Kubeflow Pipelines and KServe: Zero-Downtime Model Migration Boundary

Kubeflow Pipelines and KServe: Zero-Downtime Model Migration Boundary

Kubeflow Pipelines compiles workflow definitions into Argo Workflow YAML and executes PipelineRuns as immutable instances. KServe manages InferenceService deployments and relies on Istio for canary rollouts and traffic splitting. The integration boundary is the handoff from a completed training pipeline to a live serving endpoint.

Pipeline versioning allows multiple workflow versions to coexist and existing runs are not interrupted by new uploads, but moving active triggers between versions requires external orchestration. An unresolved behavior is metadata tracking consistency when a pipeline or serving artifact is migrated across Kubeflow namespaces.

• Is there a documented mechanism for KFP pipeline completion to trigger a KServe canary rollout without manual intervention?
• Is metadata lineage preserved for model artifacts when traffic shifts between versions in KServe?
• Are metadata tracking guarantees defined when a pipeline is migrated across Kubeflow namespaces?