[SPARK-34897][SQL] Support reconcile schemas based on index after nes…

…ted column pruning ### What changes were proposed in this pull request? It will remove `StructField` when [pruning nested columns](https://github.com/apache/spark/blob/0f2c0b53e8fb18c86c67b5dd679c006db93f94a5/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/SchemaPruning.scala#L28-L42). For example: ```scala spark.sql( """ |CREATE TABLE t1 ( | _col0 INT, | _col1 STRING, | _col2 STRUCT<c1: STRING, c2: STRING, c3: STRING, c4: BIGINT>) |USING ORC |""".stripMargin) spark.sql("INSERT INTO t1 values(1, '2', struct('a', 'b', 'c', 10L))") spark.sql("SELECT _col0, _col2.c1 FROM t1").show ``` Before this pr. The returned schema is: ``` `_col0` INT,`_col2` STRUCT<`c1`: STRING> ``` add it will throw exception: ``` java.lang.AssertionError: assertion failed: The given data schema struct<_col0:int,_col2:struct<c1:string>> has less fields than the actual ORC physical schema, no idea which columns were dropped, fail to read. at scala.Predef$.assert(Predef.scala:223) at org.apache.spark.sql.execution.datasources.orc.OrcUtils$.requestedColumnIds(OrcUtils.scala:160) ``` After this pr. The returned schema is: ``` `_col0` INT,`_col1` STRING,`_col2` STRUCT<`c1`: STRING> ```. The finally schema is ``` `_col0` INT,`_col2` STRUCT<`c1`: STRING> ``` after the complete column pruning: https://github.com/apache/spark/blob/7a5647a93aaea9d1d78d9262e24fc8c010db04d0/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileSourceStrategy.scala#L208-L213 https://github.com/apache/spark/blob/e64eb75aede71a5403a4d4436e63b1fcfdeca14d/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/PushDownUtils.scala#L96-L97 ### Why are the changes needed? Fix bug. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? Unit test. Closes #31993 from wangyum/SPARK-34897. Authored-by: Yuming Wang <[email protected]> Signed-off-by: Liang-Chi Hsieh <[email protected]>
apache · Apr 21, 2021 · e609395 · e609395
1 parent 81dbaed
commit e609395
Show file tree

Hide file tree

Showing 5 changed files with 108 additions and 57 deletions.
diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/SchemaPruning.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/SchemaPruning.scala
@@ -22,9 +22,12 @@ import org.apache.spark.sql.types._
 
 object SchemaPruning extends SQLConfHelper {
   /**
-   * Filters the schema by the requested fields. For example, if the schema is struct<a:int, b:int>,
-   * and given requested field are "a", the field "b" is pruned in the returned schema.
-   * Note that schema field ordering at original schema is still preserved in pruned schema.
+   * Prunes the nested schema by the requested fields. For example, if the schema is:
+   * `id int, s struct<a:int, b:int>`, and given requested field "s.a", the inner field "b"
+   * is pruned in the returned schema: `id int, s struct<a:int>`.
+   * Note that:
+   *   1. The schema field ordering at original schema is still preserved in pruned schema.
+   *   2. The top-level fields are not pruned here.
    */
   def pruneDataSchema(
       dataSchema: StructType,
@@ -34,11 +37,10 @@ object SchemaPruning extends SQLConfHelper {
     // in the resulting schema may differ from their ordering in the logical relation's
     // original schema
     val mergedSchema = requestedRootFields
-      .map { case root: RootField => StructType(Array(root.field)) }
+      .map { root: RootField => StructType(Array(root.field)) }
       .reduceLeft(_ merge _)
-    val dataSchemaFieldNames = dataSchema.fieldNames.toSet
     val mergedDataSchema =
-      StructType(mergedSchema.filter(f => dataSchemaFieldNames.exists(resolver(_, f.name))))
+      StructType(dataSchema.map(d => mergedSchema.find(m => resolver(m.name, d.name)).getOrElse(d)))
     // Sort the fields of mergedDataSchema according to their order in dataSchema,
     // recursively. This makes mergedDataSchema a pruned schema of dataSchema
     sortLeftFieldsByRight(mergedDataSchema, dataSchema).asInstanceOf[StructType]

diff --git a/...atalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/SchemaPruningSuite.scala b/...atalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/SchemaPruningSuite.scala
@@ -18,39 +18,36 @@
 package org.apache.spark.sql.catalyst.expressions
 
 import org.apache.spark.SparkFunSuite
-import org.apache.spark.sql.catalyst.expressions.SchemaPruning.RootField
 import org.apache.spark.sql.catalyst.plans.SQLHelper
 import org.apache.spark.sql.internal.SQLConf.CASE_SENSITIVE
 import org.apache.spark.sql.types._
 
 class SchemaPruningSuite extends SparkFunSuite with SQLHelper {
-
-  def getRootFields(requestedFields: StructField*): Seq[RootField] = {
-    requestedFields.map { f =>
+  private def testPrunedSchema(
+      schema: StructType,
+      requestedFields: Seq[StructField],
+      expectedSchema: StructType): Unit = {
+    val requestedRootFields = requestedFields.map { f =>
       // `derivedFromAtt` doesn't affect the result of pruned schema.
       SchemaPruning.RootField(field = f, derivedFromAtt = true)
     }
+    val prunedSchema = SchemaPruning.pruneDataSchema(schema, requestedRootFields)
+    assert(prunedSchema === expectedSchema)
   }
 
   test("prune schema by the requested fields") {
-    def testPrunedSchema(
-        schema: StructType,
-        requestedFields: StructField*): Unit = {
-      val requestedRootFields = requestedFields.map { f =>
-        // `derivedFromAtt` doesn't affect the result of pruned schema.
-        SchemaPruning.RootField(field = f, derivedFromAtt = true)
-      }
-      val expectedSchema = SchemaPruning.pruneDataSchema(schema, requestedRootFields)
-      assert(expectedSchema == StructType(requestedFields))
-    }
-
-    testPrunedSchema(StructType.fromDDL("a int, b int"), StructField("a", IntegerType))
-    testPrunedSchema(StructType.fromDDL("a int, b int"), StructField("b", IntegerType))
+    testPrunedSchema(
+      StructType.fromDDL("a int, b int"),
+      Seq(StructField("a", IntegerType)),
+      StructType.fromDDL("a int, b int"))
 
     val structOfStruct = StructType.fromDDL("a struct<a:int, b:int>, b int")
-    testPrunedSchema(structOfStruct, StructField("a", StructType.fromDDL("a int, b int")))
-    testPrunedSchema(structOfStruct, StructField("b", IntegerType))
-    testPrunedSchema(structOfStruct, StructField("a", StructType.fromDDL("b int")))
+    testPrunedSchema(structOfStruct,
+      Seq(StructField("a", StructType.fromDDL("a int")), StructField("b", IntegerType)),
+      StructType.fromDDL("a struct<a:int>, b int"))
+    testPrunedSchema(structOfStruct,
+      Seq(StructField("a", StructType.fromDDL("a int"))),
+      StructType.fromDDL("a struct<a:int>, b int"))
 
     val arrayOfStruct = StructField("a", ArrayType(StructType.fromDDL("a int, b int, c string")))
     val mapOfStruct = StructField("d", MapType(StructType.fromDDL("a int, b int, c string"),
@@ -60,44 +57,76 @@ class SchemaPruningSuite extends SparkFunSuite with SQLHelper {
       arrayOfStruct :: StructField("b", structOfStruct) :: StructField("c", IntegerType) ::
         mapOfStruct :: Nil)
 
-    testPrunedSchema(complexStruct, StructField("a", ArrayType(StructType.fromDDL("b int"))),
-      StructField("b", StructType.fromDDL("a int")))
     testPrunedSchema(complexStruct,
-      StructField("a", ArrayType(StructType.fromDDL("b int, c string"))),
-      StructField("b", StructType.fromDDL("b int")))
+      Seq(StructField("a", ArrayType(StructType.fromDDL("b int"))),
+        StructField("b", StructType.fromDDL("a int"))),
+      StructType(
+        StructField("a", ArrayType(StructType.fromDDL("b int"))) ::
+          StructField("b", StructType.fromDDL("a int")) ::
+          StructField("c", IntegerType) ::
+          mapOfStruct :: Nil))
+    testPrunedSchema(complexStruct,
+      Seq(StructField("a", ArrayType(StructType.fromDDL("b int, c string"))),
+        StructField("b", StructType.fromDDL("b int"))),
+      StructType(
+        StructField("a", ArrayType(StructType.fromDDL("b int, c string"))) ::
+          StructField("b", StructType.fromDDL("b int")) ::
+          StructField("c", IntegerType) ::
+          mapOfStruct :: Nil))
 
     val selectFieldInMap = StructField("d", MapType(StructType.fromDDL("a int, b int"),
       StructType.fromDDL("e int, f string")))
-    testPrunedSchema(complexStruct, StructField("c", IntegerType), selectFieldInMap)
+    testPrunedSchema(complexStruct,
+      Seq(StructField("c", IntegerType), selectFieldInMap),
+      StructType(
+        arrayOfStruct ::
+          StructField("b", structOfStruct) ::
+          StructField("c", IntegerType) ::
+          selectFieldInMap :: Nil))
   }
 
   test("SPARK-35096: test case insensitivity of pruned schema") {
-    Seq(true, false).foreach(isCaseSensitive => {
+    val upperCaseSchema = StructType.fromDDL("A struct<A:int, B:int>, B int")
+    val lowerCaseSchema = StructType.fromDDL("a struct<a:int, b:int>, b int")
+    val upperCaseRequestedFields = Seq(StructField("A", StructType.fromDDL("A int")))
+    val lowerCaseRequestedFields = Seq(StructField("a", StructType.fromDDL("a int")))
+
+    Seq(true, false).foreach { isCaseSensitive =>
       withSQLConf(CASE_SENSITIVE.key -> isCaseSensitive.toString) {
         if (isCaseSensitive) {
-          // Schema is case-sensitive
-          val requestedFields = getRootFields(StructField("id", IntegerType))
-          val prunedSchema = SchemaPruning.pruneDataSchema(
-            StructType.fromDDL("ID int, name String"), requestedFields)
-          assert(prunedSchema == StructType(Seq.empty))
-          // Root fields are case-sensitive
-          val rootFieldsSchema = SchemaPruning.pruneDataSchema(
-            StructType.fromDDL("id int, name String"),
-            getRootFields(StructField("ID", IntegerType)))
-          assert(rootFieldsSchema == StructType(StructType(Seq.empty)))
+          testPrunedSchema(
+            upperCaseSchema,
+            upperCaseRequestedFields,
+            StructType.fromDDL("A struct<A:int>, B int"))
+          testPrunedSchema(
+            upperCaseSchema,
+            lowerCaseRequestedFields,
+            upperCaseSchema)
+
+          testPrunedSchema(
+            lowerCaseSchema,
+            upperCaseRequestedFields,
+            lowerCaseSchema)
+          testPrunedSchema(
+            lowerCaseSchema,
+            lowerCaseRequestedFields,
+            StructType.fromDDL("a struct<a:int>, b int"))
         } else {
-          // Schema is case-insensitive
-          val prunedSchema = SchemaPruning.pruneDataSchema(
-            StructType.fromDDL("ID int, name String"),
-            getRootFields(StructField("id", IntegerType)))
-          assert(prunedSchema == StructType(StructField("ID", IntegerType) :: Nil))
-          // Root fields are case-insensitive
-          val rootFieldsSchema = SchemaPruning.pruneDataSchema(
-            StructType.fromDDL("id int, name String"),
-            getRootFields(StructField("ID", IntegerType)))
-          assert(rootFieldsSchema == StructType(StructField("id", IntegerType) :: Nil))
+          Seq(upperCaseRequestedFields, lowerCaseRequestedFields).foreach { requestedFields =>
+            testPrunedSchema(
+              upperCaseSchema,
+              requestedFields,
+              StructType.fromDDL("A struct<A:int>, B int"))
+          }
+
+          Seq(upperCaseRequestedFields, lowerCaseRequestedFields).foreach { requestedFields =>
+            testPrunedSchema(
+              lowerCaseSchema,
+              requestedFields,
+              StructType.fromDDL("a struct<a:int>, b int"))
+          }
         }
       }
-    })
+    }
   }
 }
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/HadoopFsRelation.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/HadoopFsRelation.scala
@@ -40,6 +40,8 @@ import org.apache.spark.sql.types.{StructField, StructType}
 case class HadoopFsRelation(
     location: FileIndex,
     partitionSchema: StructType,
+    // The top-level columns in `dataSchema` should match the actual physical file schema, otherwise
+    // the ORC data source may not work with the by-ordinal mode.
     dataSchema: StructType,
     bucketSpec: Option[BucketSpec],
     fileFormat: FileFormat,

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/PushDownUtils.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/PushDownUtils.scala
@@ -81,6 +81,10 @@ object PushDownUtils extends PredicateHelper {
       relation: DataSourceV2Relation,
       projects: Seq[NamedExpression],
       filters: Seq[Expression]): (Scan, Seq[AttributeReference]) = {
+    val exprs = projects ++ filters
+    val requiredColumns = AttributeSet(exprs.flatMap(_.references))
+    val neededOutput = relation.output.filter(requiredColumns.contains)
+
     scanBuilder match {
       case r: SupportsPushDownRequiredColumns if SQLConf.get.nestedSchemaPruningEnabled =>
         val rootFields = SchemaPruning.identifyRootFields(projects, filters)
@@ -89,14 +93,12 @@ object PushDownUtils extends PredicateHelper {
         } else {
           new StructType()
         }
-        r.pruneColumns(prunedSchema)
+        val neededFieldNames = neededOutput.map(_.name).toSet
+        r.pruneColumns(StructType(prunedSchema.filter(f => neededFieldNames.contains(f.name))))
         val scan = r.build()
         scan -> toOutputAttrs(scan.readSchema(), relation)
 
       case r: SupportsPushDownRequiredColumns =>
-        val exprs = projects ++ filters
-        val requiredColumns = AttributeSet(exprs.flatMap(_.references))
-        val neededOutput = relation.output.filter(requiredColumns.contains)
         r.pruneColumns(neededOutput.toStructType)
         val scan = r.build()
         // always project, in case the relation's output has been updated and doesn't match

diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/orc/OrcSourceSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/orc/OrcSourceSuite.scala
@@ -633,4 +633,20 @@ class OrcSourceSuite extends OrcSuite with SharedSparkSession {
       }
     }
   }
+
+  test("SPARK-34897: Support reconcile schemas based on index after nested column pruning") {
+    withTable("t1") {
+      spark.sql(
+        """
+          |CREATE TABLE t1 (
+          |  _col0 INT,
+          |  _col1 STRING,
+          |  _col2 STRUCT<c1: STRING, c2: STRING, c3: STRING, c4: BIGINT>)
+          |USING ORC
+          |""".stripMargin)
+
+      spark.sql("INSERT INTO t1 values(1, '2', struct('a', 'b', 'c', 10L))")
+      checkAnswer(spark.sql("SELECT _col0, _col2.c1 FROM t1"), Seq(Row(1, "a")))
+    }
+  }
 }