This message was deleted.
# general
s
This message was deleted.
c
We use 3 environments: development, staging, production. None of them are accessible by our engineering staff, only cloud operations (devops + infra). Each environment is a separate google project, and we have 1 cluster each. This ensures that nothing in one environment can touch another.
I do highly recommend the separate google project, if possible, but definitely the separate cluster. A bad deploy of something like istio can bring down all of your applications. Having a way to test it out first is really handy.
e
Yeah, this is how I've traditionally do it. I know alot of recent advice says you can just use one but Im not buying it. I'm going to agree with you and use seperate projects. Makes more sense to me...
Also I think I can setup dev and stage to just scale down to 0 using Preemptive nodes if I want to really optimize cost.
b
we also do entirely separate - different azure subscriptions and clusters. as nesl247 says it's easy to break a whole kube cluster and anything with CRDs or a daemonset is going to cause interference
c
Basically I see it as this: You either pay to maintain 3 separate environments, or you pay when your production environment gets broken. One is a known/fixed cost, the other, who knows.
Either way, you pay.
Also, if you don’t have separate environments,you can’t test k8s upgrades themselves.
c
We keep dev and stage together, but separate from production, at the cluster level. We keep dev and stage separate with namespaces, rather than labels. I think you'll get much stronger separation if you separate the environments that way.
We do eventually plan to separate staging from dev into its own cluster.
b
one of the benefits of pulumi is it made scaling down one cluster and bringing up two separate ones instead essentially effortless - for the same azure VM cost
we just have a one node cluster for qa/staging
e
that's also true, you could tear down and spin up dev and qa as needed.