The issue was hit in the following scenario:
The root cause has not been found as the issue is not reproducible and happened only once in the production network. The incorrect state can be cleared by flapping each master interface.