Using MCollective Command Line Applications

MCollective is designed first and foremost for the CLI. You will mostly interact with a single executable called mco which has a number of sub-commands, arguments and flags.

Note: In Puppet Enterprise, the peadmin user is the only user who can execute mco commands. To execute commands as the peadmin user, see the Puppet Enterprise documentation.

Basic Usage of the mco Command

A simple example of a mco command can be seen below:

% mco ping
dev8                                     time=126.19 ms
dev6                                     time=132.79 ms
dev10                                    time=133.57 ms

---- ping statistics ----
25 replies max: 305.58 min: 57.50 avg: 113.16

In this example the ping sub-command is referred to as an application. Mcollective provides many applications, for a list of them, type mco help. You can also create your own application to plug into the framework. The help sub-command will show you something like this:

% mco help
The Marionette Collective version 2.0.0

  controller      Control the mcollective daemon
  facts           Reports on usage for a specific fact
  find            Find hosts matching criteria
  help            Application list and help
  inventory       General reporting tool for nodes, collectives and subcollectives
  ping            Ping all nodes
  plugin          MCollective Plugin Application
  rpc             Generic RPC agent client application

You can request help for a specific application using either mco help application or mco application —help. Shown below is part of the help for the rpc application:

% mco help rpc
Generic RPC agent client application

Usage: mco rpc [options] [filters] --agent <agent> --action <action> [--argument <key=val> --argument ...]
Usage: mco rpc [options] [filters] <agent> <action> [<key=val> <key=val> ...]

        --no-results, --nr           Do not process results, just send request
    -a, --agent AGENT                Agent to call
        --action ACTION              Action to call
        --arg, --argument ARGUMENT   Arguments to pass to agent

        --np, --no-progress          Do not show the progress bar
    -1, --one                        Send request to only one discovered nodes
        --batch SIZE                 Do requests in batches
        --batch-sleep SECONDS        Sleep time between batches
        --limit-nodes, --ln COUNT    Send request to only a subset of nodes, can be a percentage
    -j, --json                       Produce JSON output
    -c, --config FILE                Load configuratuion from file rather than default
    -v, --verbose                    Be verbose
    -h, --help                       Display this screen

Common Options
    -T, --target COLLECTIVE          Target messages to a specific sub collective
        --dt, --discovery-timeout SECONDS
                                     Timeout for doing discovery
    -t, --timeout SECONDS            Timeout for calling remote agents
    -q, --quiet                      Do not be verbose
        --ttl TTL                    Set the message validity period
        --reply-to TARGET            Set a custom target for replies

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

The Marionette Collective 2.0.0

The help first shows a basic overview of the command line syntax followed by options specific to this command. Following that you will see some Common Options and Host Filters that generally apply to most applications.

Making RPC Requests

Overview of a Request

The rpc application is the main application used to make requests to your servers. It is capable of interacting with any standard Remote Procedure Call (RPC) agent. Below is an example that shows an attempt to start a webserver on several machines:

% mco rpc service start service=httpd
Determining the amount of hosts matching filter for 2 seconds .... 10

 * [ ============================================================> ] 10 / 10

dev4                                     Request Aborted
   Could not start Service[httpd]: Execution of '/sbin/service httpd start' returned 1:

Finished processing 10 / 10 hosts in 1323.61 ms

The order of events in this process are:

  • Perform discovery against the network and discover 10 servers
  • Send the request and then show a progress bar of the replies
  • Show any results that were out of the ordinary
  • Show some statistics

Mcollective client applications aim to only provide the most relevant information. In this case, the application is not showing verbose information about the nine OK results, since the most important issue is the one Failure. Keep this in mind when viewing the results of commands.

Anatomy of a Request

MCollective agents are broken up into actions and each action can take input arguments.

% mco rpc service stop service=httpd

This shows the basic make-up of an RPC command. In this case we are:

  • using the rpc application - a generic application that can interact with any agent
  • directing our request to machines with the service agent
  • sending a request to the stop action of the service agent
  • supplying a value, httpd, to the service argument of the stop action

The same command has a longer form as well:

% mco rpc --agent service --action stop --argument service=httpd

These two commands are functionally identical.

Discovering Available Agents

The above command showed you how to interact with the service agent, but how can you find out that this agent even exists? On a correctly installed MCollective system you can use the plugin application to get a list:

% mco plugin doc
Please specify a plugin. Available plugins are:

  package         Install and uninstall software packages
  puppetd         Run puppet agent, get its status, and enable/disable it
  rpcutil         General helpful actions that expose stats and internals to SimpleRPC clients
  service         Agent to manage services

The first part of this list shows all the agents this computer is aware of. In order to show up on this list, an agent must have a DDL file and be installed locally.

To find out the actions, inputs and outputs for a specific agent use the plugin application again:

% mco plugin doc service
SimpleRPC Service Agent

Agent to manage services

      Author: R.I.Pienaar
     Version: 1.2
     License: GPLv2
     Timeout: 60
   Home Page:

   restart, start, status, stop

   status action:
       Gets the status of a service

              Description: The service to get the status for
                   Prompt: Service Name
                     Type: string
               Validation: ^[a-zA-Z\-_\d]+$
                   Length: 30

              Description: The status of service
               Display As: Service Status

This shows a truncated example of the auto-generated help for the service agent. First shown is metadata such as version, author and license. This is followed by the list of actions available, in this case the restart, start, status and stop actions.

Further information is shown about each action. For example, you can see that the status action requires an input called service which is a string, has a maximum length of 30, etc. You can also see you will receive one output called status

With this information, you can request the status for a specific service:

% mco rpc service status service=httpd
Determining the amount of hosts matching filter for 2 seconds .... 10

 * [ ============================================================> ] 10 / 10

   Service Status: stopped

   Service Status: stopped


Finished processing 10 / 10 hosts in 326.01 ms

Unlike the previous example, in this case specific information is returned on the success of the action. This is because this specific action is meant to retrieve information and so mcollective assumes you would like to see complete, thorough data regardless of success or failure.

Note that this output displays Service Status as shown in the mco plugin doc service help page. Any time you need more information about a display name, the doc for the associated agent will have a Description section for every input and output.

Selecting Request Targets Using Filters

Basic Filters

A key capability of mcollective is fast discovery of network resources. Discovery rules are written using filters. For example:

% mco rpc service status service=httpd -S "environment=development or customer=acme"

This shows a filter rule that limits the RPC request to being run on machines that are either in the Puppet environment development or belong to the Customer acme.

Filtering can be based on facts, the presence of a Configuration Management Class on the node, the node’s Identity, or installed Agents on the node.

Here are a number of examples of this with short descriptions of each filter:

# all machines with the service agent
% mco ping -A service
% mco ping --with-agent service

# all machines with the apache class on them
% mco ping -C apache
% mco ping --with-class apache

# all machines with a class that match the regular expression
% mco ping -C /service/

# all machines in the UK
% mco ping -F country=uk
% mco ping --with-fact country=uk

# all machines in either UK or USA
% mco ping -F "country=/uk|us/"

# just the machines called dev1 or dev2
% mco ping -I dev1 -I dev2

# all machines in the domain
% mco ping -I /$/

As you can see, you can filter by Agent, Class and/or Fact, and you can use regular expressions almost anywhere. You can also combine filters additively in a command so that all the criteria have to be matched.

Note: You can use a shortcut to combine Class and Fact filters:

# all machines with classes matching /apache/ in the UK
% mco ping -W "/apache/ location=uk"

Complex Compound or Select Queries

While the above examples are easy to enter, they are limited in that they can only combine filters additively. If you want to create searches with more complex boolean logic use the -S switch. For example:

% mco ping -S "((customer=acme and environment=staging) or environment=development) and /apache/"

The above example shows a scenario where the development environment is usually labeled development but one customer has chosen to use staging. You want to find all machines in those customer’s environments that match the class apache. This search would be impossible using the previously shown methods, but the above command uses -S to allow the use of boolean operators such as and and or so you can easily build the logic of the search.

The -S switch also allows for negative matches using not or !:

% mco ping -S "environment=development and !customer=acme"
% mco ping -S "environment=development and not customer=acme"

Filtering Using Data Plugins

As of version 2.1.0, custom data plugins can also be used to create complex filters:

% mco ping -S "fstat('/etc/hosts').md5=/baa3772104/ and environment=development"

This will search for the md5 hash of a specific file with matches restricted to the development environment. Note that as before, regular expressions can also be used.

As with agents, you can also discover which plugins are available for use:

% mco plugin doc

Please specify a plugin. Available plugins are:


Data Queries:
  agent           Meta data about installed MColletive Agents
  augeas_match    Allows agents and discovery to do Augeas match lookups
  fstat           Retrieve file stat data for a given file
  resource        Information about Puppet managed resources
  sysctl          Retrieve values for a given sysctl

For information on the input these plugins take and output they provide use the mco plugin doc fstat command.

Currently, each data function can only accept one input while matches are restricted to a single output field per invocation.

Chaining RPC Requests

The rpc application can chain commands one after the other. The example below uses the package agent to find machines with a specific version of mcollective and then schedules Puppet runs on those machines:

% mco rpc package status package=mcollective -j|jgrep "" |mco rpc puppetd runonce

Mcollective results can also be filtered using the opensource gem, jgrep. Mcollective data output is fully compatible with jgrep.

Using with PuppetDB

Recent versions of PuppetDB has a built in query language called Puppet Query Language that you use via the puppet query command.

Much like the above example of chaining RPC requests MCollective supports reading results from Puppet Query:

% puppet query "inventory { = 'CentOS' }"| mco rpc puppetd runonce

This will run Puppet on all CentOS machines

Seeing the Raw Data

By default the rpc application will try to show human-readable data. To see the actual raw data, add the -v flag to disable the display helpers:

% mco rpc nrpe runcommand command=check_load -I dev1 -v
dev1                                    : OK
    {:exitcode=>0,     :output=>"OK - load average: 0.00, 0.00, 0.00",     :perfdata=>      "load1=0.000;1.500;2.000;0; load5=0.000;1.500;2.000;0; load15=0.000;1.500;2.000;0;"}

This data can also be returned in JSON format:

% mco rpc nrpe runcommand command=check_load -I dev1 -j
    "action": "runcommand",
    "agent": "nrpe",
    "data": {
      "exitcode": 0,
      "output": "OK - load average: 0.00, 0.00, 0.00",
      "perfdata": "load1=0.000;1.500;2.000;0; load5=0.000;1.500;2.000;0; load15=0.000;1.500;2.000;0;"
    "statuscode": 0,
    "statusmsg": "OK",
    "sender": "dev1"

Error Messaging

When an application encounters an error, it returns an explanatory string:

% mco rpc rpcutil foo
rpc failed to run: Attempted to call action foo for rpcutil but it's not declared in the DDL (MCollective::DDLValidationError)

By default only an abbreviated error string is shown that provides some insight into the nature of the problem. For more details, add the -v flag to show a full stack trace:

% mco rpc rpcutil foo -v
rpc failed to run: Attempted to call action foo for rpcutil but it's not declared in the DDL (MCollective::DDLValidationError)
        from /usr/lib/ruby/site_ruby/1.8/mcollective/ddl.rb:303:in `validate_rpc_request'
        from /usr/lib/ruby/site_ruby/1.8/mcollective/rpc/client.rb:218:in `method_missing'
        from /home/rip/.mcollective.d/lib/mcollective/application/rpc.rb:133:in `send'
        from /home/rip/.mcollective.d/lib/mcollective/application/rpc.rb:133:in `main'
        from /usr/lib/ruby/site_ruby/1.8/mcollective/application.rb:283:in `run'
        from /usr/lib/ruby/site_ruby/1.8/mcollective/applications.rb:23:in `run'
        from /usr/bin/mco:20

Exit codes

When mco finishes, it generates an exit code. The returned exit code depends on the nature of the issue:

  • 0: If nodes were discovered and all passed.
  • 0: If no discovery was performed, at least 1 response was received, and all responses were OK.
  • 1: If no nodes were discovered, or if mco encountered an issue not listed here.
  • 2: If nodes were discovered but some RPC requests failed.
  • 3: If nodes were discovered, but no responses were received.
  • 4: If no discovery was performed, and no responses were received.

Custom Applications

The rpc application should suit most needs. However, sometimes the data being returned calls for customization such as custom aggregation, summarising or complete custom display.

In such cases, a custom application may be useful For example, the package application provides concluding summaries and provides some basic safe guards for its use. The agent also provides the commonly required data. Typical package output looks like this:

% mco package status kernel
Do you really want to operate on packages unfiltered? (y/n): y

 * [ ============================================================> ] 25 / 25

 dev5                                     version = kernel-2.6.32-220.7.1.el6
 dev9                                     version = kernel-2.6.32-220.2.1.el6

---- package agent summary ----
           Nodes: 25 / 25
        Versions: 9 * 2.6.32-220.2.1.el6, 9 * 2.6.32-220.4.1.el6, 7 * 2.6.32-220.el6
    Elapsed Time: 3.95 s

Notice how this application recognises that you are acting on all possible machines, an action which might have a big impact on your YUM servers. Consequently, package prompts for confirmation and, at the end of processing, displays a brief summary of the network status.

While the behaviors of custom applications are not always consistent with each other, in general they accept the standard discovery flags. For details of which flags are accepted in a given application, use the mco help appname command.

To discover which custom applications are available, run mco or mco help.

↑ Back to top