Automatic feature enablement
Automatic feature enablement is in Public Preview. To enroll, complete this form with your account ID. No code changes or additional configuration are required after enrollment.
Automatic feature enablement (AFE) automatically upgrades Unity Catalog managed tables to use generally available recommended features without requiring code changes or manual ALTER TABLE statements. AFE also verifies that clients are compatible before turning on new features.
AFE provides the following benefits:
- Reduce the administrative effort necessary to validate individual compatibility requirements for each table and feature combination in your workspaces, especially if you have catalogs with thousands of tables.
- Automatically gain the latest performance and reliability improvements on your managed tables.
- Implement upgrades safely. AFE only turns on features after it verifies workload compatibility.
How AFE works
AFE monitors Unity Catalog access patterns at both the table and schema levels and uses a 50-day observation window to verify that access patterns are compatible before enabling any feature. AFE uses serverless compute to upgrade tables in the background.
Schemas and tables
AFE behavior depends on whether the schema and tables existed before AFE was turned on. This table has more detail:
Schema | Table | AFE behavior |
|---|---|---|
New | New | AFE sets schema-level defaults at creation time so that tables immediately inherit all supported features without an observation period. |
Existing | New | AFE turns on a feature if only verified workloads have accessed all tables in the schema within the 50-day observation window. Otherwise, AFE does not turn on features for new tables in the schema if a single unverified workload accessed any table in the schema. See verified workloads. |
Existing | Existing | AFE turns on a feature when all of the following are true:
|
Verified workloads
A workload is considered verified for a specific feature if it accessed the table from a Databricks cluster with a Databricks Runtime version at or above the feature's minimum required version.
The following workloads are considered unverified:
- External clients and third-party services such as Flink or Presto. See Unity Catalog integrations.
- Databricks services with direct or kernel-level table access, such as Zerobus, that bypass standard Databricks Runtime access patterns. See Zerobus Ingest connector overview.
If any table in a schema was accessed within the 50-day observation window by a Databricks Runtime version below the feature's minimum required version or by an external client, AFE does not turn on the corresponding feature on any table in that schema.
Supported features
AFE may automatically turn on the following features:
Feature | What it does | Minimum compatible Databricks Runtime version |
|---|---|---|
Maintains hidden row IDs for incremental processing with change data feed. | 14.1 | |
Enables you to rename and drop columns without rewriting data. | 15.3 | |
Uses a modern checkpoint format to reduce write conflicts and improve reliability on large or frequently-updated tables. | 13.3 | |
Centralizes commits in Unity Catalog to enable multi-table transactions, improves interoperability for external writes, and enables governance policies across engines. | 16.4 |
Feature availability might differ by region.
Requirements
- Serverless compute must be available in your region.
- Tables must be Unity Catalog managed tables in Delta Lake or Apache Iceberg formats.
Observe enabled features
To check if AFE turned on a feature on a table, look for a SET TBLPROPERTIES operation in the History tab in Catalog Explorer, or use DESCRIBE HISTORY <table_name>. If AFE performed the operation, the username field shows a hash value instead of a username, such as 4d137f29-62. See What is Catalog Explorer? and View table history.
After AFE enables features on tables in a new schema, view the schema defaults in the Properties tab in Catalog Explorer. For example, a schema with row tracking enabled shows a property such as catalog.schema.enableRowTracking: "true". Existing schemas don't have AFE observability properties.
Manage AFE recommended features
Administrators can manage AFE behavior and operations with various controls.
Revert changes
Use RESTORE to revert the data and metadata for a table to a version before the feature was turned on:
RESTORE TABLE <table_name> TO VERSION AS OF <version>;
RESTORE TABLE <table_name> TO TIMESTAMP AS OF <timestamp>;
See Restore a table to an earlier state for more information about table history and restore.
Turn off features for tables
To turn off a feature on an individual table:
ALTER TABLE <table_name> DROP FEATURE <feature_name>
AFE doesn't turn on a feature again after you manually turn it off.
Limitations
- Tables shared by Delta Lake Sharing, both Databricks-to-Open and Databricks-to-Databricks, are excluded from AFE. See What is Delta Sharing?.
- AFE does not have a batch rollback mechanism to turn off a feature across all tables in an account. See Manage AFE recommended features.
- Materialized views and streaming tables are not supported.
- Workloads that bypass Unity Catalog and access tables directly by file path are not tracked by AFE. If your workloads use path-based access, contact your account team to discuss compatibility.
- External tables are typically accessed by file path, bypassing Unity Catalog, and with unverified workloads from external clients. Unity Catalog cannot reliably track these access patterns, so external tables are excluded from AFE. See Work with external tables.