homelab

README

homelab

Monorepo for my personal homelab. It contains applications and kubernetes manifests for deployment.

1. Getting started
2. Project structure
3. Third party applications
4. Prometheus exporters
5. Other tools
6. User interfaces
7. External services
8. Cluster upgrades
9. Node maintenance
   1. Crontab
   2. K3s services
   3. Overclocking
   4. Multipath
10. Managed infrastructure
    1. Terraform Providers
    2. Database provisioning
    3. Docker images
11. Environment

Getting started

This assumes you have the following tools:

• go
• kubectl
• kustomize
• make
• buildx
• terraform

To start working:

• Clone the repository
• Install golang-based tools using make install-tools
• Run make to build all binaries

Project structure

• cmd - Entry points to any bespoke applications.
• hack - Node host specific config files and tweaks.
• internal - Packages used throughout the application code.
• manifests - Kubernetes manifests to run all my homelab applications.
• scripts - Bash scripts for working within the repository.
• terraform - Terraform files for managing infrastructure.
• vendor - Vendored third-party code.

Third party applications

Here's a list of third-party applications I'm using alongside my custom applications:

• longhorn - Cloud native distributed block storage for Kubernetes.
• home-assistant - Open source home automation that puts local control and privacy first.
• pihole - A black hole for Internet advertisements.
• traefik - The Cloud Native Application Proxy.
• prometheus - The Prometheus monitoring system and time series database.
• grafana - The open observability platform.
• jaeger - Open source, end-to-end distributed tracing.
• node-exporter - Exporter for machine metrics.
• minio - High Performance, Kubernetes Native Object Storage.
• postgres - The world's most advanced open source database.
• firefly - A free and open source personal finances manager.
• photoprism - Personal Photo Management powered by Go and Google TensorFlow.
• cert-manager - x509 certificate management for Kubernetes.
• docker-registry - A stateless, highly scalable server side application that stores and lets you distribute Docker images
• fluent-bit - Log processor and forwarder

Prometheus exporters

I've implemented several custom prometheus exporters in this repo that power my dashboards, these are:

• coronavirus - Exports UK coronavirus stats as prometheus metrics
• homehub - Exports statistics from my BT HomeHub as prometheus metrics
• pihole - Exports statistics from my pihole as prometheus metrics
• speedtest - Exports speedtest results as prometheus metrics
• weather - Exports current weather data as prometheus metrics
• worldping - Exports world ping times for the local host as prometheus metrics
• home-assistant - Proxies prometheus metrics from a home-assistant server.
• synology - Exports statistics from my NAS drive.
• minecraft - Exports statistics for my Minecraft server.

Other tools

Here are other tools I've implemented for use in the cluster.

• bucket-object-cleaner - Deletes objects in a blob bucket older than a configured age.
• grafana-backup - Copies all dashboards and data sources from grafana and writes them to a MinIO bucket.
• db-backup - A backup utility for databases.
• ftp-backup - Copies all files from a specified path of an FTP server and writes them to a MinIO bucket.

User interfaces

This repo contains a few homemade user interfaces for navigation/overview of the applications running in the cluster.

• directory - A simple YAML configured link page to access different services in the homelab.

• health-dashboard - A simple UI that returns the health check status of custom services using the pkg.dsb.dev flavoured health checks.

External services

These are devices/services that the cluster interacts with, without being directly installed in the cluster.

• Ring - Home security devices, connected via home-assistant
• Tailscale VPN - Used to access the cluster from anywhere
• Synology NAS - Used as the storage backend for minio, primarily used for volume backups
• Phillips Hue - Smart lighting, connected via home-assistant
• Cloudflare - DNS, used to access my applications under the *.homelab.dsb.dev domain.
• Sentry - Cloud-based error monitoring.

Cluster upgrades

Upgrading the k3s cluster itself is managed using Rancher's system-upgrade-controller. It automates upgrading the cluster through the use of a CRD. To perform a cluster upgrade, see the plans directory. Each upgrade is stored in its own directory named using the desired version, when the plan manifests get applied via kustomize jobs will be started by the controller that upgrade the master node, followed by the worker nodes. The upgrade only takes a few minutes and tools like k9s and kubectl will not be able to communicate with the cluster for a small amount of time while the master node upgrades.

Node maintenance

The hack diretory at the root of the repository contains files used on all nodes in the cluster.

Crontab

The crontab file describes scheduled tasks that clear out temporary and old files on the filesystem (/tmp, /var/log etc) and performs package upgrades on a weekly basis. It will also prune container images that are no longer in use.

The crontab file can be deployed to all nodes using the make install-cron-jobs recipe. This command will copy over the contents of the local crontab file to each node via SSH. You need to have used ssh copy-key-id for each node, so you don't get any password prompts.

K3s services

The k3s.service and k3s-agent.service files are the systemd service files that run the service and agent nodes. It sets the cluster to communicate via the Tailscale network and stops k3s from installing traefik. This is because I run traefik 2, whereas k3s comes with 1.7 by default.

Overclocking

The usercfg.txt file is stored at /boot/firmware/usercfg.txt and is used to set overclocking values for the Raspberry Pis. Pretty certain this voids my warranty, so if you're not me and planning on using this repository you should keep that in mind.

See Overclocking options in config.txt for more details on these values.

Multipath

The multipath.conf file is the configuration file for the multipath daemon. It is used to overwrite the built-in configuration table of multipathd. Any line whose first non-white-space character is a '#' is considered a comment line. Empty lines are ignored.

The sole reason for this existing, is to handle an issue with longhorn that I was experiencing.

Managed infrastructure

Some aspects of the homelab are managed using Terraform. These include DNS records via CloudFlare. To plan and apply changes, use the Makefile in the terraform directory. The make plan and make apply recipes will perform changes.

The terraform state is included in this repository. It is encrypted using strongbox, which is installed when using make install-tools.

Terraform Providers

This list contains all terraform providers used in the project:

• cloudflare
• minio
• grafana
• tailscale
• sentry

Database provisioning

New postgres databases can be provisioned using a kubernetes Job resource using the createdb binary included in standard postgres docker images. Below is an example:

apiVersion: batch/v1
kind: Job
metadata:
  name: example-db-init
spec:
  template:
    spec:
      containers:
        - image: postgres:13.1-alpine
          name: createdb
          command:
            - createdb
          env:
            - name: PGHOST
              value: postgres.storage.svc.cluster.local
            - name: PGDATABASE
              value: example
            - name: PGUSER
              valueFrom:
                secretKeyRef:
                  key: postgres.user
                  name: postgres
            - name: PGPASSWORD
              valueFrom:
                secretKeyRef:
                  key: postgres.password
                  name: postgres
      restartPolicy: Never
  backoffLimit: 0

You can view the documentation for the createdb command here.

Docker images

The cluster contains a deployment of the docker registry that is used as a pull-through proxy for any images hosted on hub.docker.com. When referencing images stored in the main library, like postgres or busybox, you use the image reference registry.homelab.dsb.dev/library. Otherwise, you just use the repository/tag combination. This increases the speed at which images are pulled and also helps with docker's recent change to add API request limits.

Environment

• 4 Raspberry Pi 4b (8GB RAM)
• Kubernetes via k3s
• Zebra Bramble Cluster Case from C4 Labs
• 4 SanDisk Ultra 32 GB microSDHC Memory Cards
• 4 SanDisk Ultra Fit 128 GB USB 3.1 Flash Drive USB Drives
• Synology DS115j NAS drive.