Azure Kuberenetes Cluster design approach

Question

Hi

In my company , we are going on onboard couple of products on to the AKS.

we are evaluating below options in how these products can be run on AKS.

Option 1: We can spin up multiple AKS clusters and each cluster we will run Pods relate to one Product., so with this approach, if we want to onboard 2
products, then we will spin up 2 AKS clusters and these cluster will have dedicated and independent Scaling, Nodes, Network and load
Balancers. Since these Products are independent and these is no dependency on one on another.

Option 2: We can have one cluster and multiple namespaces, each namespace will run pods related one product.
So with this approach we can control resources on namespace wise and ingress controllers will be for each Namespace.

we are more aligning towards to the option1, please share your opinion and give me correct way of doing it.

Answer 1

This is really going to depend on what you need from your AKS clusters. Both options are viable, option 1 will be more expensive as you are duplicating resources for running things like Ingress that potentially could be shared between productions, but it will provide more flexibility. Similarly, if you think some products will need a very high node count, or have different requirements for upgrades etc. then you may want to put them on their own cluster.

You could also look at a third option, which is to use a single AKS cluster, but with Node Pools per product. That way you can isolate each product in their own node pool, and specify different node configurations per product.

Ultimately none of these are incorrect, you need to consider what is more important, cost or flexibility.

Answer 2

Hi Sam,

Thanks for the quick and valuable response.

We are more aligned to Option 1: we will created multiple Namespaces for Dev ,QA and INT and dedicated AKS cluster for PROD.
For every product that we are bringing in AKS, will have 2 clusters(NON-Prod and PROD).

with this model we can reduce some cost and for prod and INT region we will setup node scaling for lower regions we will run with only one 1 node.