Ensure you have an existing InferenceService deployed using the v1beta1 API version that will serve as the default (stable) revision
Edit the InferenceService manifest to update the model or container image in spec.predictor to the new candidate version
Add the canaryTrafficPercent field directly within spec.predictor, set to the integer percentage of traffic to route to the new revision
Apply the updated manifest with kubectl apply; KServe will automatically retain the prior revision as the default and route the specified percentage to the new revision
Monitor metrics and logs for both the canary and default revisions to validate the new version's behavior
To promote, increase canaryTrafficPercent to 100 (or remove it); to roll back, set it to 0 or revert the predictor spec
Known gotchas
Set canaryTrafficPercent inside spec.predictor directly — there is no canarySpec field in v1beta1; using a canarySpec key will be ignored or cause a validation error
KServe automatically retains the prior revision as the default when you update the predictor; you do not need to manually manage or reference the old revision
canaryTrafficPercent must be an integer between 0 and 100; the remaining traffic percentage is automatically routed to the default (previous) revision
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp