[Spark] Remove dropped columns when running REORG PURGE (#3371)

<!--
Thanks for sending a pull request!  Here are some tips for you:
1. If this is your first time, please read our contributor guidelines:
https://github.com/delta-io/delta/blob/master/CONTRIBUTING.md
2. If the PR is unfinished, add '[WIP]' in your PR title, e.g., '[WIP]
Your PR title ...'.
  3. Be sure to keep the PR description updated to reflect all changes.
  4. Please write your PR title to summarize what this PR proposes.
5. If possible, provide a concise example to reproduce the issue for a
faster review.
6. If applicable, include the corresponding issue number in the PR title
and link it in the body.
-->

#### Which Delta project/connector is this regarding?
<!--
Please add the component selected below to the beginning of the pull
request title
For example: [Spark] Title of my pull request
-->

- [x] Spark
- [ ] Standalone
- [ ] Flink
- [ ] Kernel
- [ ] Other (fill in here)

## Description
According to #3228, we add the support to also find and remove dropped
columns when running `REORG PURGE`.

Close #3228.

<!--
- Describe what this PR changes.
- Describe why we need the change.
 
If this PR resolves an issue be sure to include "Resolves #XXX" to
correctly link and close the issue upon merge.
-->

## How was this patch tested?
Through unit test in `DeltaReorgSuite.scala`.

<!--
If tests were added, say they were added here. Please make sure to test
the changes thoroughly including negative and positive cases if
possible.
If the changes were tested in any way other than unit tests, please
clarify how you tested step by step (ideally copy and paste-able, so
that other reviewers can test and check, and descendants can verify in
the future).
If the changes were not tested, please explain why.
-->

## Does this PR introduce _any_ user-facing changes?
No.
<!--
If yes, please clarify the previous behavior and the change this PR
proposes - provide the console output, description and/or an example to
show the behavior difference if possible.
If possible, please also clarify if this is a user-facing change
compared to the released Delta Lake versions or within the unreleased
branches such as master.
If no, write 'No'.
-->

Author: xzhseh

spark/src/main/scala/org/apache/spark/sql/delta/commands/DeltaReorgTableCommand.scala

@@ -102,14 +102,25 @@ sealed trait DeltaReorgOperation {
 
 /**
  * Reorg operation to purge files with soft deleted rows.
+ * This operation will also try finding and removing the dropped columns from parquet files,
+ * if ever exists such column that does not present in the current table schema.
  */
-class DeltaPurgeOperation extends DeltaReorgOperation {
+class DeltaPurgeOperation extends DeltaReorgOperation with ReorgTableHelper {
   override def filterFilesToReorg(spark: SparkSession, snapshot: Snapshot, files: Seq[AddFile])
-    : Seq[AddFile] =
-    files.filter { file =>
-      (file.deletionVector != null && file.numPhysicalRecords.isEmpty) ||
+    : Seq[AddFile] = {
+    val physicalSchema = DeltaColumnMapping.renameColumns(snapshot.schema)
+    val protocol = snapshot.protocol
+    val metadata = snapshot.metadata
+    val filesWithDroppedColumns: Seq[AddFile] =
+      filterParquetFilesOnExecutors(spark, files, snapshot, ignoreCorruptFiles = false) {
+        schema => fileHasExtraColumns(schema, physicalSchema, protocol, metadata)
+      }
+    val filesWithDV: Seq[AddFile] = files.filter { file =>
+        (file.deletionVector != null && file.numPhysicalRecords.isEmpty) ||
         file.numDeletedRecords > 0L
     }
+    (filesWithDroppedColumns ++ filesWithDV).distinct
+  }
 }
 
 /**

spark/src/main/scala/org/apache/spark/sql/delta/commands/ReorgTableHelper.scala

@@ -16,10 +16,10 @@
 
 package org.apache.spark.sql.delta.commands
 
-import org.apache.spark.sql.delta.Snapshot
-import org.apache.spark.sql.delta.actions.AddFile
+import org.apache.spark.sql.delta.{MaterializedRowCommitVersion, MaterializedRowId, Snapshot}
+import org.apache.spark.sql.delta.actions.{AddFile, Metadata, Protocol}
 import org.apache.spark.sql.delta.commands.VacuumCommand.generateCandidateFileMap
-import org.apache.spark.sql.delta.schema.SchemaMergingUtils
+import org.apache.spark.sql.delta.schema.{SchemaMergingUtils, SchemaUtils}
 import org.apache.spark.sql.delta.util.DeltaFileOperations
 import org.apache.hadoop.conf.Configuration
 import org.apache.hadoop.fs.{FileStatus, Path}
@@ -31,8 +31,12 @@ import org.apache.spark.util.SerializableConfiguration
 
 trait ReorgTableHelper extends Serializable {
   /**
-   * Determine whether `fileSchema` has any columns that has a type that differs from
+   * Determine whether `fileSchema` has any column that has a type that differs from
    * `tablePhysicalSchema`.
+   *
+   * @param fileSchema the current parquet schema to be checked.
+   * @param tablePhysicalSchema the current table schema.
+   * @return whether the file has any column that has a different type from table column.
    */
   protected def fileHasDifferentTypes(
       fileSchema: StructType,
@@ -46,6 +50,48 @@ trait ReorgTableHelper extends Serializable {
     false
   }
 
+  /**
+   * Determine whether `fileSchema` has any column that does not exist in the
+   * `tablePhysicalSchema`, this is possible by running ALTER TABLE commands,
+   * e.g., ALTER TABLE DROP COLUMN.
+   *
+   * @param fileSchema the current parquet schema to be checked.
+   * @param tablePhysicalSchema the current table schema.
+   * @param protocol the protocol used to check `row_id` and `row_commit_version`.
+   * @param metadata the metadata used to check `row_id` and `row_commit_version`.
+   * @return whether the file has any dropped column.
+   */
+  protected def fileHasExtraColumns(
+      fileSchema: StructType,
+      tablePhysicalSchema: StructType,
+      protocol: Protocol,
+      metadata: Metadata): Boolean = {
+    // 0. get the materialized names for `row_id` and `row_commit_version`.
+    val materializedRowIdColumnNameOpt =
+      MaterializedRowId.getMaterializedColumnName(protocol, metadata)
+    val materializedRowCommitVersionColumnNameOpt =
+      MaterializedRowCommitVersion.getMaterializedColumnName(protocol, metadata)
+
+    SchemaMergingUtils.transformColumns(fileSchema) { (path, field, _) =>
+      // 1. check whether the field exists in the `tablePhysicalSchema`.
+      val fullName = path :+ field.name
+      val inTableFieldOpt = SchemaUtils.findNestedFieldIgnoreCase(
+        tablePhysicalSchema, fullName, includeCollections = true)
+
+      // 2. check whether the current `field` is `row_id` or `row_commit_version`
+      //    column; if so, we need to explicitly keep these columns since they are
+      //    not part of the table schema but exist in the parquet file.
+      val isRowIdOrRowCommitVersion = materializedRowIdColumnNameOpt.contains(field.name) ||
+        materializedRowCommitVersionColumnNameOpt.contains(field.name)
+
+      if (inTableFieldOpt.isEmpty && !isRowIdOrRowCommitVersion) {
+        return true
+      }
+      field
+    }
+    false
+  }
+
   /**
    * Apply a filter on the list of AddFile to only keep the files that have physical parquet schema
    * that satisfies the given filter function.

spark/src/test/scala/org/apache/spark/sql/delta/optimize/DeltaReorgSuite.scala

@@ -16,14 +16,20 @@
 
 package org.apache.spark.sql.delta.optimize
 
-import org.apache.spark.sql.delta.DeletionVectorsTestUtils
+import org.apache.spark.sql.delta.{DeletionVectorsTestUtils, DeltaColumnMapping, DeltaLog}
+import org.apache.spark.sql.delta.actions.AddFile
+import org.apache.spark.sql.delta.commands.VacuumCommand.generateCandidateFileMap
 import org.apache.spark.sql.delta.sources.DeltaSQLConf
 import org.apache.spark.sql.delta.test.DeltaSQLCommandTest
+import org.apache.spark.sql.delta.util.DeltaFileOperations
 import io.delta.tables.DeltaTable
+import org.apache.hadoop.fs.{FileStatus, Path}
+import org.apache.parquet.hadoop.Footer
 
 import org.apache.spark.sql.QueryTest
 import org.apache.spark.sql.functions.col
 import org.apache.spark.sql.test.SharedSparkSession
+import org.apache.spark.util.SerializableConfiguration
 
 class DeltaReorgSuite extends QueryTest
   with SharedSparkSession
@@ -150,4 +156,147 @@ class DeltaReorgSuite extends QueryTest
     // Because each deleted file has a DV associated it which gets rewritten as part of PURGE
     assert(opMetrics("numDeletionVectorsRemoved").toLong === numFilesRemoved)
   }
+
+  /**
+   * Get all parquet footers for the input `files`, used only for testing.
+   *
+   * @param files the sequence of `AddFile` used to read the parquet footers
+   *              by the data file path in each `AddFile`.
+   * @param log the delta log used to get the configuration and data path.
+   * @return the sequence of the corresponding parquet footers, corresponds to
+   *         the sequence of `AddFile`.
+   */
+  private def getParquetFooters(
+      files: Seq[AddFile],
+      log: DeltaLog): Seq[Footer] = {
+    val serializedConf = new SerializableConfiguration(log.newDeltaHadoopConf())
+    val dataPath = new Path(log.dataPath.toString)
+    val nameToAddFileMap = generateCandidateFileMap(dataPath, files)
+    val fileStatuses = nameToAddFileMap.map { case (absPath, addFile) =>
+      new FileStatus(
+        /* length */ addFile.size,
+        /* isDir */ false,
+        /* blockReplication */ 0,
+        /* blockSize */ 1,
+        /* modificationTime */ addFile.modificationTime,
+        new Path(absPath)
+      )
+    }
+    DeltaFileOperations.readParquetFootersInParallel(
+      serializedConf.value,
+      fileStatuses.toList,
+      ignoreCorruptFiles = false
+    )
+  }
+
+  test("Purge dropped columns of a table without DV") {
+    val targetDf = spark.range(0, 100, 1, numPartitions = 5)
+      .withColumn("id_dropped", col("id") % 4)
+      .toDF()
+    withTempDeltaTable(targetDf) { (_, log) =>
+      val path = log.dataPath.toString
+
+      val (addFiles1, _) = getFileActionsInLastVersion(log)
+      assert(addFiles1.size === 5)
+      val footers1 = getParquetFooters(addFiles1, log)
+      footers1.foreach { footer =>
+        val fields = footer.getParquetMetadata.getFileMetaData.getSchema.getFields
+        assert(fields.size == 2)
+        assert(fields.toArray.map { _.toString }.contains("optional int64 id_dropped"))
+      }
+
+      // enable column-mapping first
+      sql(
+        s"""
+           | ALTER TABLE delta.`$path`
+           | SET TBLPROPERTIES (
+           |   'delta.columnMapping.mode' = 'name'
+           | )
+           |""".stripMargin
+      )
+      // drop the extra column by alter table and run REORG PURGE
+      sql(
+        s"""
+           | ALTER TABLE delta.`$path`
+           | DROP COLUMN id_dropped
+           |""".stripMargin
+      )
+      executePurge(path)
+
+      val (addFiles2, _) = getFileActionsInLastVersion(log)
+      assert(addFiles2.size === 1)
+      val footers2 = getParquetFooters(addFiles2, log)
+      footers2.foreach { footer =>
+        val fields = footer.getParquetMetadata.getFileMetaData.getSchema.getFields
+        assert(fields.size == 1)
+        assert(!fields.toArray.map { _.toString }.contains("optional int64 id_dropped"))
+      }
+    }
+  }
+
+  test("Columns being renamed should not be purged") {
+    val targetDf = spark.range(0, 100, 1, numPartitions = 5)
+      .withColumn("id_before_rename", col("id") % 4)
+      .withColumn("id_dropped", col("id") % 5)
+      .toDF()
+    withTempDeltaTable(targetDf) { (_, log) =>
+      val path = log.dataPath.toString
+
+      val (addFiles1, _) = getFileActionsInLastVersion(log)
+      assert(addFiles1.size === 5)
+      val footers1 = getParquetFooters(addFiles1, log)
+      footers1.foreach { footer =>
+        val fields = footer.getParquetMetadata.getFileMetaData.getSchema.getFields
+        assert(fields.size == 3)
+        assert(fields.toArray.map { _.toString }.contains("optional int64 id_dropped"))
+        assert(fields.toArray.map { _.toString }.contains("optional int64 id_before_rename"))
+      }
+
+      // enable column-mapping first
+      sql(
+        s"""
+           | ALTER TABLE delta.`$path`
+           | SET TBLPROPERTIES (
+           |   'delta.columnMapping.mode' = 'name'
+           | )
+           |""".stripMargin
+      )
+      // drop `id_dropped` and rename `id_before_rename` via alter table and run REORG PURGE,
+      // this should remove `id_dropped` but keep `id_after_rename` in the parquet files.
+      sql(
+        s"""
+           | ALTER TABLE delta.`$path`
+           | DROP COLUMN id_dropped
+           |""".stripMargin
+      )
+      sql(
+        s"""
+           | ALTER TABLE delta.`$path`
+           | RENAME COLUMN id_before_rename TO id_after_rename
+           |""".stripMargin
+      )
+      executePurge(path)
+
+      val tableSchema = log.update().schema
+      val tablePhysicalSchema = DeltaColumnMapping.renameColumns(tableSchema)
+      val beforeRenameColStr = "StructField(id_before_rename,LongType,true)"
+      val afterRenameColStr = "StructField(id_after_rename,LongType,true)"
+      assert(tableSchema.fields.length == 2 &&
+        tableSchema.map { _.toString }.contains(afterRenameColStr))
+      assert(tablePhysicalSchema.fields.length == 2 &&
+        tablePhysicalSchema.map { _.toString }.contains(beforeRenameColStr))
+
+      val (addFiles2, _) = getFileActionsInLastVersion(log)
+      assert(addFiles2.size === 1)
+      val footers2 = getParquetFooters(addFiles2, log)
+      footers2.foreach { footer =>
+        val fields = footer.getParquetMetadata.getFileMetaData.getSchema.getFields
+        assert(fields.size == 2)
+        assert(!fields.toArray.map { _.toString }.contains("optional int64 id_dropped = 3"))
+        // do note that the actual name for the column will not be
+        // changed in parquet file level
+        assert(fields.toArray.map { _.toString }.contains("optional int64 id_before_rename = 2"))
+      }
+    }
+  }
 }