Priced from the host
Every machine carries a share of the host's monthly asset cost by its cores and memory. Used and unused split by what the 95th-percentile usage occupies.
Infrastructure
One Proxmox host in Prague, five service machines and a three-node Kubernetes cluster. CEMS reads them through node metrics and an estate data feed, prices them from the host's asset cost, and shows every figure to the person who owns it. It never writes to any of them.
How the data flows
Five hops, all of them reads. The only thing that moves back toward the estate is a question from a person.
A node exporter on every host and a DaemonSet on the Kubernetes nodes publish CPU, memory, disk and network counters.
node_exporter · hostNetwork :9100The Trim monitor scrapes them every 30 seconds into Prometheus, with Loki, Alertmanager and Wazuh beside it.
prometheus · 5-minute bucketsA publisher turns the metrics and a hand-kept estate description into the exchange-pack datasets: people, products, components, machines, usage.
Estate Data API · GET onlyThe collector pulls every fifteen minutes, lands the rows, runs the detectors and records the day's findings and each machine's cost share.
Postgres · ClickHouse · VictoriaMetricsTools compute, agents explain, a critic checks every number against the tool results, and the person sees only their own scope.
cems.fatsmash.in · Real dataThe estate
What the pane lists under Real data. Addresses, hypervisor ids and utilisation sit behind sign-in; the structure is public.
| Machine | Product | Size | Role |
|---|---|---|---|
| fatsmash-gw | Network access and DNS | 4 cores · 16 GB | routing peer, DNS, certificates for the development zone |
| fatsmash-app | Operations and monitoring | 4 cores · 12 GB | Prometheus, Loki, Grafana, Alertmanager, Wazuh, the estate data feed |
| dmz-fatsmash-authentik | Identity service | 2 cores · 4 GB | sign-in, registration with approval, multi-factor |
| dmz-fatsmash-cems | CEMS SaaS platform | 4 cores · 8 GB | development instance |
| dmz-fatsmash-cems-prod | CEMS SaaS platform | 4 cores · 8 GB | production instance, cems.fatsmash.in |
| k8s-node-181 | Kubernetes platform | 4 cores · 8 GB | control plane |
| k8s-node-182 | Kubernetes platform | 4 cores · 8 GB | worker |
| k8s-node-183 | Kubernetes platform | 4 cores · 8 GB | worker |
Every machine carries a share of the host's monthly asset cost by its cores and memory. Used and unused split by what the 95th-percentile usage occupies.
Each product has a service, a product and a cost owner. Three chains roll the same figures up to the group CEO, and each person sees only their scope.
Pods and namespaces on the Kubernetes cluster. The nodes appear as machines today; workload-level cost follows with the platform's Kubernetes source.
See it live
Sign in, switch to Real data, and open Explore in the infrastructure view: region, data centre, host and machine, each with its cost, its excess and its findings.