[SPARK-21499] [SQL] Support creating persistent function for Spark UDAF(UserDefinedAggregateFunction)

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/catalog/SessionCatalog.scala

@@ -17,13 +17,15 @@
 
 package org.apache.spark.sql.catalyst.catalog
 
+import java.lang.reflect.InvocationTargetException
 import java.net.URI
 import java.util.Locale
 import java.util.concurrent.Callable
 import javax.annotation.concurrent.GuardedBy
 
 import scala.collection.mutable
 import scala.util.{Failure, Success, Try}
+import scala.util.control.NonFatal
 
 import com.google.common.cache.{Cache, CacheBuilder}
 import org.apache.hadoop.conf.Configuration
@@ -39,7 +41,9 @@ import org.apache.spark.sql.catalyst.parser.{CatalystSqlParser, ParserInterface}
 import org.apache.spark.sql.catalyst.plans.logical.{LogicalPlan, SubqueryAlias, View}
 import org.apache.spark.sql.catalyst.util.StringUtils
 import org.apache.spark.sql.internal.SQLConf
+import org.apache.spark.sql.internal.StaticSQLConf.CATALOG_IMPLEMENTATION
 import org.apache.spark.sql.types.StructType
+import org.apache.spark.util.Utils
 
 object SessionCatalog {
   val DEFAULT_DATABASE = "default"
@@ -126,6 +130,13 @@ class SessionCatalog(
     if (conf.caseSensitiveAnalysis) name else name.toLowerCase(Locale.ROOT)
   }
 
+  /**
+   * Checks whether the Hive metastore is being used
+   */
+  private def isUsingHiveMetastore: Boolean = {
+    conf.getConf(CATALOG_IMPLEMENTATION).toLowerCase(Locale.ROOT) == "hive"
+  }
+
   private val tableRelationCache: Cache[QualifiedTableName, LogicalPlan] = {
     val cacheSize = conf.tableRelationCacheSize
     CacheBuilder.newBuilder().maximumSize(cacheSize).build[QualifiedTableName, LogicalPlan]()
@@ -1090,14 +1101,50 @@ class SessionCatalog(
   // | Methods that interact with temporary and metastore functions |
   // ----------------------------------------------------------------
 
+  /**
+   * Constructs a [[FunctionBuilder]] based on the provided class that represents a function.
+   */
+  protected def makeFunctionBuilder(name: String, functionClassName: String): FunctionBuilder = {
+    val clazz = Utils.classForName(functionClassName)
+    (children: Seq[Expression]) => {
+      try {
+        makeFunctionExpression(name, Utils.classForName(functionClassName), children).getOrElse {
+          val extraMsg =
+            if (!isUsingHiveMetastore) "Use sparkSession.udf.register(...) instead." else ""
+          throw new AnalysisException(
+            s"No handler for UDF/UDAF/UDTF '${clazz.getCanonicalName}'. $extraMsg")
+        }
+      } catch {
+        case ae: AnalysisException =>
+          throw ae
+        case NonFatal(e) =>
+          val analysisException =
+            new AnalysisException(s"No handler for UDF/UDAF/UDTF '${clazz.getCanonicalName}': $e")
+          analysisException.setStackTrace(e.getStackTrace)
+          throw analysisException
+      }
+    }
+  }
+
   /**
    * Construct a [[FunctionBuilder]] based on the provided class that represents a function.
    *
    * This performs reflection to decide what type of [[Expression]] to return in the builder.
    */
-  protected def makeFunctionBuilder(name: String, functionClassName: String): FunctionBuilder = {
-    // TODO: at least support UDAFs here
-    throw new UnsupportedOperationException("Use sqlContext.udf.register(...) instead.")
+  protected def makeFunctionExpression(
+      name: String,
+      clazz: Class[_],
+      children: Seq[Expression]): Option[Expression] = {
+    val clsForUDAF =
+      Utils.classForName("org.apache.spark.sql.expressions.UserDefinedAggregateFunction")
+    if (clsForUDAF.isAssignableFrom(clazz)) {
+      val cls = Utils.classForName("org.apache.spark.sql.execution.aggregate.ScalaUDAF")
+      Some(cls.getConstructor(classOf[Seq[Expression]], clsForUDAF, classOf[Int], classOf[Int])
+        .newInstance(children, clazz.newInstance().asInstanceOf[Object], Int.box(1), Int.box(1))
+        .asInstanceOf[Expression])
+    } else {
+      None
+    }
   }
 
   /**
@@ -1121,7 +1168,14 @@ class SessionCatalog(
     }
     val info = new ExpressionInfo(funcDefinition.className, func.database.orNull, func.funcName)
     val builder =
-      functionBuilder.getOrElse(makeFunctionBuilder(func.unquotedString, funcDefinition.className))
+      functionBuilder.getOrElse {
+        val className = funcDefinition.className
+        if (!Utils.classIsLoadable(className)) {
+          throw new AnalysisException(s"Can not load class '$className' when registering " +
+            s"the function '$func', please make sure it is on the classpath")
+        }
+        makeFunctionBuilder(func.unquotedString, className)
+      }
     functionRegistry.registerFunction(func, info, builder)
   }
 

sql/core/src/test/resources/sql-tests/inputs/udaf.sql

@@ -0,0 +1,13 @@
+CREATE OR REPLACE TEMPORARY VIEW t1 AS SELECT * FROM VALUES
+(1), (2), (3), (4)
+as t1(int_col1);
+
+CREATE FUNCTION myDoubleAvg AS 'test.org.apache.spark.sql.MyDoubleAvg';
+
+SELECT default.myDoubleAvg(int_col1) as my_avg from t1;
+
+SELECT default.myDoubleAvg(int_col1, 3) as my_avg from t1;
+
+CREATE FUNCTION udaf1 AS 'test.non.existent.udaf';
+
+SELECT default.udaf1(int_col1) as udaf1 from t1;

sql/core/src/test/resources/sql-tests/results/udaf.sql.out

@@ -0,0 +1,54 @@
+-- Automatically generated by SQLQueryTestSuite
+-- Number of queries: 6
+
+
+-- !query 0
+CREATE OR REPLACE TEMPORARY VIEW t1 AS SELECT * FROM VALUES
+(1), (2), (3), (4)
+as t1(int_col1)
+-- !query 0 schema
+struct<>
+-- !query 0 output
+
+
+
+-- !query 1
+CREATE FUNCTION myDoubleAvg AS 'test.org.apache.spark.sql.MyDoubleAvg'
+-- !query 1 schema
+struct<>
+-- !query 1 output
+
+
+
+-- !query 2
+SELECT default.myDoubleAvg(int_col1) as my_avg from t1
+-- !query 2 schema
+struct<my_avg:double>
+-- !query 2 output
+102.5
+
+
+-- !query 3
+SELECT default.myDoubleAvg(int_col1, 3) as my_avg from t1
+-- !query 3 schema
+struct<>
+-- !query 3 output
+java.lang.AssertionError
+assertion failed: Incorrect number of children
+
+
+-- !query 4
+CREATE FUNCTION udaf1 AS 'test.non.existent.udaf'
+-- !query 4 schema
+struct<>
+-- !query 4 output
+
+
+
+-- !query 5
+SELECT default.udaf1(int_col1) as udaf1 from t1
+-- !query 5 schema
+struct<>
+-- !query 5 output
+org.apache.spark.sql.AnalysisException
+Can not load class 'test.non.existent.udaf' when registering the function 'default.udaf1', please make sure it is on the classpath; line 1 pos 7

sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveSessionCatalog.scala

@@ -27,7 +27,6 @@ import org.apache.hadoop.hive.ql.exec.{UDAF, UDF}
 import org.apache.hadoop.hive.ql.exec.{FunctionRegistry => HiveFunctionRegistry}
 import org.apache.hadoop.hive.ql.udf.generic.{AbstractGenericUDAFResolver, GenericUDF, GenericUDTF}
 
-import org.apache.spark.sql.AnalysisException
 import org.apache.spark.sql.catalyst.FunctionIdentifier
 import org.apache.spark.sql.catalyst.analysis.FunctionRegistry
 import org.apache.spark.sql.catalyst.analysis.FunctionRegistry.FunctionBuilder
@@ -37,7 +36,6 @@ import org.apache.spark.sql.catalyst.parser.ParserInterface
 import org.apache.spark.sql.hive.HiveShim.HiveFunctionWrapper
 import org.apache.spark.sql.internal.SQLConf
 import org.apache.spark.sql.types.{DecimalType, DoubleType}
-import org.apache.spark.util.Utils
 
 
 private[sql] class HiveSessionCatalog(
@@ -58,54 +56,44 @@ private[sql] class HiveSessionCatalog(
       parser,
       functionResourceLoader) {
 
-  override def makeFunctionBuilder(funcName: String, className: String): FunctionBuilder = {
-    makeFunctionBuilder(funcName, Utils.classForName(className))
-  }
-
   /**
    * Construct a [[FunctionBuilder]] based on the provided class that represents a function.
    */
-  private def makeFunctionBuilder(name: String, clazz: Class[_]): FunctionBuilder = {
-    // When we instantiate hive UDF wrapper class, we may throw exception if the input
-    // expressions don't satisfy the hive UDF, such as type mismatch, input number
-    // mismatch, etc. Here we catch the exception and throw AnalysisException instead.
-    (children: Seq[Expression]) => {
-      try {
-        if (classOf[UDF].isAssignableFrom(clazz)) {
-          val udf = HiveSimpleUDF(name, new HiveFunctionWrapper(clazz.getName), children)
-          udf.dataType // Force it to check input data types.
-          udf
-        } else if (classOf[GenericUDF].isAssignableFrom(clazz)) {
-          val udf = HiveGenericUDF(name, new HiveFunctionWrapper(clazz.getName), children)
-          udf.dataType // Force it to check input data types.
-          udf
-        } else if (classOf[AbstractGenericUDAFResolver].isAssignableFrom(clazz)) {
-          val udaf = HiveUDAFFunction(name, new HiveFunctionWrapper(clazz.getName), children)
-          udaf.dataType // Force it to check input data types.
-          udaf
-        } else if (classOf[UDAF].isAssignableFrom(clazz)) {
-          val udaf = HiveUDAFFunction(
-            name,
-            new HiveFunctionWrapper(clazz.getName),
-            children,
-            isUDAFBridgeRequired = true)
-          udaf.dataType  // Force it to check input data types.
-          udaf
-        } else if (classOf[GenericUDTF].isAssignableFrom(clazz)) {
-          val udtf = HiveGenericUDTF(name, new HiveFunctionWrapper(clazz.getName), children)
-          udtf.elementSchema // Force it to check input data types.
-          udtf
-        } else {
-          throw new AnalysisException(s"No handler for Hive UDF '${clazz.getCanonicalName}'")
-        }
-      } catch {
-        case ae: AnalysisException =>
-          throw ae
-        case NonFatal(e) =>
-          val analysisException =
-            new AnalysisException(s"No handler for Hive UDF '${clazz.getCanonicalName}': $e")
-          analysisException.setStackTrace(e.getStackTrace)
-          throw analysisException
+  override def makeFunctionExpression(
+      name: String,
+      clazz: Class[_],
+      children: Seq[Expression]): Option[Expression] = {
+
+    super.makeFunctionExpression(name, clazz, children).orElse {
+      // When we instantiate hive UDF wrapper class, we may throw exception if the input
+      // expressions don't satisfy the hive UDF, such as type mismatch, input number
+      // mismatch, etc. Here we catch the exception and throw AnalysisException instead.
+      if (classOf[UDF].isAssignableFrom(clazz)) {
+        val udf = HiveSimpleUDF(name, new HiveFunctionWrapper(clazz.getName), children)
+        udf.dataType // Force it to check input data types.
+        Some(udf)
+      } else if (classOf[GenericUDF].isAssignableFrom(clazz)) {
+        val udf = HiveGenericUDF(name, new HiveFunctionWrapper(clazz.getName), children)
+        udf.dataType // Force it to check input data types.
+        Some(udf)
+      } else if (classOf[AbstractGenericUDAFResolver].isAssignableFrom(clazz)) {
+        val udaf = HiveUDAFFunction(name, new HiveFunctionWrapper(clazz.getName), children)
+        udaf.dataType // Force it to check input data types.
+        Some(udaf)
+      } else if (classOf[UDAF].isAssignableFrom(clazz)) {
+        val udaf = HiveUDAFFunction(
+          name,
+          new HiveFunctionWrapper(clazz.getName),
+          children,
+          isUDAFBridgeRequired = true)
+        udaf.dataType // Force it to check input data types.
+        Some(udaf)
+      } else if (classOf[GenericUDTF].isAssignableFrom(clazz)) {
+        val udtf = HiveGenericUDTF(name, new HiveFunctionWrapper(clazz.getName), children)
+        udtf.elementSchema // Force it to check input data types.
+        Some(udtf)
+      } else {
+        None
       }
     }
   }

sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveUDAFSuite.scala

@@ -26,6 +26,7 @@ import org.apache.hadoop.hive.ql.util.JavaDataModel
 import org.apache.hadoop.hive.serde2.objectinspector.{ObjectInspector, ObjectInspectorFactory}
 import org.apache.hadoop.hive.serde2.objectinspector.primitive.PrimitiveObjectInspectorFactory
 import org.apache.hadoop.hive.serde2.typeinfo.TypeInfo
+import test.org.apache.spark.sql.MyDoubleAvg
 
 import org.apache.spark.sql.{AnalysisException, QueryTest, Row}
 import org.apache.spark.sql.execution.aggregate.ObjectHashAggregateExec
@@ -86,6 +87,18 @@ class HiveUDAFSuite extends QueryTest with TestHiveSingleton with SQLTestUtils {
     ))
   }
 
+  test("call JAVA UDAF") {
+    withTempView("temp") {
+      withUserDefinedFunction("myDoubleAvg" -> false) {
+        spark.range(1, 10).toDF("value").createOrReplaceTempView("temp")
+        sql(s"CREATE FUNCTION myDoubleAvg AS '${classOf[MyDoubleAvg].getName}'")
+        checkAnswer(
+          spark.sql("SELECT default.myDoubleAvg(value) as my_avg from temp"),
+          Row(105.0))
+      }
+    }
+  }
+
   test("non-deterministic children expressions of UDAF") {
     withTempView("view1") {
       spark.range(1).selectExpr("id as x", "id as y").createTempView("view1")

sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/HiveUDFSuite.scala

@@ -404,59 +404,34 @@ class HiveUDFSuite extends QueryTest with TestHiveSingleton with SQLTestUtils {
   }
 
   test("Hive UDFs with insufficient number of input arguments should trigger an analysis error") {
-    Seq((1, 2)).toDF("a", "b").createOrReplaceTempView("testUDF")
+    withTempView("testUDF") {
+      Seq((1, 2)).toDF("a", "b").createOrReplaceTempView("testUDF")
+
+      def testErrorMsgForFunc(funcName: String, className: String): Unit = {
+        withUserDefinedFunction(funcName -> true) {
+          sql(s"CREATE TEMPORARY FUNCTION $funcName AS '$className'")
+          val message = intercept[AnalysisException] {
+            sql(s"SELECT $funcName() FROM testUDF")
+          }.getMessage
+          assert(message.contains(s"No handler for UDF/UDAF/UDTF '$className'"))
+        }
+      }
 
-    {
       // HiveSimpleUDF
-      sql(s"CREATE TEMPORARY FUNCTION testUDFTwoListList AS '${classOf[UDFTwoListList].getName}'")
-      val message = intercept[AnalysisException] {
-        sql("SELECT testUDFTwoListList() FROM testUDF")
-      }.getMessage
-      assert(message.contains("No handler for Hive UDF"))
-      sql("DROP TEMPORARY FUNCTION IF EXISTS testUDFTwoListList")
-    }
+      testErrorMsgForFunc("testUDFTwoListList", classOf[UDFTwoListList].getName)
 
-    {
       // HiveGenericUDF
-      sql(s"CREATE TEMPORARY FUNCTION testUDFAnd AS '${classOf[GenericUDFOPAnd].getName}'")
-      val message = intercept[AnalysisException] {
-        sql("SELECT testUDFAnd() FROM testUDF")
-      }.getMessage
-      assert(message.contains("No handler for Hive UDF"))
-      sql("DROP TEMPORARY FUNCTION IF EXISTS testUDFAnd")
-    }
+      testErrorMsgForFunc("testUDFAnd", classOf[GenericUDFOPAnd].getName)
 
-    {
       // Hive UDAF
-      sql(s"CREATE TEMPORARY FUNCTION testUDAFPercentile AS '${classOf[UDAFPercentile].getName}'")
-      val message = intercept[AnalysisException] {
-        sql("SELECT testUDAFPercentile(a) FROM testUDF GROUP BY b")
-      }.getMessage
-      assert(message.contains("No handler for Hive UDF"))
-      sql("DROP TEMPORARY FUNCTION IF EXISTS testUDAFPercentile")
-    }
+      testErrorMsgForFunc("testUDAFPercentile", classOf[UDAFPercentile].getName)
 
-    {
       // AbstractGenericUDAFResolver
-      sql(s"CREATE TEMPORARY FUNCTION testUDAFAverage AS '${classOf[GenericUDAFAverage].getName}'")
-      val message = intercept[AnalysisException] {
-        sql("SELECT testUDAFAverage() FROM testUDF GROUP BY b")
-      }.getMessage
-      assert(message.contains("No handler for Hive UDF"))
-      sql("DROP TEMPORARY FUNCTION IF EXISTS testUDAFAverage")
-    }
+      testErrorMsgForFunc("testUDAFAverage", classOf[GenericUDAFAverage].getName)
 
-    {
-      // Hive UDTF
-      sql(s"CREATE TEMPORARY FUNCTION testUDTFExplode AS '${classOf[GenericUDTFExplode].getName}'")
-      val message = intercept[AnalysisException] {
-        sql("SELECT testUDTFExplode() FROM testUDF")
-      }.getMessage
-      assert(message.contains("No handler for Hive UDF"))
-      sql("DROP TEMPORARY FUNCTION IF EXISTS testUDTFExplode")
+      // AbstractGenericUDAFResolver
+      testErrorMsgForFunc("testUDTFExplode", classOf[GenericUDTFExplode].getName)
     }
-
-    spark.catalog.dropTempView("testUDF")
   }
 
   test("Hive UDF in group by") {