Implementation:Kserve Kserve Graph Pipeline Validation

Knowledge Sources	KServe KServe InferenceGraph
Domains	Testing, Pipeline, Model_Serving
Last Updated	2026-02-13 00:00 GMT

Overview

Concrete curl-based testing pattern for validating InferenceGraph pipeline outputs with expected response formats.

Description

Pipeline validation uses HTTP POST requests to the graph URL with known test inputs. The expected response format varies by graph type. The Iris dataset sample input (iris-input.json) is commonly used for testing sklearn and xgboost graph components.

Usage

Run validation after deploying an InferenceGraph and verifying all component ISVCs are ready.

Code Reference

Source Location

• Repository: kserve
• File: docs/samples/graph/iris-input.json, Lines 1-6
• File: docs/samples/graph/README.md, Lines 60-75 (sequence), Lines 185-200 (ensemble)

Signature

curl http://<graph-name>.<namespace>.example.com/ \
  -d @iris-input.json

Import

# No import — uses curl and test input JSON

I/O Contract

Inputs

Name	Type	Required	Description
Graph URL	string	Yes	InferenceGraph entry URL from status.url
Test input	JSON	Yes	Prediction input (e.g., Iris features)

Outputs

Name	Type	Description
Sequence output	JSON	{"predictions": [1, 1]} — final step result
Ensemble output	JSON	{"sklearn-iris": {"predictions": [1,1]}, "xgboost-iris": {"predictions": [1,1]}}
Splitter output	JSON	{"treeModel": {"predictions": [1,1]}} — selected step result

Usage Examples

Validate Ensemble Graph

# Test input
cat > iris-input.json <<EOF
{
  "instances": [
    [6.8, 2.8, 4.8, 1.4],
    [6.0, 3.4, 4.5, 1.6]
  ]
}
EOF

# Send to ensemble graph
GRAPH_URL=$(kubectl get ig model-ensemble -o jsonpath='{.status.url}')
curl -s ${GRAPH_URL} -d @iris-input.json | jq .

# Expected output:
# {
#   "sklearn-iris": {"predictions": [1, 1]},
#   "xgboost-iris": {"predictions": [1, 1]}
# }

Validate Sequence Graph

GRAPH_URL=$(kubectl get ig model-chained -o jsonpath='{.status.url}')
curl -s ${GRAPH_URL} -d @iris-input.json | jq .

# Expected output (final step's output):
# {"predictions": [1, 1]}

Related Pages

Implements Principle

• Principle:Kserve_Kserve_Pipeline_Validation