The blog post didn't fully specify, but the source code suggests that the locks are tracked per-partition rather than managed centrally.