Skip to main content

Metrics

Overview​

Marmot collects various application metrics for both Prometheus monitoring and built-in dashboards in the UI.

Prometheus/OpenMetrics Endpoints​

You can enable metrics in the configuration to expose a Prometheus/OpenMetrics endpoint on /metrics. This endpoint does not have auth enabled, you should configure Prometheus to scrape the endpoints for each Marmot instance you have deployed.

values.yaml:

metrics:
  enabled: true
  port: 9090

Environment variables:

MARMOT_METRICS_ENABLED=true
MARMOT_METRICS_PORT=9090

Helm Chart​

The Helm chart creates a ServiceMonitor for Prometheus Operator:

helm install marmot ./chart --set config.metrics.enabled=true --set monitoring.serviceMonitor.enabled=true
config:
  metrics:
    enabled: true
    port: 9090

monitoring:
  serviceMonitor:
    enabled: true
    interval: 30s

Endpoints​

  • /metrics - Prometheus endpoint (no auth)
  • /api/v1/metrics - UI dashboard API (requires auth)

Implementation Details​

Storage Architecture​

The reason for 2 methods of storage is due to the seperation between opertional metrics and analytical metrics. There's some overlap, hence the need for both a Prometheus endpoint and storing data into the database.

You should use the Prometheus endpoint for monitoring the health of your Marmot instance and alerts. The UI and API is designed for analytical monitoring of your data estate.

  • Raw Metrics Table: metrics_raw with daily partitioning, 7-day retention
  • Aggregated Metrics Table: metrics_aggregated with pre-computed summaries, 6-month retention

Raw metrics are inserted immediately on collection. Aggregation runs asynchronously in background goroutines.

Aggregation Pipeline​

  1. 5-minute aggregation: Runs every 5 minutes, processes data from 10-5 minutes ago
  2. Hourly aggregation: Runs at :05 past each hour, aggregates previous full hour
  3. Daily aggregation: Runs at 2:05 AM, aggregates previous full day

Query Optimization​

API queries (/api/v1/metrics) use aggregated data:

  • Queries < 2 hours: Use 5-minute buckets
  • Queries 2-48 hours: Use hourly buckets
  • Queries > 48 hours: Use daily buckets

Prometheus queries (/metrics) serve real-time counters and gauges directly from application memory.