Health checks

When the service has no probes

If a task needs a probe and none exists, say what you would add and wait: the Cohort dependency, a -app file registering the two registries, and the two paths. Cohort is a third-party dependency, so taking it is the user's call. Where they would rather not, two hand-written routes reading a background-refreshed state still beat one route that pings the database per request — say that is the trade you made.

Flag a single path serving both liveness and readiness on sight, even when the task is something else. The next section is why.

What a probe is for

An orchestrator makes two different decisions:

• Liveness — is this process wedged beyond recovery? Failing it kills and restarts the container.
• Readiness — can this instance serve traffic right now? Failing it removes the instance from the load balancer, running.