Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Databricks automatically upgrades Unity Catalog managed tables to use recommended features that are generally available, without requiring code changes or manual ALTER TABLE statements. Automatic upgrades verify that clients are compatible before turning on new features.
Automatic upgrades provide the following benefits:
- Reduce the administrative effort necessary to validate individual compatibility requirements for each table and feature combination in your workspaces. This is especially helpful if you have catalogs with thousands of tables.
- Automatically gain the latest performance and reliability improvements on your managed tables.
- Upgrade tables safely. Automatic upgrades only turn on features after verifying workload compatibility.
For more details on the benefits associated with these features, see Automatic Upgrades: best practice features for your lakehouse tables.
How automatic upgrades work
Automatic upgrades monitor Unity Catalog managed table access patterns and use an observation window to verify that access patterns are compatible before turning on any feature. The observation window is 50 days for upgrades in Public Preview, and 100 days for upgrades that are generally available. See Supported features.
Automatic upgrades use serverless compute to upgrade tables in the background. There is no charge for this process.
Schemas and tables
The behavior of automatic upgrades depends on whether a schema is new or already existed when the feature was released. When you create a new table, it inherits features from its schema's default properties. The following table describes each case:
| Schema | Table | Behavior |
|---|---|---|
| New | New | Automatic upgrades set schema-level defaults at creation time so that tables immediately inherit all supported features. |
| Existing | New | Automatic upgrades turn on a feature only if all tables in its schema were accessed by verified workloads during the prior observation window. Otherwise, if a single unverified workload accessed any table in the schema, automatic upgrades ignore the new table. See verified workloads. |
| Existing | Existing | Automatic upgrades turn 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.
Automatic upgrades consider the following workloads unverified:
- External clients and third-party services such as Flink or Presto. See Unity Catalog integrations.
- Azure Databricks services with direct 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 observation window by a Databricks Runtime version below the feature's minimum required version or by an external client, automatic upgrades don't turn on the corresponding feature on any table in that schema.
Supported features
Automatic upgrades apply to a subset of generally available features, described in the following tables. Feature availability might differ by region.
Important
Automatic liquid clustering applies only to new tables. Unlike the other features, it is added by default when you create a table and never applies to existing tables.
Features with upgrades that are generally available
The following features are generally available, and automatic upgrades apply them to the specified table types. The automatic upgrade system is also generally available.
Each feature rolls out gradually, starting on its release date and reaching all customers within about six months.
| Feature | Table types | Minimum compatible Databricks Runtime version | What it does | Release date |
|---|---|---|---|---|
| Automatic liquid clustering |
|
15.4 LTS | Automatically organizes table data based on frequently queried columns to improve query performance without manual partitioning. Automatic upgrades don't apply this feature to existing tables. | May 22, 2026 |
| Checkpoint V2 |
|
13.3 LTS | Supports more concurrent writers and reduces write conflicts on large or frequently updated tables. | May 19, 2026 for new tables in new schemas July 13, 2026 for all tables in existing schemas |
| Row tracking |
|
14.1 | Maintains hidden row IDs for incremental processing. When row tracking is enabled, automatic change data feed is available without additional configuration. See AUTO CDC APIs. | July 25, 2026 for new tables in new schemas July 13, 2026 for all tables in existing schemas |
| Catalog commits |
|
16.4 LTS | Centralizes commits in Unity Catalog to allow multi-table transactions, improve interoperability for external writes, and allow governance policies across engines. | July 13, 2026 |
| Parquet v2 |
|
18.1 | Uses advanced Parquet encodings, data page headers, and INT64 timestamps to improve query performance and reduce storage on Delta Lake tables. |
June 25, 2026 |
Features with upgrades in Public Preview
The following features are generally available, but their automatic upgrades are in Public Preview and require enrollment.
Important
Automatic upgrades for the following features are in Public Preview. To enroll, complete this form with your account ID. No code changes or additional configuration are required after enrollment.
Automatic upgrades apply to these features only if you are enrolled in the Public Preview, and they take effect immediately after you enroll:
| Feature | Table types | Minimum compatible Databricks Runtime version | What it does |
|---|---|---|---|
| Catalog commits |
|
16.4 LTS | Centralizes commits in Unity Catalog to allow multi-table transactions, improve interoperability for external writes, and allow governance policies across engines. |
| Column mapping |
|
15.4 LTS | Enables you to rename and drop columns without rewriting data. |
| Parquet v2 |
|
18.1 | Uses advanced Parquet encodings, data page headers, and INT64 timestamps to improve query performance and reduce storage on Delta Lake tables. |
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 automatic upgrades 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 automatic upgrades 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 automatic upgrades turn on 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 turned on shows a property such as catalog.schema.enableRowTracking: "true". Existing schemas don't have automatic upgrades observability properties.
For account-wide visibility, query the automatic upgrades system table. This table records each feature that automatic upgrades add to a table, when it was added, and which table was affected, so you can audit upgrade activity across all of your workspaces. See Automatic upgrades system table reference.
Manage recommended features
Administrators can revert changes from an upgrade or turn off features on individual tables.
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>
Automatic upgrades don'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 automatic upgrades. See What is OpenSharing?.
- Automatic upgrades don't have a batch rollback mechanism to turn off a feature across all tables in an account. See Manage automatic upgrades recommended features.
- Materialized views and streaming tables are not supported.
- Workloads that bypass Unity Catalog and access tables directly by path are not tracked by automatic upgrades. If your workloads use path-based access, contact your account team to discuss compatibility.
- External tables are excluded from automatic upgrades. External tables are typically accessed by file path, bypassing Unity Catalog, and Unity Catalog cannot reliably track these access patterns. See Work with external tables.
Frequently asked questions
The following answers common questions about automatic upgrades.
What are automatic upgrades for Unity Catalog managed tables?
See the introduction at the top of this page.
How do automatic upgrades check for compatibility?
See Verified workloads.
Do automatic upgrades change my tables automatically?
Yes. After a feature is verified safe for a table, Databricks applies it through a lightweight background job. You can still turn off features on individual tables.
If I turn off a feature on a table, can automatic upgrades turn it on again later?
No. After you turn off a feature that automatic upgrades added, automatic upgrades don't turn that feature on again for that table.
Do automatic upgrades change my existing tables?
Yes, but only after the observation window confirms that every client that accessed the table supports the feature. Automatic liquid clustering is the exception: it applies only to newly created tables, never to existing ones, because doing so would change their existing data layout.
How are automatic upgrades different from predictive optimization?
Predictive optimization maintains your data layout through operations such as compaction and vacuum, with the option to use automatic liquid clustering. Automatic upgrades turn on new table features, such as row tracking or Checkpoint V2. The two are complementary: one keeps tables well maintained, and the other keeps them current. Automatic liquid clustering is applied to new tables through automatic upgrades. See Predictive optimization for Unity Catalog managed tables.
How do automatic upgrades verify that a table is safe to upgrade?
Automatic upgrades only turn on generally available features that don't materially reduce performance or raise cost. Automatic upgrades wait through the observation window, require every accessing client to be compatible, skip tables they can't fully verify, and let you turn off any feature on a table at any time.
If my table changed, how can I tell it was automatic upgrades?
Every change that automatic upgrades make appears in the table's DESCRIBE HISTORY output and the History tab in Catalog Explorer, marked distinctly from your own changes. For account-wide visibility, query system.storage.table_auto_upgrade_operations_history to see when any feature was added to any table. See View table history.
Will automatic upgrades break a table that my external or open source tools read?
No. Tables accessed by external or open source clients are out of scope. Automatic upgrades only act when they can verify that every client accessing a table supports the feature.
Which tables are eligible for automatic upgrades?
See Requirements and Limitations.
How long until my tables are upgraded? When will I see changes?
Automatic upgrades use the observation window to capture infrequent workloads, such as monthly batch jobs, quarterly reports, and ad hoc analysis, before acting. After a table is verified compatible, the feature is applied shortly afterward through a background job. When a feature rolls out for the first time, the rollout is gradual across customers and tables, so it can take about six months to reach your tables with compatible workloads.
What do I need to do to turn on table features automatically in Databricks?
No action is required. Automatic upgrades evaluate and upgrade eligible tables without configuration.
Can I turn off automatic upgrades for my whole account, or for a schema, catalog, or workspace?
Automatic upgrades can't be turned off entirely, but you can turn off any individual feature on a table at any time. After you do, automatic upgrades don't turn that feature on again for that table.
Do automatic upgrades cost anything?
There is no charge for the automatic upgrade process (the background ALTER TABLE operations).
Do features applied through automatic upgrades cost me anything?
In general, features that materially increase costs are excluded from automatic upgrades. Several applied features, such as deletion vectors and Parquet v2, lower storage and compute costs. Row tracking adds a minor one-time upfront cost on very large tables because it assigns a unique identifier to every row, but it also helps you save on Materialized views by enabling incremental refresh instead of a full recomputation.
How can I keep track of features that will roll out through automatic upgrades in the future?
Check the supported features to see which features are in Public Preview and which will roll out to all customers soon. To be notified when new features ship, set up an RSS feed on the Azure Databricks release notes page. You can also check What's coming?.