Administrator guide
Diagnose a Server
Start at the failed boundary, collect secret-free evidence, and repair only after the instance and failure mode are known.
Do not create a second instance because the first one is confusing. Identify the profile, topology, setup state, and failed boundary before you mutate anything.
Read-only first pass
Terminal
opencircle profile list opencircle status --profile circle-server --format=json opencircle logs --profile circle-server --service opencircle-server
Replace circle-server with your intended profile and compare its instance identity against the deployment receipt. status reports Compose state plus best-effort setup health — it is not a full runtime-acceptance check. Review any output for secrets before sharing it.
Symptom, check, repair
| Symptom | Check | Repair boundary |
|---|---|---|
| Wrong Server or unexpected data | profile current; compare instance ID and endpoint | Select the intended profile — do not redeploy yet |
| Containers run but login is dead | status; inspect setup state and auth-service logs | Recover the identity/setup state; do not create another owner blindly |
| Remote command cannot connect | Inspect the named profile and run the Linux guide's SSH/DNS checks | Repair DNS, SSH trust, path ownership, or Docker at its own authority |
| The Familiar cannot answer | Prove the selected provider and one model-backed reply | Repair provider configuration or routing — container health is not enough |
| Restore or upgrade failed | Preserve the exact bundle and command output | Verify the bundle before any second mutation; follow the recovery branch |
When evidence disagrees, stop. Preserve the current profile, logs, deployment identity, bundle path, and error output — a failed repair that destroys the only useful evidence is worse than a temporarily degraded Server.
What success means
Acceptance means the intended instance, ready setup state, owner sign-in, full app delivery, and one real model-backed reply. A guarded upgrade performs this acceptance pass automatically; a direct restore only health-polls the restarted Server, so run the complete pass yourself afterward rather than trusting health alone.