Rerun a run

Every run in Flyte is durable: its task, code, inputs, and configuration are all recorded on the backend. That means you can launch a brand-new run from any previous one without having the original code checked out locally. This is useful for retrying a transient failure, reproducing a result, or tweaking a few inputs and running again.

You can rerun at two levels of granularity:

• The whole run: re-execute the entry point task (the a0 action) and everything it calls.
• A single action: re-execute one nested action on its own.

And you can trigger a rerun from three places: the UI, the CLI, or programmatically with the Python SDK.

Rerun from the UI

Rerun an entire run

Open the run in the UI. In the top-right corner of the run view, click Rerun.

This opens the launch form pre-filled with the original run’s inputs, environment variables, and code. You can either:

• Trigger the run as-is to reproduce the original execution exactly, or
• Edit the inputs, environment variables, and other launch settings in the form before triggering to run a variation.

Rerun a single action

You can also rerun an individual action without rerunning the whole workflow. Navigate to the action in the run’s action list, open its details view, and use the Rerun action option (in the action menu in the top-right of the action details panel).

This launches a new run starting from that action, using the action’s recorded inputs. As with a full rerun, you can adjust the inputs in the launch form first.

Rerun from the CLI

flyte rerun fetches the prior run’s task and inputs from the backend and launches a new run. You don’t need the original code checked out locally. Everything is pulled from the platform:

# Rerun with the prior run's exact code and inputs
flyte rerun <run-name>

# Give the new run a name and stream its parent-action logs
flyte rerun <run-name> --name retry-1 --follow

Common options:

Rerun compared with a fresh run

Both start the run over from the beginning. To instead keep the successful actions of a failed run and re-execute only what failed, add --recover to flyte rerun. See Recover a failed run.

Rerun programmatically

Use flyte.rerun() to rerun from Python. Like the CLI, it fetches the prior run’s task and inputs from the backend, so no local code is required:

import flyte

flyte.init_from_config()

# Rerun a prior run with its exact inputs (task + inputs fetched from the platform):
flyte.rerun("ul56wcvgqrb9vzhzz5l2")

# Change input parameters — they are converted against the prior run's interface:
flyte.rerun("ul56wcvgqrb9vzhzz5l2", x_list=[1, 2, 3])

flyte.rerun() returns a flyte.remote.Run, just like flyte.run(), so you can monitor it, wait on it, and retrieve its outputs in the same way. See Interact with runs and actions for details.

To control launch settings (name, project, domain, environment variables, labels) use flyte.with_runcontext(...).rerun(...):

flyte.with_runcontext(
    name="retry-1",
    env_vars={"LOG_LEVEL": "20"},
).rerun("ul56wcvgqrb9vzhzz5l2")

Related

• Recover a failed run: re-run only what failed, reusing the actions that succeeded.
• Interact with runs and actions: retrieve, monitor, and inspect runs and actions.
• Run command options: the full set of flyte run options.