zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alexander Shraer <shra...@gmail.com>
Subject Re: Can an observer node be promoted to be read-write?
Date Mon, 07 Dec 2015 22:00:44 GMT
I haven't tried this, but I think it should be possible. Observer is not so
different from a voting participant, and
in 3.5.0 we can actually dynamically switch between the roles using the
reconfig command (not during downtime
as you need). But I think it should work. You should be aware of the
potential scenario where the original DC hosting
the voters isn't actually down, but just lost connectivity or something
else, and then promoting the observer to a participant would create split
brain. So you should make sure this doesn't happen.



On Sun, Dec 6, 2015 at 8:48 PM, Jason Rosenberg <jbr@squareup.com> wrote:

> I'm wondering if an observer node, can be upgraded to be a voting
> read-write node in a cluster, after a restart with a config update to
> remove the observer config?
>
> The scenario I'm thinking about is for a cross-datacenter cluster, where we
> have a master datacenter where the nodes are all read-write, and then have
> remote observer nodes in remote datacenters.  In the event that there's a
> catastrophic loss of a the 'master' datacenter, is it feasible to make a
> new dc the 'master' dc, and restart all the nodes there to no longer be
> observers, and then restart nodes in other dcs to point to the new master?
>
> The point being that it seems desirable to have the voting read-write nodes
> always co-located in closer proximity (and remote dc nodes can be
> observers), but we need a decent disaster recovery plan in the event of a
> dc loss.  This plan would be a manual recovery, of course, but it's
> workable.
>
> Thanks,
>
> Jason
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message