Furthermore, anyone spending enough on compute to warrant k8s shouldn’t balk at all at $70/mo. I think the threshold for introducing the complexity and overhead of k8s isn’t probably until at least $5-10k/mo of spend (and probably 3-10x that in the normal case). Less than that and k8s is a whole lot of overkill.
We use google cloud projects to isolate customers and environments. (Some of our clients are old school and VERY scared of cloud, and multi-tenant). so for our pretty small company, we have >60 projects, each with a k8s cluster. that is a pretty good bump in costs come this summer.
Historically, we have powered down all the compute in a project that isn't needed, but left the k8s cluster in place (with its compute nodes powered down) because we could then bring it back up in 2 min or less. That dropped our cost for each project that was powered down to ~$25-$50/month, depending on how much disk space, etc, they were using. this will more than double (or triple) the cost of those projects. If we have to rebuild a full cluster from scratch, we then have to wait for the global load balancer to build our ingress, and then go authorize a TLS cert. This adds 20-30 min to re-activating our projects, which will suck.
I do think that $70/mo is reasonable per cluster but don't dismiss the value of k8s even for small projects. I used to bring a project from Ansible to k8s and even though it use only 2 nodes (~$200/mo), the tooling, the abstraction, the snappiness of gke was very much worth the switch.
