Major Description
I was diagnosing a slow CopyTable job recently. It seems we're not exposing through hadoop counters as many statistics as we might. For example, HDFS gives things like bytes read, we apparently do not. This task is to review the counters exposed – assume you don't have access to debug logs to understand what happened – and ensure our coverage on the key points. This particular job appears to be moving data between two tables, both of which have poorly distributed region splits.
Perhaps there's follow-on to make things work better with uneven region sizes, leveraging some of the region size goodness added in HBASE-10413.