Github user wzhfy commented on a diff in the pull request:
https://github.com/apache/spark/pull/14712#discussion_r77139470
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/command/AnalyzeTableCommand.scala
---
@@ -88,24 +85,53 @@ case class AnalyzeTableCommand(tableName: String)
extends RunnableCommand {
}
}.getOrElse(0L)
- // Update the Hive metastore if the total size of the table is
different than the size
- // recorded in the Hive metastore.
- // This logic is based on
org.apache.hadoop.hive.ql.exec.StatsTask.aggregateStats().
- if (newTotalSize > 0 && newTotalSize != oldTotalSize) {
- sessionState.catalog.alterTable(
- catalogTable.copy(
- properties = relation.catalogTable.properties +
- (AnalyzeTableCommand.TOTAL_SIZE_FIELD ->
newTotalSize.toString)))
- }
+ updateTableStats(
+ catalogTable,
+ oldTotalSize =
catalogTable.stats.map(_.sizeInBytes.toLong).getOrElse(0L),
+ oldRowCount =
catalogTable.stats.flatMap(_.rowCount.map(_.toLong)).getOrElse(-1L),
+ newTotalSize = newTotalSize)
+
+ // data source tables have been converted into LogicalRelations
+ case logicalRel: LogicalRelation if
logicalRel.catalogTable.isDefined =>
+ updateTableStats(
+ logicalRel.catalogTable.get,
+ oldTotalSize = logicalRel.statistics.sizeInBytes.toLong,
+ oldRowCount =
logicalRel.statistics.rowCount.map(_.toLong).getOrElse(-1L),
+ newTotalSize = logicalRel.relation.sizeInBytes)
--- End diff --
Yes, for the first analyze they are equal. But if there's a second
analyze, they are not equal, since logicalRel is in the cachedDataSourceTables,
its statistics is the original one.
Anyway, there's a bug here. I think we can use catalogTable's statistics as
the old one and I'll restore the test case to check size.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
