Dagster 1.13.8 Release Refines IO Managers and Cloud Integration

Dagster version 1.13.8 arrived on June 4, 2026. This update introduces a defensive change to how standard IO managers handle empty data results and resolves several connectivity issues in the AWS and Databricks integrations.

The full release notes and downloads are on the GitHub release page.

A notable change in this release affects the BigQuery, Snowflake, and DuckDB IO managers. These managers now skip the table write step when an asset returns an empty DataFrame. Instead of attempting to write zero rows, the system logs a warning.

This adjustment prevents the creation of degenerate tables. When an orchestration engine attempts to infer a schema from an empty dataset, it often defaults to incorrect types or fails entirely if no columns can be typed. This issue frequently occurs in partitioned pipelines where specific time slices might contain no data. By skipping the write, the system maintains the integrity of the existing table structure and avoids errors associated with empty partitions. The warning provides enough context for operators to understand why a write was skipped without stopping the entire pipeline execution.

The dagster-aws package received a fix for the EMR PySpark step launcher. It now correctly honors the configured S3 job package path. This is a critical correction for teams using custom code locations on S3 to bundle their application logic. Without this fix, the launcher could default to incorrect paths, leading to execution failures when the EMR cluster could not find the required job artifacts.

On the cloud operations side, the dagster-cloud metrics now include a server instance id tag for code server metrics. This change improves observability for multi replica code locations. When multiple replicas are running, operators can now identify exactly which gRPC server replica answered a specific metrics ping. This data is essential for debugging performance bottlenecks or connection issues that only affect a subset of the running servers.

The dagster-clickhouse integration received an important security patch. This release fixes a SQL injection vulnerability that appeared when using dynamic partition keys. In previous versions, the system might interpolate partition key values directly into the SQL string. The updated logic now binds these values as driver parameters. This ensures that the ClickHouse driver handles the escaping and prevents malicious or malformed keys from altering the SQL query structure. This fix also applies to the dagster-clickhouse-pandas and dagster-clickhouse-polars libraries.

The dbt integration also saw multiple refinements. The DbtProject.prepare method now ensures that dbt deps runs whenever dependency files are present. This prevents situations where a project might fail to build because of missing packages in the local environment. Additionally, the UI has renamed the Docs tab to Components. This reflects the growing library of reusable dbt components within the ecosystem. Existing links to the docs path will still work through an internal redirect.

Teams using Databricks will find improved stability in workspace interactions. The dagster-databricks library now includes pagination for the workspace job list and adds automatic retries for rate limit responses from the Databricks API. These changes make the integration more resilient for large scale workspaces where the number of jobs might exceed a single API response limit or trigger throttling.

Finally, the release includes several fixes for the dg command line tool. This includes correcting an issue where the AI dispatch command would write a plan file into version control. It also resolves an error when using TypedDict fields in component configuration, ensuring better type safety for complex asset definitions. The system also now allows special characters in team owners for jobs, schedules, and sensors, giving teams more flexibility in how they organize their metadata.

The project continues to expand its visual and descriptive metadata. This version adds kinds tags and icons for Microsoft Fabric and OneLake, allowing these platforms to be represented clearly within the Dagster UI. New documentation was also added for the OpenLineage integration and the dagster-hf-datasets library, alongside an example covering various deployment strategies. These additions help engineers map their data pipelines more accurately to the underlying infrastructure.

The release and source code are available at the following locations:

• Release page: https://github.com/dagster-io/dagster/releases/tag/1.13.8
• Repository: https://github.com/dagster-io/dagster
• Version tag: 1.13.8