Service and cluster status
Connect real monitoring sources to display cluster health and load.
View service status
The public status page has two separate views:
- Global network: explore 58 city nodes on an interactive globe, filter by region, and browse the node directory, 120-second trend and request stream.
- Service status: uses the same real monitoring endpoint as the console. Without a connection, it reports that monitoring is not connected. Model-generated values never fill gaps in real monitoring.
Valid monitoring configuration selects Service status by default; otherwise Global network opens. A monitoring failure does not automatically switch to the global network's visualization metrics.
Visualization endpoint
GET /api/platform/status-demo returns a separate snapshot with source: "demo", generatedAt, qpm, chars24h, okRate, avgLatency, p95, nodes, history, and feed. Request events contain synthetic language codes, scenarios, character counts, and durations; they contain no user content.
The model adapts the legacy site's daily demand curves and load relationships. A fixed timestamp produces the same data. Visitors share a server snapshot for five seconds, so refreshing the page does not randomly reset metrics. History contains 120 one-second samples; the feed contains 14 synthetic events with ISO timestamps.
The browser refreshes every five seconds while in the foreground. Fetch failures are identified, and retained snapshots are not represented as current. Globe motion respects reduced-motion preferences. The endpoint does not contact the legacy server, access monitoring credentials, or fetch external URLs.
Monitoring-source protocol
An administrator configures STATUS_TARGETS_JSON on the server. Each item contains id, name, region, an HTTPS endpoint, and an optional tokenEnv, which names the environment variable containing the access token. Up to 12 unique monitoring targets are supported.
[
{
"id": "asr-primary",
"name": "Primary ASR cluster",
"region": "East China",
"endpoint": "https://your-monitoring.example/elejo/asr",
"tokenEnv": "ASR_MONITOR_TOKEN"
}
]Monitoring endpoints should return this format. These example values illustrate the protocol; they are not current server measurements:
{
"status": "operational",
"metrics": {
"healthyNodes": 3,
"totalNodes": 3,
"queueDepth": 0,
"utilization": 42
}
}status accepts operational, degraded, or outage. Metrics are optional; omitted values appear as “—”. Healthy nodes cannot exceed total nodes. Utilization must be between 0 and 100.
Public read endpoint
GET /api/platform/status returns normalized service status, hiding internal monitoring URLs, credentials, and raw responses. Public names and regions should be service-level labels without sensitive host information.
Each monitoring request has a 3-second timeout, rejects redirects, and limits the response to 8 KiB. Server-side caching lasts up to 30 seconds, and the frontend refreshes every 60 seconds. Monitoring errors or unreachable endpoints produce unknown; this alone does not establish that the service cluster is down.
Connect Prometheus or another monitoring system
Deploy an internal adapter endpoint that maps your existing metrics to this protocol. Do not expose a Prometheus query interface or operations credentials directly. Only trusted deployment administrators may configure target addresses; public user input must not supply URLs.
This version provides current status and metrics, not historical uptime or SLA statistics. The displayed monitoring response time is not model-inference latency.
