ORBIS Cloud launches today in eu-nl-1, our Amsterdam region. It is the managed version of the platform: we run it, we patch it, we are on call for it.
We have been running Terravek's own internal processing on it since January — every calibration campaign, every internal research job, every demonstration environment. This was deliberate. If the managed service is not good enough for our own workloads it is not good enough to sell.
The architecture is boring on purpose. Kubernetes, PostgreSQL with PostGIS, object storage, and a control plane defined entirely in Terraform. There is nothing here that a competent platform engineer has not seen before, which is a feature: it means we can hire, and it means the failure modes are well understood.
The interesting part is the data plane. Processing an ORBIS delivery is a burst workload with an enormous working set and a hard deadline, which is an awkward shape for most schedulers. We ended up building a placement layer that keeps the source imagery and the compute in the same availability zone and gives up some bin-packing efficiency to do it.
eu-de-1 in Frankfurt follows next year as a warm standby and a second residency option. Status is at status.terravek.com.