You are viewing a plain text version of this content. The canonical link for it is here.

Posted to common-dev@hadoop.apache.org by Giovanni Marzulli <gi...@ba.infn.it> on 2012/02/13 09:46:41 UTC

Writing failure test. Client and datanode on the same machine.

Hi,

I'm testing HDFS (0.20.203) on my cluster.

Particularly, I'm conducting writing tests while a datanode is down, but 
not yet marked dead by namenode.
If I write a file by client on the same machine where datanode process 
was killed, the task fails and these logs are printed:

/...INFO hdfs.DFSClient: Exception in createBlockOutputStream 
java.net.ConnectException: Connection refused... /(for every retries)/

/Inoticed that client insists on writing on local datanode exhausting 
all retries (dfs.client.write.block.retries is set to 100) until writing 
aborts!
Is this correct behavior? After N failed retries, it shouldn't contact 
another datanode to avoid writing abort?

Thanks

Gianni