How Batch-Based Commits Keep Code Migrations Safe During Interruptions

Software engineer Anton, working on migrating a PHP monolith into Go services, shares a workflow pattern for safely stopping complex multi-step tasks mid-execution. The core principle is that work should only be committed at the boundary of a 'batch' — a single executor's ordered pass through a set of iterations — rather than after each individual iteration. This ensures that any interruption leaves no uncommitted or partially reconstructed work behind, since only fully completed batches are ever committed. Tasks are organized into iterations, batches, waves, and stages, with batches within a wave running in parallel while waves themselves follow a strict sequential order. The approach addresses a specific failure mode: needing to stop mid-task without losing track of what has been done and what still needs to be completed.