Historic AtlasOBSERVABILITY / FIELD GUIDE ↘
SIGNALS WITHOUT A SINGLE POINT OF FAILURE

Observe everything.
Depend on less.

The monitoring stack may lose history. It must never take the application down with it.

An interactive architecture guide. Every signal on this page is synthetic.

The application

Own Compose project
No monitoring dependency

ServicesHealth + metrics endpoints
→
stdoutRotated Docker json-file logs
→
docker logsAvailable independently

The observers

Separate user + storage
Private access only

PrometheusPulls metrics privately
+
Alloy → LokiReads log files, not socket
→
GrafanaAdmins on private network
FAILURE REHEARSAL / LOCAL SIMULATION

Break the observer.
Not the application.

Backend readiness includes its database dependency, never Prometheus or Loki. These toggles do not touch a real service.

History is the trade-off.

Synthetic 12-sample series · illustrative units
EARLIERNOW →

THE FIRST-VERSION SIGNAL CONTRACT

The right signal.
At the right boundary.

Choose a component to inspect its responsibilities, rather than a made-up production status.

JSON lines use UTC timestamp, level, service, request_id, optional user_id and message. INFO in stage/prod; DEBUG may be used in dev. No credentials or personal data beyond a user ID.

SPECIMEN / NOT COLLECTED TELEMETRY

KEEP THE WRITE PATH SMALL

Collect. Don't intercept.

Services write stdout; Docker rotates json-file logs. Alloy mounts log files read-only, not the root-equivalent Docker socket. No Loki logging driver. If Loki is down, Alloy retries then drops lines; application writes and docker logs remain unaffected.

PRIVATE BY DESIGN

Pull, don't publish.

Prometheus joins the metrics-only network. No monitoring port is public. Backend and Keycloak use management ports; nginx and database have exporters. Only dev/stage expose backend /actuator/ through ingress. Production does not.

KNOW WHAT'S NOT HERE

Later is a real boundary.

Distributed traces, browser telemetry and alert routing are future work. Identity events stay in Keycloak's event store and metrics, not logs. Audit lines carry audit=true for separate retention. Retention periods, volume sizing and stack deploy tooling remain open.

Separate monitoring storage preserves history across application releases. Optional developer monitoring starts manually. This static page creates no containers, collectors, dashboards, traces or network connections.