During the development of the last items of
YARN-6875, it was typically difficult to extract information about the internal state of some log aggregation related classes (e.g. AppLogAggregatiorImpl and LogAggregationFileController).
On my fork I added a few more messages to those classes like:
- displaying the number of log aggregation cycles
- displaying the names of the files currently considered for log aggregation by containers
- immediately displaying any exception caught (and sent to the RM in the diagnostic messages) during the log aggregation process.
Those messages were quite useful for debugging if any issue occurs, but otherwise it flooded the NM log file with these messages that are usually not needed. I suggest to add (some of) these messages in DEBUG or TRACE level.