[HDFS-13845] RBF: The default MountTableResolver should fail resolving multi-destination paths

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.0, 3.1.0, 2.9.1
• Fix Version/s: 3.3.0, HDFS-13891
• Component/s: federation, hdfs
• Hadoop Flags: Reviewed
• Assignee: yanghuafeng
• Reporter: yanghuafeng
• Due: 22/Aug/18
• Created: 22/Aug/18 12:14
• Updated: 12/Sep/19 19:51
• Resolved: 30/Oct/18 06:05

Description

When we use the default MountTableResolver to resolve the path, we cannot get the destination paths for the default DestinationOrder.HASH.

// Some comments here
private static PathLocation buildLocation(
      ......
    List<RemoteLocation> locations = new LinkedList<>();
    for (RemoteLocation oneDst : entry.getDestinations()) {
      String nsId = oneDst.getNameserviceId();
      String dest = oneDst.getDest();
      String newPath = dest;
      if (!newPath.endsWith(Path.SEPARATOR) && !remainingPath.isEmpty()) {
        newPath += Path.SEPARATOR;
      }
      newPath += remainingPath;
      RemoteLocation remoteLocation = new RemoteLocation(nsId, newPath, path);
      locations.add(remoteLocation);
    }
    DestinationOrder order = entry.getDestOrder();
    return new PathLocation(srcPath, locations, order);
  }

The default order will be hash, but the HashFirstResolver will not be invoked to order the location.
It is ambiguous for the MountTableResolver that we will see the HASH order in the web ui for multi-destinations path but we cannot get the result.
In my opinion, the MountTableResolver will be a simple resolver to implement 1 to 1 not including the 1 to n destinations. So we should check the buildLocation. If the entry has multi destinations, we should reject it.

Issue Links

• is related to: HDFS-13224 RBF: Resolvers to support mount points across multiple subclusters