HDFS-15478: When Empty mount points, we are assigning fallback link to self. But it should not use full URI for target fs. (#2160). Contributed by Uma Maheswara Rao G.

Author: umamaheswararao

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/viewfs/ViewFileSystem.java

@@ -294,7 +294,7 @@ public void initialize(final URI theUri, final Configuration conf)
       myUri = new URI(getScheme(), authority, "/", null, null);
       boolean initingUriAsFallbackOnNoMounts =
           !FsConstants.VIEWFS_TYPE.equals(getType());
-      fsState = new InodeTree<FileSystem>(conf, tableName, theUri,
+      fsState = new InodeTree<FileSystem>(conf, tableName, myUri,
           initingUriAsFallbackOnNoMounts) {
         @Override
         protected FileSystem getTargetFileSystem(final URI uri)

hadoop-common-project/hadoop-common/src/test/java/org/apache/hadoop/fs/viewfs/TestViewFsOverloadSchemeListStatus.java

@@ -127,19 +127,30 @@ public void testListStatusACL() throws IOException, URISyntaxException {
 
   /**
    * Tests that ViewFSOverloadScheme should consider initialized fs as fallback
-   * if there are no mount links configured.
+   * if there are no mount links configured. It should add fallback with the
+   * chrootedFS at it's uri's root.
    */
   @Test(timeout = 30000)
   public void testViewFSOverloadSchemeWithoutAnyMountLinks() throws Exception {
-    try (FileSystem fs = FileSystem.get(TEST_DIR.toPath().toUri(), conf)) {
+    Path initUri = new Path(TEST_DIR.toURI().toString(), "init");
+    try (FileSystem fs = FileSystem.get(initUri.toUri(), conf)) {
       ViewFileSystemOverloadScheme vfs = (ViewFileSystemOverloadScheme) fs;
       assertEquals(0, vfs.getMountPoints().length);
-      Path testFallBack = new Path("test", FILE_NAME);
-      assertTrue(vfs.mkdirs(testFallBack));
-      FileStatus[] status = vfs.listStatus(testFallBack.getParent());
-      assertEquals(FILE_NAME, status[0].getPath().getName());
-      assertEquals(testFallBack.getName(),
-          vfs.getFileLinkStatus(testFallBack).getPath().getName());
+      Path testOnFallbackPath = new Path(TEST_DIR.toURI().toString(), "test");
+      assertTrue(vfs.mkdirs(testOnFallbackPath));
+      FileStatus[] status = vfs.listStatus(testOnFallbackPath.getParent());
+      assertEquals(Path.getPathWithoutSchemeAndAuthority(testOnFallbackPath),
+          Path.getPathWithoutSchemeAndAuthority(status[0].getPath()));
+      //Check directly on localFS. The fallBackFs(localFS) should be chrooted
+      //at it's root. So, after
+      FileSystem lfs = vfs.getRawFileSystem(testOnFallbackPath, conf);
+      FileStatus[] statusOnLocalFS =
+          lfs.listStatus(testOnFallbackPath.getParent());
+      assertEquals(testOnFallbackPath.getName(),
+          statusOnLocalFS[0].getPath().getName());
+      //initUri should not have exist in lfs, as it would have chrooted on it's
+      // root only.
+      assertFalse(lfs.exists(initUri));
     }
   }
 

hadoop-hdfs-project/hadoop-hdfs/src/site/markdown/ViewFsOverloadScheme.md

@@ -34,6 +34,8 @@ If a user wants to continue use the same fs.defaultFS and wants to have more mou
 Example if fs.defaultFS is `hdfs://mycluster`, then the mount link configuration key name should be like in the following format `fs.viewfs.mounttable.*mycluster*.link.<mountLinkPath>`.
 Even if the initialized fs uri has hostname:port, it will simply ignore the port number and only consider the hostname as the mount table name.  We will discuss more example configurations in following sections.
 If there are no mount links configured with the initializing uri's hostname as the mount table name, then it will automatically consider the current uri as fallback(`fs.viewfs.mounttable.*mycluster*.linkFallback`) target fs uri.
+If the initialized uri contains path part, it will consider only scheme and authority part, but not the path part. Example, if the initialized uri contains `hdfs://mycluster/data`, it will consider only `hdfs://mycluster` as fallback target fs uri.
+The path part `data` will be ignored.
 
 Another important improvement with the ViewFileSystemOverloadScheme is, administrators need not copy the `mount-table.xml` configuration file to 1000s of client nodes. Instead, they can keep the mount-table configuration file in a Hadoop compatible file system. So, keeping the configuration file in a central place makes administrators life easier as they can update mount-table in single place.