What causes "Invalid path: {location}: directory does not exist" in apache/hadoop?

Create the directory chain on the DN host first: mkdir -p /mnt/disk2/dn && chown hdfs:hdfs, then resubmit

apache/hadoop · error · IOException

Invalid path: {location}: directory does not exist

Error message

Invalid path: {location}: directory does not exist

What it means

checkSameDiskTieringMount runs during refreshVolumes when dfs.datanode.allow.same.disk.tiering is true and the new location's storage type participates in same-disk tiering. To find the disk mount point it climbs parent-by-parent from the location path to the first existing directory; if getParentFile() returns null before any ancestor exists (relative path, a URI java.io.File cannot walk, or an entirely missing directory chain), it throws IOException("Invalid path: <location>: directory does not exist").

Source

Thrown at hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DataNode.java:1259

   *
   * TODO: We can add feature to
   *   allow refreshing volume with capacity ratio,
   *   and solve the case of replacing volume on same mount.
   */
  private void validateVolumesWithSameDiskTiering(ChangedVolumes
      changedVolumes) throws IOException {
    if (dnConf.getConf().getBoolean(DFS_DATANODE_ALLOW_SAME_DISK_TIERING,
        DFS_DATANODE_ALLOW_SAME_DISK_TIERING_DEFAULT)
        && data.getMountVolumeMap() != null) {
      // Check if mount already exist.
      for (StorageLocation location : changedVolumes.newLocations) {
        if (StorageType.allowSameDiskTiering(location.getStorageType())) {
          File dir = new File(location.getUri());
          // Get the first parent dir that exists to check disk mount point.
          while (!dir.exists()) {
            dir = dir.getParentFile();
            if (dir == null) {
              throw new IOException("Invalid path: "
                  + location + ": directory does not exist");
            }
          }
          DF df = new DF(dir, dnConf.getConf());
          String mount = df.getMount();
          if (data.getMountVolumeMap().hasMount(mount)) {
            String errMsg = "Disk mount " + mount
                + " already has volume, when trying to add "
                + location + ". Please try removing mounts first"
                + " or restart datanode.";
            LOG.error(errMsg);
            throw new IOException(errMsg);
          }
        }
      }
    }
  }

View on GitHub (pinned to 2add963021)

Solutions

Create the directory chain on the DN host first: mkdir -p /mnt/disk2/dn && chown hdfs:hdfs, then resubmit
Use absolute paths with the file:// scheme; relative paths cannot be resolved by this check
Verify the mount is live before retrying: findmnt /mnt/disk2 or stat -f /mnt/disk2

Example fix

# before: /mnt/diskX not mounted, path has no existing ancestor
dfs.datanode.data.dir=[DISK]/mnt/diskX/dn
# after
mkdir -p /mnt/diskX/dn && mount /dev/sdb1 /mnt/diskX   # on the DN host
dfs.datanode.data.dir=[DISK]file:///mnt/diskX/dn

Defensive patterns

Strategy: validation

Validate before calling

File f = new File(location.getUri());
if (!f.isAbsolute()) throw new IllegalArgumentException("use an absolute file:// path: " + location);
File probe = f;
while (probe != null && !probe.exists()) probe = probe.getParentFile();
if (probe == null) throw new IllegalArgumentException("no existing ancestor for " + f);

Try / catch

catch (ReconfigurationException e) {
  if (String.valueOf(e.getCause()).contains("directory does not exist")) {
    // mkdir -p the path (and mount the disk) on the DN host, then resubmit
  }
}

Prevention

Pre-flight 'mkdir -p' every new data dir on the DN host before a volume refresh
Always use absolute paths with an explicit file:// scheme
Verify mounts are live (findmnt) before adding volumes under them

When it happens

Trigger: Submitting a relative data dir like data/dn1; a path under a mount that is not mounted or not yet created (typo /mnt/diskX/dn where diskX does not exist); a URI form File maps to nothing. Only checked when same-disk tiering is enabled and the location's type allows it.

Common situations: New disks cabled but not mounted when the reconfig fires; template paths referencing host-specific mounts that differ per host; NFS/network mounts flapping mid-refresh.

Related errors

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/e6d79fe50820f570. Report an issue: GitHub.