Thanks for checking the HDFS code.

>> Also it's strange that the region servers got corrupted reads when there are 
>> two more replicase available on HDFS.
>
> Corrupted reads? This is a loaded term, are you really saying that the
> region server read corrupted data from HDFS?

Sorry, it was too early to say the read data was corrupted. But the
other region servers had to shut themselves down because they detected
there were something wrong with HFiles.

"ABORTING region server serverName=ap12.secur2,60020,1298987576087,
load=(requests=0, regions=4, usedHeap=218, maxHeap=1998): Replay of
HLog required.
Forcing server shutdown"


I asked the guy to watch the data node and name node status if he can
run the same test again. He hasn't came back to me yet.

Thanks,
Tatsuya


2011/3/5 Jean-Daniel Cryans <[email protected]>:
> (heh this thread gives me a reason to look at the HDFS code)
>
>> Well, doen't the following message imply HDFS could accept writes when it 
>> has at least 1 data node available?
>>
>>> error: java.io.IOException: File 
>>> /hbase/Object_Speed_Test/1dbc1bf84b48e1145638b3a3bc3ad1cd/.tmp/1275904589980700621
>>>  could only be replicated to 0 nodes, instead of 1
>
> This is how that message is constructed:
>
>       throw new IOException("File " + src + " could only be replicated to " +
>                           targets.length + " nodes, instead of "
>                           minReplication);
>
> minReplication is the number of replicas needed in order to accept a
> write, by default 1. In this case, it wasn't able to place the block
> anywhere for an unknown reason.
>
>>
>> Also it's strange that the region servers got corrupted reads when there are 
>> two more replicase available on HDFS.
>
> Corrupted reads? This is a loaded term, are you really saying that the
> region server read corrupted data from HDFS?
>
> J-D
>



-- 
河野 達也
Tatsuya Kawano (Mr.)
Tokyo, Japan

twitter: http://twitter.com/tatsuya6502

Reply via email to