Operations¶
Running Snodra is running a stateless server, a set of client services and the stores they use. This section is the operator's view: what to watch, how to test, how to tune, and how to upgrade.
| Task | Page | First question |
|---|---|---|
| Know whether it is healthy | Monitoring | Are the server, the services and the stores up? |
| Prove a change works | Testing | Does the handler, the store or the client pass its kit? |
| Handle load and deadlines | Tuning | Which queue, permit or timeout is the limit? |
| Move to a new release | Upgrades | Is this a rolling restart or a coordinated window? |
The operational surface¶
- Server: gRPC
7000, HTTP7001, the dashboard at/ui/, the health service, JSON logs on stdout, OTLP metrics and traces, and the--health-checkprobe. - Services: the library's hub-client and processor metrics, plus the service's own telemetry.
- Stores: the broker's lag and disk, the database's transactions, indexes and retention.
- Tools: the dashboard for the live view, the event browser for what happened, and the conformance runner for a client library.
Backups and blast radius¶
The server keeps no state. Back up the message store (outbox, markers, sagas) and the event store's retention policy, not the server. Losing a server loses only the in-memory registry: clients reconnect and re-register. Losing the message store loses duplicate suppression and saga state; losing the event store loses the events the broker has not retained.
The browser's operational detail is in Server operations; the settings behind it are in Server settings.