MLMD schema mismatch after Kubeflow Pipelines database restore

MLMD schema mismatch after Kubeflow Pipelines database restore

Restoring the ML Metadata (MLMD) MySQL database for Kubeflow Pipelines should recover pipeline execution history. The restored schema must match the version expected by the deployed KFP control plane, or the UI may fail to reconcile executions.

A restored MLMD store can also reference artifacts in object storage that no longer exist, and the KFP UI may display ghost executions for workloads whose pods have been removed. Volume snapshots taken without flushing MySQL may leave the database in an inconsistent state.

Assuming Kubeflow Pipelines 2.x with MLMD backed by MySQL, which checks confirm the restored schema is compatible before the UI reads it? What procedure verifies that artifact URIs in restored MLMD records still resolve in object storage? What distinguishes a ghost execution from a valid recovered run in the KFP UI?