Details
Minor Description
The return value of Literal.references is an empty AttributeSet, so Literal is mistaken for a partition column.
org.apache.spark.sql.execution.dynamicpruning.PartitionPruning#getFilterableTableScan:
val srcInfo: Option[(Expression, LogicalPlan)] = findExpressionAndTrackLineageDown(a, plan)
srcInfo.flatMap {
case (resExp, l: LogicalRelation) =>
l.relation match {
case fs: HadoopFsRelation =>
val partitionColumns = AttributeSet(
l.resolve(fs.partitionSchema, fs.sparkSession.sessionState.analyzer.resolver))
// When resExp is a Literal, Literal is considered a partition column.
if (resExp.references.subsetOf(partitionColumns)) {
return Some(l)
} else {
None
}
case _ => None
}
