[SPARK-11303][SQL] filter should not be pushed down into sample

When sampling and then filtering DataFrame, the SQL Optimizer will push down filter into sample and produce wrong result. This is due to the sampler is calculated based on the original scope rather than the scope after filtering.

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #9294 from yanboliang/spark-11303.

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala

@@ -74,10 +74,6 @@ object DefaultOptimizer extends Optimizer {
 object SamplePushDown extends Rule[LogicalPlan] {
 
   def apply(plan: LogicalPlan): LogicalPlan = plan transform {
-    // Push down filter into sample
-    case Filter(condition, s @ Sample(lb, up, replace, seed, child)) =>
-      Sample(lb, up, replace, seed,
-        Filter(condition, child))
     // Push down projection into sample
     case Project(projectList, s @ Sample(lb, up, replace, seed, child)) =>
       Sample(lb, up, replace, seed,

sql/core/src/test/scala/org/apache/spark/sql/SQLQuerySuite.scala

@@ -1860,4 +1860,14 @@ class SQLQuerySuite extends QueryTest with SharedSQLContext {
         Row(1))
     }
   }
+
+  test("SPARK-11303: filter should not be pushed down into sample") {
+    val df = sqlContext.range(100)
+    List(true, false).foreach { withReplacement =>
+      val sampled = df.sample(withReplacement, 0.1, 1)
+      val sampledOdd = sampled.filter("id % 2 != 0")
+      val sampledEven = sampled.filter("id % 2 = 0")
+      assert(sampled.count() == sampledOdd.count() + sampledEven.count())
+    }
+  }
 }