[jira] [Created] (HDFS-10993) rename may fail without a clear message indicating the failure reason.

[Yongjun Zhang (JIRA)]

Yongjun Zhang created HDFS-10993:
------------------------------------

             Summary: rename may fail without a clear message indicating the 
failure reason.
                 Key: HDFS-10993
                 URL: https://issues.apache.org/jira/browse/HDFS-10993
             Project: Hadoop HDFS
          Issue Type: Bug
          Components: hdfs
            Reporter: Yongjun Zhang


Currently the FSDirRenameOp$unprotectedRenameTo  looks like
{code}
 static INodesInPath unprotectedRenameTo(FSDirectory fsd,
      final INodesInPath srcIIP, final INodesInPath dstIIP, long timestamp)
      throws IOException {
    assert fsd.hasWriteLock();
    final INode srcInode = srcIIP.getLastINode();
    try {
      validateRenameSource(fsd, srcIIP);
    } catch (SnapshotException e) {
      throw e;
    } catch (IOException ignored) {
      return null;
    }

    String src = srcIIP.getPath();
    String dst = dstIIP.getPath();
    // validate the destination
    if (dst.equals(src)) {
      return dstIIP;
    }

    try {
      validateDestination(src, dst, srcInode);
    } catch (IOException ignored) {
      return null;
    }

    if (dstIIP.getLastINode() != null) {
      NameNode.stateChangeLog.warn("DIR* FSDirectory.unprotectedRenameTo: " +
          "failed to rename " + src + " to " + dst + " because destination " +
          "exists");
      return null;
    }
    INode dstParent = dstIIP.getINode(-2);
    if (dstParent == null) {
      NameNode.stateChangeLog.warn("DIR* FSDirectory.unprotectedRenameTo: " +
          "failed to rename " + src + " to " + dst + " because destination's " +
          "parent does not exist");
      return null;
    }

    fsd.ezManager.checkMoveValidity(srcIIP, dstIIP, src);
    // Ensure dst has quota to accommodate rename
    verifyFsLimitsForRename(fsd, srcIIP, dstIIP);
    verifyQuotaForRename(fsd, srcIIP, dstIIP);

    RenameOperation tx = new RenameOperation(fsd, srcIIP, dstIIP);

    boolean added = false;

    INodesInPath renamedIIP = null;
    try {
      // remove src
      if (!tx.removeSrc4OldRename()) {
        return null;
      }

      renamedIIP = tx.addSourceToDestination();
      added = (renamedIIP != null);
      if (added) {
        if (NameNode.stateChangeLog.isDebugEnabled()) {
          NameNode.stateChangeLog.debug("DIR* FSDirectory" +
              ".unprotectedRenameTo: " + src + " is renamed to " + dst);
        }

        tx.updateMtimeAndLease(timestamp);
        tx.updateQuotasInSourceTree(fsd.getBlockStoragePolicySuite());

        return renamedIIP;
      }
    } finally {
      if (!added) {
        tx.restoreSource();
      }
    }
    NameNode.stateChangeLog.warn("DIR* FSDirectory.unprotectedRenameTo: " +
        "failed to rename " + src + " to " + dst);
    return null;
  }
{code}

There are several places that returns null without a clear message. Though that 
seems to be on purpose in the code, it left to user to guess what's going on.

It seems to make sense to have a warning for each failed scenario.




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-h...@hadoop.apache.org