Technology

Delta UniForm

A Delta Lake feature that automatically generates Iceberg and Hudi metadata for Delta tables, enabling cross-format reads without data copying.

6 connections2 resources1 post

Summary

What it is

A Delta Lake feature that automatically generates Iceberg and Hudi metadata for Delta tables, enabling cross-format reads without data copying.

Where it fits

UniForm is Delta Lake's native answer to the interoperability problem. Instead of using an external translator (XTable), UniForm maintains Iceberg-compatible metadata as a side-effect of every Delta write. This makes it simpler for Databricks-centric environments but limits control.

Misconceptions / Traps

UniForm does not support all Delta features in Iceberg mode. Liquid Clustering, for example, is not compatible with UniForm-generated Iceberg metadata.
Iceberg metadata is read-only. External Iceberg engines can read but cannot write to UniForm-exposed tables.

Key Connections

depends_on Delta Lake — Delta-native feature
enables Apache Iceberg — generates compatible metadata for Iceberg readers
competes_with Apache XTable — alternative interoperability approach

Definition

What it is

A Delta Lake feature that automatically generates Apache Iceberg and Apache Hudi metadata for Delta tables, enabling engines that read Iceberg or Hudi to access Delta data without any data copying.

Why it exists

Delta Lake is the default format in the Databricks ecosystem, but many query engines (Trino, Snowflake, Athena) prefer Iceberg. UniForm bridges this gap by maintaining Iceberg-compatible metadata alongside Delta's native metadata, eliminating the interoperability penalty.

Primary use cases

Multi-engine access to Delta tables via Iceberg clients, format interoperability in mixed Databricks/non-Databricks environments.

Recent developments

Latest signals

Delta 4.3: UniForm's Iceberg conversion becomes atomic and incremental (June 22, 2026). Large commits now convert atomically within the Delta transaction, and incremental conversion regenerates only the changed log range instead of the full metadata tree — the two changes that make UniForm viable on high-commit-rate production tables. 4.3 also ships IcebergCompatV3 (experimental), allowing deletion vectors and UniForm to coexist for the first time, built against Iceberg-Spark 1.11.0 with Spark 4.0/4.1 support — though IcebergCompatV3 tables are not yet readable through Unity's external REST API, so the bleeding edge trades away external-engine access for now. Per Delta Lake 4.3 release announcement and Databricks external-access docs.
The 4.0.0 regression is context worth knowing. Delta 4.0.0 (September 2025) dropped Iceberg UniForm entirely due to Spark 4.0 incompatibility; it was restored in 4.0.1+ against Iceberg 1.10.x. Teams pinning Delta majors should treat UniForm support as version-gated, not assumed. Per Data Lakehouse 2026: Iceberg vs Delta vs Hudi (birjob) and Delta Lake UniForm 2026 status (Sparking Scala, May 30, 2026).
Interop-layer maturity framing (July 2026). Practitioner synthesis now treats the interop layer as mature enough for "manageable mixed estates": UniForm publishes Delta as Iceberg, Paimon ships an Iceberg-compat mode, XTable translates all three, and a Delta 5.0 proposal would collapse the metadata boundary entirely. Per Lakehouse Table Formats in 2026 (dev.to, July 2026).
Cross-format reads matured; Delta UniForm is the canonical "Delta-but-readable-as-Iceberg" answer. Per Dremio's Iceberg-vs-Delta-Lake comparison guide, Delta's strongest features and performance are still inside the Databricks ecosystem, but UniForm has closed the cross-format read gap for Trino, Snowflake, and Athena clients — those engines can now read Delta tables as Iceberg without a separate copy. Combined with Delta Connect (the Spark Connect protocol for Delta), the UniForm pattern lets one source-of-truth Delta table serve both Databricks workloads and Iceberg-only readers without ETL duplication.
Foundational design context still authoritative. Per Databricks's launch blog, UniForm's design principle is metadata-only — no data file copying — making the cross-format read essentially free for the write side and effectively native for the read side. That blog (August 2023) remains the load-bearing reference for understanding what UniForm does and does not do; it has not been superseded by a newer architecture post.
UniForm's Iceberg-metadata translation adds roughly 5% write overhead. Reads and writes stay interoperable without manual format conversion, but generating Iceberg-compatible metadata alongside the Delta transaction log costs a measurable, if modest, tax on write throughput. Tertiary source, flagged pending better corroboration. Per How Delta UniForm works.
UniForm's real limitations in 2026: no deletion vectors, no Merge-on-Read, and commit batching that can compromise time-travel fidelity. Mainline UniForm does not support tables with deletion vectors enabled and prohibits Merge-on-Read operations outright; because the asynchronous Iceberg-metadata writer often batches multiple Delta commits into a single Iceberg commit, snapshot-level time travel on the Iceberg side can diverge from the Delta transaction log's actual commit history. (Delta 4.3's experimental IcebergCompatV3, tracked above, is the first attempt to let deletion vectors and UniForm coexist — but it isn't the mainline default, and Unity's external REST API still can't read those tables yet.) Per Iceberg REST Catalog, Actually Explained (Datastrato). Sources: Delta Lake 4.3 release announcement (delta.io, June 22, 2026) · Apache Iceberg vs Delta Lake Comparison Guide (Dremio, February 2026) · Delta UniForm — universal format for lakehouse interoperability (Databricks launch) · Lakehouse Table Formats in 2026 (dev.to)