[SPARK-37769][SQL][FOLLOWUP] Filtering files if metadata columns are …

…present in the data filter

### What changes were proposed in this pull request?
Follow-up PR of #34575. Filtering files if metadata columns are present in the data filter.

### Why are the changes needed?
Performance improvements.

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
Existing UTs and a new UT.

Closes #35055 from Yaohua628/spark-37769.

Authored-by: yaohua <yaohua.zhao@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileFormat.scala

@@ -30,6 +30,7 @@ import org.apache.spark.sql.errors.QueryExecutionErrors
 import org.apache.spark.sql.internal.SQLConf
 import org.apache.spark.sql.sources.Filter
 import org.apache.spark.sql.types.{DataType, LongType, StringType, StructField, StructType, TimestampType}
+import org.apache.spark.unsafe.types.UTF8String
 
 
 /**
@@ -192,6 +193,36 @@ object FileFormat {
 
   // create a file metadata struct col
   def createFileMetadataCol: AttributeReference = MetadataAttribute(METADATA_NAME, METADATA_STRUCT)
+
+  // create an internal row given required metadata fields and file information
+  def createMetadataInternalRow(
+      fieldNames: Seq[String],
+      filePath: Path,
+      fileSize: Long,
+      fileModificationTime: Long): InternalRow =
+    updateMetadataInternalRow(new GenericInternalRow(fieldNames.length), fieldNames,
+      filePath, fileSize, fileModificationTime)
+
+  // update an internal row given required metadata fields and file information
+  def updateMetadataInternalRow(
+      row: InternalRow,
+      fieldNames: Seq[String],
+      filePath: Path,
+      fileSize: Long,
+      fileModificationTime: Long): InternalRow = {
+    fieldNames.zipWithIndex.foreach { case (name, i) =>
+      name match {
+        case FILE_PATH => row.update(i, UTF8String.fromString(filePath.toString))
+        case FILE_NAME => row.update(i, UTF8String.fromString(filePath.getName))
+        case FILE_SIZE => row.update(i, fileSize)
+        case FILE_MODIFICATION_TIME =>
+          // the modificationTime from the file is in millisecond,
+          // while internally, the TimestampType `file_modification_time` is stored in microsecond
+          row.update(i, fileModificationTime * 1000L)
+      }
+    }
+    row
+  }
 }
 
 /**

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FileScanRDD.scala

@@ -34,7 +34,6 @@ import org.apache.spark.sql.execution.datasources.FileFormat._
 import org.apache.spark.sql.execution.vectorized.{OnHeapColumnVector, WritableColumnVector}
 import org.apache.spark.sql.types.{LongType, StringType, StructType}
 import org.apache.spark.sql.vectorized.ColumnarBatch
-import org.apache.spark.unsafe.types.UTF8String
 import org.apache.spark.util.NextIterator
 
 /**
@@ -136,18 +135,8 @@ class FileScanRDD(
        */
       private def updateMetadataRow(): Unit = {
         if (metadataColumns.nonEmpty && currentFile != null) {
-          val path = new Path(currentFile.filePath)
-          metadataColumns.zipWithIndex.foreach { case (attr, i) =>
-            attr.name match {
-              case FILE_PATH => metadataRow.update(i, UTF8String.fromString(path.toString))
-              case FILE_NAME => metadataRow.update(i, UTF8String.fromString(path.getName))
-              case FILE_SIZE => metadataRow.update(i, currentFile.fileSize)
-              case FILE_MODIFICATION_TIME =>
-                // the modificationTime from the file is in millisecond,
-                // while internally, the TimestampType is stored in microsecond
-                metadataRow.update(i, currentFile.modificationTime * 1000L)
-            }
-          }
+          updateMetadataInternalRow(metadataRow, metadataColumns.map(_.name),
+            new Path(currentFile.filePath), currentFile.fileSize, currentFile.modificationTime)
         }
       }
 

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/PartitioningAwareFileIndex.scala

@@ -27,6 +27,7 @@ import org.apache.spark.sql.SparkSession
 import org.apache.spark.sql.catalyst.{expressions, InternalRow}
 import org.apache.spark.sql.catalyst.expressions._
 import org.apache.spark.sql.catalyst.util.{CaseInsensitiveMap, DateTimeUtils}
+import org.apache.spark.sql.execution.datasources.FileFormat.createMetadataInternalRow
 import org.apache.spark.sql.types.StructType
 
 /**
@@ -71,8 +72,37 @@ abstract class PartitioningAwareFileIndex(
     def isNonEmptyFile(f: FileStatus): Boolean = {
       isDataPath(f.getPath) && f.getLen > 0
     }
+
+    // retrieve the file metadata filters and reduce to a final filter expression
+    val fileMetadataFilterOpt = dataFilters.filter(_.references.forall {
+      case MetadataAttribute(_) => true
+      case _ => false
+    }).reduceOption(expressions.And)
+
+    // - create a bound references for filters: put the metadata struct at 0 position for each file
+    // - retrieve the final metadata struct (could be pruned) from filters
+    val boundedFilterMetadataStructOpt = fileMetadataFilterOpt.map { fileMetadataFilter =>
+      val metadataStruct = fileMetadataFilter.references.head.dataType
+      val boundedFilter = Predicate.createInterpreted(fileMetadataFilter.transform {
+        case _: AttributeReference => BoundReference(0, metadataStruct, nullable = true)
+      })
+      (boundedFilter, metadataStruct)
+    }
+
+    def matchFileMetadataPredicate(f: FileStatus): Boolean = {
+      // use option.forall, so if there is no filter no metadata struct, return true
+      boundedFilterMetadataStructOpt.forall { case (boundedFilter, metadataStruct) =>
+        val row = InternalRow.fromSeq(Seq(
+          createMetadataInternalRow(metadataStruct.asInstanceOf[StructType].names,
+            f.getPath, f.getLen, f.getModificationTime)
+        ))
+        boundedFilter.eval(row)
+      }
+    }
+
     val selectedPartitions = if (partitionSpec().partitionColumns.isEmpty) {
-      PartitionDirectory(InternalRow.empty, allFiles().filter(isNonEmptyFile)) :: Nil
+      PartitionDirectory(InternalRow.empty, allFiles()
+        .filter(f => isNonEmptyFile(f) && matchFileMetadataPredicate(f))) :: Nil
     } else {
       if (recursiveFileLookup) {
         throw new IllegalArgumentException(
@@ -83,7 +113,8 @@ abstract class PartitioningAwareFileIndex(
           val files: Seq[FileStatus] = leafDirToChildrenFiles.get(path) match {
             case Some(existingDir) =>
               // Directory has children files in it, return them
-              existingDir.filter(f => matchPathPattern(f) && isNonEmptyFile(f))
+              existingDir.filter(f => matchPathPattern(f) && isNonEmptyFile(f) &&
+                matchFileMetadataPredicate(f))
 
             case None =>
               // Directory does not exist, or has no children files

sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/FileMetadataStructSuite.scala

@@ -279,16 +279,58 @@ class FileMetadataStructSuite extends QueryTest with SharedSparkSession {
   }
 
   metadataColumnsTest("filter", schema) { (df, f0, _) =>
+    val filteredDF = df.select("name", "age", METADATA_FILE_NAME)
+      .where(Column(METADATA_FILE_NAME) === f0(METADATA_FILE_NAME))
+
+    // check the filtered file
+    val partitions = filteredDF.queryExecution.sparkPlan.collectFirst {
+      case p: FileSourceScanExec => p.selectedPartitions
+    }.get
+
+    assert(partitions.length == 1) // 1 partition
+    assert(partitions.head.files.length == 1) // 1 file in that partition
+    assert(partitions.head.files.head.getPath.toString == f0(METADATA_FILE_PATH)) // the file is f0
+
+    // check result
     checkAnswer(
-      df.select("name", "age", METADATA_FILE_NAME)
-        .where(Column(METADATA_FILE_NAME) === f0(METADATA_FILE_NAME)),
+      filteredDF,
       Seq(
         // _file_name == f0's name, so we will only have 1 row
         Row("jack", 24, f0(METADATA_FILE_NAME))
       )
     )
   }
 
+  metadataColumnsTest("filter on metadata and user data", schema) { (df, _, f1) =>
+
+    val filteredDF = df.select("name", "age", "info",
+      METADATA_FILE_NAME, METADATA_FILE_PATH,
+      METADATA_FILE_SIZE, METADATA_FILE_MODIFICATION_TIME)
+      // mix metadata column + user column
+      .where(Column(METADATA_FILE_NAME) === f1(METADATA_FILE_NAME) and Column("name") === "lily")
+      // only metadata columns
+      .where(Column(METADATA_FILE_PATH) === f1(METADATA_FILE_PATH))
+      // only user column
+      .where("age == 31")
+
+    // check the filtered file
+    val partitions = filteredDF.queryExecution.sparkPlan.collectFirst {
+      case p: FileSourceScanExec => p.selectedPartitions
+    }.get
+
+    assert(partitions.length == 1) // 1 partition
+    assert(partitions.head.files.length == 1) // 1 file in that partition
+    assert(partitions.head.files.head.getPath.toString == f1(METADATA_FILE_PATH)) // the file is f1
+
+    // check result
+    checkAnswer(
+      filteredDF,
+      Seq(Row("lily", 31, Row(54321L, "ucb"),
+        f1(METADATA_FILE_NAME), f1(METADATA_FILE_PATH),
+        f1(METADATA_FILE_SIZE), f1(METADATA_FILE_MODIFICATION_TIME)))
+    )
+  }
+
   Seq(true, false).foreach { caseSensitive =>
     metadataColumnsTest(s"upper/lower case when case " +
       s"sensitive is $caseSensitive", schemaWithNameConflicts) { (df, f0, f1) =>