooting guide.
External links
- 2026-09-21, Fast Recovery for LLM Serving via Decoupled Device Memory Lifetime in Dynamo
- 2026-07-22, Forensic Analysis of Container Snapshot Chains for Post-event Reconstruction
- 2026-07-14, Accelerating Model Inference with On‑the‑Fly Snapshot Compression
- 2026-07-09, Reverse-engineering NVIDIA's cuda-checkpoint for Faster Cold Starts
- 2026-06-29, Fast vLLM: cutting cold starts by 20x
- 2026-06-25, Elastic Resource Management for Heterogenous Container-Based Clouds
- 2026-06-01, NVIDIA Dynamo Snapshot: Fast Startup for Inference Workloads on Kubernetes
- 2026-05-12, How we achieved truly serverless GPUs
- 2026-04-27, Towards On-the-Fly Snapshot Memory Compression for Low-Latency Elastic Inference Serving Systems
- 2026-04-06, Cloudburst: 70x faster cold(ish) starts for SGLang
- 2026-02-24, GPU Checkpoint/Restore Made Fast and Lightweight
- 2026-01-21, Announcing the Checkpoint/Restore Working Group
- 2025-11-17, Engine-Agnostic Model Hot-Swapping for Cost-Effective LLM Inference
- More external articles...
