Operate CALIBER
Health and readiness
The operator meaning of CALIBER liveness, readiness, and the deeper evidence surfaces to use when a runtime is up but not actually usable.
Use this page when you need the operational meaning of CALIBER health signals: what basic liveness proves, what readiness proves, and where to go next when a runtime is up but not actually usable.
At a glance
| Signal | What it answers | Use it for |
|---|---|---|
/health | Is the service process alive? | liveness and coarse orchestration checks |
/ajax-api/2.0/mlflow/caliber/readiness | Is the runtime usable with its dependencies? | operator go/no-go decisions |
| traces, logs, metrics, incidents | What happened and why? | diagnosis and recovery |
1. Readiness matters more than liveness
An alive process is not the same thing as a usable CALIBER deployment.
For operator decisions, prefer the readiness surface because it captures whether the surrounding runtime posture is actually suitable for real use.
2. What to verify after boot
After bring-up or configuration changes, verify:
- liveness responds
- readiness responds with the expected posture
- the UI is reachable
- the dependencies you actually need are healthy
3. When health is green but the system still feels broken
Use the deeper evidence surfaces: