What happens when two engines write the same Iceberg table at once: snapshot isolation, optimistic commits, conflict detection, and when retries fail.
Variant shredding turns messy JSON into Parquet columns with statistics. How the layout works, how readers reassemble values, and why some queries prune.
Managed and external Iceberg tables mean different things on every platform. Five ownership dimensions and a translation method for vendor vocabulary.
Apache Iceberg v3 deletion vectors for high-throughput streaming ingest: how bitmaps and Puffin files fix CDC write amplification and read decay.
