Sepia Lab Wiki

This is an old revision of the document!

---

Grafana Agent Setup for Lab Nodes Monitoring

### Overview

The grafana_agent Ansible role deploys Grafana Agent (static mode) on Lab Nodes to collect system metrics and forward them to a central Mimir instance.

This setup replaces or coexists with the previous Prometheus + node_exporter approach, providing a lightweight agent that pushes metrics into Mimir for long-term storage and Grafana dashboards.

### Architecture & Data Flow

[ Lab Nodes ]
      │
      ▼
node_exporter (on every lab node)
      │
      ▼
Grafana Agent (on every lab node)
      │
      ▼  (remote_write over HTTPS)
   Mimir (central instance)
      │
      ▼
 Grafana Dashboards

Components explained:

1. node_exporter

1. Installed on every host (`prometheus-node-exporter` on Debian/Ubuntu, `node_exporter` on RedHat via EPEL)
2. Exposes standard host metrics (CPU, memory, disk, network, etc.) on `http://localhost:9100/metrics`

2. Grafana Agent

1. Lightweight Prometheus-compatible agent
2. Scrapes `node_exporter` locally from `localhost:9100`
3. Relabels the instance using the host’s FQDN
4. Forwards all collected metrics to the central Mimir using the `remote_write` protocol

3. Mimir

1. Central long-term metrics storage backend (Prometheus-compatible)
2. Accepts metrics via the remote_write protocol with basic authentication

### Configuration Details

Grafana Agent config (`/etc/grafana-agent.yaml` on target hosts):

- Global scrape interval: `60s` (configurable via `scrape_interval_global`) - Scrapes the `node` job from `localhost:9100` - Applies relabeling:

1. `instance`: `<fqdn>:9100`
2. `nodename`: `<fqdn>`

- Forwards metrics to Mimir with basic auth credentials stored in the secrets repository

Secrets required (in `mimir_password.yml`):

- `mimir_username` - `mimir_password` - Grafana repository URLs (`grafana_apt_repo_key_url`, `grafana_apt_repo_url`, `grafana_rpm_repo_url`, `grafana_rpm_repo_key_url`) - `agent_mimir_url`

### Mimir Endpoint

`https://mimir-jenkins-monitoring.apps.pok.os.sepia.ceph.com/api/v1/push`