When I did my traineeship at ATLAS DAQ around 10 years ago, one issue was how to per-process the data that could be considered invalid fast enough, to be able to throw it away before it arrived into the cluster.