Github user wzhfy commented on a diff in the pull request:
https://github.com/apache/spark/pull/16401#discussion_r94774210
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/LogicalPlan.scala
---
@@ -95,6 +96,29 @@ abstract class LogicalPlan extends
QueryPlan[LogicalPlan] with Logging {
}
/**
+ * Returns the default statistics or statistics estimated by cbo based
on configuration.
+ */
+ final def planStats(conf: CatalystConf): Statistics = {
+ if (conf.cboEnabled) {
+ if (estimatedStats.isEmpty) {
+ estimatedStats = Some(cboStatistics(conf))
+ }
+ estimatedStats.get
+ } else {
+ statistics
+ }
+ }
+
+ /**
+ * Returns statistics estimated by cbo. If the plan doesn't override
this, it returns the
+ * default statistics.
+ */
+ protected def cboStatistics(conf: CatalystConf): Statistics = statistics
+
+ /** A cache for the estimated statistics, such that it will only be
computed once. */
+ private var estimatedStats: Option[Statistics] = None
--- End diff --
The conf won't change in a query, but if we cache plans across queries,
then we can't cache the calculated stats. And now I do find such plans e.g.
`cachedDataSourceTables`.
If we don't cache the stats, estimation can be a performance hit.
Then we need to use lazy val and don't depend on the conf. In the current
stage it's ok. I just thought in the future we may need some parameters in more
complicated estimation. e.g. we may need a threshold to determine whether a
column is a primary key.
---
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]
