What Apache Iceberg table changes actually cost, measured on 20M rows with Trino, MinIO and Docker Compose: which changes are metadata-only, what partition evolution does to query planning, what expire_snapshots really reclaims (zero), and why a table marked copy-on-write is written merge-on-read anyway. Claims mapped to offline tests.
sql docker-compose s3 data-engineering minio parquet query-engine schema-evolution partitioning measurement time-travel iceberg trino apache-iceberg lakehouse table-format
-
Updated
Sep 28, 2026 - Python