[SPARK-5168] Make SQLConf a field rather than mixin in SQLContext

This change should be binary and source backward compatible since we didn't change any user facing APIs.

Author: Reynold Xin <rxin@databricks.com>

Closes #3965 from rxin/SPARK-5168-sqlconf and squashes the following commits:

42eec09 [Reynold Xin] Fix default conf value.
0ef86cc [Reynold Xin] Fix constructor ordering.
4d7f910 [Reynold Xin] Properly override config.
ccc8e6a [Reynold Xin] [SPARK-5168] Make SQLConf a field rather than mixin in SQLContext

Author: rxin

sql/core/src/main/scala/org/apache/spark/sql/CacheManager.scala

@@ -91,8 +91,8 @@ private[sql] trait CacheManager {
         CachedData(
           planToCache,
           InMemoryRelation(
-            useCompression,
-            columnBatchSize,
+            conf.useCompression,
+            conf.columnBatchSize,
             storageLevel,
             query.queryExecution.executedPlan,
             tableName))

sql/core/src/main/scala/org/apache/spark/sql/SQLConf.scala

@@ -61,7 +61,7 @@ private[spark] object SQLConf {
  *
  * SQLConf is thread-safe (internally synchronized, so safe to be used in multiple threads).
  */
-private[sql] trait SQLConf {
+private[sql] class SQLConf {
   import SQLConf._
 
   /** Only low degree of contention is expected for conf, thus NOT using ConcurrentHashMap. */

sql/core/src/main/scala/org/apache/spark/sql/SQLContext.scala

@@ -17,6 +17,9 @@
 
 package org.apache.spark.sql
 
+import java.util.Properties
+
+import scala.collection.immutable
 import scala.language.implicitConversions
 import scala.reflect.runtime.universe.TypeTag
 
@@ -49,14 +52,37 @@ import org.apache.spark.sql.sources.{DataSourceStrategy, BaseRelation, DDLParser
 @AlphaComponent
 class SQLContext(@transient val sparkContext: SparkContext)
   extends org.apache.spark.Logging
-  with SQLConf
   with CacheManager
   with ExpressionConversions
   with UDFRegistration
   with Serializable {
 
   self =>
 
+  // Note that this is a lazy val so we can override the default value in subclasses.
+  private[sql] lazy val conf: SQLConf = new SQLConf
+
+  /** Set Spark SQL configuration properties. */
+  def setConf(props: Properties): Unit = conf.setConf(props)
+
+  /** Set the given Spark SQL configuration property. */
+  def setConf(key: String, value: String): Unit = conf.setConf(key, value)
+
+  /** Return the value of Spark SQL configuration property for the given key. */
+  def getConf(key: String): String = conf.getConf(key)
+
+  /**
+   * Return the value of Spark SQL configuration property for the given key. If the key is not set
+   * yet, return `defaultValue`.
+   */
+  def getConf(key: String, defaultValue: String): String = conf.getConf(key, defaultValue)
+
+  /**
+   * Return all the configuration properties that have been set (i.e. not the default).
+   * This creates a new copy of the config properties in the form of a Map.
+   */
+  def getAllConfs: immutable.Map[String, String] = conf.getAllConfs
+
   @transient
   protected[sql] lazy val catalog: Catalog = new SimpleCatalog(true)
 
@@ -212,7 +238,7 @@ class SQLContext(@transient val sparkContext: SparkContext)
    */
   @Experimental
   def jsonRDD(json: RDD[String], schema: StructType): SchemaRDD = {
-    val columnNameOfCorruptJsonRecord = columnNameOfCorruptRecord
+    val columnNameOfCorruptJsonRecord = conf.columnNameOfCorruptRecord
     val appliedSchema =
       Option(schema).getOrElse(
         JsonRDD.nullTypeToStringType(
@@ -226,7 +252,7 @@ class SQLContext(@transient val sparkContext: SparkContext)
    */
   @Experimental
   def jsonRDD(json: RDD[String], samplingRatio: Double): SchemaRDD = {
-    val columnNameOfCorruptJsonRecord = columnNameOfCorruptRecord
+    val columnNameOfCorruptJsonRecord = conf.columnNameOfCorruptRecord
     val appliedSchema =
       JsonRDD.nullTypeToStringType(
         JsonRDD.inferSchema(json, samplingRatio, columnNameOfCorruptJsonRecord))
@@ -299,10 +325,10 @@ class SQLContext(@transient val sparkContext: SparkContext)
    * @group userf
    */
   def sql(sqlText: String): SchemaRDD = {
-    if (dialect == "sql") {
+    if (conf.dialect == "sql") {
       new SchemaRDD(this, parseSql(sqlText))
     } else {
-      sys.error(s"Unsupported SQL dialect: $dialect")
+      sys.error(s"Unsupported SQL dialect: ${conf.dialect}")
     }
   }
 
@@ -323,9 +349,9 @@ class SQLContext(@transient val sparkContext: SparkContext)
 
     val sqlContext: SQLContext = self
 
-    def codegenEnabled = self.codegenEnabled
+    def codegenEnabled = self.conf.codegenEnabled
 
-    def numPartitions = self.numShufflePartitions
+    def numPartitions = self.conf.numShufflePartitions
 
     def strategies: Seq[Strategy] =
       extraStrategies ++ (

sql/core/src/main/scala/org/apache/spark/sql/api/java/JavaSQLContext.scala

@@ -52,7 +52,7 @@ class JavaSQLContext(val sqlContext: SQLContext) extends UDFRegistration {
    * @group userf
    */
   def sql(sqlText: String): JavaSchemaRDD = {
-    if (sqlContext.dialect == "sql") {
+    if (sqlContext.conf.dialect == "sql") {
       new JavaSchemaRDD(sqlContext, sqlContext.parseSql(sqlText))
     } else {
       sys.error(s"Unsupported SQL dialect: $sqlContext.dialect")
@@ -164,7 +164,7 @@ class JavaSQLContext(val sqlContext: SQLContext) extends UDFRegistration {
    * It goes through the entire dataset once to determine the schema.
    */
   def jsonRDD(json: JavaRDD[String]): JavaSchemaRDD = {
-    val columnNameOfCorruptJsonRecord = sqlContext.columnNameOfCorruptRecord
+    val columnNameOfCorruptJsonRecord = sqlContext.conf.columnNameOfCorruptRecord
     val appliedScalaSchema =
       JsonRDD.nullTypeToStringType(
         JsonRDD.inferSchema(json.rdd, 1.0, columnNameOfCorruptJsonRecord))
@@ -182,7 +182,7 @@ class JavaSQLContext(val sqlContext: SQLContext) extends UDFRegistration {
    */
   @Experimental
   def jsonRDD(json: JavaRDD[String], schema: StructType): JavaSchemaRDD = {
-    val columnNameOfCorruptJsonRecord = sqlContext.columnNameOfCorruptRecord
+    val columnNameOfCorruptJsonRecord = sqlContext.conf.columnNameOfCorruptRecord
     val appliedScalaSchema =
       Option(asScalaDataType(schema)).getOrElse(
         JsonRDD.nullTypeToStringType(

sql/core/src/main/scala/org/apache/spark/sql/columnar/InMemoryColumnarTableScan.scala

@@ -82,7 +82,7 @@ private[sql] case class InMemoryRelation(
     if (batchStats.value.isEmpty) {
       // Underlying columnar RDD hasn't been materialized, no useful statistics information
       // available, return the default statistics.
-      Statistics(sizeInBytes = child.sqlContext.defaultSizeInBytes)
+      Statistics(sizeInBytes = child.sqlContext.conf.defaultSizeInBytes)
     } else {
       // Underlying columnar RDD has been materialized, required information has also been collected
       // via the `batchStats` accumulator, compute the final statistics, and update `_statistics`.
@@ -233,7 +233,7 @@ private[sql] case class InMemoryColumnarTableScan(
   val readPartitions = sparkContext.accumulator(0)
   val readBatches = sparkContext.accumulator(0)
 
-  private val inMemoryPartitionPruningEnabled = sqlContext.inMemoryPartitionPruning
+  private val inMemoryPartitionPruningEnabled = sqlContext.conf.inMemoryPartitionPruning
 
   override def execute() = {
     readPartitions.setValue(0)

sql/core/src/main/scala/org/apache/spark/sql/execution/Exchange.scala

@@ -123,7 +123,7 @@ case class Exchange(newPartitioning: Partitioning, child: SparkPlan) extends Una
  */
 private[sql] case class AddExchange(sqlContext: SQLContext) extends Rule[SparkPlan] {
   // TODO: Determine the number of partitions.
-  def numPartitions = sqlContext.numShufflePartitions
+  def numPartitions = sqlContext.conf.numShufflePartitions
 
   def apply(plan: SparkPlan): SparkPlan = plan.transformUp {
     case operator: SparkPlan =>

sql/core/src/main/scala/org/apache/spark/sql/execution/ExistingRDD.scala

@@ -69,7 +69,7 @@ case class LogicalRDD(output: Seq[Attribute], rdd: RDD[Row])(sqlContext: SQLCont
   @transient override lazy val statistics = Statistics(
     // TODO: Instead of returning a default value here, find a way to return a meaningful size
     // estimate for RDDs. See PR 1238 for more discussions.
-    sizeInBytes = BigInt(sqlContext.defaultSizeInBytes)
+    sizeInBytes = BigInt(sqlContext.conf.defaultSizeInBytes)
   )
 }
 
@@ -106,6 +106,6 @@ case class SparkLogicalPlan(alreadyPlanned: SparkPlan)(@transient sqlContext: SQ
   @transient override lazy val statistics = Statistics(
     // TODO: Instead of returning a default value here, find a way to return a meaningful size
     // estimate for RDDs. See PR 1238 for more discussions.
-    sizeInBytes = BigInt(sqlContext.defaultSizeInBytes)
+    sizeInBytes = BigInt(sqlContext.conf.defaultSizeInBytes)
   )
 }

sql/core/src/main/scala/org/apache/spark/sql/execution/SparkPlan.scala

@@ -51,7 +51,7 @@ abstract class SparkPlan extends QueryPlan[SparkPlan] with Logging with Serializ
   // sqlContext will be null when we are being deserialized on the slaves.  In this instance
   // the value of codegenEnabled will be set by the desserializer after the constructor has run.
   val codegenEnabled: Boolean = if (sqlContext != null) {
-    sqlContext.codegenEnabled
+    sqlContext.conf.codegenEnabled
   } else {
     false
   }

sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala

@@ -35,8 +35,8 @@ private[sql] abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
   object LeftSemiJoin extends Strategy with PredicateHelper {
     def apply(plan: LogicalPlan): Seq[SparkPlan] = plan match {
       case ExtractEquiJoinKeys(LeftSemi, leftKeys, rightKeys, condition, left, right)
-        if sqlContext.autoBroadcastJoinThreshold > 0 &&
-          right.statistics.sizeInBytes <= sqlContext.autoBroadcastJoinThreshold =>
+        if sqlContext.conf.autoBroadcastJoinThreshold > 0 &&
+          right.statistics.sizeInBytes <= sqlContext.conf.autoBroadcastJoinThreshold =>
         val semiJoin = joins.BroadcastLeftSemiJoinHash(
           leftKeys, rightKeys, planLater(left), planLater(right))
         condition.map(Filter(_, semiJoin)).getOrElse(semiJoin) :: Nil
@@ -81,13 +81,13 @@ private[sql] abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
 
     def apply(plan: LogicalPlan): Seq[SparkPlan] = plan match {
       case ExtractEquiJoinKeys(Inner, leftKeys, rightKeys, condition, left, right)
-        if sqlContext.autoBroadcastJoinThreshold > 0 &&
-           right.statistics.sizeInBytes <= sqlContext.autoBroadcastJoinThreshold =>
+        if sqlContext.conf.autoBroadcastJoinThreshold > 0 &&
+           right.statistics.sizeInBytes <= sqlContext.conf.autoBroadcastJoinThreshold =>
         makeBroadcastHashJoin(leftKeys, rightKeys, left, right, condition, joins.BuildRight)
 
       case ExtractEquiJoinKeys(Inner, leftKeys, rightKeys, condition, left, right)
-        if sqlContext.autoBroadcastJoinThreshold > 0 &&
-           left.statistics.sizeInBytes <= sqlContext.autoBroadcastJoinThreshold =>
+        if sqlContext.conf.autoBroadcastJoinThreshold > 0 &&
+           left.statistics.sizeInBytes <= sqlContext.conf.autoBroadcastJoinThreshold =>
           makeBroadcastHashJoin(leftKeys, rightKeys, left, right, condition, joins.BuildLeft)
 
       case ExtractEquiJoinKeys(Inner, leftKeys, rightKeys, condition, left, right) =>
@@ -215,7 +215,7 @@ private[sql] abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
         InsertIntoParquetTable(table, planLater(child), overwrite) :: Nil
       case PhysicalOperation(projectList, filters: Seq[Expression], relation: ParquetRelation) =>
         val prunePushedDownFilters =
-          if (sqlContext.parquetFilterPushDown) {
+          if (sqlContext.conf.parquetFilterPushDown) {
             (predicates: Seq[Expression]) => {
               // Note: filters cannot be pushed down to Parquet if they contain more complex
               // expressions than simple "Attribute cmp Literal" comparisons. Here we remove all
@@ -237,7 +237,7 @@ private[sql] abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
           ParquetTableScan(
             _,
             relation,
-            if (sqlContext.parquetFilterPushDown) filters else Nil)) :: Nil
+            if (sqlContext.conf.parquetFilterPushDown) filters else Nil)) :: Nil
 
       case _ => Nil
     }
@@ -270,7 +270,7 @@ private[sql] abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
         // This sort only sorts tuples within a partition. Its requiredDistribution will be
         // an UnspecifiedDistribution.
         execution.Sort(sortExprs, global = false, planLater(child)) :: Nil
-      case logical.Sort(sortExprs, global, child) if sqlContext.externalSortEnabled =>
+      case logical.Sort(sortExprs, global, child) if sqlContext.conf.externalSortEnabled =>
         execution.ExternalSort(sortExprs, global, planLater(child)):: Nil
       case logical.Sort(sortExprs, global, child) =>
         execution.Sort(sortExprs, global, planLater(child)):: Nil

sql/core/src/main/scala/org/apache/spark/sql/execution/commands.scala

@@ -94,7 +94,7 @@ case class SetCommand(
       logWarning(
         s"Property ${SQLConf.Deprecated.MAPRED_REDUCE_TASKS} is deprecated, " +
           s"showing ${SQLConf.SHUFFLE_PARTITIONS} instead.")
-      Seq(Row(s"${SQLConf.SHUFFLE_PARTITIONS}=${sqlContext.numShufflePartitions}"))
+      Seq(Row(s"${SQLConf.SHUFFLE_PARTITIONS}=${sqlContext.conf.numShufflePartitions}"))
 
     // Queries a single property.
     case Some((key, None)) =>