Hudi & the format landscape

1. "When would you actually choose Hudi over Iceberg or Delta?"

Choose Hudi when the workload is mutable and upsert-heavy (CDC ingestion, frequent primary-key updates). Also choose it when you want incremental pulls: "give me only what changed since T". The record-level index of Hudi maps keys to file groups. So an upsert touches only the affected files. Its timeline drives incremental queries. For append-only analytics, that machinery is only overhead.

2. "Iceberg vs Delta for a greenfield batch lakehouse — how do you decide?"

For batch, they have comparable features. So decide on ecosystem fit. Databricks + Unity Catalog

• liquid clustering → Delta (native). Vendor-neutral, many query engines, and you want free partition evolution and hidden partitioning → Iceberg. I would not decide on a benchmark that ran on a workload unlike ours.

3. "You run Iceberg but a new tool only reads Delta. Migrate?"

Usually no. Use interop. Use Delta UniForm (it exposes Iceberg-compatible metadata) or Apache XTable (it translates metadata between formats). Then you can read one copy of the data as both formats. Migration is costly. Interop solves the real need, which is another engine that reads the data. It does this without moving the data.

4. "Explain Copy-on-Write vs Merge-on-Read and when each wins."

CoW rewrites whole data files on update. Writes are more expensive, reads are cheap, and it is the simplest. MoR appends delete or log files and merges them at read time. Writes are cheap and fast. But reads do extra merge work until compaction. Upsert-heavy and streaming workloads favor MoR. Read-heavy workloads with infrequent updates favor CoW. All three formats offer both.

5. "What's the single most overrated factor in the format debate, and the most underrated?"

Overrated: micro-benchmarks. They rarely match your workload, and the formats converge. Underrated: the catalog and governance layer (Unity, Polaris, Glue, Nessie) and your existing engine ecosystem. They drive day-to-day operability far more than the on-disk format.