Anti-entropy
A background process in distributed databases that compares replicas and repairs differences, so they converge over time toward the same data. "Entropy" = data that has drifted apart; anti-entropy pulls them back in sync.
Why: in leaderless/eventually-consistent systems (Dynamo, Cassandra) replicas can fall out of sync due to network failures, nodes that were down, or missed writes. Anti-entropy (often as "repair") systematically compares what each replica has and fills in what's missing. To avoid comparing all data byte by byte, Merkle trees are used: nodes exchange hashes level by level and only drill down where the hashes differ → efficient detection of exactly which ranges diverged. Distinct from read repair (which fixes lazily, on read) — anti-entropy is proactive and comprehensive. In Cassandra it runs as scheduled "repair", critical especially because of tombstones. Related to Merkle tree and read repair.