apache/hadoop · error · IOException

Cannot concatenate source file {} because it is located in a

Error message

Cannot concatenate source file {} because it is located in a different namespace with nameservice {} from the target file with nameservice {}

What it means

HDFS concat physically appends blocks to the target file, which only works within a single NameNode namespace. The Router therefore checks every source's resolved nameservice against the target's; on mismatch it throws IOException naming the source file, its nameservice, and the target's nameservice. This is a structural limitation of concat under federation, not a transient error.

Source

Thrown at hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/router/RouterClientProtocol.java:701

    // Concat only effects when all files in the same namespace.
    RemoteLocation targetDestination = getFileRemoteLocation(trg);
    if (targetDestination == null) {
      throw new IOException("Cannot find target file - " + trg);
    }
    String targetNameService = targetDestination.getNameserviceId();

    String[] sourceDestinations = new String[src.length];
    for (int i = 0; i < src.length; i++) {
      String sourceFile = src[i];
      RemoteLocation srcLocation = getFileRemoteLocation(sourceFile);
      if (srcLocation == null) {
        throw new IOException("Cannot find source file - " + sourceFile);
      }
      sourceDestinations[i] = srcLocation.getDest();

      if (!targetNameService.equals(srcLocation.getNameserviceId())) {
        throw new IOException("Cannot concatenate source file " + sourceFile
            + " because it is located in a different namespace" + " with nameservice "
            + srcLocation.getNameserviceId() + " from the target file with nameservice "
            + targetNameService);
      }
    }
    // Invoke
    RemoteMethod method = new RemoteMethod("concat",
        new Class<?>[] {String.class, String[].class},
        targetDestination.getDest(), sourceDestinations);
    rpcClient.invokeSingle(targetDestination, method, Void.class);
  }

  @Override
  public boolean truncate(String src, long newLength, String clientName)
      throws IOException {
    rpcServer.checkOperation(NameNode.OperationCategory.WRITE);

    final List<RemoteLocation> locations =

View on GitHub (pinned to 2add963021)

Solutions

1. Ensure target and ALL sources live in the same nameservice (same mount subtree) before concat
2. If data must be combined across subclusters, use distcp/copy to move sources into the target's namespace first, then concat
3. Re-examine mount layout: put the directory tree that needs concat behind a single mount

Example fix

# before: rejected, /mntB is a different nameservice
hdfs dfs -fs hdfs://router -concat /mntA/trg /mntB/part-0

# after: copy source into target's namespace, then concat
hadoop distcp hdfs://router/mntB/part-0 hdfs://router/mntA/part-0
hdfs dfs -fs hdfs://router -concat /mntA/trg /mntA/part-0

// Enforce single-namespace concat before calling the Router
RemoteLocation t = clientProtocol.getFileRemoteLocation(trg);
for (String s : srcs) {
  RemoteLocation l = clientProtocol.getFileRemoteLocation(s);
  if (l == null || !l.getNameserviceId().equals(t.getNameserviceId())) {
    throw new IOException(s + " is in "
        + (l == null ? "?" : l.getNameserviceId())
        + ", target is in " + t.getNameserviceId());
  }
}
clientProtocol.concat(trg, srcs);

try {
  clientProtocol.concat(trg, srcs);
} catch (IOException e) {
  if (e.getMessage() != null && e.getMessage().contains("located in a different namespace")) {
    // copy the offending source into the target's namespace (distcp), then concat there
  } else { throw e; }
}

When it happens

Trigger: hdfs dfs -concat /mntA/trg /mntB/src where /mntA and /mntB are mounts into different subclusters (different nameservice IDs); multi-destination mounts where the target resolves to ns0 but a source resolves to ns1.

Common situations: Users assume Router federation merges namespaces transparently and try to concat across mount points; data pipeline writing parts to two different subclusters then attempting to concat.

Related errors

• Cannot find target file - {}
• Cannot concatenate source file {sourceFile} because it is lo
• Permission denied: {} is not allowed to change quota of {}
• Mount table state store is not available.
• File not found in downstream nameservices: {}

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/3ba2231ca0d6ea11. Report an issue: GitHub.