manage/docs/observability-runbooks.md

# Observability Runbooks

Operational playbooks for the Manage self-hosted observability stack (Prometheus, Grafana, Loki, Alertmanager).

## Service Overview

| Service | Compose name | Internal URL | Health check |
|---------|--------------|--------------|--------------|
| Prometheus | `prometheus` | `http://prometheus:9090` | `/-/healthy` |
| Grafana | `grafana` | `http://grafana:3000` | `/api/health` |
| Loki | `loki` | `http://loki:3100` | `/ready` |
| Alloy | `alloy` | `http://alloy:12345` | `/-/healthy` |
| Alertmanager | `alertmanager` | `http://alertmanager:9093` | `/-/healthy` |
| Node Exporter | `node-exporter` | `http://node-exporter:9100` | `/` |
| Manage backend | `backend` | `http://backend:8000` | `/api/health` |

---

## Alert: `BackupJobFailed`

**Severity**: critical  
**Meaning**: A backup job reported `status=failed` within the last hour.

### Steps

1. Open **Manage → Backups** and identify the failed job/run.
2. Check the run output / logs for the failure reason.
3. Search Loki for `{container="backend"} | json | message=~"(?i)backup"` around the failure time.
4. If transient (network, lock file), retry the job.
5. If persistent, open a task to fix the backup script or credentials.

---

## Alert: `BackupJobStuck`

**Severity**: warning  
**Meaning**: No successful backup run has been recorded for a job in the last 24 hours.

### Steps

1. Confirm the job is still scheduled and expected to run.
2. Check whether the backup scheduler/host is running.
3. Verify the job can still report success to `POST /api/backups/reports`.
4. Inspect Prometheus graph for `manage_backup_runs_last_success_timestamp` by `job_name`.
5. If the job was intentionally retired, remove or disable its reporting.

---

## Alert: `PrometheusTargetMissing`

**Severity**: warning  
**Meaning**: A Prometheus scrape target is down (`up == 0`) for more than 2 minutes.

### Steps

1. Identify `job` and `instance` from the alert labels.
2. Check the container/process status:
   - `docker compose ps <service>`
   - `docker compose logs --tail 100 <service>`
3. Verify network reachability from the Prometheus container:
   - `docker compose exec prometheus wget -qO- http://<instance>/`
4. If the target is a remote Node Exporter:
   - Check the machine is reachable over SSH.
   - Verify Node Exporter is installed and running (`systemctl status node_exporter`).
   - Confirm the scrape host/port in Manage → Settings for that machine.
5. Restart if needed: `docker compose restart <service>`.

---

## Alert: `AlertmanagerDown`

**Severity**: critical  
**Meaning**: Prometheus cannot scrape Alertmanager; new alerts may not be delivered.

### Steps

1. Check container status: `docker compose ps alertmanager`
2. Review logs: `docker compose logs --tail 200 alertmanager`
3. Validate config syntax:
   - `docker compose exec alertmanager amtool check-config /etc/alertmanager/alertmanager.yml`
4. Verify SMTP environment variables are present if using email receivers.
5. Restart: `docker compose restart alertmanager`

---

## Alert: `GrafanaDown`

**Severity**: warning  
**Meaning**: Grafana is unreachable; dashboards and iframe panels in Manage are unavailable.

### Steps

1. Check container status and logs.
2. Verify the OAuth client configuration is correct (`GF_AUTH_GENERIC_OAUTH_*`).
3. If embedded panels are blank, confirm Grafana `allow_embedding = true` and cookie settings.
4. Restart: `docker compose restart grafana`

---

## Routine Maintenance

### Check overall health

```bash
cd /path/to/manage
docker compose ps
docker compose exec prometheus wget -qO- http://127.0.0.1:9090/-/healthy
docker compose exec grafana wget -qO- http://127.0.0.1:3000/api/health
docker compose exec loki wget -qO- http://127.0.0.1:3100/ready
docker compose exec alertmanager wget -qO- http://127.0.0.1:9093/-/healthy
```

### Reload Prometheus after rule/config changes

Prometheus is started with `--web.enable-lifecycle`, so a SIGHUP or HTTP call reloads config:

```bash
curl -X POST http://localhost:9090/-/reload
```

### Inspect logs

```bash
# All backend logs in Loki via Grafana Explore, or locally:
docker compose logs --tail 500 backend

# Specific service:
docker compose logs -f prometheus
```

### Storage usage

```bash
docker system df -v
docker compose exec prometheus du -sh /prometheus
docker compose exec loki du -sh /loki
docker compose exec grafana du -sh /var/lib/grafana
```

---

## Backup and Disaster Recovery

The observability data lives in named volumes:

- `prometheus_data`
- `loki_data`
- `grafana_data`
- `alertmanager_data`

### Backup volumes

```bash
# Stop the stack to ensure consistency
docker compose down

# Back up each volume to a tarball
docker run --rm -v manage_prometheus_data:/data -v $(pwd)/backups:/backups alpine \
  tar czf /backups/prometheus-$(date +%F).tar.gz -C /data .
docker run --rm -v manage_loki_data:/data -v $(pwd)/backups:/backups alpine \
  tar czf /backups/loki-$(date +%F).tar.gz -C /data .
docker run --rm -v manage_grafana_data:/data -v $(pwd)/backups:/backups alpine \
  tar czf /backups/grafana-$(date +%F).tar.gz -C /data .
docker run --rm -v manage_alertmanager_data:/data -v $(pwd)/backups:/backups alpine \
  tar czf /backups/alertmanager-$(date +%F).tar.gz -C /data .

# Start the stack again
docker compose up -d
```

> Replace `manage_` with your actual Docker Compose project name if different.

### Restore a volume

```bash
docker compose down
docker volume rm manage_prometheus_data
docker volume create manage_prometheus_data
docker run --rm -v manage_prometheus_data:/data -v $(pwd)/backups:/backups alpine \
  tar xzf /backups/prometheus-YYYY-MM-DD.tar.gz -C /data
docker compose up -d
```

---

## Scaling Notes

- The current `deploy.resources` blocks are tuned for a small homelab. Raise memory limits if you monitor many machines or retain logs longer than 30 days.
- Loki is configured for single-node filesystem storage. For larger deployments, migrate to object storage (S3/GCS/MinIO) and a shared index.
- Prometheus remote-write or Thanos/Cortex can be added later for long-term metrics without changing application instrumentation.