Configure Health Probes

Konifer exposes separate endpoints for liveness and readiness:

Probe	Endpoint	Use
Liveness	GET /health/live	Restart an instance that cannot serve HTTP requests.
Readiness	GET /health/ready	Remove an instance from traffic until Konifer and its dependencies can serve requests.

Use both probes in an orchestrated deployment. A dependency outage should make an instance unready without triggering a restart loop.

Kubernetes configuration

The following configuration uses the container's HTTP port:

livenessProbe:
  httpGet:
    path: /health/live
    port: 8080
  periodSeconds: 10
  timeoutSeconds: 1
  failureThreshold: 3

readinessProbe:
  httpGet:
    path: /health/ready
    port: 8080
  periodSeconds: 5
  timeoutSeconds: 1
  failureThreshold: 2

Adjust the periods and failure thresholds to match your platform's rollout and traffic-management behavior. Konifer returns both responses with an empty body, so the probe client only needs to inspect the HTTP status.

Liveness behavior

The liveness endpoint returns 200 OK when Konifer's HTTP server can handle the request. It does not query PostgreSQL, S3, or filesystem storage.

Readiness behavior

The readiness endpoint returns 200 OK when every registered health indicator reports healthy. It returns 503 Service Unavailable when any indicator reports unhealthy.

Konifer checks these components:

Component	Check
Application lifecycle	The Konifer application has started and is not stopping.
PostgreSQL	Executes SELECT 1.
In-memory data store	Reports healthy without an external check.
S3 object store	Sends HeadBucket for the effective default bucket.
Filesystem object store	Confirms that the mount path exists, is a directory, and is writable.
In-memory object store	Reports healthy without an external check.

The S3 check covers the default bucket resolved for /. It does not enumerate buckets assigned to specific path configurations. Create and validate every configured bucket as part of deployment.

PostgreSQL and object-store checks run in the background every five seconds with a two-second timeout. Readiness requests read this cached result. A new instance remains unready until its first checks succeed, and a dependency change may take up to one refresh interval to appear.

Konifer logs a warning when a cached indicator first becomes unhealthy and an informational message when it recovers.

See the Health API reference for the API reference.