Stall time is calculated by adding up the time that a processor spends waiting on input after it has asked for the next row from its upstream processor. For example, in the
HashJoiner/19 case the
5.5s waiting on its left input in total since the
HashJoiner started execution. This includes most of the time spent reading rows, execution, and sending them across the network before the
HashJoiner gets a row back. However, note that stall time does not necessarily include all of this time as processors on different nodes may have been started at different times and rows are pushed, not pulled over the network.
I hope this answers your question, let me know if you would like me to clarify anything further.