Kubeflow Pipelines Server API

2.16.0 · active · verified Fri Apr 10

The `kfp-server-api` library provides a Python client for interacting directly with the Kubeflow Pipelines (KFP) v2 server API. It offers low-level access to KFP resources like pipelines, runs, and experiments, suitable for advanced integrations or custom automation. It is part of the KFP SDK v2 ecosystem, currently at version 2.16.0, and typically released in lockstep with the main `kfp` SDK.

Warnings

breaking The `kfp-server-api` library is built for Kubeflow Pipelines v2. Code written for KFP v1 client libraries (e.g., `kubeflow-pipelines` or older `kfp` versions) will likely be incompatible due to significant changes in API contracts, models, and service endpoints.
Fix: Rewrite client code to use the KFP v2 API patterns and models. Refer to the KFP v2 SDK documentation for updated API specifications and client usage.
gotcha Direct usage of `kfp-server-api` requires explicit management of authentication and API endpoints. Unlike the higher-level `kfp` SDK, which often infers these from the environment or provides simpler `Client` objects, you must manually configure bearer tokens, cookies, or other authentication methods.
Fix: Set `Configuration.host` to the KFP API endpoint and populate `Configuration.access_token` or other credential fields. For GCP, consider `google-auth` mechanisms. Ensure proper network access to the KFP backend.
gotcha For stability, it is crucial to ensure that the `kfp-server-api` version aligns with the version of the `kfp` SDK and the KFP backend server you are interacting with. Mismatched versions can lead to deserialization errors, unexpected API behavior, or missing features.
Fix: Always install `kfp-server-api` at the exact same version as your `kfp` SDK package (e.g., `pip install kfp-server-api==2.16.0 kfp==2.16.0`). Regularly update both together to match your KFP server deployment.

Install

pip install kfp-server-api==2.16.0 Install specific version
pip install kfp # This will pull in kfp-server-api as a dependency Install KFP SDK (recommended for most users)

Imports

Configuration
```
from kfp_server_api import Configuration
```
Top-level import for Configuration is preferred for v2 client.
ApiClient
```
from kfp_server_api import ApiClient
```

PipelineServiceApi

from kfp_server_api.api.pipeline_service_api import PipelineServiceApi

RunServiceApi

from kfp_server_api.api.run_service_api import RunServiceApi

ExperimentServiceApi

from kfp_server_api.api.experiment_service_api import ExperimentServiceApi

V2beta1Pipeline

from kfp_server_api.models import V2beta1Pipeline

Quickstart

This quickstart demonstrates how to initialize the `kfp-server-api` client, configure its host, and make a basic call to list pipelines. It assumes KFP is running and accessible at the specified host. For production deployments, robust authentication (e.g., OAuth2, Bearer tokens from service accounts) would be required, which needs to be configured in the `Configuration` object.

import os
from kfp_server_api import Configuration, ApiClient
from kfp_server_api.api.pipeline_service_api import PipelineServiceApi

# Replace with your KFP API endpoint. E.g., 'http://localhost:8080/pipeline'
# or 'https://<your-kfp-host>/pipeline'
KFP_HOST = os.environ.get('KFP_SERVER_API_HOST', 'http://localhost:8080')

# Configure API key authorization: Bearer
# For most KFP deployments, you might need a service account token or session cookie.
# Example for no auth or basic local setup:
configuration = Configuration(host=KFP_HOST)

# For token-based auth (e.g., from a service account or user session):
# KFP_API_TOKEN = os.environ.get('KFP_SERVER_API_TOKEN')
# if KFP_API_TOKEN:
#     configuration.access_token = KFP_API_TOKEN

# Create an API client
with ApiClient(configuration) as api_client:
    # Create an instance of the PipelineServiceApi
    pipeline_api = PipelineServiceApi(api_client)

    try:
        # List pipelines (requires appropriate authentication and permissions)
        list_response = pipeline_api.list_pipelines()
        print(f"Successfully connected to KFP API host: {KFP_HOST}")
        print(f"Found {len(list_response.pipelines or [])} pipelines.")
        for p in (list_response.pipelines or []):
            print(f"  - {p.display_name} (ID: {p.pipeline_id})")
    except Exception as e:
        print(f"Error connecting to KFP API or listing pipelines: {e}")
        print("Make sure KFP_SERVER_API_HOST is correctly set and you have necessary authentication.")

view raw JSON →