CXB Core exposes two health endpoints: one per-worker and one fleet-wide aggregated view.
Per-worker health
Returns the calling status of the individual worker that handles the request.
{
"status": "ok",
"worker_calls": 0,
"worker_max": 1,
"total_capacity": 16,
"total_available": 16,
"workers": 16,
"worker_id": "3",
"worker_pid": 12345
}
| Field | Type | Description |
|---|
status | string | Always "ok" if the worker is responding |
worker_calls | int | Active calls on this worker (0 or 1) |
worker_max | int | MAX_CONCURRENT_CALLS setting (always 1) |
total_capacity | int | worker_max × workers (fleet total) |
total_available | int | total_capacity - worker_calls (estimated) |
workers | int | CXBCORE_WORKERS setting |
worker_id | string | Worker number (derived from socket path) |
worker_pid | int | OS process ID |
When called through nginx, total_available is an estimate from this worker’s perspective. Use /health/fleet for accurate totals.
Fleet health
Queries all workers on the local server over their Unix sockets and aggregates their status.
{
"status": "ok",
"fleet_calls": 3,
"fleet_max": 16,
"fleet_available": 13,
"workers": [
{"id": "1", "calls": 1, "max": 1, "pid": 12345},
{"id": "2", "calls": 0, "max": 1, "pid": 12346},
{"id": "3", "calls": 1, "max": 1, "pid": 12347},
{"id": "4", "calls": 0, "max": 1, "pid": 12348},
...
]
}
| Field | Type | Description |
|---|
fleet_calls | int | Total active calls across all workers |
fleet_max | int | Total capacity across all workers |
fleet_available | int | Available call slots |
workers | list | Per-worker breakdown |
{"id": "N", "status": "unreachable"}.
The fleet_available value is parsed by the HAProxy WSS ingress capacity health check, which matches it against the PCRE2 pattern "fleet_available":([1-9]|1[0-6])[,}] (bounded to 1-16). Two coupling hazards:
- Changing the
/health/fleet JSON shape (renaming the field, reordering keys, or altering how the number is serialized) can silently break the ingress check, marking healthy fleets DOWN.
- Scaling a fleet past 16 workers makes
fleet_available exceed 16, the pattern stops matching, and the higher-capacity fleet is marked DOWN — during the exact “add capacity” operation. When you raise maxconn past 16, raise this upper bound in the HAProxy template and README in the same change, and re-validate against a captured body.
Monitoring commands
Fleet-wide view (from anywhere):
curl -s https://fleet.example.com/health/fleet | python3 -m json.tool
curl -s https://fleet.example.com/health
