Dagster 1.13.16 was published on July 30, 2026. Its main change is a preview path for applying Declarative Automation to jobs, while the same release fixes transient Airbyte API requests that ignored configured retries.
The full release notes and downloads are on the GitHub release page.
Declarative Automation can now launch jobs ¶
The preview extends Declarative Automation to jobs. Pass an automation_condition to define_asset_job, then wrap an asset level condition with AutomationCondition.any_job_root_assets_match or AutomationCondition.all_job_root_assets_match. When the resulting condition becomes true, Dagster launches one run of the job.
That single run is the important execution unit. The condition applies to the job after the root asset test; it does not launch a separate run for each matching root asset.
The two wrappers encode different orchestration policies. AutomationCondition.any_job_root_assets_match permits a job run when any root asset matches the wrapped condition. AutomationCondition.all_job_root_assets_match requires all job root assets to match it. That distinction matters when a job contains roots with different arrival times or readiness criteria.
Evaluation history appears in a new Automation tab on job pages. Operators can inspect the condition history at the same job scope as the run it controls.
This feature is explicitly a preview. The release notes give no migration steps or compatibility promises for it. Test both wrapper semantics with representative asset states before replacing an existing schedule or sensor, especially for jobs whose root assets become ready at different times.
Airbyte retries now honor the configured limit ¶
In dagster-airbyte, transient Airbyte API requests now retry instead of failing after one error when retries are configured. Before this fix, a sync could fail on the first transient API failure even when request_max_retries allowed more attempts. The setting and the observed request behavior were out of agreement.
The operational effect is narrow but useful. A configured retry budget once again participates in request failure handling, so a brief Airbyte API disturbance does not necessarily become an immediate failed sync. This reduces false failures only for the transient request path described in the release notes.
The notes do not specify which exceptions qualify as transient, the delay between attempts, or the backoff policy. This is a focused bug fix, not a broad redesign of Airbyte resilience. Keep existing alerts, and verify the failure modes that matter to the deployed connector before changing run retry policy around it.
Component inventory and metadata guidance ¶
The Components tab for a code location now lists every component instance in that location. Earlier behavior showed only app managed instances. For platform operators, the tab is now a more complete inventory when checking what a code location actually contains.
The remaining changes are documentation. Dagster now clarifies the difference between definition time metadata and runtime metadata, including access to asset definition metadata from a custom I/O manager. The release notes do not claim a runtime metadata behavior change, so this should be read as corrected guidance rather than a compatibility fix.
Configuration options for the Soda integration are also documented. No individual Soda keys are named in the release summary. Use the linked release material and its documentation references before changing integration configuration.