Principle:Kserve Kserve Graph Component Deployment
| Knowledge Sources | |
|---|---|
| Domains | MLOps, Pipeline, Model_Serving |
| Last Updated | 2026-02-13 00:00 GMT |
Overview
A deployment pattern that provisions multiple individual InferenceService endpoints as building blocks for inference graph pipelines.
Description
Graph Component Deployment establishes the individual model endpoints that an InferenceGraph routes between. Each component is a standalone InferenceService with its own model, scaling, and resources. The InferenceGraph references these components by name or URL to build multi-model pipelines.
This pattern follows the microservices principle: each model is an independently deployable, scalable unit. The graph layer composes these units into complex inference workflows.
Usage
Use this as the prerequisite for any InferenceGraph pipeline. All component InferenceServices must be running and ready before the graph can route traffic to them.
Theoretical Basis
# Component deployment model (NOT implementation code)
For each model in the pipeline:
1. Create InferenceService with model-specific config
2. Wait for Ready status
3. Note the service name (used in graph spec)
Example pipeline components:
sklearn-iris → Classification model
xgboost-iris → Alternative classification model
preprocessor → Custom container for data transformation