CheckpointPlusLog

CheckpointPlusLog is a fault-tolerant persistence approach that combines periodic state checkpoints with an append-only change log. The blended model enables fast recovery from crashes while preserving the exact sequence of state-altering operations since the last checkpoint, supporting deterministic replay to the current state.

The architecture typically includes a checkpoint store, a log store, a recovery module, and a coordinating component that triggers checkpoints and manages log flushes. During normal operation, the application periodically captures a snapshot of its in-memory state as a checkpoint and writes it to durable storage. In parallel, all state-changing operations are appended to a log in a strictly ordered fashion. The log may be used to replay events after a checkpoint to reconstruct the latest state.

Recovery proceeds by loading the most recent successful checkpoint and then replaying log records that followed that checkpoint. This combination often provides faster recovery than checkpoint-only approaches and reduces the amount of recomputation needed during rollback. Log compaction or pruning can be employed after a checkpoint to remove entries that are already reflected in the checkpoint, thereby controlling storage growth.

Benefits include faster recovery times, potential support for incremental checkpoints, and clearer separation between state snapshots and event history. Trade-offs involve increased storage overhead and write latency due to dual persistence, as well as added complexity in ensuring consistent replay and effective log management across failure modes.

Use cases span stateful stream processing, distributed databases, and long-running services that require reliable durability and quick recovery. Implementations typically rely on durable storage backends (such as object stores or distributed filesystems) and may use a consensus protocol to maintain log order across nodes. Related concepts include write-ahead logs and traditional checkpointing.