Built for real infrastructure

Design principles, not afterthoughts.

Every one of these is a hard constraint on the architecture, not a marketing claim — they hold for every specification and every line of code in the platform.

Fail open, always

No GreenShift failure can stop or modify a VM, change Kubernetes scheduling, or touch application availability.

Outbound-only

No inbound hypervisor ports by default. Nothing in a sensitive environment listens for the outside world.

Explicit freshness

Every time-sensitive value carries a window, a source timestamp, a confidence, and an expiry. Nothing stale is silently reused.

Server-side authorization

Identity comes from a signed cluster token or a Kubernetes service account token — never a request-body field.

Target architecture — not yet in the pilot mTLS and OIDC/OAuth2 identities.

Prometheus-native

Every metric is Prometheus format, greenshift_-prefixed, ready to scrape from day one — no custom exporter required.

Data minimization

No application payloads, packet contents, secrets or credentials are ever collected. Process command lines are read locally only to identify VM processes, then discarded — our message format cannot carry them.

Scaphandre+PDU / BMC →Kepler →GreenShift →Prometheus metrics, per tenant
How it fits together

From power signals to tenants.

Datacenter power Host / process (Scaphandre) PDU / BMC power Cluster telemetry Kepler pod / container metrics GreenShift reconcile + attribute VMs & nodes Namespaces & pods Workloads Tenants