Description
Due to the way Spark's architected (SQL is defined on top of the RDD API), there are two separate metrics system used in core vs SQL. Ideally, we'd want to be able to get the shuffle metrics for each of the exchange operator independently, e.g. blocks read, number of records.
Attachments
Issue Links
- Is contained by
-
SPARK-26221 Improve Spark SQL instrumentation and metrics
- Resolved