docs(authoring): restructure authoring guides

Mainly because we have two module specs now and they will remain
valid for a while. We need to keep the older format documented for a while
This commit is contained in:
Johannes Kirschbauer
2025-04-08 21:32:19 +02:00
parent da164696ea
commit b4f63b8835
7 changed files with 202 additions and 208 deletions

View File

@@ -1,94 +0,0 @@
!!! Danger ":fontawesome-solid-road-barrier: Under Construction :fontawesome-solid-road-barrier:"
Currently under construction use with caution
:fontawesome-solid-road-barrier: :fontawesome-solid-road-barrier: :fontawesome-solid-road-barrier:
## Structure
A disk template consists of exactly two files
- `default.nix`
- `README.md`
```sh
└── single-disk
├── default.nix
└── README.md
```
## `default.nix`
Placeholders are filled with their machine specific options when a template is used for a machine.
The user can choose any valid options from the hardware report.
The file itself is then copied to `machines/{machineName}/disko.nix` and will be automatically loaded by the machine.
`single-disk/default.nix`
```
{
disko.devices = {
disk = {
main = {
device = "{{mainDisk}}";
...
};
};
};
}
```
## Placeholders
Each template must declare the options of its placeholders depending on the hardware-report.
`api/disk.py`
```py
templates: dict[str, dict[str, Callable[[dict[str, Any]], Placeholder]]] = {
"single-disk": {
# Placeholders
"mainDisk": lambda hw_report: Placeholder(
label="Main disk", options=hw_main_disk_options(hw_report), required=True
),
}
}
```
Introducing new local or global placeholders requires contributing to clan-core `api/disks.py`.
### Predefined placeholders
Some placeholders provide predefined functionality
- `uuid`: In most cases we recommend adding a unique id to all disks. This prevents the system to false boot from i.e. hot-plugged devices.
```
disko.devices = {
disk = {
main = {
name = "main-{{uuid}}";
...
}
}
}
```
## Readme
The readme frontmatter must be of the same format as modules frontmatter.
```markdown
---
description = "Simple disk schema for single disk setups"
---
# Single disk
Use this schema for simple setups where ....
```
The format and fields of this file is not clear yet. We might change that once fully implemented.

View File

@@ -1,12 +1,7 @@
# Services
# Instances
First of all it might be needed to explain what we mean by the term *distributed service*
!!! Note
Currently there are two ways of using such services.
1. via `inventory.services` **Will be deprecated**
2. via `inventory.instances` **Will be the new `inventory.services` once everyone has migrated**
## What is considered a service?
A **distributed service** is a system where multiple machines work together to provide a certain functionality, abstracting complexity and allowing for declarative configuration and management.
@@ -17,7 +12,7 @@ The term **Multi-host-service-abstractions** was introduced previously in the [n
## How to use such a Service in Clan?
In clan everyone can provide services via modules. Those modules must comply to a certain [specification](#service-module-specification), which is discussed later.
In clan everyone can provide services via modules. Those modules must be [`clan.service` modules](../authoring/clanServices/index.md).
To use a service you need to create an instance of it via the `clan.inventory.instances` attribute:
The source of the module must be specified as a simple string.
@@ -131,189 +126,3 @@ The following example shows how to use remote modules and configure them for use
);
}
```
## Service Module Specification
This section explains how to author a clan service module. As decided in [01-clan-service-modules](https://git.clan.lol/clan/clan-core/src/branch/main/decisions/01-ClanModules.md)
!!! Warning
The described modules are fundamentally different to the existing [clanModules](../clanmodules/index.md)
Most of the clanModules will be migrated into the described format. We actively seek for contributions here.
### Minimal module
!!! Tip
Unlike previously modules can now be inlined. There is no file-system structure needed anymore.
First of all we need to hook our module into the `inventory.modules` attribute. Make sure to choose a unique name so the module doesn't have a name collision with any of the core modules.
While not required we recommend to prefix your module attribute name.
If you export the module from your flake, other people will be able to import it and use it within their clan.
i.e. `@hsjobeki/customNetworking`
```nix title=flake.nix
# ...
outputs = inputs: flake-parts.lib.mkFlake { inherit inputs; } ({
imports = [ inputs.clan-core.flakeModules.default ];
# ...
clan = {
inventory = {
# We could also inline the complete module spec here
# For example
# {...}: { _class = "clan.service"; ... };
modules."@hsjobeki/customNetworking" = import ./service-modules/networking.nix;
};
# If needed: Exporting the module for other people
modules."@hsjobeki/customNetworking" = import ./service-modules/networking.nix;
};
})
```
The imported module file must fulfill at least the following requirements:
- Be an actual module. That means: Be either an attribute set; or a function that returns an attribute set.
- Required `_class = "clan.service"
- Required `manifest.name = "<name of the provided service>"`
```nix title="/service-modules/networking.nix"
{
_class = "clan.service";
manifest.name = "zerotier-networking";
# ...
}
```
### Adding functionality to the module
While the very minimal module is valid in itself it has no way of adding any machines to it, because it doesn't specify any roles.
The next logical step is to think about the interactions between the machines and define *roles* for them.
Here is a short guide with some conventions:
- [ ] If they all have the same relation to each other `peer` is commonly used. `peers` can often talk to each other directly.
- [ ] Often machines don't necessarily have direct relation to each other and there is one elevated machine in the middle classically know as `client-server`. `clients` are less likely to talk directly to each other than `peers`
- [ ] If your machines don't have any relation and/or interactions to each other you should reconsider if the desired functionality is really a multi-host service.
```nix title="/service-modules/networking.nix"
{
_class = "clan.service";
manifest.name = "zerotier-networking";
# Define what roles exist
roles.peer = {};
roles.controller = {};
# ...
}
```
Next we need to define the settings and the behavior of these distinct roles.
```nix title="/service-modules/networking.nix"
{
_class = "clan.service";
manifest.name = "zerotier-networking";
# Define what roles exist
roles.peer = {
interface = {
# These options can be set via 'roles.client.settings'
options.ipRanges = mkOption { type = listOf str; };
};
# Maps over all instances and produces one result per instance.
perInstance = { instanceName, settings, machine, roles, ... }: {
# Analog to 'perSystem' of flake-parts.
# For every instance of this service we will add a nixosModule to a client-machine
nixosModule = { config, ... }: {
# Interaction examples what you could do here:
# - Get some settings of this machine
# settings.ipRanges
#
# - Get all controller names:
# allControllerNames = lib.attrNames roles.controller.machines
#
# - Get all roles of the machine:
# machine.roles
#
# - Get the settings that where applied to a specific controller machine:
# roles.controller.machines.jon.settings
#
# Add one systemd service for every instance
systemd.services.zerotier-client-${instanceName} = {
# ... depend on the '.config' and 'perInstance arguments'
};
};
}
};
roles.controller = {
interface = {
# These options can be set via 'roles.server.settings'
options.dynamicIp.enable = mkOption { type = bool; };
};
perInstance = { ... }: {};
};
# Maps over all machines and produces one result per machine.
perMachine = { instances, machine, ... }: {
# Analog to 'perSystem' of flake-parts.
# For every machine of this service we will add exactly one nixosModule to a machine
nixosModule = { config, ... }: {
# Interaction examples what you could do here:
# - Get the name of this machine
# machine.name
#
# - Get all roles of this machine across all instances:
# machine.roles
#
# - Get the settings of a specific instance of a specific machine
# instances.foo.roles.peer.machines.jon.settings
#
# Globally enable something
networking.enable = true;
};
};
# ...
}
```
## Using values from a NixOS machine inside the module
!!! Example "Experimental Status"
This feature is experimental and should be used with care.
Sometimes a settings value depends on something within a machines `config`.
Since the `interface` is defined completely machine-agnostic this means values from a machine cannot be set in the traditional way.
The following example shows how to create a local instance of machine specific settings.
```nix title="someservice.nix"
{
# Maps over all instances and produces one result per instance.
perInstance = { instanceName, extendSettings, machine, roles, ... }: {
nixosModule = { config, ... }:
let
# Create new settings with
# 'ipRanges' defaulting to 'config.network.ip.range' from this machine
# This only works if there is no 'default' already.
localSettings = extendSettings {
ipRanges = lib.mkDefault config.network.ip.range;
};
in
{
# ...
};
};
}
```
!!! Danger
`localSettings` are a local attribute. Other machines cannot access it.
If calling extendSettings is done that doesn't change the original `settings` this means if a different machine tries to access i.e `roles.client.settings` it would *NOT* contain your changes.
Exposing the changed settings to other machines would come with a huge performance penalty, thats why we don't want to offer it.