programagain

Programagain is a conceptual framework and collection of tools designed to manage the lifecycle of running programs by enabling automatic re-execution and state preservation after failures or preemption. It is described in software engineering discussions as a means to improve reliability and reproducibility of long-running tasks.

At its core, programagain combines checkpoints, deterministic replay, and retry policies. A program run is instrumented with a lightweight runtime that can capture input state, environment, and outputs at defined points, and can trigger a subsequent run if the process exits abnormally or is interrupted. Users can configure backoff strategies, maximum retries, and conditions for retries.

The typical architecture consists of an orchestrator component, a runtime agent tied to the target program, and a plugin interface for integration with scheduling systems, cloud storage, and logging. Checkpoints may be stored locally or remotely, enabling resume from the most recent checkpoint. The design emphasizes idempotence and minimal intrusion into the target program's code.

Common use cases include batch data processing, scientific simulations, workflow pipelines, and continuous integration tasks where long runtimes risk interruption. Limitations include added overhead, potential non-idempotence in some programs, and the need for careful configuration to avoid repeated failures. It is not a standardized or widely implemented technology, and several projects implement similar ideas under different names.

Related concepts include checkpointing, retry logic, fault tolerance, idempotence, and workflow orchestration.