The problem

Every engineering team we talked to described the same pain: the architecture was sound, but the operational details — retry budgets, cache invalidation, drain windows — ate the engineering time the project was supposed to save.

How it works

Bergwerk Cloud solves this with a control plane that treats configuration as data: every change is a versioned record, every rollout is a diff, and every region converges on the same state within seconds.

The architecture

  • A write-ahead configuration log replicated per region
  • Stateless workers that reconcile observed vs desired state
  • Drain windows coordinated through the same primitive as maintenance
  • Per-tenant rate limits enforced at the edge, not the database

Lessons learned

The team's biggest surprise was organizational, not technical: once rollouts became boring, the escalation path moved from on-call rotations to ordinary code review — exactly where it belongs.

Outlook

Expect the next milestone in the coming quarter. CloudRadar will track the follow-through — pricing pages, SLA text, and real region availability — and update this piece if anything shifts.