[SPARK-24356][CORE] Duplicate strings in File.path managed by FileSegmentManagedBuffer

This patch eliminates duplicate strings that come from the 'path' field of
java.io.File objects created by FileSegmentManagedBuffer. That is, we want
to avoid the situation when multiple File instances for the same pathname
"foo/bar" are created, each with a separate copy of the "foo/bar" String
instance. In some scenarios such duplicate strings may waste a lot of memory
(~ 10% of the heap). To avoid that, we intern the pathname with
String.intern(), and before that we make sure that it's in a normalized
form (contains no "//", "///" etc.) Otherwise, the code in java.io.File
would normalize it later, creating a new "foo/bar" String copy.
Unfortunately, the normalization code that java.io.File uses internally
is in the package-private class java.io.FileSystem, so we cannot call it
here directly.

## What changes were proposed in this pull request?

Added code to ExternalShuffleBlockResolver.getFile(), that normalizes and then interns the pathname string before passing it to the File() constructor.

## How was this patch tested?

Added unit test

Author: Misha Dmitriev <[email protected]>

Closes #21456 from countmdm/misha/spark-24356.

Author: misha-cloudera

common/network-shuffle/src/main/java/org/apache/spark/network/shuffle/ExternalShuffleBlockResolver.java

@@ -24,6 +24,8 @@
 import java.util.concurrent.ExecutionException;
 import java.util.concurrent.Executor;
 import java.util.concurrent.Executors;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
 
 import com.fasterxml.jackson.annotation.JsonCreator;
 import com.fasterxml.jackson.annotation.JsonProperty;
@@ -59,13 +61,16 @@ public class ExternalShuffleBlockResolver {
   private static final Logger logger = LoggerFactory.getLogger(ExternalShuffleBlockResolver.class);
 
   private static final ObjectMapper mapper = new ObjectMapper();
+
   /**
    * This a common prefix to the key for each app registration we stick in leveldb, so they
    * are easy to find, since leveldb lets you search based on prefix.
    */
   private static final String APP_KEY_PREFIX = "AppExecShuffleInfo";
   private static final StoreVersion CURRENT_VERSION = new StoreVersion(1, 0);
 
+  private static final Pattern MULTIPLE_SEPARATORS = Pattern.compile(File.separator + "{2,}");
+
   // Map containing all registered executors' metadata.
   @VisibleForTesting
   final ConcurrentMap<AppExecId, ExecutorShuffleInfo> executors;
@@ -302,7 +307,8 @@ static File getFile(String[] localDirs, int subDirsPerLocalDir, String filename)
     int hash = JavaUtils.nonNegativeHash(filename);
     String localDir = localDirs[hash % localDirs.length];
     int subDirId = (hash / localDirs.length) % subDirsPerLocalDir;
-    return new File(new File(localDir, String.format("%02x", subDirId)), filename);
+    return new File(createNormalizedInternedPathname(
+        localDir, String.format("%02x", subDirId), filename));
   }
 
   void close() {
@@ -315,6 +321,28 @@ void close() {
     }
   }
 
+  /**
+   * This method is needed to avoid the situation when multiple File instances for the
+   * same pathname "foo/bar" are created, each with a separate copy of the "foo/bar" String.
+   * According to measurements, in some scenarios such duplicate strings may waste a lot
+   * of memory (~ 10% of the heap). To avoid that, we intern the pathname, and before that
+   * we make sure that it's in a normalized form (contains no "//", "///" etc.) Otherwise,
+   * the internal code in java.io.File would normalize it later, creating a new "foo/bar"
+   * String copy. Unfortunately, we cannot just reuse the normalization code that java.io.File
+   * uses, since it is in the package-private class java.io.FileSystem.
+   */
+  @VisibleForTesting
+  static String createNormalizedInternedPathname(String dir1, String dir2, String fname) {
+    String pathname = dir1 + File.separator + dir2 + File.separator + fname;
+    Matcher m = MULTIPLE_SEPARATORS.matcher(pathname);
+    pathname = m.replaceAll("/");
+    // A single trailing slash needs to be taken care of separately
+    if (pathname.length() > 1 && pathname.endsWith("/")) {
+      pathname = pathname.substring(0, pathname.length() - 1);
+    }
+    return pathname.intern();
+  }
+
   /** Simply encodes an executor's full ID, which is appId + execId. */
   public static class AppExecId {
     public final String appId;

common/network-shuffle/src/test/java/org/apache/spark/network/shuffle/ExternalShuffleBlockResolverSuite.java

@@ -17,6 +17,7 @@
 
 package org.apache.spark.network.shuffle;
 
+import java.io.File;
 import java.io.IOException;
 import java.io.InputStream;
 import java.io.InputStreamReader;
@@ -135,4 +136,23 @@ public void jsonSerializationOfExecutorRegistration() throws IOException {
       "\"subDirsPerLocalDir\": 7, \"shuffleManager\": " + "\"" + SORT_MANAGER + "\"}";
     assertEquals(shuffleInfo, mapper.readValue(legacyShuffleJson, ExecutorShuffleInfo.class));
   }
+
+  @Test
+  public void testNormalizeAndInternPathname() {
+    assertPathsMatch("/foo", "bar", "baz", "/foo/bar/baz");
+    assertPathsMatch("//foo/", "bar/", "//baz", "/foo/bar/baz");
+    assertPathsMatch("foo", "bar", "baz///", "foo/bar/baz");
+    assertPathsMatch("/foo/", "/bar//", "/baz", "/foo/bar/baz");
+    assertPathsMatch("/", "", "", "/");
+    assertPathsMatch("/", "/", "/", "/");
+  }
+
+  private void assertPathsMatch(String p1, String p2, String p3, String expectedPathname) {
+    String normPathname =
+        ExternalShuffleBlockResolver.createNormalizedInternedPathname(p1, p2, p3);
+    assertEquals(expectedPathname, normPathname);
+    File file = new File(normPathname);
+    String returnedPath = file.getPath();
+    assertTrue(normPathname == returnedPath);
+  }
 }