Health & Readiness

Conduit exposes two distinct HTTP probes, following the standard liveness/readiness split used by orchestrators such as Kubernetes:

• /healthz (liveness) — is the Conduit process alive and able to reach its state store? Use it to decide whether to restart the process.
• /readyz (readiness) — is the engine finished starting up and able to serve? Use it to decide whether to route traffic.

Both are served on the HTTP API address (:8080 by default; see Configuration).

Liveness — /healthz

The liveness check verifies that Conduit can successfully connect to the database it was configured with (BadgerDB, PostgreSQL, SQLite, or the in-memory one). It's available at the /healthz path:

$ curl "http://localhost:8080/healthz"
{"status":"SERVING"}

You can also check an individual service within Conduit. The following example checks whether the PipelineService is running:

$ curl "http://localhost:8080/healthz?service=PipelineService"
{"status":"SERVING"}

The services that can be checked for health are PipelineService, ConnectorService, ProcessorService, and PluginService.

Readiness — /readyz

The readiness check reports whether the engine can serve requests. It returns:

• 503 Service Unavailable with {"status":"starting"} while the engine is still starting up (services initializing, pipelines provisioning, API coming online).
• 503 Service Unavailable with {"status":"unavailable","reason":"…"} if the engine has started but its state store is unreachable.
• 200 OK with {"status":"ready", …} once the engine can serve.

$ curl "http://localhost:8080/readyz"
{
  "status": "ready",
  "pipelines": {
    "total": 3,
    "running": 2,
    "degraded": 1,
    "degradedPipelines": [
      {"id": "pg-to-kafka", "status": "degraded", "error": "source connector error"}
    ]
  }
}

note

A degraded pipeline does not make Conduit "not ready." The engine can still serve requests, so degraded pipelines are reported in the response body rather than flipping the probe to 503. Watch them through the pipelines block above (and through metrics); use /readyz only to decide whether the engine itself can accept traffic.

Using the probes with Kubernetes

Map the probes directly onto the Kubernetes probe types:

livenessProbe:
  httpGet:
    path: /healthz
    port: 8080
readinessProbe:
  httpGet:
    path: /readyz
    port: 8080

/healthz decides restarts; /readyz decides whether traffic is routed to the pod. Because a degraded pipeline keeps /readyz at 200, a single failing pipeline won't pull the whole instance out of service.

Metrics

Prometheus metrics are exposed at /metrics on the same HTTP address. See Metrics for the full list.