sequenceDiagram
    participant Host as ML Host Controller
    participant State as ML Host State
    participant Worker as ML Worker Thread
    participant Dim as ML Dimension State

    Note over Host, Dim: Host Shutdown Sequence (ml_host_stop)

    Host->>State: NEW: Set ml_running = false
    Note right of State: Prevents new ML tasks from starting immediately

    par Concurrent Execution
        Worker->>Dim: ml_dimension_update_models()
        Dim->>Dim: Lock slock
        Dim->>State: Check ml_running status
        alt NEW: ml_running is false
            Dim->>Dim: Set training_in_progress = false
            Dim-->>Worker: Early Exit
        else ml_running is true
            Dim->>Dim: Standard model update flow
        end
        Dim->>Dim: Unlock slock
    and State Cleanup
        Host->>Dim: Lock slock
        Host->>Dim: CHANGED: Clear km_contexts
        Note right of Dim: Prevents use of stale models during reset
        Host->>Dim: Unlock slock
    end

    Note over Worker, Dim: Model Inlining Flow (ml_worker_add_existing_model)

    Worker->>Worker: Prepare model data
    Worker->>Dim: CHANGED: Assign Dim->kmeans (Outside slock)
    Note over Worker, Dim: Safe: per-host work is serialized via worker queue
    Worker->>Dim: ml_dimension_update_models()