Changelog
This changelog provides a detailed record of updates and enhancements we have made to improve your Dremio Cloud experience.
April 23, 2026
You can now create and manage service users for automated tools, jobs, and applications. Service users are non-human Dremio accounts that enable secure programmatic access to Dremio services and datasets.
April 21, 2026
Fixed an issue where trial-ended and project-hibernation notification emails were not being sent to customers from the scheduled batch deactivation path due to a missing tenant context when looking up org details.
Fixed credential validation failure when configuring the native Anthropic or Amazon Bedrock AI provider with Claude Opus 4.7, caused by Dremio sending the deprecated temperature parameter that the model rejects.
April 14, 2026
Iceberg v3 Timestamp Granularity (Preview): You can now read and write data source tables that have nanosecond data types, allowing you to operate over timestamps with greater precision than ever before.
Iceberg v3 Column Default Values: Dremio now supports column default values as specified in the Iceberg v3 specification, enhancing schema management and data consistency.
You can now update Okta and Microsoft Entra ID authentication credentials directly in the Dremio console by using the new Edit option, without deleting and re-adding the identity provider.
You can now securely update and manage credit card details in the Dremio console.
Reflections now support a Source Consistency property that specifies whether refresh is managed automatically by Dremio (Dremio Managed) or manually by users (User Managed). This property is shown in the Acceleration section of a raw profile, helping users understand refresh behavior and troubleshoot why a Reflection may or may not have the latest data.
Fixed an issue where view schemas did not get updated in Open Catalog when underlying tables were changed.
Fixed an issue where the search shortcut (Cmd+K) did not work while typing in the SQL editor.
Renamed Users to User management in the organization settings sidebar, and removed Role from the summary view on the User management page.
Fixed an issue where APPROX_PERCENTILE with a FILTER clause failed with an internal error, and LISTAGG with a FILTER clause silently dropped the delimiter argument.
Fixed an issue where Dremio would generate invalid SQL syntax when pushing down constant values to Amazon Redshift.
Fixed a crash that could occur during queries involving hash aggregation with large variable-length columns when the accumulated data in a single batch exceeded 2 GB.
Added fail-fast write-permission validation for Azure Storage sources that use Microsoft Entra ID service principals or Azure Managed Identity. CREATE TABLE, CTAS, and INSERT now fail immediately with a clear "No write permission" error instead of failing later with downstream Azure writer errors.
Improved concurrency safety in S3, Azure Storage, and Glue connectors by removing externally observable synchronization patterns in client creation and credential refresh flows.
Added cleanup for object store source initialization failures so failed setups no longer leave partial resources behind.
Fixed an issue where the SQL planner could apply duplicate runtime filters to the same scan when multiple hash joins share equivalent build subtrees.
Folders that are not queryable no longer show an incorrect job count of "0" on the Open Catalog page.
Fixed an issue where source icons appeared stretched in the lineage view.
Fixed an issue where the cursor and text selection did not align correctly in the SQL Runner.
Fixed an issue where hovering over the UPLOAD_FILE privilege in project privileges caused the UI to error out.
Fixed an issue where granting ownership on Open Catalog tables or views without specifying a branch would fail with an error instead of using the default branch.
Fixed a bug where the incorrect Iceberg snapshot rollback was targeted when a Reflection OPTIMIZE job is cancelled or fails.
Reflection names are now unique within a project, allowing you to reference Reflections by name (instead of ID) in REST APIs, SQL commands, Reflection hints, and table functions. During the upgrade, already-unique names are preserved, and duplicates are automatically renamed using the format <originalName>_<datasetPath>_<type>_<index>. ID-based references continue to work, and new Reflections must have unique names.
Fixed an issue where non-admin users could incorrectly access the Semantic Layer.
The AI agent now correctly handles datasets whose names contain dots or special characters when referenced via @mention.
Fixed an incorrect sample query in the in-app documentation for the DECRYPT function.
Fixed an issue with long planning times for SQL queries with many joins when planner statistics were disabled.
Glue Iceberg REST Catalog sources can now authenticate using an IAM role via AWS assume-role, enabling more secure and centralized access management without long-lived credentials.
Dremio's Open Catalog now supports cross-region S3 access, allowing you to query data stored in S3 buckets across different AWS regions.
You can now configure billing event alerts to monitor usage and costs in your Dremio Cloud organization.
First-time users now see a guided provisioning banner on the Home page to help them get started with Dremio Cloud.
AI configurations are now listed on a separate page in the Dremio console settings, where you can access and configure AI model providers if you have an ADMIN role or CREATE MODEL PROVIDER permission.
A model provider selector is now available when configuring AI features, allowing you to select and switch between available AI model providers.
The Dremio MCP server now exposes OAuth metadata for dynamic client registration, improving compatibility with OAuth 2.0-aware tools and agents.
The Dremio MCP server now includes improved authentication observability, detailed tool error logging, and a 60-second token expiry buffer to reduce unexpected session timeouts.
Dremio now supports Amazon Bedrock Guardrails for requests sent to models through Bedrock. You can configure a centrally managed guardrail ID, which Dremio enforces at runtime to help meet governance requirements (for example, personally identifiable information (PII) protection and policy controls).
Fixed an issue where the owner_id and owner_type values in the sys.project."tables" and sys.project.views system tables were not being populated correctly.
Fixed an issue where the Created By field in project information showed as "Unavailable" for public users with the USAGE privilege.
Fixed an issue where adding an engine schedule and modifying other engine properties could not be done in a single update.
Fixed a SQL compatibility issue where certain queries failed with "Incorrect syntax near the keyword 'AS'".
Fixed an issue where CASE or IF expressions with a union output type containing a scalar or non-union reader branch threw a compile exception.
Fixed an issue where a Reflection displayed the warning "Reflection cannot be accelerated. 1 refresh job failed" even though the refresh history contained no failed jobs.
Fixed an issue where queries on a promoted file-based dataset returned a "/ by zero" error after the source files were deleted, instead of a clear message indicating the files were missing.
Fixed a regression where the _utf8() function failed when executed via Arrow Flight after an upgrade.
Fixed an issue where the AI agent incorrectly recommended creating a table in a space, which is not supported in Dremio Cloud, when the intended catalog source was the appropriate target.
Fixed an issue that prevented the engine dropdown menu from working as expected when modifying a view.
Fixed an issue where date/time extraction functions returned inconsistent output types depending on whether Gandiva vectorized execution was applied.
Fixed an issue where the format_number function could cause an OutOfMemoryError when called with a very large number of decimal places.
Fixed an issue where CREATE OR REPLACE VIEW failed with "Column not found" for struct sub-fields when replacing an existing view with a stale schema.
Fixed an issue where REFRESH DATASET jobs could become stuck and could not be cancelled.
Fixed an issue where Learn More links from SQL functions in the Dremio console directed users to the legacy documentation site instead of the current documentation.
Fixed an error that caused queries on externally created Apache Iceberg tables to fail when Parquet column IDs did not match the Iceberg schema.
Fixed an issue where querying dremio_catalog returned a 403 error.
Fixed an issue where the lineage graph displayed a job count of 0 for datasets that had existing jobs.
Fixed an issue where queries could fail with an IllegalStateException due to an operator allocator being closed while buffers were still allocated.
Reflection refresh job profiles now include the refresh trigger reason and trigger type as a comment, making it easier to understand why a refresh was initiated.
Fixed an issue where interval literal values of 100 or greater (for example, INTERVAL '100' SECOND) failed with an "exceeds precision" error.
Fixed an issue where views created using a catalog context before a catalog migration could fail to load after the migration.
Fixed an issue where system queries exposed internal catalog names in query results.
Fixed an issue where tables or views that were created or deleted did not appear in the AI agent's autocomplete suggestions until the page was manually refreshed.
Fixed an issue where creating a serverless project with a malformed IAM Role ARN returned a generic error instead of a specific validation message.
Fixed an issue where the AI agent showed no error message when users attempted to interact with it while the AI toggle was disabled at the project level.
Fixed a regression where assigned roles for users could not be viewed on the Users page in the Dremio console.
April 7, 2026
Fixed an issue where the NVL function returned an error when used with mixed DATE and TIMESTAMP operands.
March 31, 2026
Lineage nodes now display a Created timestamp.
Fixed an issue where writing an Iceberg table with positional deletes to a REST catalog source could fail when running DELETE, UPDATE, or MERGE DML queries.
Fixed an issue where writing an Iceberg table with deletion vectors to Open Catalog could fail when running DELETE, UPDATE, or MERGE DML queries.
Fixed an issue that prevented users from exchanging External Token Provider (ETP) JWT tokens for Dremio access tokens via the OAuth token exchange grant.
Fixed an issue rendering bar charts with temporal data.
Fixed an issue where visualizations could not be viewed by non-admin users who had permissions on the underlying objects.
Fixed an issue where reading Parquet files with DATA_PAGE_V2 pages marked as uncompressed under a compressed column codec could fail with a decode error.
Fixed an issue where tooltips could display raw HTML.
Fixed an issue where visualizations could show stale data after the underlying view was updated.
Fixed an issue where visualization search results did not return the correct entry.
March 25, 2026
Fixed an issue where the Start new chat button and Clear chat action in the AI agent were not working as expected. Support for Start new chat has been removed at this time and will be reintroduced in a future update.
March 24, 2026
Iceberg v3 Data Types (Preview): Iceberg v3 data types including VARIANT and UUID are now supported in Preview. VARIANT enables flexible storage and querying of semi-structured data like JSON. UUID allows you to read and write universally unique identifiers.
Iceberg v3 Features (Preview):
Row lineage adds row_id and last_updated_sequence_number metadata columns to v3 tables, tracking each row's identity and modification history across updates and compaction. This enables CDC, audit trails, incremental processing, and regulatory compliance.
Iceberg v3 deletion vectors improve performance when updating or deleting data in large tables.
OPTIMIZE TABLE now processes tables in resource-aware batches, preventing out-of-memory errors on large datasets. Multiple runs may be needed for very large tables. Use automatic table maintenance for automatic optimization.
OpenAI-compatible models can be used with Dremio for the AI agent and AI functions.
For users who have permissions to create and alter objects in the catalog, write operations (DML/DDL) are enabled from the agent of your choice via the Dremio MCP Server. The agent will ask for the user to verify and approve the action.
Preview: Google BigLake Iceberg tables can now be queried through the Dremio Iceberg REST Catalog, enabling seamless access to BigLake-managed datasets.
Typing "@" into the Agent chat now allows searching through catalog objects.
The AI agent is available on the Catalog and Jobs page.
SQL blocks in an AI agent chat now have a Run in SQL Runner button to simplify SQL execution.
The AI related project configurations have been moved out of the Preferences page and into a separate AI pages under Admin > Project.
Queries using AI functions with a LIMIT clause now apply the limit before invoking AI functions, improving performance and reducing token usage.
Fixed an issue where jobs were missing end timestamp.
Fixed an issue where Iceberg REST Catalog sources on Azure Data Lake Storage (ADLS) Gen2 failed during query execution.
The namespaces section will now have a max height and is scrollable to allow for easier access of the connections section.
Improve the user experience when adding/deleting namespaces by handling loading states in a smoother manor and adding a go to namespace link with the success notification.
Fixed an issue where the Catalog API returned a 500 error when requesting permissions on the top-level catalog listing.
Fixed a regression where legacy spaces could be created via the Catalog API on projects using Open Catalog.
Fixed an issue where Open Catalog appended /catalog to the base location.
Updated the privileges required to run OPTIMIZE on Open Catalog to SELECT and WRITE.
Fixed an issue where Reflection creation could fail due to validation errors.
Reflection results on Iceberg tables are now consistently up‑to‑date with the latest table snapshot, even with external writers.
Fixed an issue where the LEAST and GREATEST functions returned incorrect results when comparing DECIMAL values with different precision and NULL values.
Improved error messages when running metadata queries (such as table_snapshot or table_history) on views.
Fixed an issue where the Data panel only displayed namespaces instead of all sources.
Removed the "Query Type" column from the SQL Runner's results table.
Updated ChannelClosedException query failure errors to surface the underlying exception that caused the channel to close, i.e. out of memory errors.
Fixed a direct memory leak that occurs in certain scenarios such as query cancellation when the query has a JDBC scan in one of its fragments.
Previously a user didn't see any region selection validation when trying to create a project. Now when they should see a warning under the region selector.
Reduced API response time for Dremio console trial status endpoint.
March 18, 2026
Fixed an issue where folders in Open Catalog could become inaccessible with contents disappearing from view due to an internal error.
February 24, 2026
The AI Agent can now answer questions about upstream sources, downstream dependencies, and how a dataset was created or transformed. The AI Agent can also add a wiki and labels to a dataset.
Lineage information for Iceberg views in Open Catalog is now automatically updated when views are created, modified, or refreshed using ALTER VIEW ... REFRESH METADATA.
In SQL Runner, if a query fails, an Explain button is displayed that uses the AI Agent to provide more information about the error and how to fix it.
Result sets returned by Dremio's AI Agent can now be downloaded as a CSV file.
The Visual Profile now shows information about C3 cache utilization across scan operator threads, helping identify cache configuration issues.
Fixed an issue where S3 connections using an HTTP proxy could ignore the proxy or use an incorrect scheme/port due to proxy SSL configuration handling.
Fixed an issue where Tableau project files downloaded from the Dremio console for Open Catalog datasets contained duplicate catalog names in the path.
Fixed an issue where the total-file-size field in Iceberg table metadata was incorrectly set to zero, which could cause compatibility issues with external tools like Databricks.
Fixed an issue where Redshift sources did not display databases created from datashares.
Fixed an issue where query plans could fail when querying a user-defined function with a Reflection hint.
Fixed an issue where queries on complex data types (structs, lists, unions) using KVGEN and FLATTEN with a WHERE clause could return NULL values.
Fixed an issue where grouping sets with a single column caused the WHERE clause to not apply to NULL values.
Fixed an issue where inserting a NULL element into a list of structs threw an exception.
The Open Catalog page now features an updated layout and branding that highlights Open Catalog powered by Apache Polaris.
The side navigation bar now features labeled menus with expand and collapse functionality.
Column data types now display tooltips in the Open Catalog tree, on metadata cards, and in the Details panel.
You can view and grant project privileges on the Project Settings page in the Dremio console.
Fixed an issue where the lineage graph did not render all parent datasets for public users with SELECT privileges.
February 14, 2026
Fixed an issue where the MCP server and search API returned dataset paths with catalog names included.
Fixed an issue where Tableau and Power BI live connections downloads (.tds and .pbids files) from the Dremio console contained incorrect dataset paths.
February 5, 2026
Fixed an issue where the AI Agent could generate incorrect SQL syntax for function mappings, ILIKE syntax, reserved keyword quoting, and interval/array access patterns.
Fixed a thread-safety issue in the catalog HTTP client that could cause table list retrieval failures during concurrent operations.
Native OAuth applications now support refresh token rotation for improved security.
Fixed an issue where the letter "c" could not be typed in the Create Table from File dialog.
January 27, 2026
Fixed an issue where incorrect files could be deleted when cleaning up old materializations for Reflections.
January 22, 2026
Inactive projects now automatically hibernate after 15 days of inactivity to optimize resource usage and reduce costs.
The AI Agent now supports Mermaid diagrams for visualizing workflows, architectures, and data relationships.
You can now easily create an Iceberg table from a file in object storage with the Create Table from File functionality.
Fixed an issue where multiple LEAD functions partitioned on the same columns could return incorrect results.
Fixed an issue where queries lost the DESC collation when row/column policies were used.
Fixed an issue where created views had extra columns from ORDER BY when ORDER BY expressions were not in the SELECT list.
Fixed an issue where INSERT INTO operations failed with a privilege error.
Fixed an issue where semantic search did not match column names with different capitalization.
January 14, 2026
Fixed an issue where Reflections were incorrectly matched into user queries, producing the wrong results.
December 18, 2025
Dremio Cloud is now available in AWS Europe (Ireland) and Germany (Frankfurt), offering improved performance, lower latency, and enhanced compliance with EU data residency and GDPR requirements.
Creating functions on another project's Open Catalog without navigating to that project first is now properly disallowed.
Setting and unsetting row-filtering and column-masking policies on another project's Open Catalog (added as a source) is now properly disallowed with an improved error message.
Fixed an issue where full refresh Reflections could stop working after repeated refreshes with no data changes.
Reflection refresh job profiles now include a new "Source Table Refresh State" phase that displays tables and other Reflection scans that were planned to build the new materialization.
Fixed an issue that resulted in a truncated file when exporting a query result containing a large number of rows or columns from the Dremio console.
The Catalog API now returns isMetadataExpired and lastMetadataRefreshAt in responses for tables and views.
Fixed an issue where the GET /reflection/recommendations/job-based/{id}/results/ REST API response contained incorrect endpoint URLs.
Fixed an issue where the output rows count was incorrect on the Jobs page for queries accelerated by results cache.
Fixed an issue where the Catalog Usage admin monitor failed to load due to an error processing queries containing a LIST_FILES function call.
When navigating to the SQL Runner, a loading indicator is now shown in the Data Panel.
December 4, 2025
Fixed an issue where the planner threw a "cannot translate call DO_NOT_PUSHDOWN" exception.
Fixed an issue in the vectorized copier where an exception was incorrectly thrown during column vector compaction.
Queries involving evaluation of large numbers of expressions are now more responsive to cancellation.
Fixed an issue where position delete files for Apache Iceberg v2 tables were not being sorted by file_path and pos according to the Iceberg specification.
Fixed an issue where an invalid range error would be thrown when reading a Delta Lake table that used column mapping.
Fixed an issue with validation during the PREPARE statement from the JDBC driver for some DML queries with parameters. Querying complex types with parameters and certain DML queries selected from a subquery with parameters are not supported.
Fixed an issue where SQL queries using tables from the sys source intermittently failed with a "Failure getting source [sys]" error.
Fixed an issue of Dremio process aborting when complex data type columns were involved in hash join queries.
Fixed an issue with runtime filters on DECIMAL columns incorrectly filtering out values.
Fixed an issue where filters on VARBINARY columns could produce inconsistent results.
In the Events tab on the Engines page, the last query time now shows the correct datetime or "--" instead of an invalid date.
Added engine replica numbers to the Engines page for failed events.
Fixed an issue where deleting a namespace you were viewing resulted in a not-found error.
The Maintenance filter in the Attributes dropdown on the Jobs page now appears automatically upon entering the page.
November 20, 2025
Fixed an issue where Iceberg table write operations could create invalid Parquet files when nullability checks were disabled in Dremio, leading to DATA_READ ERROR failures during subsequent queries.
Fixed an issue where sources in a broken state could not be reconfigured.
Reflections are now blocked on JDBC sources (such as Microsoft SQL Server and Oracle) with user impersonation enabled. If user impersonation is enabled after a Reflection is created, users should delete existing Reflections to prevent unauthorized data access, as these Reflections will fail to refresh and expire eventually according to their corresponding policy.
All values in the Attributes filter on the Jobs page now act as OR filters, showing jobs that match any of the selected criteria.
November 12, 2025
Introducing the new Dremio Cloud, featuring the built-in Open Catalog (Powered by Polaris) for seamless data governance and interoperability, Dremio's Query Engine to deliver on fast performance and Dremio's AI Agent so you can discover, analyze, and visualize data using natural language with no SQL expertise required. Try it out with our 30-day free trial by signing up at dremio.com/get-started.