Runtime Monitoring Dashboard

NFWeb Runtime Monitoring is a read-only dashboard for the current NORFAB fabric. It shows the broker, every worker known to the broker, and the local NFWeb client without deploying another service or telemetry system.

What It Shows

The dashboard provides:

• current broker, worker, and local-client health;
• compact broker and local-client CPU and resident-memory summaries;
• aligned worker CPU and resident-memory comparisons, ranked on every sample by CPU plus memory relative to the largest worker;
• stacked fabric-wide CPU and resident-memory time series that show the broker and every worker's contribution to the combined totals;
• separate CPU and memory trends for the selected worker;
• broker worker and service totals;
• worker keepalive transmit/receive totals and current hold time;
• NFWeb client outbound-queue state;
• local client job-database totals, last-24-hour volume, event count, average completion time, and stacked job-status activity over time;
• a full-width database summary for the selected worker, including recent job, status, and task counts returned by that worker's existing job_list task.

The browser receives new samples over a WebSocket. The Refresh button requests an immediate sample; if shared collection is already running, the request reuses the latest completed sample instead of returning a conflict. A dashboard selector reads the latest shared sample every 5, 10, 30, or 60 seconds without starting a second collection. Periodic server collection remains shared by all browser tabs. Worker comparison charts show ten workers at a time and share one vertical scroll/zoom position when more workers are available.

The job activity chart compares each retained database snapshot with the previous sample and stacks positive increases for FAILED, STARTED, COMPLETED, and STALE. The client database stores current aggregate counts rather than a status transition journal, so decreases are not presented as invented transitions.

Data Sources

Monitoring reuses normal NORFAB interfaces:

Component data	Existing interface
Broker state and process resources	mmi.service.broker / show_broker
Worker registration and keepalives	mmi.service.broker / show_workers
Worker process resources and uptime	get_watchdog_stats on all workers
Local client counters and process resources	The NFWeb process and its native NFPClient
Job and event statistics	The native client's existing job database
Selected-worker job statistics	Existing job_list task on that worker

The monitoring request traffic itself is included in the local client's job and message accounting. The dashboard reads aggregate database statistics; it does not capture or retain message payloads or create a separate monitoring database. Selecting a worker makes one targeted job_list request and summarizes at most the latest 1,000 returned records. The card labels a full window as truncated; it does not present the bounded result as an all-time worker total.

In-memory History

Samples are held in a bounded Python deque for up to 180 minutes. They are not written to SQLite, logs, or another database. Restarting NFWeb clears all monitoring history, which is intentional.

The history length also has a sample-count bound derived from the collection interval. The browser uses the server's retained samples to draw trends and then appends live WebSocket samples.

Configuration

Configure monitoring under client.nfweb.monitoring:

client:
  nfweb:
    monitoring:
      collection_interval: 5
      retention_minutes: 180
      request_timeout: 10

Field	Default	Meaning
collection_interval	5	Seconds between samples; minimum 5 seconds
retention_minutes	180	In-memory history window; maximum 180 minutes
request_timeout	10	Seconds allowed for each NORFAB status request

Unknown settings are rejected.

Failure Behavior

If a worker does not return watchdog data, it remains visible with the registration state provided by the broker and unknown resource values. A sample is marked partial when some calls fail and failed when broker state is unavailable. Unknown values are shown as unknown rather than inferred. Collection warnings are shown in the compact Fabric observability toolbar rather than expanding the dashboard with alert cards. When a worker response is absent or omits CPU or memory data, the warning names that worker and states that it did not respond during the sample interval.

Monitoring is polling-based and read-only, has no persistence, and shares NFWeb's existing security boundary. Restrict NFWeb to a trusted administrative network as described in the NFWeb overview.