Platform Team/Server Kit/Mace

From Sugar Labs
Jump to: navigation, search


The mace is a tool to macke final configuration using source templates. Mace is supposed to help with configuration of services on Server based school servers.

These are the core differences compared with tools like Puppet or Cfengine that makes mace a different niche project:

  • mace provides a simple and straightforward usage workflow, i.e., with tools like Puppet, users need to "code" a configuration (starting from regular coding in Ruby for missing functions or types, and ending with handling the relationships between all configured resources). It is different in mace; users only need to enter a configuration for the particular service, the rest (like dependencies between configured services) is already coded in mace;
  • mace doesn't provide new metaphors; people need to follow the standard configuration syntax for any particular service;
  • mace is not intended to be a unified system like Puppet or Cfengine; it supports only a limited set of services (of those that the Server based solution provides), but does it well, e.g., for iptables, just write rules, and the rest will be done by mace;
  • mace doesn't function like a daemon; it just converts configuration sources to the final configuration on the final server, e.g., as a post procedure after installing packages;
  • mace is designed to ease support of intermediate customizing, i.e., the original configuration, provided by an upstream project, might be supplemented (not patched) in the downstream product before deploying it to the final users.
  • within the Sugar Server Kit initiative, mace is intended to support, mostly, only the Dumb school servers functional model.

Configuration sources

Sources for mace are stored in a GNU/Linux distribution agnostic manner, in this form:


The service-confile is optional and makes sense only if the configured service has several of them, e.g., Bind, or, if a singular configuration file was split at the Mace level into several parts to make it easy to configure, e.g., per IPTables table service-confiles. The configuration-file is a configuration file in a particular service configuration's syntax. All configuration-file files will be merged by mace to the singular service-confile file and placed in the appropriate distribution-specific directory. The purpose in having arbitrary-path is that there might be several directories with the same service-names to make the configuration more flexible, i.e., it allows having several high-level configuration components in a project that configure the same service.

The following example shows how IPTables and Squid proxy might be configured:

<service>  | +iptables.d
<confile>  |   +nat
<config>   |     +010.conf
<dir>      +020.proxy
<service>    +iptables.d
<confile>    | +nat
<config>     |   +020.squid.conf
<service>    +squid.d
<confile>      +access
<config>          +010.acl.conf

See sugar-server-templates sources for a more complex examples.

Configuration variables

In addition to configuration files, Mace can process files with a .env suffix. These files contain variable declarations in Bash syntax. The values of these variables might be entered in configuration files in the form of @VARIABLE@, in which case they will be expanded to real values while applying the configuration.

In addition to custom variables, there is a list of special ones:

  • variables exported by providers;
  • <provider>_SERVICE variables with paths to /etc/init.d services for all providers that support services;
  • SRC_ROOT, path to the root directory with configuration sources;
  • DST_ROOT, path to the root directory to place final configuration;
  • INCLUDE, if this variable is set in .env files, it will be treated as a colon separated list of directories to search for additional configuration sources; if INCLUDE directories are not absolute, they will be referenced from the sugar-server-templates root;
  • PWD, path to the directory where current source configuration file is located.


Source configuration file names might contain substrings in format


which will tell Mace that this source file can be processed only if this particular keyword is enabled in Mace by mentioning the keyword name in the --keywords command-line argument.

By default, Mace has these keywords enabled:

  • production

For example, if there is a sources tree,

<dir>      +squid.d
<config>   | +010.base.conf
<config>   | +020.log[production].conf
<config>   | +020.log[test].conf

the final configuration tree will be as follows:

<dir>      +squid.d
<config>   | +010.base.conf
<config>   | +020.log[production].conf

Configuration application

Mace will apply a configuration sources tree to the real system by doing the following:

  1. walk through the sources tree to collect all configuration source files for each supported service;
  2. apply the configuration by writing the sources content to the proper configuration files for a particular service on the particular OS, the backup copies will be kept;
  3. ask the mace provider of a particular service to make sure that everything, related to this service, is good, e.g., check if Prosody SSL keys/certificates exist and are not expired, or check if Squid's swap is created;
  4. if a new configuration is different from a previous one, or if step 3 changed something in the system, restart this service;
  5. make sure that service will be started at boot time;
  6. if all the previous steps aborted due to failures, revert to the original configuration.

To start the applying process:

sudo mace apply

If the sources path is not the default, /etc/mace, use the --input argument to specify the right one.


Providers is the way how Mace can configure particular services. Providers don't contain high-level configuration logic, configuration will happen only basing on configuration sources passed to the Mace to process by these providers. At the same time, providers might do some extra work, e.g., to make sure that service's directory are created. See providers documentation for more information how particular provider might affect final system after configuration application.

Mace supports only a limited number of GNU/Linux distributions and services. That's because Mace is supposed to be used mostly as a school server configuration tool.


The project sugar-server-templates is intended to provide most of basic configurations that might be useful for schools servers. After installing it from packages, the final configuration might be composed by symlinking templates from /usr/share/sugar-server-templates directory to the directory where Mace will find it, by default in /etc/mace.


By default, configuration occurs based on several sources (sorted by applied order):

  • /etc/mace.conf system-wide configuration file,
  • ~/.local/mace/config user-wide configuration file,
  • mace's command-line arguments.

Configuration files contain option names equal to command-line arguments. To get the current configuration, call:

mace config


  • By default, if mace can't start configuring services with new configuration, it falls back to the old configuration. It might be not very useful in case of investigating new configuration errors. To prevent such behavior use -R command-line argument and try to start affected service manually (with new configuration):
mace apply -R
/etc/init.d/<SERVICE> start

Testing routines

Before applying a configuration on a real system, it might be applied to a temporary directory (do not forget about --dry-services argument to not restart services):

mace apply -o /tmp/test -v -S

Mace also supports several listing commands that just walk though the sources tree and print useful information about it:

  • mace ls, list sources status for a particular directory;
  • mace dirs, list sources status by services;
  • mace show, interpret services status given by mace dirs command;
  • mace files, list all configuration files.

The status legend is:

Symbol Note
- there are configuration files
* configuration files for the same service were found in more than one directory
o some of configuration files are overridden by files with the same name but from a different directory
O all configuration files are overridden by files with the same name but from a different directory
h some of configuration files are hidden by empty files with the same name but from a different directory
H all configuration files are hidden by empty files with the same name but from a different directory

Getting involved