I’ve just listened to a wonderful Screaming in the Cloud podcast episode with Corey Quinn and Google VP Chen Goldberg who leads the Cloud Runtimes (CR) product area. This means she’s in charge of Kubernetes and Google Run (GCP’s serverless for those who didn’t know – I didn’t know).
This episode expertly negotiates the Kubernetes elephant that’s been in the room forever around how organisations:
- Have to be a certain size to make the most of Kubernetes
- Can’t find the good people to build the k8s infra
- And those people (the best people) should be working on applications, not infra
The biggest takeaway I got is this:
- If your organisation is too small a scale for Kubernetes, and it’s so hard to get it how you want (through no people and complexity) then just do everything serverless.
What’s “Too Small A Scale”?
I’ve heard the following rule of thumb for scale for your application. If you use more than 80 cores in total (if you can work that out for your use case) then you might want to think about Kubernetes.
Simple as that?
It sounds a lot to me that there is a tacit agreement in this episode however it’s also not a simple thing to move from serverless to Kubernetes. You could take this approach as a starting point. Always default to serverless and see how it goes? A way of winning without betting the farm?