Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/2923#discussion_r91081832
--- Diff:
flink-libraries/flink-table/src/main/scala/org/apache/flink/api/table/plan/nodes/dataset/BatchScan.scala
---
@@ -45,7 +45,8 @@ abstract class BatchScan(
override def computeSelfCost (planner: RelOptPlanner, metadata:
RelMetadataQuery): RelOptCost = {
val rowCnt = metadata.getRowCount(this)
- planner.getCostFactory.makeCost(rowCnt, rowCnt, 0)
+ val columnCnt = getRowType.getFieldCount
--- End diff --
The first cost parameter is number of processed rows, second parameter
indicates CPU cost, and third IO cost. I would not change the number of
processed rows but add IO costs of `rowCnt * rowSize` where `rowSize` can be
computed by `DataSetRel.estimateRowSize()`.
Have a look at `DataSetCost` to check how costs are compared. The cost
model aims at minimizing IO costs. So making here a difference results in the
choice of the plan with the most projection.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
