Principle:Kserve Kserve Canary Deployment Initialization
| Knowledge Sources | |
|---|---|
| Domains | MLOps, Deployment_Strategy, Model_Serving |
| Last Updated | 2026-02-13 00:00 GMT |
Overview
A deployment pattern that establishes an initial baseline model version serving 100% of traffic before introducing canary changes.
Description
Canary Deployment Initialization creates the stable baseline for progressive rollout. An InferenceService is deployed without any canaryTrafficPercent field, meaning all traffic routes to the single active revision. This establishes the "golden" version that canary updates will be compared against.
This step is critical because the canary pattern requires a known-good baseline. Without it, there is no stable revision to fall back to if the canary fails validation.
Usage
Use this as the first step in any canary rollout strategy. Deploy the initial model version and confirm it is serving correctly before proceeding with canary updates.
Theoretical Basis
# Canary initialization state (NOT implementation code)
State: Single revision, 100% traffic
Revision: <name>-predictor-default-00001
Traffic: 100% → revision-00001
No canaryTrafficPercent → Knative routes all traffic to latest revision