Snapshot
A Kubernetes-native checkpoint and restore system for NVIDIA GPU workloads. It saves a fully initialized pod's process and GPU memory, then restores that state on another node so new replicas skip cold start.
Snapshot is a Kubernetes-native checkpoint and restore system for NVIDIA GPU workloads. It pauses a running pod and saves its complete execution state, including CPU and GPU memory, as a portable artifact, then restores that artifact on any node with matching GPU hardware and driver versions so the process resumes where it left off, skipping model loading and warm-up.
It provides checkpoint and restore primitives only; orchestration decisions such as which pods to checkpoint and when are left to systems that integrate it. Typical uses include speeding up autoscaling for inference serving, parking idle models at scale-to-zero, and recovering a pod's state after a restart or a move to another node.
It is written in Go, licensed under Apache-2.0, and its APIs may still change, so the project is not yet recommended for production-critical workloads.