Hello,

I ran into what a weird problem creating files and for the minute I only have a 
shaky conclusion:

logged in as a vanilla user on a datanode the simple command "hdfs dfs -put 
/etc/motd motd" reproducibly bails out with

WARN hdfs.DFSClient: DataStreamer Exception
org.apache.hadoop.ipc.RemoteException(java.io.IOException): File 
/user/tobbicke/motd._COPYING_ could only be replicated to 0 nodes instead of 
minReplication (=1).  There are 17 datanode(s) running and no node(s) are 
excluded in this operation.


Restarting datanodes did not help, the namenode logs were rather inconclusive. 
Trying to follow the only hint in there which were authentication problems from 
other users (we're using Kerberos) I happend to log in to another datanode, and 
to my surprise (!) there everything worked smoothly.

Trying on all of them with a mix of success and failures the only conclusion I 
came up with is that putting the datanode into "decommisionning" somehow 
affects client write access (no problem for -get), even from ordinary users.

Is this possible? Intended even, and if yes, what would be the logic behind 
that (after all I don't care on which datanodes the file ends up, there are 
plenty).

We're on Cloudera CDH 5.2.0, hadoop 2.5.0 in case that matters.

Any ideas?


Reply via email to