hosting/selfhostblocks
1
0
Fork 0
Code Issues Activity
selfhostblocks/README.md
ibizaman 4533763553 add tip to speed up CI
2023-11-23 01:09:57 -08:00

26 KiB
Raw Blame History

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 opinionated building blocks fitting together to self-host any service you'd want. Some common services are provided out of the box.

The building blocks allow you to easily setup:

The provided services will have all those integrated. Progress is detailed in the Supported Features section.

You should know that although I am using everything in this repo for my personal production server, this is really just a one person effort for now and there are most certainly bugs that I didn't discover yet.

TOC

Supported Features

Currently supported services and features are:

  • Authelia as SSO provider.
    • Export metrics to Prometheus.
  • LDAP server through lldap, it provides a nice Web UI.
    • Administrative UI only accessible from local network.
  • Backup with Restic or BorgBackup
    • UI for backups.
    • Export metrics to Prometheus.
    • Alert when backups fail or are not done on time.
  • Reverse Proxy with Nginx.
    • Export metrics to Prometheus.
    • Log slow requests.
    • SSL support.
    • Backup support.
  • Monitoring through Prometheus and Grafana.
    • Export systemd services status.
    • Provide out of the box dashboards and alerts for common tasks.
    • LDAP auth.
    • SSO auth.
  • Vaultwarden
    • UI only accessible for vaultwarden_user LDAP group.
    • /admin only accessible for vaultwarden_admin LDAP group.
    • [WIP] True SSO support, see dani-garcia/vaultwarden/issues/246. For now, Authelia protects access to the UI but you need to login afterwards to Vaultwarden. So there are two login required.
  • Nextcloud
    • LDAP auth, unfortunately we need to configure this manually.
      • Declarative setup.
    • SSO auth.
      • Declarative setup.
    • Backup support.
    • Optional tracing debug.
    • Export traces to Prometheus.
    • Export metrics to Prometheus.
  • Home Assistant.
    • Export metrics to Prometheus.
    • LDAP auth through homeassistant_user LDAP group.
    • SSO auth.
    • Backup support.
  • Jellyfin
    • Export metrics to Prometheus.
    • LDAP auth through jellyfin_user and jellyfin_admin LDAP groups.
    • SSO auth.
    • Backup support.
  • Hledger
    • Export metrics to Prometheus.
    • LDAP auth through hledger_user LDAP group.
    • SSO auth.
    • Backup support.
  • Database Postgres
    • Slow log monitoring.
    • Export metrics to Prometheus.
  • VPN tunnel
  • Arr suite
    • SSO auth (one account for all users).
    • VPN support.
  • Mount webdav folders
  • Gitea to deploy
  • Scrutiny to monitor hard drives health
    • Export metrics to Prometheus.
  • QoL
    • Unit tests for modules.
    • Running in CI.
    • Integration tests with real nodes.
    • Self published documentation for options.
    • Examples for all building blocks.

Building Blocks

The building blocks are the foundation selfhostblocks intend to provide to allow you to self host easily and with best practices any service of your choosing. Some services are already provided out of the box but you might not want to use those if for example you want to integrate with existing services or slowly transition to NixOS.

Following somewhat the Unix principle, each block has one goal and does it correctly. They also are independent of each other, you can use only one or combine them to your liking.

Although these blocks provide options that encourage best practices, these are just NixOS modules that configure other modules provided by nixpkgs. Would you need to make tweaks, you can always access those underlying modules directly, like for any NixOS module.

The best way for now to understand how to use those modules is to read the code linked above and see how they are used in the provided services and in the demos. Also, here are a few examples taken from my personal usage of selfhostblocks.

Add SSL configuration

This is pretty much a prerequisite for all services.

shb.ssl = {
  enable = true;
  domain = "example.com";
  adminEmail = "me@example.com";
  sopsFile = ./secrets/linode.yaml;
  dnsProvider = "linode";
};

The configuration above assumes you own the example.com domain and the DNS is managed by Linode.

The sops file must be in the following format:

acme: |-
    LINODE_HTTP_TIMEOUT=10
    LINODE_POLLING_INTERVAL=10
    LINODE_PROPAGATION_TIMEOUT=240
    LINODE_TOKEN=XYZ...

For now, linode is the only supported DNS provider as it's the one I'm using. I intend to make this module more generic so you can easily use another provider not supported by selfhostblocks. You can skip setting the shb.ssl options and roll your own. Feel free to look at the ssl.nix for inspiration.

Add LDAP and Authelia services

These too are prerequisites for other services. Not all services support LDAP and SSO just yet, but I'm working on that.

shb.ldap = {
  enable = true;
  domain = "example.com";
  subdomain = "ldap";
  ldapPort = 3890;
  httpPort = 17170;
  dcdomain = "dc=example,dc=com";
  sopsFile = ./secrets/ldap.yaml;
  localNetworkIPRange = "192.168.1.0/24";
};

shb.authelia = {
  enable = true;
  domain = "example.com";
  subdomain = "authelia";

  ldapEndpoint = "ldap://127.0.0.1:${builtins.toString config.shb.ldap.ldapPort}";
  dcdomain = config.shb.ldap.dcdomain;

  smtpHost = "smtp.mailgun.org";
  smtpPort = 587;
  smtpUsername = "postmaster@mg.example.com";

  secrets = {
    jwtSecretFile = config.sops.secrets."authelia/jwt_secret".path;
    ldapAdminPasswordFile = config.sops.secrets."authelia/ldap_admin_password".path;
    sessionSecretFile = config.sops.secrets."authelia/session_secret".path;
    notifierSMTPPasswordFile = config.sops.secrets."authelia/smtp_password".path;
    storageEncryptionKeyFile = config.sops.secrets."authelia/storage_encryption_key".path;
    identityProvidersOIDCHMACSecretFile = config.sops.secrets."authelia/hmac_secret".path;
    identityProvidersOIDCIssuerPrivateKeyFile = config.sops.secrets."authelia/private_key".path;
  };
};
sops.secrets."authelia/jwt_secret" = {
  sopsFile = ./secrets/authelia.yaml;
  mode = "0400";
  owner = config.shb.authelia.autheliaUser;
  restartUnits = [ "authelia.service" ];
};
sops.secrets."authelia/ldap_admin_password" = {
  sopsFile = ./secrets/authelia.yaml;
  mode = "0400";
  owner = config.shb.authelia.autheliaUser;
  restartUnits = [ "authelia.service" ];
};
sops.secrets."authelia/session_secret" = {
  sopsFile = ./secrets/authelia.yaml;
  mode = "0400";
  owner = config.shb.authelia.autheliaUser;
  restartUnits = [ "authelia.service" ];
};
sops.secrets."authelia/smtp_password" = {
  sopsFile = ./secrets/authelia.yaml;
  mode = "0400";
  owner = config.shb.authelia.autheliaUser;
  restartUnits = [ "authelia.service" ];
};
sops.secrets."authelia/storage_encryption_key" = {
  sopsFile = ./secrets/authelia.yaml;
  mode = "0400";
  owner = config.shb.authelia.autheliaUser;
  restartUnits = [ "authelia.service" ];
};
sops.secrets."authelia/hmac_secret" = {
  sopsFile = ./secrets/authelia.yaml;
  mode = "0400";
  owner = config.shb.authelia.autheliaUser;
  restartUnits = [ "authelia.service" ];
};
sops.secrets."authelia/private_key" = {
  sopsFile = ./secrets/authelia.yaml;
  mode = "0400";
  owner = config.shb.authelia.autheliaUser;
  restartUnits = [ "authelia.service" ];
};

This sets up lldap under https://ldap.example.com and authelia under https://authelia.example.com.

The lldap sops file must be in the following format:

lldap:
    user_password: XXX...
    jwt_secret: YYY...

You can format the Authelia sops file as you wish since you can give the path to every secret independently. For completeness, here's the format expected by the snippet above:

authelia:
    ldap_admin_password: AAA...
    smtp_password: BBB...
    jwt_secret: CCC...
    storage_encryption_key: DDD...
    session_secret: EEE...
    storage_encryption_key: FFF...
    hmac_secret: GGG...
    private_key: |
        -----BEGIN PRIVATE KEY-----
        MII...MDQ=
        -----END PRIVATE KEY-----

Add backup to LDAP:

shb.backup.instances.lldap = {
  # Can also use "borgmatic".
  backend = "restic";

  keySopsFile = ./secrets/backup.yaml;

  # Backs up to 2 repositories.
  repositories = [
    "/srv/backup/restic/nextcloud"
    "s3:s3.us-west-000.backblazeb2.com/myserver-backup/nextcloud"
  ];

  retention = {
    keep_within = "1d";
    keep_hourly = 24;
    keep_daily = 7;
    keep_weekly = 4;
    keep_monthly = 6;
  };

  consistency = {
    repository = "2 weeks";
    archives = "1 month";
  };

  environmentFile = true;  # Needed for the s3 repository
}

This will backup the ldap users and groups to two different repositories. It assumes you have a backblaze account.

The backup sops file format is:

restic:
    passphrases:
        lldap: XYZ...
    environmentfiles:
        lldap: |-
            AWS_ACCESS_KEY_ID=XXX...
            AWS_SECRET_ACCESS_KEY=YYY...

The AWS keys are those provided by Backblaze.

See the ldap.nix and authelia.nix modules for more info.

Deploy the full Grafana, Prometheus and Loki suite

This is not a prerequisite for anything and could be enabled just for debugging.

shb.monitoring = {
  enable = true;
  subdomain = "grafana";
  inherit domain;
};

With that, Grafana, Prometheus, Loki and Promtail are setup! You can access Grafana at grafana.example.com.

A few Prometheus metrics scrapers are setup automatically:

  • node - cpu, memory, disk I/O, network I/O and a few others of the computer
  • smartctl - hard drive health
  • prometheus_internal - scraping jobs health
  • nginx
  • dnsmasq (if the service is enabled)

The following Loki logs scraper is setup automatically:

  • systemd journal

I intend to provide more options so that you could for example tweak data retention.

Also, since all logs are now stored in Loki, you can probably reduce the systemd journal retention time with:

# See https://www.freedesktop.org/software/systemd/man/journald.conf.html#SystemMaxUse=
services.journald.extraConfig = ''
SystemMaxUse=2G
SystemKeepFree=4G
SystemMaxFileSize=100M
MaxFileSec=day
'';

Set up network tunnel with VPN and Proxy

shb.vpn.nordvpnus = {
  enable = true;
  # Only "nordvpn" supported for now.
  provider = "nordvpn";
  dev = "tun1";
  # Must be unique per VPN instance.
  routingNumber = 10;
  # Change to the one you want to connect to
  remoteServerIP = "1.2.3.4";
  sopsFile = ./secrets/vpn.yaml;
  proxyPort = 12000;
};

This sets up a tunnel interface tun1 that connects to the VPN provider, here NordVPN. Also, if the proxyPort option is not null, this will spin up a tinyproxy instance that listens on the given port and redirects all traffic through that VPN.

$ curl 'https://api.ipify.org?format=json'
{"ip":"107.21.107.115"}

$ curl --interface tun1 'https://api.ipify.org?format=json'
{"ip":"46.12.123.113"}

$ curl --proxy 127.0.0.1:12000 'https://api.ipify.org?format=json'
{"ip":"46.12.123.113"}

Provided Services

The services above are those I am using myself. I intend to add more.

The best way for now to understand how to use those modules is to read the code linked above and see how they are used in the demos. Also, here are a few examples taken from my personal usage of selfhostblocks.

Common Options

Some common options are provided for all services.

  • enable (bool). Set to true to deploy and run the service.
  • subdomain (string). Subdomain under which to serve the service.
  • domain (string). Domain under which to server the service.

Some other common options are the following. I am not satisfied with how those are expressed so those will most certainly change.

  • LDAP and OIDC options for SSO, authentication and authorization.
  • Secrets.
  • Backups.

Note that for backups, every service exposes what directory should be backed up, you must merely choose when those backups will take place and where they will be stored.

Deploy a Nextcloud Instance

shb.nextcloud = {
  enable = true;
  domain = "example.com";
  subdomain = "nextcloud";
  sopsFile = ./secrets/nextcloud.yaml;
  localNetworkIPRange = "192.168.1.0/24";
  debug = false;
};

# Only needed if you want to override some default settings.
services.nextcloud = {
  datadir = "/srv/nextcloud";
  poolSettings = {
    "pm" = "dynamic";
    "pm.max_children" = 120;
    "pm.start_servers" = 12;
    "pm.min_spare_servers" = 6;
    "pm.max_spare_servers" = 18;
  };
};

# Backup the Nextcloud data.
shb.backup.instances.nextcloud = # Same as for the Authelia one above;

# For onlyoffice
nixpkgs.config.allowUnfreePredicate = pkg: builtins.elem (pkgs.lib.getName pkg) [
  "corefonts"
];

The snippet above sets up:

  • The nginx reverse proxy to listen on requests for the nextcloud.example.com domain.
  • An onlyoffice instance listening at oo.example.com that only listens on the local nextwork; you still need to setup manually the onlyoffice plugin in Nextcloud.
  • All the required databases and secrets.

The sops file format is:

nextcloud:
    adminpass: XXX...
    onlyoffice:
        jwt_secret: YYY...

See the nextcloud-server.nix module for more info.

You can enable tracing with:

shb.nextcloud.debug = true;

See my blog post for how to look at the traces.

Enable verbose Nginx logging

In case you need more verbose logging to investigate an issue:

shb.nginx.accessLog = true;
shb.nginx.debugLog = true;

See the nginx.nix module to see the effect of those options.

Deploy an hledger Instance with LDAP and SSO support

shb.hledger = {
  enable = true;
  subdomain = "hledger";
  domain = "example.com";
  oidcEndpoint = "https://authelia.example.com";
  localNetworkIPRange = "192.168.1.0/24";
};
shb.backup.instances.hledger = # Same as the examples above

This will setup:

  • The nginx reverse proxy to listen on requests for the hledger.example.com domain.
  • Backup of everything.
  • Only allow users of the hledger_user group to be able to login.
  • All the required databases and secrets.

See hledger.nix module for more details.

Deploy a Jellyfin instance with LDAP and SSO support

shb.jellyfin = {
  enable = true;
  domain = "example.com";
  subdomain = "jellyfin";

  sopsFile = ./secrets/jellyfin.yaml;
  ldapHost = "127.0.0.1";
  ldapPort = 3890;
  dcdomain = config.shb.ldap.dcdomain;
  oidcEndpoint = "https://${config.shb.authelia.subdomain}.${config.shb.authelia.domain}";
  oidcClientID = "jellyfin";
  oidcUserGroup = "jellyfin_user";
  oidcAdminUserGroup = "jellyfin_admin";
};
shb.backup.instances.jellyfin = # Same as the examples above

This sets up, as usual:

  • The nginx reverse proxy to listen on requests for the jellyfin.example.com domain.
  • Backup of everything.
  • Only allow users of the jellyfin_user or jellyfin_admin ldap group to be able to login.
  • All the required databases and secrets.

The sops file format is:

jellyfin:
    ldap_password: XXX...
    sso_secret: YYY...

Although the configuration of the LDAP and SSO plugins is done declaratively in the Jellyfin preStart step, they still need to be installed manually at the moment.

See jellyfin.nix module for more details.

Deploy a Home Assistant instance with LDAP support

SSO support is WIP.

shb.home-assistant = {
  enable = true;
  subdomain = "ha";
  inherit domain;
  ldapEndpoint = "http://127.0.0.1:${builtins.toString config.shb.ldap.httpPort}";
  backupCfg = # Same as the examples above
  sopsFile = ./secrets/homeassistant.yaml;
};
services.home-assistant = {
  extraComponents = [
    "backup"
    "esphome"
    "jellyfin"
    "kodi"
    "wyoming"
    "zha"
  ];
};
services.wyoming.piper.servers = {
  "fr" = {
    enable = true;
    voice = "fr-siwis-medium";
    uri = "tcp://0.0.0.0:10200";
    speaker = 0;
  };
};
services.wyoming.faster-whisper.servers = {
  "tiny-fr" = {
    enable = true;
    model = "medium-int8";
    language = "fr";
    uri = "tcp://0.0.0.0:10300";
    device = "cpu";
  };
};

This sets up everything needed to have a Home Assistant instance available under ha.example.com. It also shows how to have a piper and whisper server for respectively text to speech and speech to text. The integrations must still be setup in the web UI.

The sops file must be in the following format:

home-assistant: |
    country: "US"
    latitude_home: "0.01234567890123"
    longitude_home: "-0.01234567890123"

Demos

Demos that start and deploy a service on a Virtual Machine on your computer are located under the demo folder. These show the onboarding experience you would get if you deployed one of the services on your own server.

Import selfhostblocks

Ready to start using selfhostblocks? Thank you for trusting selfhostblocks. Please raise any question you have or hurdle you encounter by creating an issue.

The top-level flake.nix just outputs a nixos module that gathers all other modules from the modules/ directory. Use this repo as a flake input to your own repo. The inputs field of your flake.nix file in your repo should look like so:

inputs = {
  nixpkgs.url = "github:nixos/nixpkgs/nixos-unstable";
  sops-nix.url = "github:Mic92/sops-nix";

  selfhostblocks.url = "github:ibizaman/selfhostblocks";
  selfhostblocks.inputs.nixpkgs.follows = "nixpkgs";
  selfhostblocks.inputs.sops-nix.follows = "sops-nix";
};

sops-nix is used to setup passwords and secrets. Currently selfhostblocks has a strong dependency on it but I'm working on removing that so you could use any secret provider.

The snippet above makes selfhostblocks' inputs follow yours. This is not maintainable though because options that selfhostblocks rely on can change or disappear and you have no control on that. Later, I intend to make selfhostblocks provide its own nixpkgs input and update it myself through CI.

How you actually deploy using selfhostblocks depends on what system you choose. If you use colmena, this is what your outputs field could look like:

outputs = inputs@{ self, nixpkgs, ... }: {
  colmena = {
    meta = {
      nixpkgs = import inputs.nixpkgs {
        system = "x86_64-linux";
      };
      specialArgs = inputs;
    };

    myserver = import ./machines/myserver.nix;
  };
}

Now, what goes inside this ./machines/myserver.nix file? First, import selfhostblocks and sops-nix:

imports = [
  selfhostblocks.nixosModules.x86_64-linux.default
  sops-nix.nixosModules.default
]

For the rest, see the above building blocks, provided services and demos sections.

Community

All issues and PRs are welcome. For PRs, if they are substantial changes, please open an issue to discuss the details first.

Come hang out in the Matrix channel. :)

Along the way, I made quite a few changes to the ubderlying nixpkgs module I'm using. I intend to upstream to nixpkgs as much of those as makes sense.

Tips

Run tests

Run all tests:

$ nix build .#checks.${system}.all
# or
$ nix flake check

Run one group of tests:

$ nix build .#checks.${system}.modules
$ nix build .#checks.${system}.vm_postgresql_peerAuth

Speed up CI

Github actions do not have hardware acceleration and tests could timeout when running there. The easiest way to speed up CI is to push the test results to cachix. Instructions are given here.

Deploy using colmena

$ nix run nixpkgs#colmena -- apply

Use a local version of selfhostblocks

This works with any flake input you have. Either, change the .url field directly in you flake.nix:

selfhostblocks.url = "/home/me/projects/selfhostblocks";

Or override on the command line:

$ nix flake lock --override-input selfhostblocks ../selfhostblocks

I usually combine the override snippet above with deploying:

$ nix flake lock --override-input selfhostblocks ../selfhostblocks && nix run nixpkgs#colmena -- apply

Diff changes

First, you must know what to compare. You need to know the path to the nix store of what is already deployed and to what you will deploy.

What is deployed

To know what is deployed, either just stash the changes you made and run build:

$ nix run nixpkgs#colmena -- build
...
Built "/nix/store/yyw9rgn8v5jrn4657vwpg01ydq0hazgx-nixos-system-baryum-23.11pre-git"

Or ask the target machine:

$ nix run nixpkgs#colmena -- exec -v readlink -f /run/current-system
baryum | /nix/store/77n1hwhgmr9z0x3gs8z2g6cfx8gkr4nm-nixos-system-baryum-23.11pre-git

What will get deployed

Assuming you made some changes, then instead of deploying with apply, just build:

$ nix run nixpkgs#colmena -- build
...
Built "/nix/store/16n1klx5cxkjpqhrdf0k12npx3vn5042-nixos-system-baryum-23.11pre-git"

Get the full diff

With nix-diff:

$ nix run nixpkgs#nix-diff -- \
  /nix/store/yyw9rgn8v5jrn4657vwpg01ydq0hazgx-nixos-system-baryum-23.11pre-git \
  /nix/store/16n1klx5cxkjpqhrdf0k12npx3vn5042-nixos-system-baryum-23.11pre-git \
  --color always | less

Get version bumps

A nice summary of version changes can be produced with:

$ nix run nixpkgs#nvd -- diff \
  /nix/store/yyw9rgn8v5jrn4657vwpg01ydq0hazgx-nixos-system-baryum-23.11pre-git \
  /nix/store/16n1klx5cxkjpqhrdf0k12npx3vn5042-nixos-system-baryum-23.11pre-git \

Generate random secret

$ nix run nixpkgs#openssl -- rand -hex 64

TODOs

  • Add examples that sets up services in a VM.
  • Do not depend on sops.
  • Add more options to avoid hardcoding stuff.
  • Make sure nginx gets reloaded when SSL certs gets updated.
  • Better backup story by taking optional LVM or ZFS snapshot before backing up.
  • Many more tests.
  • Tests deploying to real nodes.
  • DNS must be more configurable.
  • Fix tests on nix-darwin.

Links that helped

While creating NixOS tests:

While creating an XML config generator for Radarr:

License

I'm following the Nextcloud license which is AGPLv3. See this article from the FSF that explains what this license adds to the GPL one.