Replaying runs and steps

Re-run pipelines or individual steps using artifacts from a previous execution.

A replay re-executes a pipeline or step using the same input artifacts (and parameters) as a previous run. This is useful when you want to:

• Debug a failed step without re-running the entire pipeline.

• Re-run a pipeline after a code fix, skipping steps that already succeeded.

• Test local changes against production data by replaying in debug mode.

---

Replaying a pipeline

Replay a pipeline either from local pipeline code (.replay()) or directly via the server (Client().replay_pipeline_run(...)).

ZenML resolves the run to replay using the first match:

1. pipeline_run — replay that exact run.

2. pipeline — replay the latest run of that pipeline name/ID.

3. Neither — replay the latest run whose name matches the current pipeline instance.

from zenml import pipeline, step

@step
def load_data() -> dict:
    ...

@step
def train(data: dict) -> None:
    ...

@pipeline
def training_pipeline(data: dict):
    train(data=load_data())

# Replay the latest run of this pipeline
training_pipeline.replay()

# Replay a specific run
training_pipeline.replay(pipeline_run="run_name_or_id")

Skipping steps

You can skip steps that don't need to be re-executed. Skipped steps reuse their output artifacts from the original run.

A step can only be skipped if all of its upstream dependencies are also skipped.

Overriding pipeline parameters

Pass input_overrides to change parameters for the replayed run. Any parameters you don't override are carried over from the original run.

Overriding step inputs

Use step_input_overrides to replace specific step inputs for a replayed pipeline run.

step_input_overrides expects a mapping of invocation_id -> input_name -> value.

---

Replaying a single step

Call .replay() on any @step-decorated function. ZenML loads the original input artifacts, feeds them to your (potentially updated) step code, and runs it as a single-step pipeline on the active stack.

The step to replay is resolved using the first match:

1. step_run_id — replay that exact step run.

2. pipeline_run + optional invocation_id — find the step in that run.

3. pipeline + optional invocation_id — find the step in the latest run of that pipeline.

Overriding inputs

You can replace any of the original input artifacts with new values:

Using invocation_id

If a step appears more than once in a pipeline, specify which invocation to replay:

---

Debug mode

Both pipeline and step replays accept debug=True. This runs the replay on a local orchestrator while keeping the rest of your active stack (artifact store, etc.), so you can iterate quickly without waiting for remote infrastructure.