Graphs

One environment. One file.

Several named machines plus the edges that say who can reach whom. An edge is an authorization, not “the network”.

name: tienda
nodes:
  db:    {from: pg16-golden, ports: [5432], wake: lazy, idle_freeze: 120}
  api:   {from: api-node, ports: [8081], egress: allowlist, allow_domains: [api.stripe.com]}
  web:   {from: web-static, ports: [8000]}
edges:
  - {from: api, to: db, kind: credential, port: 5432, user: app, database: shop, env: PGPASSWORD, secret_env: SHOP_PG_PASS}
  - {from: web, to: api, kind: link, port: 8081}

SHOP_PG_PASS=... kling graph up tienda.yaml

Output

graph tienda (4f2a9c1e0b7d) up in 412ms
  NODE  STATE          MACHINE       WAKE   FROM        PORTS
  api   running        9c1e0b7d4f2a  eager  api-node    8081
  db    (not started)  -             lazy   pg16-golden 5432
  web   running        0b7d4f2a9c1e  eager  web-static  8000

Inside web, http://api.graph:8081 reaches api.
Inside api, db.graph:5432 reaches db with the password filled in by the proxy: api only sees a placeholder.
db is lazy: it does not exist until the first psql.

The whole graph at once

kling graph snapshot tienda -name t0   # one template per node, from the same instant
kling graph fork tienda -n 3           # 3 new graphs that cannot see each other
kling graph freeze tienda ; kling graph thaw tienda
kling graph audit tienda -since 10m    # every connection per edge, on a timeline

Limits: 32 nodes, 64 edges.

In the repo: docs/grafos.md · docs/grafos-diseno.md