Oct 23, 2026: Atomic deployments, centralized configuration, and advanced branching for task graphs (Preview)¶

With this preview release, you can manage task graph settings centrally, update pipelines without suspending them, and control when tasks run based on upstream outcomes.

Centralized configuration and atomic deployments¶

The new TASK GRAPH SQL commands let you:

• Define the schedule, retry behavior, and default settings for member tasks in one place.
• Update the graph and its member tasks without suspending the pipeline.
• Stage changes and commit them together as a new version.

Subsequent runs use the newly committed version. Runs already in progress continue using the version they started with.

For more information, see Create and manage task graphs with SQL.

Advanced branching¶

By default, a child task in a task graph runs only when all of its predecessors succeed. Advanced branching lets a task run on other outcomes, such as when all of its predecessors failed, when at least one succeeded, or when a specific upstream task reached a particular state. Use it to build error handling, fallback paths, and cleanup logic directly into a task graph.

Two new system functions support advanced branching in a task’s WHEN clause:

• SYSTEM$PREDECESSORS_STATE evaluates predecessor completion states against a condition, such as ALL_FAILED or ALL_DONE.
• SYSTEM$TASK_COMPLETION_STATE returns the completion state of a specific task in the graph, such as SUCCEEDED, FAILED, or SKIPPED.

A task evaluates these conditions after all of its resumed predecessors finish.

For more information, see Control when tasks run with advanced branching.