Puppet's services: Puppet apply

This version is out of date. For current versions, see Puppet packages and versions.

Puppet apply is an application that compiles and manages configurations on nodes. It acts like a self-contained combination of the Puppet master and Puppet agent applications. For more info about Puppet’s architecture, see Overview of Puppet’s Architecture — in particular, read the note at the end about differences and trade-offs between agent/master and puppet apply.

For details about invoking the Puppet apply command, see the puppet apply man page.

Supported platforms

Puppet apply runs similarly on both *nix and Windows systems.

Not all operating systems can manage the same resources with Puppet; some resource types are OS-specific, and others have OS-specific features. See the resource type reference for details.

Puppet apply’s run environment

Unlike Puppet agent, Puppet apply never runs as a daemon or service. It always runs as a single task in the foreground, which compiles a catalog, applies it, files a report, and exits.

By default, it never initiates outbound network connections, although it can be configured to do so. It never accepts inbound network connections.

Main manifest

Like the Puppet master application, Puppet apply uses its settings (such as basemodulepath) and the configured environments to locate the Puppet code and configuration data it will use when compiling a catalog.

The one exception is the main manifest. Puppet apply always requires a single command line argument, which acts as its main manifest. It ignores the main manifest from its environment.

(Alternately, you can write a tiny main manifest directly on the command line, with the -e option. See the puppet apply man page for details.)


Puppet apply runs as whichever user executed the Puppet apply command.

To manage a complete system, you should run Puppet apply as:

  • root on *nix systems
  • Either LocalService or a member of the Administrators group on Windows systems

Puppet apply can also run as a non-root user. This will restrict the resources that Puppet can manage (see below).

Resource types for non-root Puppet apply runs

When running without root permissions, most of Puppet’s resource providers cannot use sudo to elevate permissions. This means Puppet can only manage resources that its user can modify without using sudo.

Out of the core resource types listed in the resource type reference, the following are available to non-root Puppet apply runs:

  • cron (only non-root cron jobs can be viewed or set)
  • exec (cannot run as another user or group)
  • file (only if the non-root user has read/write privileges)
  • notify
  • schedule
  • ssh_key
  • ssh_authorized_key
  • service (for services that don’t require root. You can also use the start, stop, and status attributes to specify how non-root users should control the service; see the tips and examples for the service type for more info.)
  • augeas

If you need to install packages into a directory controlled by a non-root user, you can either use an exec to unzip a tarball or use a recursive file resource to copy a directory into place.

Network access

By default, Puppet apply does not communicate over the network. It uses its local collection of modules for any file sources, and does not submit reports to a central server.

Depending on your system and the resources you are managing, it might download packages from your configured package repositories or access files on UNC shares.

If you have configured an external node classifier (ENC), your ENC script might create an outbound HTTP connection. Additionally, if you’ve configured the HTTP report processor, Puppet agent sends reports via HTTP or HTTPS.

If you have configured PuppetDB, Puppet apply will create outbound HTTPS connections to PuppetDB.


By default, Puppet apply logs directly to the terminal. This is good for interactive use, and less good when running as a scheduled task or cron job.

You can adjust how verbose the logs are with the log_level setting, which defaults to notice. Setting it to info is equivalent to running with the --verbose option, and setting it to debug is equivalent to --debug. You can also make logs quieter by dialing back to warning or lower.

When started with the --logdest syslog option, Puppet apply logs to the *nix syslog service. Your syslog configuration dictates where these messages will be saved, but the default location is /var/log/messages on Linux, /var/log/system.log on Mac OS X, and /var/adm/messages on Solaris.

When started with the --logdest eventlog option, it logs to the Windows Event Log. You can view its logs by browsing the Event Viewer. (Control Panel → System and Security → Administrative Tools → Event Viewer)

When started with the --logdest <FILE> option, it logs to the file specified by <FILE>.


In addition to local logging, Puppet apply will process a report using its configured report handlers, like a Puppet master does. You can enable different reports with the reports setting; see the list of available reports for more info. For more about reporting, see the docs on reporting.

To disable reporting and avoid taking up disk space with the store report handler, you can set report = false in puppet.conf.

Managing systems with Puppet apply

In a normal Puppet site, every node should periodically do configuration runs, to revert unwanted changes and to pick up recent updates.

Since Puppet apply doesn’t run as a service, you must manually create a scheduled task or cron job if you want it to run on a regular basis.

On *nix, you can use the Puppet resource command to set up a cron job. Below is an example that runs Puppet once an hour; adjust the path to the Puppet command if you are not using Puppet Enterprise.

sudo puppet resource cron puppet-apply ensure=present user=root minute=30 command='/opt/puppetlabs/bin/puppet apply /etc/puppetlabs/puppet/manifests --logdest syslog'

Configuring Puppet apply

Puppet apply should be configured with puppet.conf, using the [user] and/or [main] section. For notes on which settings are most relevant to puppet apply, see the short list of important settings.

Puppet sites use proprietary and third-party cookies. By using our sites, you agree to our cookie policy.