Skip to content

[SPARK-33798][SQL] Add new rule to push down the foldable expressions through CaseWhen/If #30790

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wangyum wants to merge 12 commits into from
Closed

[SPARK-33798][SQL] Add new rule to push down the foldable expressions through CaseWhen/If #30790

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
19b0a83
Simplify EqualTo(CaseWhen/If, Literal) always false
wangyum Dec 16, 2020
831bf66
Address comments
wangyum Dec 16, 2020
859893d
Null value should be handled by NullPropagation.
wangyum Dec 16, 2020
db584ef
Handle Null values
wangyum Dec 16, 2020
55f4528
address comments
wangyum Dec 16, 2020
0a48048
address comments
wangyum Dec 17, 2020
f9f622f
Push down EqualTo through CaseWhen/If
wangyum Dec 17, 2020
21ef0c3
Push down EqualTo through CaseWhen/If
wangyum Dec 17, 2020
d5135ec
Fix
wangyum Dec 17, 2020
8ccc3c1
fix
wangyum Dec 17, 2020
45b56fc
fix
wangyum Dec 18, 2020
cfac0e8
fix
wangyum Dec 18, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Null value should be handled by NullPropagation.
  • Loading branch information
@wangyum
wangyum committed Dec 16, 2020
commit 859893d0a8629d689ee72db9e9897f772e5ef4a5
8 changes: 6 additions & 2 deletions sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/expressions.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -472,6 +472,7 @@ object SimplifyConditionals extends Rule[LogicalPlan] with PredicateHelper {

private def isAlwaysFalse(exps: Seq[Expression], equalTo: Literal): Boolean = {
Copy link
Member

@dongjoon-hyun dongjoon-hyun Dec 16, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

isAlwaysFalse seems to be mismatched a little with the behavior of this function. Technically, this function resembles contains function where expr is a list and equalTo literal is the item to be check the containment.

scala> Seq(1, 2).contains(3)
res0: Boolean = false

scala> Seq(1, 2).contains(2)
res1: Boolean = true

Copy link
Member Author

@wangyum wangyum Dec 17, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, do you have a recommended function name?

Copy link
Member

@dongjoon-hyun dongjoon-hyun Dec 17, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe, something like notEqualForAll instead of isAlwaysFalse?

exps.forall(_.isInstanceOf[Literal]) &&
exps.forall(_.asInstanceOf[Literal].value != null) &&
exps.forall(!EqualTo(_, equalTo).eval(EmptyRow).asInstanceOf[Boolean])
Copy link
Contributor

@cloud-fan cloud-fan Dec 16, 2020 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

EqualTo may return null, we need to take care of it, because null is not false in SQL.

Copy link
Member Author

@wangyum wangyum Dec 16, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes. Null value should be handled by NullPropagation.

}

Expand Down Expand Up @@ -529,12 +530,15 @@ object SimplifyConditionals extends Rule[LogicalPlan] with PredicateHelper {
e.copy(branches = branches.take(i).map(branch => (branch._1, elseValue)))
}

// Null value should be handled by NullPropagation.
Copy link
Contributor

@cloud-fan cloud-fan Dec 16, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's better to make optimizer rules orthogonal and don't rely on each other. Let's make isAlwaysFalse stricter:

exps.forall  { e =>
  val res = EqualTo(_, equalTo).eval(EmptyRow)
  res != null && !res.asInstanceOf[Boolean]
}

Copy link
Member Author

@wangyum wangyum Dec 16, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How about:

exps.forall {
  case l: Literal =>
    val res = EqualTo(l, equalTo).eval(EmptyRow)
    res != null && !res.asInstanceOf[Boolean]
  case _ => false
}

otherwise this test will fail:

val nonFoldable = CaseWhen(Seq(normalBranch, (a, b)), None)
assertEquivalent(EqualTo(nonFoldable, Literal(1)), EqualTo(nonFoldable, Literal(1)))

case EqualTo(i @ If(_, trueValue, falseValue), right: Literal)
if i.deterministic && isAlwaysFalse(trueValue :: falseValue :: Nil, right) =>
if i.deterministic && right.value != null &&
isAlwaysFalse(trueValue :: falseValue :: Nil, right) =>
FalseLiteral

case EqualTo(c @ CaseWhen(branches, elseValue), right: Literal)
Copy link
Contributor

@cloud-fan cloud-fan Dec 17, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A more general idea is to push down EqualTo(or other binary expressions) through If/CaseWhen.

For the example case, (CASE WHEN id = 1 THEN 'b' WHEN id = 3 THEN 'c' END) = 'a', we go through the following steps

  1. CASE WHEN id = 1 THEN 'b' = 'a' WHEN id = 3 THEN 'c' = 'a' END
  2. CASE WHEN id = 1 THEN false WHEN id = 3 THEN false END
  3. false

Step 3 may need more work in SimplifyConditionals to make it possible.

Copy link
Member Author

@wangyum wangyum Dec 17, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

+1 for this more general idea.

Copy link
Contributor

@cloud-fan cloud-fan Dec 17, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We can create a new rule to do it, e.g. PushFoldableIntoBranches

Copy link
Member Author

@wangyum wangyum Dec 17, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

OK.

if c.deterministic && isAlwaysFalse(branches.map(_._2) ++ elseValue, right) =>
if c.deterministic && right.value != null &&
isAlwaysFalse(branches.map(_._2) ++ elseValue, right) =>
FalseLiteral
}
}
Expand Down
14 changes: 6 additions & 8 deletions ...yst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/SimplifyConditionalSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -220,14 +220,13 @@ class SimplifyConditionalSuite extends PlanTest with ExpressionEvalHelper with P
assert(!nonDeterministic.deterministic)
assertEquivalent(EqualTo(nonDeterministic, Literal(-1)), EqualTo(nonDeterministic, Literal(-1)))

// null check, SPARK-33798 will change the following two behaviors.
// Null value should be handled by NullPropagation.
assertEquivalent(
EqualTo(If(a === Literal(1), Literal(null, IntegerType), Literal(1)), Literal(2)),
FalseLiteral)
EqualTo(If(a === Literal(1), Literal(null, IntegerType), Literal(1)), Literal(2)))
assertEquivalent(
EqualTo(If(a =!= Literal(1), Literal(1), Literal(2)), Literal(null, IntegerType)),
FalseLiteral)

EqualTo(If(a =!= Literal(1), Literal(1), Literal(2)), Literal(null, IntegerType)))
assertEquivalent(
EqualTo(If(a === Literal(1), Literal(null, IntegerType), Literal(1)), Literal(1)),
EqualTo(If(a === Literal(1), Literal(null, IntegerType), Literal(1)), Literal(1)))
Expand Down Expand Up @@ -273,14 +272,13 @@ class SimplifyConditionalSuite extends PlanTest with ExpressionEvalHelper with P
assert(!nonDeterministic.deterministic)
assertEquivalent(EqualTo(nonDeterministic, Literal(-1)), EqualTo(nonDeterministic, Literal(-1)))

// null check, SPARK-33798 will change the following two behaviors.
// Null value should be handled by NullPropagation.
assertEquivalent(
EqualTo(CaseWhen(Seq((a, Literal(null, IntegerType))), Some(Literal(1))), Literal(2)),
FalseLiteral)
EqualTo(CaseWhen(Seq((a, Literal(null, IntegerType))), Some(Literal(1))), Literal(2)))
assertEquivalent(
EqualTo(CaseWhen(Seq((a, Literal(1))), Some(Literal(2))), Literal(null, IntegerType)),
FalseLiteral)

EqualTo(CaseWhen(Seq((a, Literal(1))), Some(Literal(2))), Literal(null, IntegerType)))
assertEquivalent(
EqualTo(CaseWhen(Seq((a, Literal(null, IntegerType))), Some(Literal(1))), Literal(1)),
EqualTo(CaseWhen(Seq((a, Literal(null, IntegerType))), Some(Literal(1))), Literal(1)))
Expand Down