Keep previous executions of dependent tasks when retrying a failed task in a DAG

[fstaudt]

We use DAG with multiple tasks to run multiple dependent tasks.
Some tasks are configured with continueOn.failed=true to execute dependent tasks even when they failed.

Here is an example of such DAG before retry:

Using restart-successful and node-field-selector options of Argo CLI, we can retry the failed task in the DAG:

argo retry training-batch-plan-dpgn7 --restart-successful --node-field-selector name=training-batch-plan-dpgn7.training-batch-non-critical

We notice however that previous executions of all dependent tasks have been removed from the DAG (training-after-non-critical and training-after-all in example above).
Dependent tasks are actually removed from workflow status nodes when task is retried and will be executed again when retried task completes.

This is problematic for us because we want to keep those previous executions visible in the UI.
More specifically, we want to keep the output artifacts of those tasks.

Is there a way to keep the previous executions of the dependent tasks in a DAG when retrying a failed task ?

[jswxstw]

Not supported yet, but there are already similar feature requests for it, #13839 and #13692 etc.

[fstaudt]

@jswxstw ,

#13839 looks indeed to describe exactly the feature that we need.