File formats & storage · Formats & Table Formats

What is Apache Hudi? Compare Delta / Iceberg / Hudi

MediumHudiDeltaIcebergupsertCDCcomparison

Interview question

What is Apache Hudi, and how does it compare to Delta Lake and Iceberg?

How to answer

The short answer

Apache Hudi (Hadoop Upserts Deletes and Incrementals) is a lakehouse table format aimed at incremental data and upsert-heavy pipelines: landing, near-real-time tables, and efficient upserts on cloud storage.

Hudi table types (know these names)

  • Copy-on-Write (CoW): updates rewrite Parquet files; reads stay simple/fast
  • Merge-on-Read (MoR): base files + log/delta files; merge at read (or compact later) for faster writes
Output
CDC / Kafka
    -> Hudi upserts (CoW rewrite OR MoR log append)
    -> queries see latest view after merge/compaction

Comparison table (Delta vs Iceberg vs Hudi)

CapabilityDelta LakeIcebergHudi
Core ideaTx log on ParquetSnapshots + manifestsUpsert/incremental first
ACID / time travelYesYesYes
Upsert / CDC focusStrong MERGEStrong, engine-dependentCore design goal
Write patternsBatch + streaming mergesBatch + streaming writersCoW / MoR trade-offs
CompactionOPTIMIZE / vacuumrewrite / expire snapshotsCompaction critical for MoR
Multi-engineGrowingVery strongStrong Spark/Flink story

Interview tip: Frame Hudi as "built for incremental upserts and MoR/CoW trade-offs," Delta as "transaction log lakehouse," Iceberg as "open metadata + multi-engine." Pick based on CDC intensity and engine mix, not brand loyalty.