From 6d06ff6f7e2dd72ba8fe96cd875e83eda6ebb2a9 Mon Sep 17 00:00:00 2001
From: Josh Rosen <joshrosen@databricks.com>
Date: Wed, 14 Sep 2016 10:10:01 -0700
Subject: [SPARK-17514] df.take(1) and df.limit(1).collect() should perform the
 same in Python

## What changes were proposed in this pull request?

In PySpark, `df.take(1)` runs a single-stage job which computes only one partition of the DataFrame, while `df.limit(1).collect()` computes all partitions and runs a two-stage job. This difference in performance is confusing.

The reason why `limit(1).collect()` is so much slower is that `collect()` internally maps to `df.rdd.<some-pyspark-conversions>.toLocalIterator`, which causes Spark SQL to build a query where a global limit appears in the middle of the plan; this, in turn, ends up being executed inefficiently because limits in the middle of plans are now implemented by repartitioning to a single task rather than by running a `take()` job on the driver (this was done in #7334, a patch which was a prerequisite to allowing partition-local limits to be pushed beneath unions, etc.).

In order to fix this performance problem I think that we should generalize the fix from SPARK-10731 / #8876 so that `DataFrame.collect()` also delegates to the Scala implementation and shares the same performance properties. This patch modifies `DataFrame.collect()` to first collect all results to the driver and then pass them to Python, allowing this query to be planned using Spark's `CollectLimit` optimizations.

## How was this patch tested?

Added a regression test in `sql/tests.py` which asserts that the expected number of jobs, stages, and tasks are run for both queries.

Author: Josh Rosen <joshrosen@databricks.com>

Closes #15068 from JoshRosen/pyspark-collect-limit.
---
 sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala  |  8 ++++++--
 .../apache/spark/sql/execution/python/EvaluatePython.scala  | 13 +------------
 2 files changed, 7 insertions(+), 14 deletions(-)

(limited to 'sql/core')

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala b/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala
index 3b3cb82078..9cfbdffd02 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala
@@ -29,7 +29,7 @@ import org.apache.commons.lang3.StringUtils
 import org.apache.spark.annotation.{DeveloperApi, Experimental}
 import org.apache.spark.api.java.JavaRDD
 import org.apache.spark.api.java.function._
-import org.apache.spark.api.python.PythonRDD
+import org.apache.spark.api.python.{PythonRDD, SerDeUtil}
 import org.apache.spark.broadcast.Broadcast
 import org.apache.spark.rdd.RDD
 import org.apache.spark.sql.catalyst._
@@ -2567,8 +2567,12 @@ class Dataset[T] private[sql](
   }
 
   private[sql] def collectToPython(): Int = {
+    EvaluatePython.registerPicklers()
     withNewExecutionId {
-      PythonRDD.collectAndServe(javaToPython.rdd)
+      val toJava: (Any) => Any = EvaluatePython.toJava(_, schema)
+      val iter = new SerDeUtil.AutoBatchedPickler(
+        queryExecution.executedPlan.executeCollect().iterator.map(toJava))
+      PythonRDD.serveIterator(iter, "serve-DataFrame")
     }
   }
 
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/python/EvaluatePython.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/python/EvaluatePython.scala
index cf68ed4ec3..724025b464 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/execution/python/EvaluatePython.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/execution/python/EvaluatePython.scala
@@ -24,9 +24,8 @@ import scala.collection.JavaConverters._
 
 import net.razorvine.pickle.{IObjectPickler, Opcodes, Pickler}
 
-import org.apache.spark.api.python.{PythonRDD, SerDeUtil}
+import org.apache.spark.api.python.SerDeUtil
 import org.apache.spark.rdd.RDD
-import org.apache.spark.sql.DataFrame
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.catalyst.expressions._
 import org.apache.spark.sql.catalyst.util.{ArrayBasedMapData, ArrayData, GenericArrayData, MapData}
@@ -34,16 +33,6 @@ import org.apache.spark.sql.types._
 import org.apache.spark.unsafe.types.UTF8String
 
 object EvaluatePython {
-  def takeAndServe(df: DataFrame, n: Int): Int = {
-    registerPicklers()
-    df.withNewExecutionId {
-      val iter = new SerDeUtil.AutoBatchedPickler(
-        df.queryExecution.executedPlan.executeTake(n).iterator.map { row =>
-          EvaluatePython.toJava(row, df.schema)
-        })
-      PythonRDD.serveIterator(iter, s"serve-DataFrame")
-    }
-  }
 
   def needConversionInPython(dt: DataType): Boolean = dt match {
     case DateType | TimestampType => true
-- 
cgit v1.2.3