Check readiness

Returns 200 when the orchestrator can serve traffic, 503 otherwise. Checks the state store with a bounded ListVMs call (default 2s timeout) and verifies the FleetManager is initialized. The 503 body includes a per-check breakdown so operators can see which dependency is unhealthy. Mounted outside auth — k8s readinessProbe does not carry credentials.

Failure messages are clipped to the first 200 bytes, with the literal marker …(truncated) appended — so a truncated check value is 214 bytes, not 200. Truncation keeps driver errors and stack traces from blowing out the probe body. Use this endpoint — not /health — for a load balancer’s routing decision; /health only reports that the process is up, not that it can serve a request end to end.

When restate is configured (FUSE_RESTATE_MODE is not off), a restate check reports whether restate ingress answers. Restate being down is degraded with a 200, not a 503: nothing runs on restate yet, so it must not take the orchestrator out of rotation.

curl -X GET "http://localhost:8080/ready"

const response = await fetch("http://localhost:8080/ready", {
  method: "GET"
});

import requests

response = requests.get(
    "http://localhost:8080/ready",
)

{
  "status": "ready",
  "checks": {
    "state_store": "ok",
    "fleet": "ok",
    "restate": "ok"
  }
}

{
  "status": "not_ready",
  "checks": {
    "state_store": "connection refused",
    "fleet": "ok"
  }
}