hadoop-hdfs-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From hadoop <had...@wangzw.org>
Subject Re:Re: potential bug when read directly from local block
Date Thu, 09 Aug 2012 13:30:00 GMT
<span ></span>Hi,&nbsp;<div><br></div><div>I run this
test as superuser.&nbsp;</div><div><br></div><div>The reason
of this warning is that, hdfs client buffer the block information and then the block is appended.
For example, the generation stamp buffered on the client is 1000, after append, the real generation
stamp of block is 1001. When the client ask datanode to return the local block path, client
send 1000 to datanode(in the block information in the rpc call), and datanode return metadata
file path which is "blockid_1000.meta", and this metadata file does not exist, it should be
"blockid_1001.meta"</div><div><br></div><div><br></div><div><br></div><div><br><pre>-----
Original Message -----From: Harsh J &lt;harsh@cloudera.com&gt;
To: hdfs-user@hadoop.apache.org

Hi,

If you're running this as a non-superuser (i.e. not the user thats
running the DataNodes), have you set your dfs.datanode.data.dir.perm
to be 755 instead of the default 700?

On Thu, Aug 9, 2012 at 3:06 PM, hadoop &lt;hadoop@wangzw.org&gt; wrote:
&gt;
&gt; Hi all
&gt;
&gt;
&gt; I found some warning in the log, and it may the bug of hdfs. Is there anyone can
confirm that?
&gt;
&gt;
&gt; I make a simple test case and the output is attached.
&gt;
&gt;
&gt; I test it on hadoop-dist-2.0.0-alpha
&gt;
&gt;
&gt;
&gt;
&gt; Best Regards
&gt;
&gt;
&gt; --
&gt; Zhanwei Wang
&gt;



-- 
Harsh J
</pre></div>
Mime
View raw message