Telemetry Model

Agent metrics include CPU, memory, process counts, uptime, load, network and disk summaries, plus host metadata.

Dashboard pressure lanes use latest per-server samples and freshness windows for online/offline classification.

Current agents report runtime and delivery health, including prior submission outcomes, failure streaks, queue mode, and dropped-sample counters.

Agent health contract v1 reports heartbeat, disk, network, Docker, logs, SMART, and package collector cadence, duration, status, and bounded errors. Configuration reporting identifies argument, environment, or default origins without returning values or credentials.

The authenticated agent health read model derives healthy, degraded, offline, and unknown states from the latest heartbeat, delivery failures, and collector freshness.

Current metric cards retain field-level provenance. CPU, memory, process, and load values show their selected Agent or SNMP source, exact sample time, and stale state instead of inheriting one page-wide label.

SNMP ingestion now persists enriched hardware snapshot fields including discovered hostname, processor details, and disk device snapshots.

Server hostname backfill can occur from discovered SNMP hostname for generated placeholders when safe.

Traffic rates are derived from adjacent successful octet counters. Reboot and reset intervals are excluded, while detected 32-bit wraps are corrected before rate calculation.

Server charts mark SNMP counter wraps and resets at the affected poll. Hybrid charts mark Agent-to-SNMP field fallback at the selected SNMP sample.

• Use dashboard triage lanes for priority ordering.
• Use server detail pages for protocol-specific diagnostics.
• Use notification policy with transition-based delivery for incident routing.