GitLab Runner Orchestrator for Hetzner Cloud

Automatic provisioning of Hetzner Cloud servers as GitLab CI runners - pay only when you need it.

Features

• Automatic server creation when pipelines are pending/running
• Cost-optimized deletion - server is deleted 5 minutes before the next billing cycle
• Polls all projects - one runner for the entire GitLab instance
• State persistence - survives restarts without data loss
• CSV logging - documents all server starts/stops with reason and duration
• Rotating log files - daily rotation

Architecture

flowchart TB
    subgraph main [Main Loop]
        Poll[Polling Loop]
        Poll --> FetchProjects[GitLab: Fetch all projects]
        FetchProjects --> CheckPipelines[Per project: Pending/Running pipelines?]
        CheckPipelines -->|Yes + No server| LogStart[CSV: Log reason]
        LogStart --> CreateServer[Hetzner: Create server]
        CheckPipelines -->|No everywhere + Server exists| CheckDelete[Check delete logic]
        CheckDelete -->|Conditions met| LogStop[CSV: Log usage]
        LogStop --> DeleteServer[Hetzner: Delete server]
    end

    subgraph config [Config]
        ConfigToml[config/config.toml]
        RunnerToml[config/runner.toml]
    end

    subgraph logs [Logging]
        CSV[logs/runner_usage.csv]
        LogFile[logs/orchestrator.log]
    end

    ConfigToml --> main
    RunnerToml --> CreateServer
    main --> CSV
    main --> LogFile

Quick Start

1. Build the binary

cargo build --release

2. Create configuration

On first start, config/config.example.toml is automatically created. Copy and customize it:

cp config/config.example.toml config/config.toml
# Edit config/config.toml with your API keys

3. Create runner configuration

Create config/runner.toml with your GitLab Runner configuration. You can register the GitLab Runner with the following command:

docker run -it -v /var/run/docker.sock:/var/run/docker.sock -v ./config:/etc/gitlab-runner gitlab/gitlab-runner:latest register --url https://gitlab.example.com --token YOUR_TOKEN

IMPORTANT: Under [runners.docker] there is a pull_policy setting. Set it to:

pull_policy = ["if-not-present"]

Otherwise the runner will do too many docker image pull requests and your IP will be banned!

4. Start

cargo run --release
# or
./target/release/hetzner_gitlab_runner

Configuration

config/config.toml

[gitlab]
url = "https://gitlab.example.com"
token = "glpat-xxxxxxxxxxxxxxxxxxxx"  # read_api scope

[hetzner]
token = "xxxxxxxxxxxxxxxxxxxxxxxx"
server_type = "ccx23"      # AMD dedicated CPU
location = "nbg1"          # Nuremberg
image = "ubuntu-24.04"
ssh_key_name = "my-ssh-key"

[runner]
name = "flexi-runner"
min_lifetime_minutes = 20  # Minimum runtime
poll_interval_seconds = 30

Debug vs Release

Billing Optimization

Hetzner charges per started hour from server creation.

Example:

• Server created at 14:47
• Pipeline finished at 15:10 (server ran 23min)
• Next billing cycle: 15:47
• Server is deleted at 15:42 (5min buffer) This ensures that if the server is needed again, the runner doesn't have to be set up again. Hetzner charges per started hour, meaning 5x10min = 5x60min billed.

Logs

• logs/orchestrator.log - Detailed logs (daily rotation)
• logs/runner_usage.csv - Server usage documentation

CSV Format

timestamp,event,server_id,project,pipeline_id,reason,duration_minutes
2026-01-14T10:30:00Z,START,12345678,mygroup/myproject,9876,pipeline_pending,
2026-01-14T11:15:00Z,STOP,12345678,,,all_pipelines_done,45