Multiple nodes leaving cluster

Magnus Kessler mkessler at basho.com
Wed Feb 8 04:00:21 EST 2017


On 7 February 2017 at 23:25, Daniel Miller <dmiller at dimagi.com> wrote:

> Hi Riak Users,
>
> In the documentation
> <http://docs.basho.com/riak/kv/2.2.0/using/admin/commands/#leave> for riak-admin
> cluster leave it says “You can stage multiple leave command before
> planning/committing.” This implies that it is safe to stage multiple nodes
> leaving the cluster simultaneously. Is that true? Will all data in the
> cluster be continuously available during the removal period if, for
> example, I setup and commit a plan for 3 nodes to be leave a 9-node cluster
> (assuming there is enough space for the data on the remaining 6 nodes)?
>
> I had asked a similar question on IRC a couple weeks ago. In that case I
> was asking about replacing multiple nodes simultaneously using riak-admin
> cluster replace. The answer I got there left some doubt in my mind as to
> whether it is safe (i.e., will not result in a period data availability) to
> have multiple nodes leaving the cluster at once. The documentation for
> replace implies that it is safe to replace multiple nodes simultaneously as
> well: “You can stage multiple replace actions before planning/committing.”
>
> Note that I am not asking about force-remove or force-replace, which I
> would expect to result in permanent data loss if multiple nodes are
> force-removed/replaced simultaneously.
>
> My cluster is running Riak 2.1.1 with standard nval of 3.
>
> Thanks!
> Daniel
>>

Hi Daniel,

Yes, staging several riak-admin cluster leave steps before  riak-admin
cluster commit is safe. The leaving nodes will perform an ownership handoff
of all their partitions to other nodes in the cluster before shutting
themselves down. While this is happening, these nodes remain in the cluster
as fully functional nodes. The same is true for  riak-admin cluster replace.

The reason we recommend staging of multiple leave or join operations is
that this minimises reshuffling data throughout the cluster. Only set of
ownership handoffs needs to happen, whereas if you were to perform several
consecutive leave or join operations, on each commit a substantial amount
of handoffs throughout the cluster is needed.

Kind Regards,

Magnus


-- 
Magnus Kessler
Client Services Engineer
Basho Technologies Limited

Registered Office - 8 Lincoln’s Inn Fields London WC2A 3BP Reg 07970431
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.basho.com/pipermail/riak-users_lists.basho.com/attachments/20170208/97bda168/attachment-0002.html>


More information about the riak-users mailing list