If you are designing a fairly large platform to support upwards of 100 dev teams, what is the best approach to think about cloud accounts, would you have one platform account for everything or would you have an account per dev team or product? Any learning from those who have walked this path?

This depends a bit on what you are building, the cloud provider and if teams should be able to interact with the cloud provider directly. I've struggle in the past to cleanly separate permissions inside a single AWS account as not all services support tag based permissions etc. Multiple accounts make this way easier (because the boundary is the account), but make other things more challenging. E.g. SSO to 50+ accounts, keeping security requirements inline etc.

Maybe good question also for @Maryann Bell @Bryan Finster @Galo Navarro @Brian Leathem

on aws multiaccount is the only way (otherwise you drown in policy complexity if you want to have any change of least privilege - also at scale API limits become an issue) - https://docs.aws.amazon.com/whitepapers/latest/organizing-your-aws-environment/organizing-your-aws-environment.html gives a good overview about sensible options

Thank you @Joern Barthel for this, really helpful- much appreciated 🙂

Morning!

Out of interest, what do people here use for cost tracking? Tagging is often a losing battle, ownership changes, etc, so a way of dynamically creating cost groups outside of AWS feels like the way to go. Yet to find the right tool to do that, along with monitoring, reporting, anomaly detection, recommendations, all in a single pane of glass. We have fine-grained accounts which helps a lot, but within that (or even across accounts) cost management is tricky.

depends - if you want / can afford (highly dependent on your profile & how you negotiate) an external cost management tool (typically requires substantial investment into understanding how to use it) can help here the standard 1 account per environment and workload / service structure + individuals sandboxes style i described above also tends to simplify this aspect a lot - for what kind of (shared?) costs do you need insights in your fine grained accounts? or are they cut by sth. other than workload types?

Cut by business groups and environments, but within each account there is still some splitting of costs required.

for resources provisioned by the users of that account? or centrally (stacksets, lz, adf, ctc, whatever) provisioned?