Skip to content

ip1981/nixsap

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

About

Nixsap is a set of modules built on top of NixOS/Nixpkgs. Nixsap provides NixOS modules in the nixsap "namespace", e. g. nixsap.apps.mariadb or nixsap.system.users, and adds or overrides some packages in Nixpkgs. From vanila Nixpkgs, Nixsap relies only on basic services like systemd, ssh, ntpd and package set (extending and overriding it).

Features

Plug & Play

Each module under the modules directory is automatically available. When creating a new machine just use

imports = [ <nixsap> ];

Each package foo under the ./pkgs is automatically available as pkgs.foo. For example:

pkgs/writeXML.nix        => pkgs.writeXML
pkgs/rdsdump/default.nix => pkgs.rdsdump

You can use this techniques in your own projects. You can take out any modules or packages and put them into your project with your modifications without maintaning a fork of Nixsap. When taking modules you have to change the nixsap namespace to something different to avoid conflicts.

Package Overlay

Nixsap modules use a package overlay located in ./pkgs/default.nix. It is also possible to build and install packages from this overlay independently, for example:

nix-build -E '(import <nixpkgs> { overlays = [ (import ./pkgs/default.nix) ]; })' -A icinga2

Automatic unix user id

To create daemon users just add their names into the list nixsap.system.users.daemons. List nixsap.system.users.normal does the same for users with login shell, and nixsap.system.groups for unix groups. Users and groups will automatically get their ids based on their names in a deterministic manner. See examples in the applications directory and implementation in modules/system/users.nix. This feature is used throughout nixsap.apps.

Examples:

# id icinga
uid=1240920351(icinga) gid=100(users) groups=21(proc),100(users)

# id pashev
uid=1141737888(pashev) gid=100(users) groups=100(users),21(proc),62(systemd-journal),1061782283(sysops)

# id jenkins-dumpoo
uid=1201814562(jenkins-dumpoo) gid=1201814562(jenkins-dumpoo) groups=96(keys),1201814562(jenkins-dumpoo)

# id mariadb
uid=1213117043(mariadb) gid=1213117043(mariadb) groups=96(keys),1213117043(mariadb)

Keyrings

Keyrings provide a means of deploying secret files with sensitive content. It's inspired by NixOps and relies on it as on reference implementation. Most applications from nixsap.apps recognize keys from their parameters or extract them from configuration files and automatically build their keyrings.

Multi-instance applications

For most applications it is possible to run multiple instances of them on the same machine. Each instance may have its own home directory and user. For example jenkins-foo with home directory /jenkins/foo and user jenkins-foo. Not all applications allow that purely for historical reasons, and this will be fixed eventually.

Design

Static analysis

There are a handful of tools used thoughout the applications to ensure correctness at build time: writeBashScript, writeXML, writePHPFile. writeBashScript uses shellcheck. writeXML runs xmllint. writePHPFile relies on PHP's built-in syntax checker.

The objective it to do static analysis and linting for every single file at build time. What is wanted: configuration files for Nginx, Icinga; MySQL and PostgreSQL scripts, etc.

Parametrization

Everything that can be used at build time should have a parameter (integer, string, path, etc.). Examples are TCP port, data directory, UNIX user. TCP port can be used for configuring firewall or HTTP proxy, data directory can be used for setting up mount points, UNIX user can be included into extra groups, etc. When we have it all parametrized we do not repeat ourselves.

Parametrization also helps modularity. I. e. you can define default set of values and override only some of them in specific setups.

Some applications accept only discrete set of options, in that case we should parametrize them all. Examples are memcached, php-fpm and sproxy2.

Parametrization should give access to all application features. Ideally, parameters should exactly match to the application options, including their names and meanings. Examples are MariaDB and PostgreSQL. This makes documentation unnecessary, because each parameter is documented somewhere else.

Almost every parameter, if it's not required by application (i. e. has a built-in default value), should have value null by default. If such parameter is not set, it is not passed to the application. This is twofold: more transparency because we use application's defaults (not ours), and it is safer to use different versions of application, when particular options may be added or removed.

Even though, if the value of parameter is required at build time, the parameter should have default value, preferably application's default. Example is MariaDB's TCP port. We need it to configure firewall, thus we define it to be 3306 by default.

If application default value is known to be insecure, we should set our own, secure, default value.

Recommendations

License

This project is under the MIT license (see LICENSE), unless stated otherwise in individual files.