multi master across different AZs

Originally reported by @rbritoda https://its.cern.ch/jira/browse/OS-15901

Enforce split of multiple masters across AZs.

We don't have a downstream ticket for this, it was being tracked upstream only.

assigned to @strigazi

assigned to @strigazi

Needs a doc update (FAQ) once this is there: https://kubernetes.docs.cern.ch/docs/faq/#can-i-have-multiple-masters-across-different-availability-zones

@strigazi will this make it? Can i help?

I will try to finish it in the after noon, i should manage

Brilliant! Thanks really!

We want:

1. users to select the AZs (this requires some user input validation)

Or we can leave it to error in heat like so:

1.1 non existent AZ: Resource CREATE failed: BadRequest: resources.kube_masters.resources[2].resources.kube-master: The requested availability zone is not available (HTTP 400) (Request-ID: req-8fe7d3aa-bacf-42e8-be5e-ec9390542f6a)

1.2 asked N masters, passed less than N AZs: Resource CREATE failed: resources[2]: resources.kube_masters.Property error: resources.kube-master.properties.availability_zone: tuple index out of range

2. force split across the hard-coded list of "cern-geneva-a,cern-geneva-b,cern-geneva-c"
3. force split across a list defined in magnum.conf

Would it be easy to have a default of all (the three) but still let the users override the list? This could give some flexibility in case we start having AZs full, or unavailable, etc. As well as the special cases for special AZs.

mentioned in merge request !206 (merged)

@rbritoda The only issue is that you can not create a cluster with:

• multi-master in different AZs
• select an AZ for the default nodegroup

What would the default nodegroup AZ be in that case? Random?

We can still do:

• multi-master in different AZs
• multiple nodegroups one in each AZ

so i think we're good.

changed milestone to /kubernetes%"R-2022-Q2-1"

added A-cluster-lifecycle A-networking p2 labels

closed with merge request !206 (merged)