@ancient-wolf-19325 re-run one specific task from a Flyte run, using the inputs from a previous Flyte cloud run

Flyte does not natively support re-running a single task from a previous workflow execution directly; you cannot re-execute just one task in-place within a failed workflow run. However, you can manually launch a new execution of the task using the same inputs as before by retrieving the task and its inputs from the previous execution and launching it independently via the UI or CLI. This is a commonly requested feature, and partial support exists in the UI for launching a task with copied inputs, but it is not fully automated for all cases, especially for dynamically generated tasks or in-place recovery within the same workflow execution (GitHub Issue #545). For full workflow recovery (i.e., resuming from failed nodes), Flyte supports "recovery mode," which re-runs only the failed nodes and their downstream dependencies, but not an arbitrary single task in isolation (Build Indestructible Pipelines with Flyte). Would you like more detail on how to manually re-run a task or use recovery mode? Sources: - GitHub Issue #545 - Build Indestructible Pipelines with Flyte 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.

@ancient-wolf-19325 manually re-run a task

To manually re-run a task in Flyte, you can launch a new execution of the task by providing the same inputs as the previous run. This can be done via the Flyte UI, CLI, or programmatically using the SDK. In the UI, you can select the task, view its inputs, and use the launch form to re-execute it. In the CLI, you can use the

flyte run

command with the task and its inputs (Run and deploy tasks). For already deployed tasks, use the

deployed-task

keyword with the full task reference, e.g.,

flyte run deployed-task my_env.my_task --arg1 value1

(Persistent deployment). Programmatically, you can use

flyte.run()

to execute the task with the desired inputs:

import flyte

env = flyte.TaskEnvironment(name="my_env")

@env.task
async def my_task(arg1: str) -> str:
    return f"Hello, {arg1}!"

if __name__ == "__main__":
    flyte.init_from_config()
    result = flyte.run(my_task, arg1="World")
    print(result)

Would you like more detail on retrieving previous inputs or automating this process? Sources: - Run and deploy tasks - Persistent deployment 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.