Does whirr support secondarynamenode instances for a hadoop cluster? If so,
what are the configuration parameters in the whirr config file?
--
Thanks,
John C
The secondary namenode is not started by the hadoop install /
configure scripts.
-- Andrei Savu / andreisavu.ro
On Wed, Jul 13, 2011 at 2:05 PM, John Conwell wrote:
> Does whirr support secondarynamenode instances for a hadoop cluster? If so,
> what are the configuration parameters in the whirr
I'm not sure what you mean by that. Let me rephrase...can I use whirr to
launch a cluster that has a secondarynamenode?
On Wed, Jul 13, 2011 at 2:14 PM, Andrei Savu wrote:
> The secondary namenode is not started by the hadoop install /
> configure scripts.
>
> -- Andrei Savu / andreisavu.ro
>
>
I just took a look at the source code. It seems like there is some
code in configure scripts that can be used to start the secondary
namenode unfortunately there is no hadoop-secondarynamenode role
defined.
So I believe it's not possible to start a Hadoop cluster that has a
secondarynamenode.
Tom
Its odd because the configure_cdh_hadoop.sh file actually has code in it to
launch a secondarynamenode, its just that there is no way to configure whirr
to pass in the role to the shell file.
for role in $(echo "$ROLES" | tr "," "\n"); do
case $role in
hadoop-namenode)
start_
You are right. A possible quick hack would be to update line 92 in
HadoopNameNodeClusterActionHandler.java.
We should consider refactoring that code so that the namenode & the
jobtracker can be started on different machine and also add support
for secondary namenode as discussed.
-- Andrei
On We
On Wed, Jul 13, 2011 at 2:43 PM, Andrei Savu wrote:
> You are right. A possible quick hack would be to update line 92 in
> HadoopNameNodeClusterActionHandler.java.
>
> We should consider refactoring that code so that the namenode & the
> jobtracker can be started on different machine and also add
Yea, that would be awesome!
On Wed, Jul 13, 2011 at 3:06 PM, Tom White wrote:
> On Wed, Jul 13, 2011 at 2:43 PM, Andrei Savu
> wrote:
> > You are right. A possible quick hack would be to update line 92 in
> > HadoopNameNodeClusterActionHandler.java.
> >
> > We should consider refactoring that c