AgentsNet Ops

Quick green/red banner + per-node table + cross-checks. Refreshes 30s. Open when: first thing each morning + during paged incidents.

User feedback messages delivered to M9WB3CET4NQEJWGQ. Open when: daily morning skim + during incident if users reporting issues.

70+ alert rules (cluster · host · capacity · offline · self-health · coturn · expiry · blackbox). Routes severity=page → @MWopsbot Telegram. Open when: Telegram pages you for details, OR before maintenance to silence noise.

12 dashboards incl. User Traffic, Capacity Forecast, Monitoring Self-Health, Offline Store, Host Resources, Cluster Health, Relay Baseline. Open when: debugging week-over-week trends OR after incident for time-series context. Direct links below ↓

Raw PromQL + scrape targets + alert rule eval. Open when: writing new PromQL OR debugging why an alert fired.

Online connections (total + per region) · messages/hour · new & total unique mailboxes · connection growth rate · route-cache health. Open when: tracking usage trends or after a traffic alert. Aggregate-only (R4 content-blind).

Disk days-to-full (predict_linear) · CPU/mem/disk headroom per node · egress trend vs provisioned specs. Open when: planning scale-up or after a disk/egress alert.

Who watches the watchdog: scrape-target up · blackbox/AM/Grafana up · probe/backup/expiry-check freshness · cert/GTM countdown. Open when: a panel elsewhere shows No-data, or daily.

Offline-message spool (2nd raft group): rows/blobs/bytes · eviction & reject rate · effective TTL · persist latency · orphan sweep. Open when: recipients offline en masse or an OfflineStore* alert fires.

Edge coturn load: STUN binding requests (all 4 edges) · TURN relay allocations + bandwidth (mostly SG#2) · packet drop ratio. Open when: P2P connect failures spike or relay egress looks high. Aggregate-only (R4 content-blind).

Anonymous GoAccess report (volume + geography) over this host's Caddy access log, refreshed every 15 min. Open when: reviewing access trends. No per-visitor tracking.