Implementation:Kserve Kserve Initial InferenceService Deployment

Knowledge Sources	KServe KServe Canary Rollout
Domains	MLOps, Deployment_Strategy, Model_Serving
Last Updated	2026-02-13 00:00 GMT

Overview

Concrete YAML pattern for deploying the initial baseline InferenceService that serves as the stable reference for canary rollouts.

Description

This pattern creates an InferenceService with a single predictor and no canaryTrafficPercent field. The absence of canary configuration means Knative routes 100% of traffic to the single active revision. This establishes the baseline that subsequent canary updates will be compared against.

Usage

Use this pattern as the first step in a canary rollout workflow. Apply this YAML, verify the service is ready and serving correct predictions, then proceed with canary updates.

Code Reference

Source Location

• Repository: kserve
• File: docs/samples/v1beta1/rollout/default.yaml, Lines 1-8

Signature

apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
  name: my-model
spec:
  predictor:
    tensorflow:
      storageUri: "gs://kfserving-examples/models/tensorflow/flowers"

Import

kubectl apply -f default.yaml

I/O Contract

Inputs

Name	Type	Required	Description
metadata.name	string	Yes	InferenceService name
spec.predictor.tensorflow.storageUri	string	Yes	Model artifact URI

Outputs

Name	Type	Description
Revision	Knative Revision	Single revision <name>-predictor-default-00001
Traffic	100%	All traffic to the single revision
status.url	URL	Prediction endpoint ready for requests

Usage Examples

Deploy Baseline Model

# 1. Deploy initial model
kubectl apply -f - <<EOF
apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
  name: my-model
spec:
  predictor:
    tensorflow:
      storageUri: "gs://kfserving-examples/models/tensorflow/flowers"
EOF

# 2. Wait for readiness
kubectl wait inferenceservice my-model --for=condition=Ready --timeout=120s

# 3. Verify single revision
kubectl get revisions -l serving.knative.dev/service=my-model-predictor-default
# NAME                                   READY
# my-model-predictor-default-00001       True

Related Pages

Implements Principle

• Principle:Kserve_Kserve_Canary_Deployment_Initialization