checkpointrestore

Checkpoint/restore, often written checkpoint/restore, is a technique for saving the complete state of a running computation or execution environment to durable storage so that it can be restored and resumed at a later time. It is used to provide fault tolerance, support long-running workflows, and enable live workload migration.

A checkpoint typically captures the memory contents, processor state, and the state of runtime resources such as open file descriptors, sockets, and in-flight I/O. Restoring reconstructs the execution context and reestablishes resources. Checkpointing can be performed in full, recording all state, or incrementally, recording only changes since the previous checkpoint. Coordinated checkpoints require application-wide agreement on a consistent save point, while uncoordinated approaches save independently and rely on recovery mechanisms to resolve inconsistencies.

Common domains include process and container management as well as virtual machines. In Linux user space, tools like CRIU (Checkpoint/Restore In Userspace) enable checkpointing and restoring of individual processes and containers, supporting live migration and fault-tolerance workflows. DMTCP is another user-space checkpointing framework for multi-threaded applications. For virtual machines, hypervisors such as QEMU/KVM implement live migration by transferring a VM’s state to another host.

Applications include high-performance computing, database systems, and cloud platforms where paused workloads must be resumed without data loss. Checkpointing also serves as a maintenance mechanism to pause and resume long-running computations during upgrades or hardware changes.

Challenges include handling non-deterministic state, external resources, and network connections; interacting with kernels and device drivers; ensuring portability across architectures; and managing the performance overhead and security implications of saving and restoring state.

See also: checkpoint, process migration, live migration, fault tolerance, persistence.