Apache Iceberg

• Metadata layer: tracks data files + partitions + statistics
• ACID: snapshot isolation, write-audit-publish pattern
• Schema evolution: add/drop columns without rewriting data
• Time travel: query as of specific snapshot / timestamp
• Hidden partitioning: partition by transform (year(ts)), no user impact

-- Spark + Iceberg
CREATE TABLE prod.db.sales (
  id bigint,
  date date,
  amount decimal(18,2)
) USING iceberg
PARTITIONED BY (month(date));

-- Time travel
SELECT * FROM prod.db.sales
FOR TIMESTAMP AS OF '2026-03-01 00:00:00';

-- Schema evolution
ALTER TABLE prod.db.sales ADD COLUMN region string;

• Data Lake
• Parquet
• dbt