Home | About | Sematext search-lucene.com search-hadoop.com
NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB
 Search Hadoop and all its subprojects:

Switch to Threaded View
Zookeeper >> mail # user >> Can Observers be reconfigured in a DR situation?


Copy link to this message
-
Re: Can Observers be reconfigured in a DR situation?
I've changed a server from Observer to Leader under a similar scenario
(data migration form old DC to new DC) without any problem. Just stop all
servers, change the configuration and start up all servers. Everything
should work fine.
/César.

On Tue, Jul 10, 2012 at 4:40 PM, Raj N <[EMAIL PROTECTED]> wrote:

> I am purely talking about a DR scenario. As in we can safely assume that
> the primary DC is no longer available. Also before the DR, the clients were
> connecting to the ZK ensemble in primary and after the DR they will connect
> to the ZK ensemble in secondary DC. You wont have a split brain problem.
> The only problem I see is potential loss of some data. The question really
> was if there are any gotchas in reconfiguring a Zookeeper node which was a
> observer to now be a leader/follower.
>
> Thanks
> -Raj
>
> On Tue, Jul 10, 2012 at 6:31 AM, David Nickerson <
> [EMAIL PROTECTED]> wrote:
>
> > Hi Raj,
> >
> > I've been considering this same solution for my project. Let's say that
> the
> > secondary DC automatically makes its own quorum when it can no-longer see
> > the primary. Then you have a case where a network partition can result in
> > your quorum splitting into two autonomously running quorums. (ZooKeeper's
> > majority rule exists to prevent this situation.) Just make sure that you
> > have some additional logic that invalidates the primary once it can
> > no-longer be found.
> >
> > On Tue, Jul 10, 2012 at 3:39 AM, César Álvarez Núñez <
> > [EMAIL PROTECTED]> wrote:
> >
> > > Yes you can. Just edit the zoo.cfg from the  backup DC, comment the
> lines
> > > relative to the nodes in the primary DC and the observer peer type.
> > >
> > > In this way you can start a new ZK ensemble with just the 3 servers on
> > > backup DC.
> > >
> > > But be aware that due to latency issues the backup DC could not have
> the
> > > latest changes from the active DC.
> > >
> > > You will also need to update each client "connection string" to use
> only
> > > the servers from the backup DC.
> > >
> > > /Sent from my GP0.
> > > El 09/07/2012 23:16, "Raj N" <[EMAIL PROTECTED]> escribió:
> > >
> > > > Let's say you have 2 DCs, primary and secondary. The primary DC has
> a 3
> > > > node zookeeper quorum. And the secondary DC has 3 nodes that are
> > > observers.
> > > > Is it possible in a DR scenario(primary down) to make the 3 observer
> > > nodes
> > > > the new quorum?
> > > >
> > > > Thanks
> > > > -Raj
> > > >
> > >
> >
>
NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB