General documentation


The mgmt tool is a next generation config management prototype. It’s not yet ready for production, but we hope to get there soon. Get involved today!

Project Description

The mgmt tool is a distributed, event driven, config management tool, that supports parallel execution, and librarification to be used as the management foundation in and for, new and existing software.

For more information, you may like to read some blog posts from the author:

There is also an introductory video available. Older videos and other material is available.


You’ll probably want to read the quick start guide to get going.


This section details the numerous features of mgmt and some caveats you might need to be aware of.


Automatic edges, or AutoEdges, is the mechanism in mgmt by which it will automatically create dependencies for you between resources. For example, since mgmt can discover which files are installed by a package it will automatically ensure that any file resource you declare that matches a file installed by your package resource will only be processed after the package is installed.

Controlling autoedges

Though autoedges is likely to be very helpful and avoid you having to declare all dependencies explicitly, there are cases where this behaviour is undesirable.

Some distributions allow package installations to automatically start the service they ship. This can be problematic in the case of packages like MySQL as there are configuration options that need to be set before MySQL is ever started for the first time (or you’ll need to wipe the data directory). In order to handle this situation you can disable autoedges per resource and explicitly declare that you want my.cnf to be written to disk before the installation of the mysql-server package.

You can disable autoedges for a resource by setting the autoedge key on the meta attributes of that resource to false.

Blog post

You can read the introductory blog post about this topic here:


Automatic grouping or AutoGroup is the mechanism in mgmt by which it will automatically group multiple resource vertices into a single one. This is particularly useful for grouping multiple package resources into a single resource, since the multiple installations can happen together in a single transaction, which saves a lot of time because package resources typically have a large fixed cost to running (downloading and verifying the package repo) and if they are grouped they share this fixed cost. This grouping feature can be used for other use cases too.

You can disable autogrouping for a resource by setting the autogroup key on the meta attributes of that resource to false.

Blog post

You can read the introductory blog post about this topic here:

Automatic clustering

Automatic clustering is a feature by which mgmt automatically builds, scales, and manages the embedded etcd cluster which is compiled into mgmt itself. It is quite helpful for rapidly bootstrapping clusters and avoiding the extra work to setup etcd.

If you prefer to avoid this feature. you can always opt to use an existing etcd cluster that is managed separately from mgmt by pointing your mgmt agents at it with the --seeds variable.

Blog post

You can read the introductory blog post about this topic here:

Remote (”agent-less”) mode

Remote mode is a special mode that lets you kick off mgmt runs on one or more remote machines which are only accessible via SSH. In this mode the initiating host connects over SSH, copies over the mgmt binary, opens an SSH tunnel, and runs the remote program while simultaneously passing the etcd traffic back through the tunnel so that the initiators etcd cluster can be used to exchange resource data.

The interesting benefit of this architecture is that multiple hosts which can’t connect directly use the initiator to pass the important traffic through to each other. Once the cluster has converged all the remote programs can shutdown leaving no residual agent.

This mode can also be useful for bootstrapping a new host where you’d like to have the service run continuously and as part of an mgmt cluster normally.

In particular, when combined with the --converged-timeout parameter, the entire set of running mgmt agents will need to all simultaneously converge for the group to exit. This is particularly useful for bootstrapping new clusters which need to exchange information that is only available at run time.

This existed in earlier versions of mgmt as a --remote option, but it has been removed and is being ported to a more powerful variant where you can remote execute via a remote resource.

Blog post

You can read the introductory blog post about this topic here:

Puppet support

You can supply a Puppet manifest instead of creating the (YAML) graph manually. Puppet must be installed and in mgmt’s search path. You also need the ffrank-mgmtgraph Puppet module.

Invoke mgmt with the --puppet switch, which supports 3 variants:

  1. Request the configuration from the Puppet Master (like puppet agent does)

    mgmt run puppet --puppet agent

  2. Compile a local manifest file (like puppet apply)

    mgmt run puppet --puppet /path/to/my/manifest.pp

  3. Compile an ad hoc manifest from the commandline (like puppet apply -e)

    mgmt run puppet --puppet 'file { "/etc/ntp.conf": ensure => file }'

For more details and caveats see

Blog post

An introductory post on the Puppet support is on Felix’s blog.


Please note that there are a number of undocumented options. For more information on these options, please view the source at: If you feel that a well used option needs documenting here, please patch it!

Overview of reference

Meta parameters

These meta parameters are special parameters (or properties) which can apply to any resource. The usefulness of doing so will depend on the particular meta parameter and resource combination.


Boolean. Should we generate auto edges for this resource?


Boolean. Should we attempt to automatically group this resource with others?


Boolean. Should the Apply portion of the CheckApply method of the resource make any changes? Noop is a concatenation of no-operation.


Integer. The number of times to retry running the resource on error. Use -1 for infinite. This currently applies for both the Watch operation (which can fail) and for the CheckApply operation. While they could have separate values, I’ve decided to use the same ones for both until there’s a proper reason to want to do something differently for the Watch errors.


Integer. Number of milliseconds to wait between retries. The same value is shared between the Watch and CheckApply retries. This currently applies for both the Watch operation (which can fail) and for the CheckApply operation. While they could have separate values, I’ve decided to use the same ones for both until there’s a proper reason to want to do something differently for the Watch errors.


Integer. Number of seconds to wait between CheckApply checks. If this is greater than zero, then the standard event based Watch mechanism for this resource is replaced with a simple polling mechanism. In general, this is not recommended, unless you have a very good reason for doing so.

Please keep in mind that if you have a resource which changes every I seconds, and you poll it every J seconds, and you’ve asked for a converged timeout of K seconds, and I <= J <= K, then your graph will likely never converge.

When polling, the system detects that a resource is not converged if its CheckApply method returns false. This allows a resource which changes every I seconds, and which is polled every J seconds, and with a converged timeout of K seconds to still converge when J <= K, as long as I > J || I > K, which is another way of saying that if the resource finally settles down to give the graph enough time, it can probably converge.


Float. Maximum rate of CheckApply runs started per second. Useful to limit an especially eventful process from causing excessive checks to run. This defaults to +Infinity which adds no limiting. If you change this value, you will also need to change the Burst value to a non-zero value. Please see the rate package for more information.


Integer. Burst is the maximum number of runs which can happen without invoking the rate limiter as designated by the Limit value. If the Limit is not set to +Infinity, this must be a non-zero value. Please see the rate package for more information.


List of string ids. Sema is a P/V style counting semaphore which can be used to limit parallelism during the CheckApply phase of resource execution. Each resource can have N different semaphores which share a graph global namespace. Each semaphore has a maximum count associated with it. The default value of the size is 1 (one) if size is unspecified. Each string id is the unique id of the semaphore. If the id contains a trailing colon (:) followed by a positive integer, then that value is the max size for that semaphore. Valid semaphore id’s include: some_id, hello:42, not:smart:4 and :13. It is expected that the last bare example be only used by the engine to add a global semaphore.


Boolean. Rewatch specifies whether we re-run the Watch worker during a graph swap if it has errored. When doing a graph compare to swap the graphs, if this is true, and this particular worker has errored, then we’ll remove it and add it back as a new vertex, thus causing it to run again. This is different from the Retry metaparam which applies during the normal execution. It is only when this is exhausted that we’re in permanent worker failure, and only then can we rely on this metaparam.


Boolean. Realize ensures that the resource is guaranteed to converge at least once before a potential graph swap removes or changes it. This guarantee is useful for fast changing graphs, to ensure that the brief creation of a resource is seen. This guarantee does not prevent against the engine quitting normally, and it can’t guarantee it if the resource is blocked because of a failed pre-requisite resource. XXX: This is currently not implemented!


Boolean. Reverse is a property that some resources can implement that specifies that some “reverse” operation should happen when that resource “disappears”. A disappearance happens when a resource is defined in one instance of the graph, and is gone in the subsequent one. This disappearance can happen if it was previously in an if statement that then becomes false.

This is helpful for building robust programs with the engine. The engine adds a “reversed” resource to that subsequent graph to accomplish the desired “reverse” mechanics. The specifics of what this entails is a property of the particular resource that is being “reversed”.

It might be wise to combine the use of this meta parameter with the use of the realize meta parameter to ensure that your reversed resource actually runs at least once, if there’s a chance that it might be gone for a while.

Lang metadata file

Any module must have a metadata file in its root. It must be named metadata.yaml, even if it’s empty. You can specify zero or more values in yaml format which can change how your module behaves, and where the mcl language looks for code and other files. The most important top level keys are: main, path, files, and license.


The main key points to the default entry point of your code. It must be a relative path if specified. If it’s empty it defaults to main.mcl. It should generally not be changed. It is sometimes set to main/main.mcl if you’d like your modules code out of the root and into a child directory for cases where you don’t plan on having a lot deeper imports relative to main.mcl and all those files would clutter things up.


The path key specifies the modules import search directory to use for this module. You can specify this if you’d like to vendor something for your module. In general, if you use it, please use the convention: path/. If it’s not specified, you will default to the parent modules directory.


The files key specifies some additional files that will get included in your deploy. It defaults to files/.


The license key allows you to specify a license for the module. Please specify one so that everyone can enjoy your code! Use a “short license identifier”, like LGPLv3+, or MIT. The former is a safe choice if you’re not sure what to use.

Graph definition file

graph.yaml is the compiled graph definition file. The format is currently undocumented, but by looking through the examples/ you can probably figure out most of it, as it’s fairly intuitive. It’s not recommended that you use this, since it’s preferable to write code in the mcl language front-end.

Command line

The main interface to the mgmt tool is the command line. For the most recent documentation, please run mgmt --help.

--converged-timeout <seconds>

Exit if the machine has converged for approximately this many seconds.

--max-runtime <seconds>

Exit when the agent has run for approximately this many seconds. This is not generally recommended, but may be useful for users who know what they’re doing.


Globally force all resources into no-op mode. This also disables the export to etcd functionality, but does not disable resource collection, however all resources that are collected will have their individual noop settings set.

--sema <size>

Globally add a counting semaphore of this size to each resource in the graph. The semaphore will get given an id of :size. In other words if you specify a size of 42, you can expect a semaphore if named: :42. It is expected that consumers of the semaphore metaparameter always include a prefix to avoid a collision with this globally defined semaphore. The size value must be greater than zero at this time. The traditional non-parallel execution found in config management tools such as Puppet can be obtained with --sema 1.


Allow interactive prompting for SSH passwords if there is no authentication method that works.


Specify the path for finding SSH keys. This defaults to ~/.ssh/id_rsa. To never use this method of authentication, set this to the empty string.


The maximum number of concurrent remote ssh connections to run. This defaults to 0, which means unlimited.


Don’t allow remote caching of the remote execution binary. This will require the binary to be copied over for every remote execution, but it limits the likelihood that there is leftover information from the configuration process.

--prefix <path>

Specify a path to a custom working directory prefix. This directory will get created if it does not exist. This usually defaults to /var/lib/mgmt/. This can’t be combined with the --tmp-prefix option. It can be combined with the --allow-tmp-prefix option.


If this option is specified, a temporary prefix will be used instead of the default prefix. This can’t be combined with the --prefix option.


If this option is specified, we will attempt to fall back to a temporary prefix if the primary prefix couldn’t be created. This is useful for avoiding failures in environments where the primary prefix may or may not be available, but you’d like to try. The canonical example is when running mgmt with remote execution there might be a cached copy of the binary in the primary prefix, but if there’s no binary available continue working in a temporary directory to avoid failure.

Compilation options

You can control some compilation variables by using environment variables.

Disable libvirt support

If you wish to compile mgmt without libvirt, you can use the following command:

GOTAGS=novirt make build

Disable augeas support

If you wish to compile mgmt without augeas support, you can use the following command:

GOTAGS=noaugeas make build

Disable docker support

If you wish to compile mgmt without docker support, you can use the following command:

GOTAGS=nodocker make build

Combining compile-time flags

You can combine multiple tags by using a space-separated list:

GOTAGS="noaugeas novirt nodocker" make build


For example configurations, please consult the examples/ directory in the git source repository. It is available from:


See misc/mgmt.service for a sample systemd unit file. This unit file is part of the RPM.

To specify your custom options for mgmt on a systemd distro:

sudo mkdir -p /etc/systemd/system/mgmt.service.d/

cat > /etc/systemd/system/mgmt.service.d/env.conf <<EOF
# Environment variables:

# Other CLI options if necessary.

sudo systemctl daemon-reload


This is a project that I started in my free time in 2013. Development is driven by all of our collective patches! Dive right in, and start hacking! Please contact me if you’d like to invite me to speak about this at your event.

You can follow along on my technical blog.

To report any bugs, please file a ticket at:


Copyright (C) 2013-2023+ James Shubin and the project contributors

Please see the AUTHORS file for more information.