Remove prefetch feature, V2 scan dead code, and deprecate BatchReader

BatchReader is annotated @IcebergApi and kept for Iceberg compatibility,
but Comet's own production code no longer uses it. The prefetch feature
was entirely built on BatchReader and is dead code now that the
native_iceberg_compat path uses NativeBatchReader. V2 Parquet scan
acceleration (CometParquetScan) is also no longer active.

This commit:
- Marks BatchReader as @deprecated (since 0.14.0)
- Removes all prefetch internals from BatchReader (fields, methods, inner class)
- Removes COMET_SCAN_PREFETCH_ENABLED and COMET_SCAN_PREFETCH_THREAD_NUM configs
- Removes CometPrefetchThreadPool
- Deletes CometParquetPartitionReaderFactory and CometParquetScan
- Simplifies CometScanExec.prepareRDD to always use newFileScanRDD
- Cleans up EliminateRedundantTransitions V2 dead code path
- Removes prefetch tests and BatchReader benchmark case
- Cleans up CometParquetScan references in tests

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: andygrove

common/src/main/java/org/apache/comet/parquet/BatchReader.java

@@ -24,10 +24,6 @@
 import java.net.URI;
 import java.net.URISyntaxException;
 import java.util.*;
-import java.util.concurrent.Callable;
-import java.util.concurrent.ExecutorService;
-import java.util.concurrent.Future;
-import java.util.concurrent.LinkedBlockingQueue;
 
 import scala.Option;
 
@@ -36,9 +32,7 @@
 
 import org.apache.arrow.memory.BufferAllocator;
 import org.apache.arrow.memory.RootAllocator;
-import org.apache.commons.lang3.tuple.Pair;
 import org.apache.hadoop.conf.Configuration;
-import org.apache.hadoop.fs.FileSystem;
 import org.apache.hadoop.fs.Path;
 import org.apache.hadoop.mapreduce.InputSplit;
 import org.apache.hadoop.mapreduce.RecordReader;
@@ -87,7 +81,10 @@
  *     reader.close();
  *   }
  * </pre>
+ *
+ * @deprecated since 0.14.0. This class is kept for Iceberg compatibility only.
  */
+@Deprecated
 @IcebergApi
 public class BatchReader extends RecordReader<Void, ColumnarBatch> implements Closeable {
   private static final Logger LOG = LoggerFactory.getLogger(FileReader.class);
@@ -110,8 +107,6 @@ public class BatchReader extends RecordReader<Void, ColumnarBatch> implements Cl
   protected AbstractColumnReader[] columnReaders;
   private CometSchemaImporter importer;
   protected ColumnarBatch currentBatch;
-  private Future<Option<Throwable>> prefetchTask;
-  private LinkedBlockingQueue<Pair<PageReadStore, Long>> prefetchQueue;
   private FileReader fileReader;
   private boolean[] missingColumns;
   protected boolean isInitialized;
@@ -363,26 +358,7 @@ public void init() throws URISyntaxException, IOException {
       }
     }
 
-    // Pre-fetching
-    boolean preFetchEnabled =
-        conf.getBoolean(
-            CometConf.COMET_SCAN_PREFETCH_ENABLED().key(),
-            (boolean) CometConf.COMET_SCAN_PREFETCH_ENABLED().defaultValue().get());
-
-    if (preFetchEnabled) {
-      LOG.info("Prefetch enabled for BatchReader.");
-      this.prefetchQueue = new LinkedBlockingQueue<>();
-    }
-
     isInitialized = true;
-    synchronized (this) {
-      // if prefetch is enabled, `init()` is called in separate thread. When
-      // `BatchReader.nextBatch()` is called asynchronously, it is possibly that
-      // `init()` is not called or finished. We need to hold on `nextBatch` until
-      // initialization of `BatchReader` is done. Once we are close to finish
-      // initialization, we notify the waiting thread of `nextBatch` to continue.
-      notifyAll();
-    }
   }
 
   /**
@@ -436,51 +412,13 @@ public ColumnarBatch currentBatch() {
     return currentBatch;
   }
 
-  // Only for testing
-  public Future<Option<Throwable>> getPrefetchTask() {
-    return this.prefetchTask;
-  }
-
-  // Only for testing
-  public LinkedBlockingQueue<Pair<PageReadStore, Long>> getPrefetchQueue() {
-    return this.prefetchQueue;
-  }
-
   /**
    * Loads the next batch of rows.
    *
    * @return true if there are no more rows to read, false otherwise.
    */
   public boolean nextBatch() throws IOException {
-    if (this.prefetchTask == null) {
-      Preconditions.checkState(isInitialized, "init() should be called first!");
-    } else {
-      // If prefetch is enabled, this reader will be initialized asynchronously from a
-      // different thread. Wait until it is initialized
-      while (!isInitialized) {
-        synchronized (this) {
-          try {
-            // Wait until initialization of current `BatchReader` is finished (i.e., `init()`),
-            // is done. It is possibly that `init()` is done after entering this while loop,
-            // so a short timeout is given.
-            wait(100);
-
-            // Checks if prefetch task is finished. If so, tries to get exception if any.
-            if (prefetchTask.isDone()) {
-              Option<Throwable> exception = prefetchTask.get();
-              if (exception.isDefined()) {
-                throw exception.get();
-              }
-            }
-          } catch (RuntimeException e) {
-            // Spark will check certain exception e.g. `SchemaColumnConvertNotSupportedException`.
-            throw e;
-          } catch (Throwable e) {
-            throw new IOException(e);
-          }
-        }
-      }
-    }
+    Preconditions.checkState(isInitialized, "init() should be called first!");
 
     if (rowsRead >= totalRowCount) return false;
     boolean hasMore;
@@ -547,7 +485,6 @@ public void close() throws IOException {
     }
   }
 
-  @SuppressWarnings("deprecation")
   private boolean loadNextRowGroupIfNecessary() throws Throwable {
     // More rows can be read from loaded row group. No need to load next one.
     if (rowsRead != totalRowsLoaded) return true;
@@ -556,21 +493,7 @@ private boolean loadNextRowGroupIfNecessary() throws Throwable {
     SQLMetric numRowGroupsMetric = metrics.get("ParquetRowGroups");
     long startNs = System.nanoTime();
 
-    PageReadStore rowGroupReader = null;
-    if (prefetchTask != null && prefetchQueue != null) {
-      // Wait for pre-fetch task to finish.
-      Pair<PageReadStore, Long> rowGroupReaderPair = prefetchQueue.take();
-      rowGroupReader = rowGroupReaderPair.getLeft();
-
-      // Update incremental byte read metric. Because this metric in Spark is maintained
-      // by thread local variable, we need to manually update it.
-      // TODO: We may expose metrics from `FileReader` and get from it directly.
-      long incBytesRead = rowGroupReaderPair.getRight();
-      FileSystem.getAllStatistics().stream()
-          .forEach(statistic -> statistic.incrementBytesRead(incBytesRead));
-    } else {
-      rowGroupReader = fileReader.readNextRowGroup();
-    }
+    PageReadStore rowGroupReader = fileReader.readNextRowGroup();
 
     if (rowGroupTimeMetric != null) {
       rowGroupTimeMetric.add(System.nanoTime() - startNs);
@@ -608,48 +531,4 @@ private boolean loadNextRowGroupIfNecessary() throws Throwable {
     totalRowsLoaded += rowGroupReader.getRowCount();
     return true;
   }
-
-  // Submits a prefetch task for this reader.
-  public void submitPrefetchTask(ExecutorService threadPool) {
-    this.prefetchTask = threadPool.submit(new PrefetchTask());
-  }
-
-  // A task for prefetching parquet row groups.
-  private class PrefetchTask implements Callable<Option<Throwable>> {
-    private long getBytesRead() {
-      return FileSystem.getAllStatistics().stream()
-          .mapToLong(s -> s.getThreadStatistics().getBytesRead())
-          .sum();
-    }
-
-    @Override
-    public Option<Throwable> call() throws Exception {
-      // Gets the bytes read so far.
-      long baseline = getBytesRead();
-
-      try {
-        init();
-
-        while (true) {
-          PageReadStore rowGroupReader = fileReader.readNextRowGroup();
-
-          if (rowGroupReader == null) {
-            // Reaches the end of row groups.
-            return Option.empty();
-          } else {
-            long incBytesRead = getBytesRead() - baseline;
-
-            prefetchQueue.add(Pair.of(rowGroupReader, incBytesRead));
-          }
-        }
-      } catch (Throwable e) {
-        // Returns exception thrown from the reader. The reader will re-throw it.
-        return Option.apply(e);
-      } finally {
-        if (fileReader != null) {
-          fileReader.closeStream();
-        }
-      }
-    }
-  }
 }

common/src/main/java/org/apache/comet/parquet/IcebergCometBatchReader.java

@@ -24,9 +24,11 @@
 import org.apache.spark.sql.types.StructType;
 import org.apache.spark.sql.vectorized.ColumnarBatch;
 
+import org.apache.comet.IcebergApi;
 import org.apache.comet.vector.CometVector;
 
 /** This class is a public interface used by Apache Iceberg to read batches using Comet */
+@IcebergApi
 public class IcebergCometBatchReader extends BatchReader {
   public IcebergCometBatchReader(int numColumns, StructType schema) {
     this.columnReaders = new AbstractColumnReader[numColumns];

common/src/main/scala/org/apache/comet/CometConf.scala

@@ -651,23 +651,6 @@ object CometConf extends ShimCometConf {
       .doubleConf
       .createWithDefault(1.0)
 
-  val COMET_SCAN_PREFETCH_ENABLED: ConfigEntry[Boolean] =
-    conf("spark.comet.scan.preFetch.enabled")
-      .category(CATEGORY_SCAN)
-      .doc("Whether to enable pre-fetching feature of CometScan.")
-      .booleanConf
-      .createWithDefault(false)
-
-  val COMET_SCAN_PREFETCH_THREAD_NUM: ConfigEntry[Int] =
-    conf("spark.comet.scan.preFetch.threadNum")
-      .category(CATEGORY_SCAN)
-      .doc(
-        "The number of threads running pre-fetching for CometScan. Effective if " +
-          s"${COMET_SCAN_PREFETCH_ENABLED.key} is enabled. Note that more " +
-          "pre-fetching threads means more memory requirement to store pre-fetched row groups.")
-      .intConf
-      .createWithDefault(2)
-
   val COMET_NATIVE_LOAD_REQUIRED: ConfigEntry[Boolean] = conf("spark.comet.nativeLoadRequired")
     .category(CATEGORY_EXEC)
     .doc(

common/src/main/scala/org/apache/comet/parquet/CometReaderThreadPool.scala

@@ -54,11 +54,6 @@ abstract class CometReaderThreadPool {
 
 }
 
-// A thread pool used for pre-fetching files.
-object CometPrefetchThreadPool extends CometReaderThreadPool {
-  override def threadNamePrefix: String = "prefetch_thread"
-}
-
 // Thread pool used by the Parquet parallel reader
 object CometFileReaderThreadPool extends CometReaderThreadPool {
   override def threadNamePrefix: String = "file_reader_thread"