Quick answer: Start Spark while the lag is actively happening, capture long enough to include the slow event, then read the report from the server tick downward. The widest hot path is where time was spent, not automatically the component to delete. Confirm the same path appears across repeated...