Invoking actions

You can invoke MCollective actions on both *nix and Windows machines, based on commands you run from the command line on a Linux-based Puppet master servers.

MCollective has its own section of the documentation site, which includes more complete details and examples for command line orchestration usage.

The following topics cover basic CLI usage and all PE-specific information; for more details, see the following pages from the full docs:

  • MCollective command line usage
  • Filtering

Logging into MCollective

To run commands, you must log in to the Puppet master server as the special peadmin user account, which PE creates during installation.

Note: PE does not support adding more MCollective user accounts.

By default, the peadmin account cannot log in with a password. We recommend you log in using sudo or by adding SSH keys to the peadmin account.

Logging in with sudo

Anyone able to log into the Puppet master server as an admin user with full root sudo privileges can become the peadmin user by running:

$ sudo -i -u peadmin

This is the default way to log in as the peadmin user. It means that MCollective commands can only be issued by the group of users who can fully control the Puppet master.

Logging in with SSH keys

If you wish to allow other users to run commands without giving them full control over the Puppet master, you can add their public SSH keys to peadmin's authorized keys file.

You can use Puppet's ssh_authorized_key resource type to do this, or add keys manually to the /var/lib/peadmin/.ssh/authorized_keys file.

The mco command

All MCollective actions are invoked with the mco executable. The mco command always requires a subcommand to invoke actions.

Important: For security, the mco command relies on a config file (/var/lib/peadmin/.mcollective) which is only readable by the peadmin user. PE automatically configures and manages this file, so do not modify it.

mco Subcommands

The mco command has several subcommands, and it's possible to add more --- run mco help for a list of all available subcommands. The following default subcommands are available:

  • main
    Command Use
    rpc This is the general purpose client, which can invoke actions from any agent plugin.
  • special purpose
    Command Use
    puppet These subcommands only invoke certain kinds of actions, but have some extra UI enhancements to make them easier to use than the equivalent mco rpc command.
  • Help and support subcommands
    Command Use
    help Displays help for subcommands.
    plugin The mco plugin doc command can display help for agent plugins.
    completion A helper for shell completion systems.
  • Inventory and reporting subcommands
    Command Use
    ping Pings all matching nodes and reports on response times
    facts Displays a summary of values for a single fact across all systems
    inventory General reporting tool for nodes, collectives and subcollectives
    find Like ping, but doesn't report response times

Getting help on the command line

You can get information about subcommands, actions, and other plugins on the command line.

Subcommand Help

Use one of the following commands to get help for a specific subcommand:

$ mco help <SUBCOMMAND>
$ mco <SUBCOMMAND> --help

List of plugins

To get a list of the available plugins, which includes agent plugins, data query plugins, discovery methods, and validator plugins, run mco plugin doc.

Agent plugin help

Related actions are bundled together in agent plugins. For example, Puppet-related actions are all in the puppet plugin.

To get detailed info on a given plugin's actions and their required inputs, run:

$ mco plugin doc <PLUGIN>

If there is also a data plugin with the same name, you may need to prepend agent/ to the plugin name to disambiguate:

$ mco plugin doc agent/<PLUGIN>

Invoking actions

MCollective actions are invoked with either the general purpose rpc subcommand or one of the special-purpose subcommands. Note that unless you specify a filter, commands will be run on every server in your Puppet Enterprise deployment; make sure you know what will happen before confirming any potentially disruptive commands.

The rpc subcommand

The most useful subcommand is mco rpc. This is the general purpose client, which can invoke actions from any agent plugin. See the list of built-in actions for more information about agent plugins.


$ mco rpc service restart service=httpd

The general form of an mco rpc command is:


For a list of available agent plugins, actions, and their required inputs, see the list of built-in actions actions or the information on getting help.

Special-purpose subcommands

Although mco rpc can invoke any action, sometimes a special-purpose application can provide a more convenient interface.


$ mco puppet runall 5

The puppet subcommand's special runall action is able to run many nodes without exceeding a certain load of concurrent runs. It does this by repeatedly invoking the puppet agent's status action, and only sending a runonce action to the next node if there's enough room in the concurrency limit.

This uses the same actions that the mco rpc command can invoke, but since rpc doesn't know that the output of the status action is relevant to the timing of the runonce action, it can't provide that improved UI.

Each special-purpose subcommand (puppet, service, and package) has its own CLI syntax. For example, mco service puts the name of the service before the action, to mimic the format of the more common platform-specific service commands:

$ mco service httpd status

Run mco help <SUBCOMMAND> to get specific help for each subcommand.

Filtering actions

By default, orchestration actions affect all managed nodes. You can limit any action to a smaller set of nodes by specifying a filter.

For example, the following filter checks the status of the pe-nginx service running on the node defined as the console.

$ mco service pe-nginx status --with-fact fact_is_puppetconsole=true

Note: For more details about filters, see the following pages from the MCollective docs:

All command line actions can accept the same filter options, which are listed under the "Host Filters" section of any mco help <SUBCOMMAND> text:

Host Filters
    -W, --with FILTER                Combined classes and facts filter
    -S, --select FILTER              Compound filter combining facts and classes
    -F, --wf, --with-fact fact=val   Match hosts with a certain fact
    -C, --wc, --with-class CLASS     Match hosts with a certain config management class
    -A, --wa, --with-agent AGENT     Match hosts with a certain agent
    -I, --wi, --with-identity IDENT  Match hosts with a certain configured identity

Each type of filter lets you specify a type of metadata and a desired value. The action will only run on nodes where that data has that desired value.

Any number of fact, class, and agent filters can also be combined in a single command; this will make it so nodes must match every filter to run the action.

Matching strings and regular expressions

Filter values are usually simple strings. These must match exactly and are case-sensitive.

Most filters can also accept regular expressions as their values; these are surrounded by forward slashes, and are interpreted as standard Ruby regular expressions. (You can even turn on various options for a subpattern, such as case insensitivity --- -F "osfamily=/(?i:redhat)/".) Unlike plain strings, they accept partial matches.

Filtering by identity

A node's "identity" is the same as its certname, as specified during installation. Identities will almost always be unique per node.

$ mco puppet runonce -I
  • You can use the -I or --with-identity option multiple times to create a filter that matches multiple specific nodes.
  • You cannot combine the identity filter with other filter types.
  • The identity filter accepts regular expressions.

Filtering by fact, class, and agent

  • Facts are the standard Puppet facts, which are available in your Puppet manifests. A list of the core facts is available here. Use the -F or --with-fact option with a fact=value pair to filter on facts.
  • Classes are the Puppet classes that are assigned to a node. This includes classes assigned in the console, assigned via Hiera, declared in site.pp, or declared indirectly by another class. Use the -C or --with-class option with a class name to filter on classes.
  • Agents are MCollective agent plugins. PE's default plugins are available on every node, so filtering by agent makes more sense if you are distributing custom plugins to only a subset of your nodes. For example, if you made an emergency change to a custom plugin that you distribute with Puppet, you could filter by agent to trigger an immediate run on all affected systems. (mco puppet runall 5 -A my_agent) Use the -A or --with-agent option to filter on agents.

Since mixing classes and facts is so common, you can also use the -W or --with option to supply a mixture of class names and fact=value pairs.

Compound "select" filters

The -S or --select option accepts arbitrarily complex filters. Like -W, it can accept a mixture of class names and fact=value pairs, but it has two extra tricks:

  • Boolean logic

    The -W filter always combines facts and classes with "and" logic --- nodes must match all of the criteria to match the filter.

    The -S filter lets you combine values with nested Boolean "and"/"or"/"not" logic:

    $ mco service httpd restart -S "((customer=acme and osfamily=RedHat) or and /apache/"
  • Data plugins

    In addition, the -S filter lets you use data plugin queries as an additional kind of metadata.

    Data plugins can be tricky, but are very powerful. To use them effectively, you must:

    1. Check the list of data plugins with mco plugin doc.
    2. Read the help for the data plugin you want to use, with mco plugin doc data/<NAME>. Note any required input and the available outputs.
    3. Use the rpcutil plugin's get_data action on a single node to check the format of the output you're interested in. This action requires source (the plugin name) and query (the input) arguments:
      $ mco rpc rpcutil get_data source="fstat" query="/etc/hosts" -I web01

      This will show all of the outputs for that plugin and input on that node.

    4. Construct a query fragment of the format <PLUGIN>('<INPUT>').<OUTPUT>=<VALUE> --- note the parentheses, the fact that the input must be in quotes, the .output notation, and the equals sign. Make sure the value you're searching for matches the expected format, which you saw when you did your test query.
    5. Use that fragment as part of a -S filter:
      $ mco find -S "fstat('/etc/hosts').md5=/baa3772104/ and osfamily=RedHat"

You can specify multiple data plugin query fragments per -S filter.

The MCollective documentation includes a page on writing custom data plugins. Installing custom data plugins is similar to installing custom agent plugins; see Adding new actions for details.

Testing filters with mco find

Before invoking any potentially disruptive action, like a service restart, you should test the filter with mco find or mco ping, to make sure your command will act on the nodes you expect.

Batching and limiting actions

By default, actions run simultaneously on all of the targeted nodes.

This is fast and powerful, but is sometimes not what you want:

  • Sometimes you want the option to cancel out of an action with control-C before all nodes have run it.
  • Sometimes, like when retrieving inventory data, you want to run a command on just a sample of nodes and don't need to see the results from everything that matches the filter.
  • Certain actions may consume limited capacity on a shared resource (such as the Puppet master server), and invoking them on a "thundering herd" of nodes can disrupt that resource.

In these cases, you can batch actions, to run all of the matching nodes in a controlled series, or limit them, to run only a subset of the matching nodes.


  • Use the --batch <SIZE> option to invoke an action on only <SIZE> nodes at once. PE will invoke it on the first <SIZE> nodes, wait briefly, invoke it on the next batch, and so on.
  • Use the --batch-sleep <SECONDS> option to control how long PE should sleep between batches.


  • Use the --limit <COUNT> option to invoke an action on only <COUNT> matching nodes. <COUNT> can be an absolute number or a percentage. The nodes will be chosen randomly.
  • Use the -1 or --one option to invoke an action on just one matching node, chosen randomly.