apache/hadoop · error · IOException

Har: create not allowed.

Error message

Har: create not allowed.

What it means

HarFileSystem is a read-only view: a .har is an immutable package (index + part files) produced by the `hadoop archive` tool, and every mutating FileSystem API is overridden to throw. FileSystem.create(...) throws 'Har: create not allowed' — new files cannot be written into an archive.

Source

Thrown at hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/HarFileSystem.java:718

  public FSDataInputStream open(PathHandle fd, int bufferSize)
      throws IOException {
    throw new UnsupportedOperationException();
  }

  /**
   * Used for delegation token related functionality. Must delegate to
   * underlying file system.
   */
  @Override
  public FileSystem[] getChildFileSystems() {
    return new FileSystem[]{fs};
  }

  @Override
  public FSDataOutputStream create(Path f, FsPermission permission,
      boolean overwrite, int bufferSize, short replication, long blockSize,
      Progressable progress) throws IOException {
    throw new IOException("Har: create not allowed.");
  }

  @Override
  public FSDataOutputStream createNonRecursive(Path f, boolean overwrite,
      int bufferSize, short replication, long blockSize, Progressable progress)
      throws IOException {
    throw new IOException("Har: create not allowed.");
  }

  @Override
  public FSDataOutputStream append(Path f, int bufferSize, Progressable progress) throws IOException {
    throw new IOException("Har: append not allowed.");
  }

  @Override
  public void close() throws IOException {
    super.close();
    if (fs != null) {

View on GitHub (pinned to 2add963021)

Solutions

1. Write to the underlying filesystem (hdfs://...) instead, then package with `hadoop archive -archiveName out.har -p <dir> <parentDir>`
2. Fix the configuration: output, staging, and temp directories must never use the har scheme
3. To expose new data as har, write to HDFS, build a fresh .har, and publish its har:// URI

Example fix

// before
FSDataOutputStream out = fs.create(new Path("har://hdfs-nn:8020/a/data.har/newFile"));

// after
FSDataOutputStream out = hdfs.create(new Path("hdfs://nn:8020/a/data/newFile"));
// then: hadoop archive -archiveName data.har -p /a/data /a

static void requireWritablePath(Path p) {
  URI u = p.toUri();
  if ("har".equals(u.getScheme())) {
    throw new UnsupportedOperationException("har:// is read-only; write to the underlying filesystem instead: " + p);
  }
}

static boolean isReadOnlyHarFs(FileSystem fs) {
  return fs instanceof HarFileSystem;
}

When it happens

Trigger: fs.create(new Path("har://...")) directly, or any framework opening an output stream on an output path under har://: MapReduce/Spark output committers, Hive/Spark INSERT writes, DistCp with a har:// destination.

Common situations: Setting mapreduce.output.fileoutputformat.outputdir, a Spark write path, or a Hive staging location to har://; generic copy utilities that write to whatever filesystem the destination URI resolves to.

Related errors

• Har: append not allowed.
• Har: setReplication not allowed
• Har: rename not allowed
• Har: truncate not allowed
• Har: delete not allowed

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/157a626a03e23bc8. Report an issue: GitHub.