RowKeyJoin operator by itself incurs negligible CPU and I/O cost since it is not doing a real join. The actual cost is attributed to the skip-scan (random I/O). The RK join will hold 1 batch in memory but since it is not a buffering operator, we typically don't assign memory costs to them. I can add the explanation.
[ Full content available at: https://github.com/apache/drill/pull/1466 ] This message was relayed via gitbox.apache.org for [email protected]
