Server Groups

[benofben]

Right now the template is deploying an autoscaling group across every AZ in a region. We probably want it to split an even number of nodes over two.

There's a wider question of how MDS impacts this that I don't know the answer to yet.

[benofben]

Not quite sure if this is the right approach. Need more discussion with Karthik and Tony. I'm going to continue using every available AZ until then.

[benofben]

I wonder what would happen if we had a replica count of 1 but a server group per AZ, which would mean 2 copies of the data are spread across 2, 3 or more server groups.

[benofben]

Perry confirmed this'll work well:

I’m debating adding some code to the various cloud templates where (let’s take AWS as an example) a node would call the metadata service on boot to check what AZ it’s in. It would then assign itself to a server group with the name of that AZ. Assuming our replica placement takes advantage of that, it would allow us to get the maximum spread across AZs in a region, rather than making use of only two as seems to be the recommendation from some conversations I’ve had. An added benefit is that is way easier to code and maintain that restricting to two regions.

[benofben]

https://developer.couchbase.com/documentation/server/current/cli/cbcli/group-manage.html

[jackharp]

This is a high priority.