#906 Add flyte.replay() for re-executing runs with original or updated code Pull request opened by kumare3 Replay lets users re-run a previous execution using the same inputs and RunSpec, optionally swapping in a new TaskTemplate with updated code. This is useful for: • Debugging: reproduce a failed run locally with the exact same inputs • Iterating: re-run with modified task code against production inputs • Testing: verify a fix against the inputs that caused a failure Usage examples:

# Replay remotely with original task template
flyte.replay("my-run-name")

# Replay a specific action within a run
flyte.replay("my-run-name", action_name="a1")

# Replay with updated code
flyte.replay("my-run-name", task_template=my_updated_task)

# Replay locally (requires a task_template with executable code)
flyte.with_replaycontext(mode="local").replay(
    "my-run-name", task_template=my_task
)

# Replay remotely with explicit mode
flyte.with_replaycontext(mode="remote").replay("my-run-name")

Implementation notes: • Follows the same _Runner/with_runcontext pattern: _Replayer holds config, with_replaycontext() returns a _Replayer, top-level replay() uses defaults • Local replay always requires a task_template (validated early, before network calls) since there is no Python function to execute without one • Extracts shared local execution logic into run_task_locally() in _run.py, eliminating duplication between _Runner._run_local and _Replayer._replay_local flyteorg/flyte-sdk