Skip to main content
Version: 5.0.x 🚧

Grafana Integration

MTE Relay v5 emits structured JSON logs (see the Event Schema Reference). Two prebuilt Grafana dashboards are provided, one per common log backend.

v4 dashboards

The dashboards under /grafana/aws and /grafana/azure target MTE Relay v4 field names and do not work with v5 logs. Use the v5 dashboards below.

Artifacts​

FileBackend
mte-relay-v5-cloudwatch-dashboard.jsonCloudWatch Logs Insights (AWS deployments)
mte-relay-v5-loki-dashboard.jsonGrafana Loki (Kubernetes/on-prem deployments)

Both include: requests/sec, latency percentiles, the decrypt/validate/proxy/encrypt phase breakdown, status codes, security events by type, high/critical security event feeds, and error/panic feeds.

CloudWatch (AWS)​

  1. Ship relay container logs to CloudWatch (ECS awslogs driver or FireLens → CloudWatch).
  2. In Grafana (e.g. Amazon Managed Grafana), add a CloudWatch data source with Logs Insights permissions (AWS instructions).
  3. Dashboards → New → Import, upload mte-relay-v5-cloudwatch-dashboard.json.
  4. Set the region and loggroup variables (defaults: us-east-1, /ecs/mte-relay).

Loki (Kubernetes / on-prem)​

  1. Ship container logs to Loki with Promtail, Grafana Alloy, or Fluent Bit.
  2. Add the Loki data source in Grafana.
  3. Dashboards → New → Import, upload mte-relay-v5-loki-dashboard.json.
  4. Set the container variable to the label value your collector assigns the relay container (default regex mte-relay.*). If your setup labels streams differently (e.g. app or job instead of container), edit the stream selector in the panel queries once — everything else is derived via | json.

Alerting​

Grafana-managed alert rules can be built directly on the dashboard queries. Recommended starters (mirroring the Datadog/New Relic alert sets):

  • Any security_event.severity in high|critical — 5 min window
  • Any error_event.event_type = panic — 5 min window
  • state_save_failed / state_not_found count > 5 — 10 min window
  • Avg performance_event.total_time_ms > 500 — 10 min window (tune)
  • client_pairs_exhausted count > 50 — 15 min window (tune)

Cost control​

Per-request performance events dominate log volume. On high-traffic deployments set PERF_EVENT_SAMPLE_PERCENT=10 on the relay; dashboard rates and percentiles remain statistically sound.