Hi,

i have a datanode maintenance question.

if i bring a data node for memory upgrade down for several hours and then
return it back, i don't need to run decommission? decommission is only if i
want to get rid of that node for good and now want to re-distribute its
blocks, right?

Or i will be risking some sort of 'split brain' inconsistency?

Thank you.
-Dmitriy

On Sep 22, 2010, at 12:36 PM, Dmitriy Lyubimov wrote:
> i have a datanode maintenance question.
>
> if i bring a data node for memory upgrade down for several hours and then return it back, i don't need to run decommission? decommission is only if i want to get rid of that node for good and now want to re-distribute its blocks, right?
Hadoop will treat it as a failure and will start re-replicating the blocks on that node.   When the node is added back into the system, it will delete the extra replications.  Those extra replications may or may not be on the same node that you re-added.
> Or i will be risking some sort of 'split brain' inconsistency?

Nah, should be fine.

Thank you. just suspected as much.

-Dmitriy

On Wed, Sep 22, 2010 at 12:42 PM, Allen Wittenauer <[EMAIL PROTECTED]
> wrote:

>
> On Sep 22, 2010, at 12:36 PM, Dmitriy Lyubimov wrote:
> > i have a datanode maintenance question.
> >
> > if i bring a data node for memory upgrade down for several hours and then
> return it back, i don't need to run decommission? decommission is only if i
> want to get rid of that node for good and now want to re-distribute its
> blocks, right?
>
>
>         Hadoop will treat it as a failure and will start re-replicating the
> blocks on that node.   When the node is added back into the system, it will
> delete the extra replications.  Those extra replications may or may not be
> on the same node that you re-added.
>
>
> > Or i will be risking some sort of 'split brain' inconsistency?
>
>         Nah, should be fine.
>
>