Github user bersprockets commented on a diff in the pull request:
https://github.com/apache/spark/pull/21950#discussion_r212719073
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/PruneFileSourcePartitions.scala
---
@@ -76,4 +78,16 @@ private[sql] object PruneFileSourcePartitions extends
Rule[LogicalPlan] {
op
}
}
+
+ private def calcPartSize(catalogTable: Option[CatalogTable],
sizeInBytes: Long): Long = {
+ val conf: SQLConf = SQLConf.get
+ val factor = conf.sizeDeserializationFactor
+ if (catalogTable.isDefined && factor != 1.0 &&
+ // TODO: The serde check should be in a utility function, since it
is also checked elsewhere
+ catalogTable.get.storage.serde.exists(s => s.contains("Parquet") ||
s.contains("Orc"))) {
--- End diff --
@mgaido91 Good point. Also, I notice that even when the table's files are
not compressed (say, a table backed by CSV files), the LongToUnsafeRowMap or
BytesToBytesMap that backs the relation is roughly 3 times larger than the
total file size. So even under the best of circumstances (i.e., the table's
files are not compressed), Spark will get it wrong by several multiples.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
