Implementation:Kubeflow Pipelines Client Create Run
| Sources | Domains | Last Updated |
|---|---|---|
| Kubeflow Pipelines, KFP Client | ML_Pipelines, Execution | 2026-02-13 |
Overview
Concrete tool for submitting pipeline runs to a Kubeflow Pipelines backend provided by the KFP Client class.
Description
kfp.Client().create_run_from_pipeline_func() compiles and submits a pipeline in one step. The Client connects to the KFP API server and creates a run. Returns a RunPipelineResult that can optionally be waited on with .wait_for_run_completion().
Usage
Use to programmatically execute pipelines against a running KFP deployment. Provide the host endpoint; use None for in-cluster access.
Code Reference
Source Location: Repository: kubeflow/pipelines, File: samples/core/caching/caching_sample.py (L34-37), samples/core/XGBoost/xgboost_sample.py (L80-82)
Signature:
class Client:
def __init__(self, host: Optional[str] = None, ...): ...
def create_run_from_pipeline_func(
self,
pipeline_func: Callable,
arguments: dict = {},
run_name: str = None,
experiment_name: str = None,
namespace: str = None,
) -> RunPipelineResult: ...
Import: from kfp import Client or import kfp
I/O Contract
Inputs:
| Parameter | Type | Required | Description |
|---|---|---|---|
| host | str | No | API endpoint URL |
| pipeline_func | Callable | Yes | Pipeline function |
| arguments | dict | No | Runtime parameters |
| run_name | str | No | Name for the run |
| experiment_name | str | No | Name for the experiment |
Outputs:
- RunPipelineResult -- with
.wait_for_run_completion(timeout=N)method
Usage Examples
Example 1 -- Submit and wait:
kfp.Client(host=kfp_endpoint).create_run_from_pipeline_func(
caching_pipeline,
arguments=dict(seconds=60),
).wait_for_run_completion(timeout=999)
Example 2 -- Direct submission:
kfp.Client(host=kfp_endpoint).create_run_from_pipeline_func(
xgboost_pipeline, arguments={})