Why it matters
ACID enables slowly-changing dimension patterns, GDPR-required deletions, and correction workflows without rewriting whole tables. This was a huge unlock for enterprise Hive adoption.
The architecture
ACID tables must be bucketed and stored in ORC format. Each transaction writes a new delta file rather than modifying existing files. Base files hold the original data; delta files hold inserts and updates for each transaction.
The transaction manager tracks open, committed, and aborted transactions. Reads filter out uncommitted or aborted data via valid transaction id lists.
How it works end to end
The lock manager (Zookeeper-backed) coordinates concurrent transactions. Row-level locking is not enforced; snapshot isolation prevents conflicts.
Minor compaction merges delta files. Major compaction merges base + all deltas into new base files, applying all mutations.
MERGE INTO enables UPSERT-style operations. UPDATE and DELETE work on any ACID table.