crashconsistent

Crash-consistent is a property of a storage or data-management system that ensures a valid, well-defined state after an unexpected crash or power loss. A crash-consistent system guarantees that completed updates are durable and partially completed updates do not leave data structures or on-disk metadata in an invalid state. In practice, this means that operations appear atomic with respect to crashes, and recovery procedures restore invariants.

Techniques used to achieve crash consistency include journaling or write-ahead logging, copy-on-write, and log-structured file systems; atomic commits; and the use of barriers and flushes to enforce the ordering of writes to non-volatile storage. Metadata-only transactions, redo/undo logging, and controlled replay during recovery also help maintain consistency. The goal is to ensure that after a crash, the system either reflects all of a requested update or none of it, avoiding partially applied states.

In practice, crash consistency interacts with durability and atomicity and depends on both software mechanisms and hardware support. In databases and file systems, crash-consistent designs enable correct recovery by replaying logs or reconstructing consistent metadata. In persistent-memory and modern storage systems, developers must explicitly persist and order writes to maintain crash consistency, often via dedicated APIs or transactional libraries, and rely on hardware features such as non-volatile memory and cache barriers to enforce persistence guarantees.