Hello, Log aggregation is great. However, if a yarn application runs a large number of tasks which generate large logs, it takes some finite amount of time for all of the logs to be collected and written to the HDFS.
Currently our client code runs the equivalent of the "yarn logs" command once all tasks have completed. This works fine provided log aggregation is complete. But it fails in a variety of ways if aggregation is not complete. This includes one case where the "yarn logs" code encounters no exceptions and no non-zero return codes from methods, but returns an empty string. So, is there a way to determine if log aggregation is complete? Thanks, Geoff