Highly Available, Barely Running
Multi-region HA that doesn't cost twice: Flux, KEDA and K8GB for a standby that costs close to nothing until you need it. With Pedro Cachaldora.
Multi-Region HA That Doesn’t Cost Twice. A 30-minute talk and live demo with Pedro Cachaldora at KCD Porto × DevOps Days Portugal, Friday 20 November 2026, 11:00, DevOps Days Room. Tracks: Operations + Performance, Platform Engineering.
High availability is usually designed as an architecture problem and paid for as a finance problem. The reference diagrams look clean: two regions, two providers, traffic balanced between them. Then somebody runs the numbers on keeping a second environment permanently warm.
The compromise most organisations settle on is worse than either extreme. A standby region gets provisioned, sized for a peak it will probably never see, left running at partial capacity, and quietly never exercised. It costs real money every month, and nobody is entirely confident it works.
What we cover
An open-source approach to multi-region and multi-cloud high availability where standby capacity costs close to nothing until the moment it is needed:
- Flux keeps every cluster reconciled to the same declarative state.
- KEDA lets workloads sit at zero and come up on demand.
- K8GB handles global DNS-based failover between clusters and providers, with no proprietary global load balancer in the failover path.
We are specific about the trade-offs: what DNS-based failover can and cannot guarantee, what recovery time looks like when the standby starts from zero, how stateful workloads change the picture, the failure modes we ran into, and where a managed global load balancer is still the better answer.
What you take away
A vendor-neutral pattern for multi-region resilience, a way to reason about the cost and recovery-time trade-off on purpose rather than by default, and a failover cheap enough that you can afford to test it regularly.
Slides and the demo repository will be linked here after the talk.