weird crash corner cases

[Ryan Rawson <ry...@gmail.com>]

i messed my cluster up tonight by killing off hdfs before hbase, oops.

i had to do a staged start, i first started hbase, but only 4 regionservers
came up and the master.  so i then kill -9ed all the regionservers, and
restarted them.  at this point, the cluster is DOA because the -ROOT- table
is not where it is supposed to be.

just a FYI for you zookeeper folks....

Re: weird crash corner cases

[Nitay <ni...@gmail.com>]

When you restarted the servers, shouldn't everything be in safe mode? The
root region server ZNode is not ephemeral so it will not go away when all
the servers die. The reason for this was so that things can continue working
if the master dies. The master has an ephemeral out of safe mode ZNode
however that the regionservers should be waiting for when they startup. Can
you file a JIRA with some more details on this (logs, etc)?

Thanks,
-n

On Wed, Apr 8, 2009 at 1:01 AM, Ryan Rawson <ry...@gmail.com> wrote:

> i messed my cluster up tonight by killing off hdfs before hbase, oops.
>
> i had to do a staged start, i first started hbase, but only 4 regionservers
> came up and the master.  so i then kill -9ed all the regionservers, and
> restarted them.  at this point, the cluster is DOA because the -ROOT- table
> is not where it is supposed to be.
>
> just a FYI for you zookeeper folks....
>