[relay] Phase 5 — observability: Grafana dashboards, failure alerts, end-to-end lag

[Dipraise1]

Context

The relay exposes relay_client.status() (dlq_pending, dlq_abandoned, total_ok, last_ok_hash) and logs submissions to SQLite, but nothing is visualized or alerting.

Tasks

• Grafana dashboard: relay latency, submission success rate, payload volume
• Alert on 3+ consecutive relay failures
• Track XERIS acknowledgement vs relay submission timestamp (end-to-end lag)

Notes

Validator runs on the VPS (/opt/engram); a Prometheus exporter or simple pushgateway off the SQLite relay log is probably the lightest path.

From TASKS.md — Relay Phase 5.