Delta Lake and UniForm Trial

Overview

Delta Lake remains the ACID table format anchor for lakehouses; UniForm and related interoperability efforts expose Iceberg or Hudi compatibility without duplicate data copies (Delta Lake, UniForm).

Trial when multiple engines must read the same governed tables for analytics and AI feature pipelines. Validate performance and compatibility on your workload mix before committing catalog policies.

On Databricks the feature is now documented as "Iceberg reads"; it requires Unity Catalog registration, column mapping and Databricks Runtime 14.3 LTS or above, and Iceberg clients get read-only access (Read Delta tables with Iceberg clients). In open-source Delta Lake, Iceberg support is generally available and Hudi support is still marked as preview (Delta UniForm docs).

Adoption Signals

• UniForm with Iceberg compatibility reached general availability on Databricks in June 2024, with M Science quoted as writing "a single copy of our data that can be queried by any engine that supports Delta or Iceberg" (Databricks UniForm GA announcement).
• In May 2026 Databricks made Iceberg v3 generally available across managed, foreign and UniForm-enabled tables and named Spark, Trino, Flink, Snowflake, DuckDB and pandas as engines that can use Unity Catalog's Iceberg tables (Unity Catalog and the next era of Apache Iceberg).
• Snowflake can query UniForm tables through a catalog integration against Unity Catalog's Iceberg REST endpoint, a pattern a Databricks employee documents step by step on the Databricks Community technical blog (Read Delta tables with Snowflake via Unity Catalog).
• The open-source project shipped Delta 4.1.0 (February 2026), 4.2.0 (April), 4.3.0 (June) and 4.4.0 (August 2026); 4.3.0 made Iceberg conversion atomic and incremental and added Spark 4.1 support for UniForm (Delta 4.3.0 release notes, Delta releases).
• The same Delta-to-Iceberg pattern is built into Microsoft Fabric: OneLake stores tables in Delta Lake and generates virtual Iceberg v2 metadata for Iceberg readers (Use Iceberg tables with OneLake).
• The delta-io/delta repository had about 9,000 GitHub stars and 2,200 forks at the end of September 2026 (Delta Lake GitHub).

Risks

• Read-only interoperability. Iceberg and Hudi clients can only read UniForm tables; in the open-source variant this cannot be enforced, so an unaware external writer or cleanup job can corrupt the Delta table (Delta UniForm docs).
• Stale metadata for path-based readers. Databricks documents that path-based Iceberg clients may need manual metadata refreshes and can fail on out-of-date versions after VACUUM removes data files (Read Delta tables with Iceberg clients).
• Version-matrix breakage. UniForm was unavailable on Spark 4.0 in Delta 4.0.0 and, as of Delta 4.4, the delta-iceberg module does not support Spark 4.2 (Delta 4.4.0 release notes); upgrading Spark can silently remove Iceberg interoperability.
• Catalog coupling. Open-source UniForm is deprecating Hive Metastore in favour of Unity Catalog as the Iceberg committing catalog (Delta 4.2.0 release notes), so the table-format decision is effectively also a catalog decision.
• Feature restrictions on v2 readers. Tables served as Iceberg v2 cannot use deletion vectors, and materialized views and streaming tables cannot enable Iceberg reads (Read Delta tables with Iceberg clients); check which Iceberg version each consumer engine supports.

Recommendation

Trial Delta Lake with UniForm when Delta is already the write format, typically on Databricks or Fabric, and other engines such as Snowflake, Trino or DuckDB need read access to the same governed tables. It is a good fit for serving gold-layer and feature tables to downstream consumers without export jobs. Before the pilot, register the tables in Unity Catalog, enable column mapping, decide per consumer whether it reads Iceberg v2 or v3, and test the exact Spark, Delta and Iceberg version combination you will run in production.

Do not use UniForm as a multi-writer strategy: when several engines must write the same tables, native Iceberg tables (see Apache Iceberg and Apache Polaris and Unity Catalog on this radar) are the better starting point. Pin the delta-iceberg module version in the build, add a check that Iceberg metadata is current after each pipeline run, and document which catalog commits Iceberg metadata so that catalog migrations do not break external readers.

Sources

• Delta Lake docs
• Delta UniForm docs
• Read Delta tables with Iceberg clients (Databricks)
• Apache Iceberg in Databricks
• Delta Lake GitHub
• Delta Lake releases
• Delta 4.4.0 release notes
• Delta 4.3.0 release notes
• Delta 4.2.0 release notes
• Databricks UniForm GA announcement
• Unity Catalog and the next era of Apache Iceberg
• Read Delta tables with Snowflake via Unity Catalog
• Use Iceberg tables with OneLake
• Apache XTable