Etcd with two nodes

[Maxim Trusov]

Hello,

We think about using etcd in our project, but it's needed to support different cluster configurations, including two nodes clusters. Would be it possible to run etcd with such envoirement? Thanks a lot.

[Rob Szumski]

Yes, you can, but its worse in every way than just having a single node or 3+. Basically, you get additional operational overhead but none of the HA benefits.

If you share some more details about your project we can brainstorm the best solution for you.

- Rob

> On Dec 9, 2016, at 10:43 AM, Maxim Trusov <maxim....@gmail.com> wrote:
>
> Hello,
>
> We think about using etcd in our project, but it's needed to support different cluster configurations, including two nodes clusters. Would be it possible to run etcd with such envoirement? Thanks a lot.
>

> --
> You received this message because you are subscribed to the Google Groups "CoreOS User" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to coreos-user...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Maxim]

Rob, thanks a lot for the quick response. Generally speaking, we've got 2 up to 8 nodes cluster and we need a way how to distribute and maintain cluster configuration among all nodes in a reliable way. Also it should support active/active mode, i.e. cluster configuration requests can come to any node on any time and the system should be able handle them properly. So since it's a management part we don't need a high performance or support of petabytes of data. You can see it's something like "Clustered Logical Volume Manager (CLVM)" but for any type of configuration. Another thing to mention is that we use IBM Power8 hardware so the system should support it. Before reinventing the wheel we decided to have a look on existing solutions. 

We've played a bit with corosync and DLM but frankly speaking not very happy with it.

Returning to etcd, functionality that etcd provides is exactly what we need but there are several thing to clarify. Am I right in thinking that in case of 2 nodes cluster, if there is a problem with network for a long time or one node crashed, the etcd cluster stops any activities and manual intervention is needed (snapshot save/snapshot restore recovery process is needed)?  The same situation would be with N nodes cluster if any N/2 nodes crashed (i.e. the quorum lost), right?

Thank you very much,

Maxim

пятница, 9 декабря 2016 г., 23:46:18 UTC+3 пользователь Rob Szumski написал: