On Fri, Jul 15, 2011 at 7:36 AM, Adam Shook <[email protected]> wrote:
> I am running a MapReduce job using standard input and output formats and 
> using an HTable as a reference data set in my Mapper code.  I am using a 
> small cluster of around 10 nodes.  In my setup phase I am using an HTablePool 
> to get a reference to a table.  On all but two nodes, the call to getting the 
> table hangs and eventually causes the task to fail.  However, on two of the 
> 10 machines, it retrieves the table and its business as usual.  (I also just 
> tried creating a new HTable without the pool - no dice).
>

Any exception thrown?

> It just so happens that the two machines I can successfully get the table on 
> are configured in the hbase-site.xml file for the hbase.zookeeper.quorum file.
>

Any other configuration differences on these machines?  Are these
using localhost to find zk and finding something because zk instance
is actually running locally (whereas the others fail to find a zk
member)?

>  I was told that the other machines don't need to be in this file - ZooKeeper 
> will handle everything and I should be able to get a table just fine.  The 
> cluster is configured so HBase is not managing ZooKeeper.
>
This should be fine.

> If I ssh into any of the 8 machines that don't work, I am able to use the 
> HBase shell and scan through a table.
>

If you look at your UI are all ten machines showing, all with regions
loaded?  If you run a count from the shell, it runs through all of
your table contents? (Could take a while if big)

St.Ack

> Any help would be very much appreciated.
>
> Thanks!
> Adam
>

Reply via email to