Keep Parallel and Dag Tasks Running After a Failure with onChildFailure

For the complete documentation index, see llms.txt. For a full content snapshot, see llms-full.txt. Append .md to any kestra.io/docs/* URL for plain Markdown.

Available on:Open Source EditionEnterprise Edition

Parallel and Dag tasks now interrupt their other running tasks when one task fails. The new onChildFailure property controls this behavior, and its default is FAIL on both tasks.

TaskonChildFailure default
io.kestra.plugin.core.flow.ParallelFAIL from 2.0.0
io.kestra.plugin.core.flow.DagCONTINUE in 2.0.0 to 2.0.7, FAIL from 2.0.8

In 1.x, the other tasks kept running to completion after a failure.

What changes after upgrading

When a task inside a Parallel or Dag fails with no retry left, and onChildFailure is FAIL:

  • Tasks that are still running are interrupted and end in the FAILED state. This includes tasks nested at any depth, such as tasks inside a Sequential branch.
  • Each interrupted task logs a warning that names the task that caused the interruption: Task terminated because sibling task '<task_id>' failed.
  • The Parallel or Dag task itself ends in the FAILED state, and its errors and finally tasks run as before.

Migration steps

  1. Search your flows for tasks of type io.kestra.plugin.core.flow.Parallel and io.kestra.plugin.core.flow.Dag.
  2. For each task that should let its other tasks finish after one fails, add onChildFailure: CONTINUE.
  3. Check any existing onChildFailure values. The accepted values are CONTINUE, CANCEL, and FAIL. Replace CANCELLED with CANCEL and FAILED with FAIL.

Before: in 1.x, extract_customers kept running after extract_orders failed.

id: parallel_flow
namespace: company.team


tasks:
  - id: parallel
    type: io.kestra.plugin.core.flow.Parallel
    tasks:
      - id: extract_orders
        type: io.kestra.plugin.core.execution.Fail


      - id: extract_customers
        type: io.kestra.plugin.core.flow.Sleep
        duration: PT1M

After: set onChildFailure: CONTINUE to keep the same behavior.

id: parallel_flow
namespace: company.team


tasks:
  - id: parallel
    type: io.kestra.plugin.core.flow.Parallel
    onChildFailure: CONTINUE
    tasks:
      - id: extract_orders
        type: io.kestra.plugin.core.execution.Fail


      - id: extract_customers
        type: io.kestra.plugin.core.flow.Sleep
        duration: PT1M

With onChildFailure: CONTINUE, extract_customers runs for its full minute after extract_orders fails. Without it, Kestra interrupts extract_customers as soon as extract_orders fails.

The same property works on Dag:

  - id: dag
    type: io.kestra.plugin.core.flow.Dag
    onChildFailure: CONTINUE
    tasks:
      - task:
          id: extract_orders
          type: io.kestra.plugin.core.execution.Fail
      - task:
          id: extract_customers
          type: io.kestra.plugin.core.flow.Sleep
          duration: PT1M

To set interrupted tasks to CANCELLED instead of FAILED, use onChildFailure: CANCEL.

Was this page helpful?