Zookeeper upgrade

[Jai Bheemsen Rao Dhanwada <ja...@gmail.com>]

Hello All,

I am using zookeeper 3.4.5 version in production and would like to upgrade
to the latest version 3.4.9

Can someone point me to the correct upgrade(with out having a downtime)
path?.

Re: Zookeeper upgrade

[Jai Bheemsen Rao Dhanwada <ja...@gmail.com>]

Thank you Rakesh.
I will try it out

On Thu, Nov 10, 2016 at 8:30 PM, Rakesh Radhakrishnan <ra...@apache.org>
wrote:

> Probably you could do rolling restarts. You have to do this manually by,
>
>
> Pre-requisite:- Ensure all the servers in the cluster is running. For
> example, if you have three node cluster A, B and C. Assume, A & B are
> Followers and C is the Leader. This is to ensure that, if one of the server
> stopped(assume A) during upgrade step then other servers(B & C) would be
> able to provide service to the ZooKeeper clients.
>
> step-1) stop one server at a time,
> step-2) then upgrade new software,
> step-3) finally restart the server back.
> The upgrade is now complete on this server and you can proceed to the next.
>
> Now, the restarted server will join the quorum. The side effect of this is,
> since this server is getting stopped all the clients(sessions) connected to
> this server will be Disconnected and then Reconnected with other server in
> the ZK cluster.
>
> When you do the above steps in LEADER server, then surely the quorum will
> be lost and all the other running PARTICIPANTS will enter into quorum
> reformation phase and elect one among them as new LEADER. Here also, you
> will see all the clients(sessions) connected to this cluster will be
> Disconnected and then Reconnected back safely. AFAIK, two times clients
> will get Disconnected and then Reconnected events, if you have watcher
> registered.
>
> I'd prefer to start rolling upgrade from OBSERVERs, then FOLLOWERs and at
> the end will pick the LEADER server. This way you could get chance to
> monitor your newly upgraded server for any inconsistencies with less client
> Disconnections. Perhaps, do rolling back if any issues are encountered.
>
> Imp Note*:*- I hope you will first experiment rolling upgrade steps in your
> test cluster and then do it in production cluster.
>
> References:-
> https://wiki.apache.org/hadoop/ZooKeeper/FAQ#A6
> https://www.cloudera.com/documentation/enterprise/5-5-
> x/topics/cdh_ig_zookeeper_earlier_cdh5_upgrade.html
>
> Thanks,
> Rakesh
>
> On Fri, Nov 11, 2016 at 9:22 AM, Jai Bheemsen Rao Dhanwada <
> jaibheemsen@gmail.com> wrote:
>
> > Hello All,
> >
> > I am using zookeeper 3.4.5 version in production and would like to
> upgrade
> > to the latest version 3.4.9
> >
> > Can someone point me to the correct upgrade(with out having a downtime)
> > path?.
> >
>

Re: Zookeeper upgrade

[Rakesh Radhakrishnan <ra...@apache.org>]

Probably you could do rolling restarts. You have to do this manually by,


Pre-requisite:- Ensure all the servers in the cluster is running. For
example, if you have three node cluster A, B and C. Assume, A & B are
Followers and C is the Leader. This is to ensure that, if one of the server
stopped(assume A) during upgrade step then other servers(B & C) would be
able to provide service to the ZooKeeper clients.

step-1) stop one server at a time,
step-2) then upgrade new software,
step-3) finally restart the server back.
The upgrade is now complete on this server and you can proceed to the next.

Now, the restarted server will join the quorum. The side effect of this is,
since this server is getting stopped all the clients(sessions) connected to
this server will be Disconnected and then Reconnected with other server in
the ZK cluster.

When you do the above steps in LEADER server, then surely the quorum will
be lost and all the other running PARTICIPANTS will enter into quorum
reformation phase and elect one among them as new LEADER. Here also, you
will see all the clients(sessions) connected to this cluster will be
Disconnected and then Reconnected back safely. AFAIK, two times clients
will get Disconnected and then Reconnected events, if you have watcher
registered.

I'd prefer to start rolling upgrade from OBSERVERs, then FOLLOWERs and at
the end will pick the LEADER server. This way you could get chance to
monitor your newly upgraded server for any inconsistencies with less client
Disconnections. Perhaps, do rolling back if any issues are encountered.

Imp Note*:*- I hope you will first experiment rolling upgrade steps in your
test cluster and then do it in production cluster.

References:-
https://wiki.apache.org/hadoop/ZooKeeper/FAQ#A6
https://www.cloudera.com/documentation/enterprise/5-5-x/topics/cdh_ig_zookeeper_earlier_cdh5_upgrade.html

Thanks,
Rakesh

On Fri, Nov 11, 2016 at 9:22 AM, Jai Bheemsen Rao Dhanwada <
jaibheemsen@gmail.com> wrote:

> Hello All,
>
> I am using zookeeper 3.4.5 version in production and would like to upgrade
> to the latest version 3.4.9
>
> Can someone point me to the correct upgrade(with out having a downtime)
> path?.
>