Projects and own apps
1Four projects, four boundaries.
Jonas creates four projects: shop, orders and backoffice for the own apps, company for the company sign-in Velohafen's people will log in through. In the three app projects he takes the project admin role and creates a namespace in each, a space of its own with its own network rules. The platform delivers each app project from a path of its own in Jonas's Git repository on GitHub, where the apps' code and container images live too.
2Two kinds may cross.
Out of the box no connection crosses a project boundary. So Jonas, from IT administration, allows exactly two kinds in the organisation's connection rules: web interfaces between own apps, and the sign-in. He consents to that change with his passkey, by fingerprint or security key, and the consent holds for exactly this change.
3Jonas runs the database himself.
The orders live in a PostgreSQL database in the order API's namespace. Jonas runs it himself, as a ready-made database package (a Helm chart) from his repository, on a volume, lasting storage the platform creates for him. That makes the backups his job too: gh0stservice does not back up what a customer runs itself.
4Each app says what it needs.
For each own app Jonas writes a score.yaml, a short description in the Score format: what the app offers and what it needs from others. The shop needs the order API, and the order API offers its web interface on port 8080, the number other apps reach it on. gh0stcli, gh0stcloud's command-line tool, reads only these details from it and never the containers, which can hold secrets. Jonas then proposes both apps as one change, and nothing changes until someone consents.
5The shop waits for the order API.
The portal shows the change on one page: two own apps and the connection between them. Jonas consents, and the environment, the map of all apps and connections, shows both as own apps in their projects. The connection from the shop to the order API waits, though: it leads into the orders project, which has not granted it yet. While it waits it stays closed, and it turns ready once the reason goes away.
6First the grant, then the path.
On the map the connection reads: waits for the grant of project orders. Jonas grants exactly this case, the order API for the shop project, and consents again with his passkey. The connection turns ready, and only now does the network path appear: a rule from the shop's pods, the running parts of the app, to the order API's pods, on port 8080 only. The shop receives the order API's address as connection details in its namespace.
Online and on the company sign-in
7Online on automatic addresses.
The shop and the back office first get automatic addresses under gh0stcloud.app, with an HTTPS certificate the platform issues itself. Jonas assigns each address to its namespace, and the platform delivers the Ingress rule, which leads the traffic to the app, from his repository. Only through these rules does traffic come in from outside.
8One door out.
Outbound, everything starts closed. One rule opens outgoing HTTPS to the internet, port 443, for the order API's pods and the backup job: the order API reaches the payment provider, and the backup job reaches storage Velohafen books itself, where it writes a copy of the database every night. Jonas restores it once as a test. Everything else stays inside.
9gh0stservice runs the sign-in.
The back office needs a company sign-in. Jonas picks the central sign-in in the catalog, a Managed App: gh0stservice runs it with backups, updates and monitoring, and a company account with a payment method on file can order it. Because the order costs money and Ida has not delegated that yet, she consents as the owner with her own passkey. The sign-in is handed over only after the first backup and a restore test in which a backup is really played back, and the service fee starts on the handover day.
10The back office on the company sign-in.
The back office's score.yaml names one need: the sign-in. Jonas adds the grant from project company to backoffice, and gh0stcli proposes it all as one change with three parts: the own app, its connection to the company sign-in and the grant. Once Jonas consents, the namespace holds the back office's own access to the sign-in, a client with its credentials, and everyone in the company sign-in logs in there with a passkey. Inside the sign-in itself the back office may do nothing on that person's behalf. That holds for every app, Jonas's own code too.
Own domain and cost
11The own domain, reviewed by gh0stservice.
The shop is to run at www.velohafen.example. First a record in DNS, the internet's address book, proves that the domain belongs to Velohafen. Jonas creates it at his DNS provider, and access to that provider stays with Velohafen. An own domain lies outside the limits an account has for public addresses out of the box. So Jonas requests an extension, and gh0stservice reviews it before it applies.
12Two records, then the shop runs on its own domain.
In the public access settings Jonas assigns www.velohafen.example to the shop's namespace and enters the hostname in his repository. Then he creates two more records at his DNS provider: one that points the address at the shop, and one for the certificate. gh0stservice names the target of that second record when the own domain is set up. Once the platform has checked both, it issues the certificate and renews it itself. From then on the orders come in through the own domain.
13Cost by project.
Ida sees the month's cost by project. Project company holds the company sign-in's service fee and its resources; shop, orders and backoffice hold only the computing power and storage of the own apps, at the price list. The base fee belongs to no project. Code, updates and backups of the own apps stay Jonas's job; the sign-in is run by gh0stservice.
Velohafen is a made-up example.