Services and ports
One table with every Service: name, port, whether it needs the public network, and which path its probe hits.
All Services are ClusterIP, the external port is uniformly 80, and the container port is uniformly 3000 — the only exception is
tethys-shipping (.NET, defaults to 8080, see below). The host in configuration is always a Service name,
and port is always 80 (across namespaces, use <service>.<tenant>.svc.cluster.local).
| Service | Product line | Container port | External | Probe / health path | Stateful |
|---|---|---|---|---|---|
rhea-email |
rhea | 3000 | Cluster-internal only | /api/v1/livez, /api/v1/readyz |
No |
rhea-payment |
rhea | 3000 | Cluster-internal only (via the gateway) | /livez, /readyz |
PostgreSQL |
rhea-push |
rhea | 3000 | Cluster-internal only | /api/v1/livez, /api/v1/readyz |
No |
rhea-gateway |
rhea | 3000 | Public domain | /livez, /readyz |
No |
rhea-web |
rhea | 3000 | Public domain | None (TCP 3000) | No |
themis-auth |
themis | 3000 | Cluster-internal only (proxied by the gateway) | /api/v1/livez, /api/v1/readyz |
PostgreSQL |
themis-gateway |
themis | 3000 | Public domain (accounts.titanides.com) |
/livez, /readyz |
No |
themis-web |
themis | 3000 | Public domain (accounts.titanides.com) |
None (TCP 3000) | No |
thia-agent |
thia | 3000 | Cluster-internal only | /livez, /readyz |
PostgreSQL (pgvector) |
thia-gateway |
thia | 3000 | Public domain | /livez, /readyz |
No |
thia-web |
thia | 3000 | Public domain | None (TCP 3000) | No |
mcp-vision |
thia | 3000 | Cluster-internal only | None (TCP 3000) | No |
mcp-perception |
thia | 3000 | Cluster-internal only | None (TCP 3000) | No |
mcp-motion |
thia | 3000 | Cluster-internal only | None (TCP 3000) | No |
mcp-fetch |
thia | 3000 | Cluster-internal only | /health |
No |
mcp-analysis |
thia | 3000 | Cluster-internal only | /health |
No (read-only database connection) |
tethys-shipping |
tethys | 8080 | Cluster-internal only | None (TCP) | SQLite file (mounted PVC) |
tethys-gateway |
tethys | 3000 | Public domain | /livez, /readyz |
No |
tethys-web |
tethys | 3000 | Public domain | None (TCP 3000) | No |
thia-jobs-tick is not a Service; it is a CronJob: every 5 minutes it POSTs to
http://thia-agent/api/v1/jobs/tick, and it does not listen on any port itself.
Three easy traps
Probe paths are not uniform. All three styles exist, and this is the easiest thing to get wrong when copying YAML:
- Root path:
thia-*,tethys-gateway,themis-gateway,rhea-gateway,rhea-payment; - Under
/api/v1:themis-auth,rhea-email,rhea-push; - None at all: the four Web sites (
thia-web/themis-web/rhea-web/tethys-web) andmcp-vision/mcp-perception/mcp-motion. These use TCP probes; do not writehttpGet /livez— a 404 will make the Pod restart repeatedly while the service itself is fine.
Only two MCPs have a health route. mcp-fetch and mcp-analysis provide GET /health; the other three do not.
The container port of tethys-shipping is not 3000. The .NET image defaults to 8080 (--urls=http://*:${PORT:-8080}),
the Service points 80 at it, and targetPort must be the port the container actually listens on.
External exposure boundary
What needs the public network is the read-only aggregating gateways and sites: thia-gateway, thia-web, themis-gateway, themis-web,
rhea-gateway, rhea-web, plus the optional tethys-gateway, tethys-web.
Everything else carries only ClusterIP, because at the application layer they do no authentication and trust the identity headers minted upstream (thia-agent, mcp-*),
or they have no authentication at all (the admin interfaces of rhea-email, rhea-push, themis-auth). Being able to connect means being able to operate —
a single public Ingress wipes that boundary away.