Databricks Makes Multi-Statement Transactions Generally Available on the Lakehouse
Databricks announced general availability of multi-statement transactions across Delta Lake and Apache Iceberg tables on its lakehouse platform.
Quick answer
What did Databricks announce about multi-statement transactions on the lakehouse?
Databricks announced that multi-statement transactions are now generally available across both Delta Lake and Apache Iceberg tables, letting organizations run atomic, multi-table warehouse workflows directly on the lakehouse instead of relying on separate transactional warehouse systems.
Key takeaways
- Databricks said multi-statement transactions are now generally available for Unity Catalog managed tables in both Delta Lake and Apache Iceberg formats.
- Databricks said it is the first lakehouse architecture to support multi-statement transactions natively across both table formats.
- The company said the feature uses row-level concurrency, allowing concurrent transactions to write to the same table as long as they do not touch the same row, which Databricks said delivers stronger performance than traditional warehouses that use locks to serialize writes.
- Databricks said transactions respect Unity Catalog attribute-based access controls, including row-level security and column masking, on both reads and writes.
- Fivetran and Accolade described using the feature in production, including Fivetran removing reconciliation code for its History Mode connector and Accolade applying it to roughly 1,700 daily claims and liabilities transactions.
Databricks said multi-statement transactions are now generally available for Unity Catalog managed tables across both Delta Lake and Apache Iceberg formats. The company said this makes it the first lakehouse architecture to support multi-statement transactions natively across both open table formats, allowing organizations to run atomic, multi-step warehouse workflows without standing up a separate transactional warehouse system.
The problem Databricks says it addresses
According to Databricks, traditional and cloud warehouses provide transactional guarantees but often require a fragmented architecture. The company listed three specific costs: additional data copies and movement created by shuttling data into and out of a separate warehouse, fragmented governance from duplicating and synchronizing access policies and auditing across systems, and complex consistency workarounds that require custom orchestration and retry logic to keep related changes aligned. Databricks said multi-statement transactions bring these guarantees directly to the open, governed lakehouse instead.
What the feature includes
Databricks described several components of the new capability. Transactions work across both Delta Lake and Apache Iceberg tables, so teams do not need to move governed data into a proprietary format or maintain a separate transactional copy. The company said the lakehouse's row-level concurrency model allows concurrent transactions to write to the same table as long as they do not update the same row, which it said performs significantly better than traditional warehouses that use locks to serialize writes.
Databricks also said transactions are composable with existing warehouse features, including stored procedures, temporary tables, and SQL scripting, and can write to streaming tables so downstream pipelines receive consistent updates rather than partial data. On governance, the company said transactions respect Unity Catalog attribute-based access controls, including row-level security and column masking, on both reads and writes. For production use, Databricks pointed to commands including DESCRIBE TABLE, SHOW COLUMNS, and GET DIAGNOSTICS for inspecting tables and transactional execution.
Customer accounts cited by Databricks
Databricks included statements from two companies. Egidio Terra, a staff product manager at Fivetran, said the company's History Mode feature requires two destination writes per source change, and that it previously defended against partial failure with reconciliation logic. Terra said that with multi-statement, multi-table transactions, Fivetran now commits both writes as one atomic unit and removed what he called an entire class of consistency-repair code.
Akshay Ajgaonkar, an engineering manager at Accolade, said the company processes roughly 1,700 transactions daily in production to handle incoming claims and liabilities. Ajgaonkar said multi-statement, multi-table transactions let Accolade run those operations with full transactional guarantees directly on its governed lakehouse as part of a broader migration of workloads off a separate warehouse system.
Use cases outlined
Databricks described three patterns it said the feature supports: giving ETL jobs multi-table transactional guarantees so related updates, such as order status and inventory balance, cannot fall out of sync; migrating stored procedures and temporary-table workflows from traditional warehouses to the lakehouse while preserving Unity Catalog governance; and implementing a Write-Audit-Publish pattern that stages data, records quality check results, and publishes validated records in one atomic operation, rolling back if validation fails.
Databricks said users can get started with transactions through its documentation or by prompting the in-product Assistant.
Source: Databricks blog, "Transactions are Generally Available, advancing warehousing on the Lakehouse," published October 16, 2026.
Frequently asked questions
- What exactly became generally available?
- Databricks said multi-statement transactions are generally available for Unity Catalog managed Delta Lake and Apache Iceberg tables, letting multiple statements across multiple tables commit or roll back as one atomic unit.
- Why does Databricks say this matters for warehouse workloads?
- Databricks said organizations previously needed separate warehouse systems for transactional guarantees, which added data movement, duplicate data copies, and fragmented governance; the company said transactions now bring those guarantees directly to governed lakehouse data.
- How does Databricks say performance compares under concurrent writes?
- Databricks said its row-level concurrency model lets multiple transactions write to the same table in parallel as long as they do not update the same row, which it said performs significantly better than traditional warehouses that lock entire tables or partitions to serialize writes.
- Which customers did Databricks cite using the feature?
- Databricks cited Fivetran, which said it uses multi-table transactions to commit paired writes for its History Mode connector atomically, and Accolade, which said it processes about 1,700 daily claims and liabilities transactions using the capability.