revert unnecessary changes in DataSourceRDD.scala

Author: peter-toth

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/DataSourceRDD.scala

@@ -135,11 +135,6 @@ class DataSourceRDD(
   }
 }
 
-/**
- * An iterator that wraps a PartitionReader and periodically updates custom metrics.
- * Metrics are updated every `CustomMetrics.NUM_ROWS_PER_UPDATE` rows to balance
- * between metric accuracy and performance overhead.
- */
 private class PartitionIterator[T](
     reader: PartitionReader[T],
     customMetrics: Map[String, SQLMetric]) extends Iterator[T] {
@@ -169,12 +164,7 @@ private class PartitionIterator[T](
   }
 }
 
-/**
- * Handles updates to Spark's built-in input metrics (records read and bytes read).
- * Bytes read are updated periodically based on Spark's update interval to avoid
- * excessive overhead from filesystem calls.
- */
-private[spark] class MetricsHandler extends Logging with Serializable {
+private class MetricsHandler extends Logging with Serializable {
   private val inputMetrics = TaskContext.get().taskMetrics().inputMetrics
   private val startingBytesRead = inputMetrics.bytesRead
   private val getBytesRead = SparkHadoopUtil.get.getFSBytesReadOnThreadCallback()
@@ -189,15 +179,7 @@ private[spark] class MetricsHandler extends Logging with Serializable {
   }
 }
 
-/**
- * Base class for iterators that track metrics and save the final custom metrics values
- * when iteration completes. The saved metrics are stored in a ThreadLocal so the task
- * completion listener can access them for final reporting.
- *
- * When `hasNext` returns false (iteration complete), this iterator saves the current
- * custom metrics to the ThreadLocal before forcing a final metrics update.
- */
-private[spark] abstract class MetricsIterator[I](iter: Iterator[I]) extends Iterator[I] {
+private abstract class MetricsIterator[I](iter: Iterator[I]) extends Iterator[I] {
   protected val metricsHandler = new MetricsHandler
 
   override def hasNext: Boolean = {