Documentation overview
A public overview of what the bundle is and how you run it. The full guides (installation, configuration, alerts, dashboards, notifications, backup/restore, upgrades, troubleshooting, security, and per-edition walkthroughs) ship inside the download.
Architecture
A base docker compose stack with profiles. Core runs Grafana,
Alertmanager, node-exporter, and cAdvisor. You add exactly one metrics backend —
Prometheus or VictoriaMetrics — and optional logs
(Loki + Grafana Alloy), blackbox probing, gpu,
image-age (image-age exporter + read-only socket proxy), and an optional
ai incident-summary profile.
Both backends speak PromQL, and Grafana's datasource is a variable, so the 9 dashboards
and 65 alert rules work unchanged on either. VictoriaMetrics scrapes the same config;
vmalert evaluates the same rules.
Requirements & compatibility
| Docker Engine | 20.10+ (24.x recommended) |
| Docker Compose | v2 (required) — `docker compose`, not `docker-compose` |
| Operating system | Linux (Ubuntu 22.04/24.04 LTS tested); any modern systemd distro |
| Metrics backend | Prometheus v3.13.x OR VictoriaMetrics v1.139.x |
| Architecture | x86_64 / arm64 (GPU edition: x86_64 + NVIDIA driver + Container Toolkit) |
| Grafana | 13.x (provisioned automatically) |
Default ports
Bound to 127.0.0.1 by default — reach Grafana over an SSH tunnel or a reverse proxy you control; never expose the metrics backend publicly.
Edition comparison
| Edition | License | What it monitors |
|---|---|---|
| single-docker-host | Starter+ | One host: node-exporter, cAdvisor, daemon checks, certs, disk, image age. |
| docker-compose-application | Professional+ | Group by Compose project, expected services, HTTP/TCP probes, per-service dashboards. |
| web-hosting-server | Professional+ | HTTP/HTTPS uptime, response time, status codes, TLS expiry, log error rates. |
| gitlab-runners | Team | Runner availability, job concurrency/failures, cache & disk (official runner metrics). |
| gpu-ai-workloads | Team | NVIDIA DCGM: utilization, memory, temp, power, ECC/XID, AI-service health. |
| small-openstack | Team | Control/compute hosts, Kolla containers, API probes, RabbitMQ/MariaDB/HAProxy. |
Installation workflow
- Unzip the bundle on a Linux host with Docker + Compose v2.
./scripts/configure.sh— choose backend + edition, generate.env../scripts/install.sh --edition <name> --backend <prometheus|victoriametrics>— validates the host, pulls pinned images, starts, and health-checks.- Add hosts/endpoints under
config/prometheus/targets/(file_sd, auto-reload), then./scripts/validate.sh. - Open Grafana and wire notifications (
docs/notifications.md).
Upgrade model
Images are pinned; a new release bumps the tags. ./scripts/upgrade.sh backs up
first, pulls the new pinned images, applies, validates, and auto-rolls-back
from the backup if validation fails. Your .env and any compose.override.yml
are never overwritten.
FAQ
Can I switch backends later? Yes — dashboards and alerts are backend-agnostic.
Do I need Kubernetes? No. It's plain Docker Compose v2.
Where do secrets live? Only in .env, which is never committed and is excluded from the release archive and the (redacted) support bundle.
Ready to deploy it?
Get the bundle — from $19