Github user dongjoon-hyun commented on a diff in the pull request:
https://github.com/apache/spark/pull/14132#discussion_r71202822
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/catalyst/SQLBuilder.scala ---
@@ -425,6 +452,49 @@ class SQLBuilder(logicalPlan: LogicalPlan) extends
Logging {
}
}
+ /**
+ * Merge and move upward to the nearest Project.
+ * A broadcast hint comment is scattered into multiple nodes inside
the plan, and the
+ * information of BroadcastHint resides its current position inside
the plan. In order to
+ * reconstruct broadcast hint comment, we need to pack the information
of BroadcastHint into
+ * Hint("BROADCAST", _, _) and collect them up by moving upward to the
nearest Project node.
+ */
+ object NormalizeBroadcastHint extends Rule[LogicalPlan] {
+ override def apply(plan: LogicalPlan): LogicalPlan = plan
transformUp {
+ // Capture the broadcasted information and store it in Hint.
+ case BroadcastHint(child @ SubqueryAlias(_, Project(_,
SQLTable(database, table, _, _)))) =>
+ Hint("BROADCAST", Seq(table), child)
+
+ // Nearest Project is found.
+ case p @ Project(_, Hint(_, _, _)) => p
+
+ // Merge BROADCAST hints up to the nearest Project.
+ case Hint("BROADCAST", params1, h @ Hint("BROADCAST", params2, _))
=>
+ h.copy(parameters = params1 ++ params2)
+ case j @ Join(h1 @ Hint("BROADCAST", p1, left), h2 @
Hint("BROADCAST", p2, right), _, _) =>
+ h1.copy(parameters = p1 ++ p2, child = j.copy(left = left, right
= right))
+
+ // Bubble up BROADCAST hints to the nearest Project.
+ case j @ Join(h @ Hint("BROADCAST", _, hintChild), _, _, _) =>
+ h.copy(child = j.copy(left = hintChild))
+ case j @ Join(_, h @ Hint("BROADCAST", _, hintChild), _, _) =>
+ h.copy(child = j.copy(right = hintChild))
+ case s @ SubqueryAlias(_, h @ Hint("BROADCAST", _, hintChild)) =>
+ h.copy(child = s.copy(child = hintChild))
+ case ll @ LocalLimit(_, h @ Hint("BROADCAST", _, hintChild)) =>
+ h.copy(child = ll.copy(child = hintChild))
+ case f @ Filter(_, h @ Hint("BROADCAST", _, hintChild)) =>
+ h.copy(child = f.copy(child = hintChild))
+ case a @ Aggregate(_, _, h @ Hint("BROADCAST", _, hintChild)) =>
+ h.copy(child = a.copy(child = hintChild))
+ case s @ Sort(_, _, h @ Hint("BROADCAST", _, hintChild)) =>
+ h.copy(child = s.copy(child = hintChild))
+ case g @ Generate(_, _, _, _, _, h @ Hint("BROADCAST", _,
hintChild)) =>
+ h.copy(child = g.copy(child = hintChild))
+ // Set operation is not allowed to be across.
UNION/INTERCEPT/EXCEPT
--- End diff --
It's not a restriction, SET operation UNION/INTERCETP/EXCEPT has `Project`
inside it. So, I skip those.
As I mentioned before, theoretically, the Hint can not be restored into the
original `Project`. In addition to that,
since `Canonicalizer` adds a number of extra things, we can not move upward
enough, too. We need to stop of nearest `Project`.
---
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]
