apache/hadoop · error · NotInMountpointException

getServerDefaults on empty path is invalid

Error message

getServerDefaults on empty path is invalid

What it means

ViewFileSystem (viewfs) is a client-side mount table that forwards each operation to the backing FileSystem of the mount point that owns the path. It has no single underlying server, so the no-argument getServerDefaults() (and the sibling no-arg getDefaultBlockSize()/getDefaultReplication()) cannot return meaningful values and always throws NotInMountpointException, an UnsupportedOperationException. Server defaults only exist per target cluster, so the path-aware overloads must be used instead.

Source

Thrown at hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/viewfs/ViewFileSystem.java:963

        throw new RuntimeException(errMsg, ex);
      }
    }
    return fsMap;
  }

  @Override
  public long getDefaultBlockSize() {
    throw new NotInMountpointException("getDefaultBlockSize");
  }

  @Override
  public short getDefaultReplication() {
    throw new NotInMountpointException("getDefaultReplication");
  }

  @Override
  public FsServerDefaults getServerDefaults() throws IOException {
    throw new NotInMountpointException("getServerDefaults");
  }

  @Override
  public long getDefaultBlockSize(Path f) {
    try {
      InodeTree.ResolveResult<FileSystem> res =
        fsState.resolve(getUriPath(f), true);
      return res.targetFileSystem.getDefaultBlockSize(res.remainingPath);
    } catch (FileNotFoundException e) {
      throw new NotInMountpointException(f, "getDefaultBlockSize");
    } catch (IOException e) {
      throw new RuntimeException("Not able to initialize fs in "
          + " getDefaultBlockSize for path " + f + " with exception", e);
    }
  }

  @Override
  public short getDefaultReplication(Path f) {

View on GitHub (pinned to 2add963021)

Solutions

1. Switch to the path-aware overloads on a path that resolves into a mount point: fs.getServerDefaults(new Path("/mounted/dir")), fs.getDefaultBlockSize(path), fs.getDefaultReplication(path)
2. Add a root mount so every path resolves: set fs.viewfs.mounttable.default.linkMergeSlash=hdfs://nameservice1/ in core-site.xml
3. If only default values are needed, read them from the target cluster config (dfs.blocksize, dfs.replication) or from getServerDefaults() of the target HDFS FileSystem instead of the viewfs handle
4. As a last resort catch NotInMountpointException (or its parent UnsupportedOperationException) and fall back to known cluster defaults

Example fix

// before
FileSystem fs = FileSystem.get(new URI("viewfs://cluster"), conf);
long bs = fs.getDefaultBlockSize(); // throws NotInMountpointException

// after
FileSystem fs = FileSystem.get(new URI("viewfs://cluster"), conf);
long bs = fs.getDefaultBlockSize(new Path("/data/somefile")); // path inside a mount point
// or, for cluster-wide defaults, ask the target FS directly:
// FsServerDefaults d = fs.getServerDefaults(new Path("/data"));

// No-arg server-default calls on viewfs are invalid by design; there is nothing to validate.
// Route the call through a mounted path instead:
Path mounted = new Path("viewfs://cluster/data");
FileSystem fs = mounted.getFileSystem(conf);
FsServerDefaults d = fs.getServerDefaults(mounted);

import org.apache.hadoop.fs.viewfs.NotInMountpointException;
try {
  long bs = fs.getDefaultBlockSize();
} catch (NotInMountpointException | UnsupportedOperationException e) {
  // viewfs has no cluster-wide defaults; use target-cluster config
  long bs = conf.getLongBytes("dfs.blocksize", 128L * 1024 * 1024);
}

When it happens

Trigger: Calling fs.getServerDefaults(), fs.getDefaultBlockSize(), or fs.getDefaultReplication() on a FileSystem whose URI is viewfs://<cluster> (e.g. new Path("viewfs://cluster/").getFileSystem(conf)). Typical producers are framework code (distcp, job submission, Hive/Spark writers) that derives default block size or replication from the default FileSystem when fs.defaultFS points at viewfs.

Common situations: fs.defaultFS switched to viewfs://cluster during an HDFS federation/router migration while tools still assume a plain HDFS handle; mount table without linkMergeSlash so even the root path resolves nowhere; old client code using the deprecated no-arg methods.

Related errors

• getServerDefaults on path `<f>' is not within a mount point
• ViewFs: Non absolute mount name in config:{src}
• Path {nextInode.fullPath} already exists as link
• Path {strB} already exists as dir; cannot create link here
• Unexpected mount table link entry '{key}'. Use linkMergeSlas

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/671e582a11ee24ee. Report an issue: GitHub.