Self Host Blocks

Building blocks for self-hosting with battery included.

SHB's (Self Host Blocks) goal is to provide a lower entry-bar for self-hosting. I intend to achieve this by providing building blocks promoting best practices to self-host a wide range of services. Also, the design will be extendable to allow users to add services not provided by SHB.

As far as features and best practices go, I intend to provide, for all services:

• Protection and single sign-on using Keycloak, where sensible.
• Automated backup of data and databases with Borgmatic.
• Encrypted external backup with Rclone.
• Central logging, monitoring and dashboards with Prometheus and Grafana.
• Integration with external services that are hard to self-host, like email sending.
• Deployment of every services on the same or different machines.
• Home dashboard with Dashy.
• Vault to store passwords and api keys using Password Store, those shouldn't be stored in config or on disk.
• Test changes using local virtual machines to avoid botching prod.
• Automated CI tests that can be run locally using virtual machines.

Implementation is made with the disnix suite - Disnix, Dysnomia, NixOps - built on top of the nix ecosystem.

Progress Status

Currently, this repo is WIP and the first two services I intend to provide are Tiny Tiny RSS and Vaultwarden. Vaultwarden was chosen as it's IMO the first stepping stone to enable self-hosting. Tiny Tiny RSS was chosen because it is somewhat lightweight.

• Haproxy
  • Systemd service
• Keycloak
  • Provision using keycloak-cli-config
  • Behind haproxy
  • Integration tests
    • Check DB is setup correctly
    • Make a curl request to assert service is up
    • Provision a user and attempt login
  • Backup
• Grafana/Alertmanager/Prometheus
  • Systemd service
  • Behind haproxy
  • Behind keycloak with oauth2proxy
  • Integration tests
  • Backup
• Vaultwarden
  • Systemd service
  • Behind haproxy
    • Under vaultwarden subdomain by default
  • Behind keycloak with oauth2proxy
    • /admin path only allowed for admins
    • /api not protected
    • rest is allowed for any authenticated user
  • Integration tests
    • Assert endpoints are correctly protected
  • Backup
  • Dashboard with Grafana
  • Alerts with Alertmanager
• TTRSS
  • Systemd service
  • Behind haproxy
  • Behind keycloak with oauth2proxy
  • Integration tests
  • Backup
  • Dashboard with Grafana
  • Alerts with Alertmanager

Some other "dev" oriented TODOs can be found at the end of the README.

Getting Started

If you know your way around disnix, feel free to skip this section. If not and you'd rather read the disnix manual instead of reading my rambling, that's probably a good idea too.

Hey, you're still here! Let's start.

First, you need at least one deploy target where NixOS is installed. I'll refer you to the official guide for how to do this. You can install on a could machine or a self-hosted server.

Second, you need a machine where Nix is installed, to drive the deploy. It can be Nix or NixOS here. To install Nix, see the official guide.

Assuming this is done, you need to create a folder which will hold 3 files:

• network.nix explains how to provision each deploy target. For example, you'd tell here which user or package should exist. That being said, the goal here is to keep this file minimal and instead use the service.nix.
• services.nix is used to install any service - a database, a reverse proxy, an app, etc. The goal here is to make the install procedure machine independent.
• distribution.nix is used to tell which service goes to which deployment target.

Please see the integration tests for examples.

Advised Workflow

The workflow is the following:

1. make a change,
2. add or modify tests,
3. run the unit tests and integration tests,
4. deploy to staging environment
5. and deploy to production environment.

The first two bullets are very general so I can't realistically enumerate all possibilities. I'll possibly provide examples in a following update.

The following three are explained in the linked documents.

Useful commands

List deployments

To get the list of deployments, run:

nixops list

List machines

To know what machines exist on a deployment, run:

nixops info -d <deployment>

Ssh into a machine

export NIXOPS_DEPLOYMENT=<deployment>

nixops ssh <machine>

Delete a deployment

nixops delete -d <deployment>

Garbage collect old derivations

disnixos-env -s services.nix -n network-prod.nix -d distribution.nix --delete-generations=old

Create manifest file

disnixos-manifest -s services.nix -n network-virtualbox.nix -d distribution.nix

Create graph of service deployment

disnix-visualize /nix/store/cjiw9s257dpnvss2v6wm5a0iqx936hpq-manifest.xml | dot -Tpng > dot.png

Test Hercules CI locally

NIX_PATH="" nix-instantiate default.nix

See https://docs.hercules-ci.com/hercules-ci/getting-started/repository for more info.

Troubleshoot

Derivation not copied correctly

Sometimes, when aborting at the wrong moment, something does not get copied over correctly from your local machine to the <machine> you deploy on. If that happens, copy the manifest name from running the disnixos-env command (something like /nix/var/nix/profiles/per-user/.../disnix-coordinator/default-319-link) and run:

disnix-distribute <manifest>

Another way is to identify what path is missing by running ls /nix/store/<path> on both the host machine and the deploy machine. That path should exist on the former but not the latter. To copy over, run:

nix-store --export /nix/store/<path> | \
  bzip2 | \
  nixops ssh <machine> "bunzip2 | nix-store --import"

Cannot lock services

If you canceled a disnixos-env invocation, you could end up with locked services and the next invocation will fail. To unlock the services manually, run:

disnix-lock -u

Dev TODOs

In rough order of highest to lowest priority.

• Misc
  • Function to generate haproxy config
  • Documentation for setting up on Linode
  • Documentation for getting started
  • Add configuration examples
  • Merge all keycloak services into one definition
  • Run tests on Hercules-CI
• Dev
  • Automatically pull client credentials from keycloak to populate oauth2proxy's clientsecret key.
  • Automatic DNS setup of linode, probably using https://github.com/kubernetes-sigs/external-dns.
  • Add LDAP server.
  • Use LDAP server with vaultwarden using "Directory Connector".
  • Currently, there's a hack with a dnsmasq config in configuration.nix to redirect every request for <subdomain>.<dev-domain> to <machine>. This is not maintainable as the configuration does not rely on information provided by distribution.nix.
  • Add dependencies to systemd service files. I'm sure some of them are lacking the correct After= and Wants= fields.
  • Merge configs with systemd units. (remaining: keycloak)
  • Make haproxy resolve hostnames. For now, I hardcorded 127.0.0.1.
  • Auto-login into vaultwarden using SSO. Depends on https://github.com/dani-garcia/vaultwarden/pull/3154 being merged.
  • Go through https://xeiaso.net/blog/paranoid-nixos-2021-07-18 and https://nixos.wiki/wiki/Security
  • Move a few packages installed through network.nix into services.nix.
  • Use something else than pass to retrieve secrets. Or better, allow multiple options.
  • Explain how to setup secret keys.