clad - Parallel SSH client


version 1.11


 clad [options] <cluster> <command>
 clad --list
 clad --help


Clad provides the ability to run the same command on several hosts at once. The output is displayed unbuffered as the various hosts run the command. The list of hosts is determined by reading a configuration file which may also contain command aliases and environment settings.



Dry run, just show the command that would be executed and each host.


Do not colorize the host names in the output.

-l user

Specify a login name for all ssh connections.


Print out a lot of debugging information which may be useful in debugging issues with clad.


Force clad to wait for the command to finish on each host before continuing to the next. This will be slower, but may be easier to read the output.

--config name

Specify the name of an alternate configuration. For example if you use --config MyClad then the configuration file ~/etc/MyClad.conf will be used instead of ~/etc/Clad.conf.


Send the server code with the payload and feed into Perl on the remote end. This makes the total payload much larger, but it allows you to use clad with servers that do not have App::clad installed. The remote end must have Perl 5.6.1 or better in the PATH.

--max number

Limit the maximum number of simultaneous connections to number

--file filename

Copy files to the remote end as part of the payload. May be specified multiple times. The names of the files are available as environment variables FILE1, FILE2, etc. The files will automatically be removed on the remote end when the command completes. An example usage for this would be to install rpm packages:

 % clad --file Database-Server-0.01-1.noarch.rp mycluster 'rpm -U $FILE1'
--dir directory

Recursively copy the directory to the remote end as part of the payload. The name of the directory is available as an environment variable DIR. The directory will automatically be removed on the remote end when the command completes. For example if you are installing a directory full of rpm packages:

 % clad --dir ~/rpmbuild/RPMS/noarch mycluster 'rpm -U $DIR/*'

Do not print out standard output and standard input, just the exit values or signals returned from each host.

--log-dir dir

Specify a directory to write log files to. Each host will have its own log file.


Same as --log-dir, but the location is ~/clad/log.


Purge any logs that have collected under your home directory from using the --log option.


List the clusters and aliases defined in your configuration.


Print help and exit.


Print the version and exit.


The configuration file is a Clustericious::Config style configuration file. See "EXAMPLES" for an example configuration. It contains these sections and configuration items:


Environment hash to override environment variables on all hosts that run the command.


Hash to define the clusters. This is a hash of lists, where the keys are the cluster names and the lists are the host names. For example:

     - host1
     - host2
     - host3
     - host4

The old key (now deprecated) clusters is also recognized, if cluster is not specified. This deprecated key will be removed on or after January 31 2016.

You can use a single hostname not in the cluster section to specify a cluster of one host, so long as it is a legal hostname understood by ssh.


Hash of aliases. This is a useful place to specify common shortcuts. The values in this hash may be either strings or lists, allowing you to use the list or scalar form of system.

The old key (now deprecated) aliases is also recognized, if alias is not specified. This deprecated key will be removed on or after January 31 2016.


clad runs on both the client and the server. This specifies the command used to communicate with the client on the server end. Unless you are testing clad you probably won't need to change this.


Include the server code as part of the payload. This is useful for hosts that do not already have App::clad installed. This is the same as the --fat option above.


The command to execute on the server side when using the --fat command line option or the fat configuration option. The default is simply perl.


This is the ssh command to use on the client side. It is ssh by default.


These are the ssh options used when opening a connection to the server. The default may change as needed.


Extra ssh command line options to be added after ssh_options. If you just want to add a few options without replacing the existing set, this is the way to go.


A list of colors as understood by Term::ANSIColor which are used in alteration for each host to help separate the output visually.


Color to use if clad determined the remote call failed

exit with non zero
killed by signal
failed to start (usually due to a bad command)

The default is bold red.


Color to use for output to standard error. The default is bold yellow.


A hash of inline scripts. The keys are the script name and the values are the script bodies. For example, with

     for i in $( ls ); do
       echo item: $i

You can get directory listing with

 % clad cluster dir_listing


Here is an example configuration

   PATH: /home/starscream/perl5/bin:/usr/local/bin:/usr/bin:/bin
   PERL5LIB: /home/starscream/perl5/lib
     - mail1
     - mail2
     - www1
     - www2
     - www3
     - db1
     - db2
     - db3
     - db4
   config_init: git clone git1:/cm/config-$CLUSTER.git ~/etc
   config_update: cd ~/etc && git pull
   config_destory: rm -rf ~/etc


To find the uptime of the mailservers:

 % clad webservers uptime
 [mail1 out ]  21:27:04 up 4 days, 12:22,  0 users,  load average: 0.96, 1.01, 1.04
 [mail2 out ]  21:24:09 up 93 days, 12:52,  0 users,  load average: 1.25, 1.33, 1.29

To find the uptime of all servers in any cluster:

 % clad mailservers,webservers,databases
 [mail1 out ]  21:27:04 up 4 days, 12:22,  0 users,  load average: 0.96, 1.01, 1.04
 [mail2 out ]  21:24:09 up 93 days, 12:52,  0 users,  load average: 1.25, 1.33, 1.29
 [www1  out ]  21:24:37 up 93 days, 12:52,  0 users,  load average: 2.60, 2.34, 2.21
 [www2  out ]  21:23:06 up 93 days, 12:51,  0 users,  load average: 0.60, 0.50, 0.50
 [www3  out ]  21:24:05 up 93 days, 12:52,  0 users,  load average: 3.99, 3.62, 3.55
 [db1   out ]  21:24:53 up 93 days, 12:47,  0 users,  load average: 11.71, 12.15, 12.23
 [db2   out ]  21:26:07 up 93 days, 12:52,  0 users,  load average: 14.13, 13.91, 13.05
 [db3   out ]  21:29:06 up 93 days, 12:53,  0 users,  load average: 1.99, 1.59, 1.14
 [db4   out ]  21:24:55 up 93 days, 12:48,  0 users,  load average: 4.99, 4.83, 4.03

(note that the output in this example is displayed in order, though in practice it will usually be jumbled

log into hosts with different user

By default clad will login to the remote servers with what ever user is default for ssh (this is usually determined by the local user and / or the ssh configuration). You can use the -l option to specify a user name for all clusters in the command

 % clad -l foo mailservers,webservers,databases whoami
 [mail1 out ] foo
 [mail2 out ] foo
 [www1  out ] foo
 [www2  out ] foo
 [www3  out ] foo
 [db1   out ] foo
 [db2   out ] foo
 [db3   out ] foo
 [db4   out ] foo

or you can prefix individual clusters with a user name using the @ sign.

 % clad foo@mailservers,bar@webservers,baz@database whoami
 [mail1 out ] foo
 [mail2 out ] foo
 [www1  out ] bar
 [www2  out ] bar
 [www3  out ] bar
 [db1   out ] baz
 [db2   out ] baz
 [db3   out ] baz
 [db4   out ] baz

running Perl remotely

In the configuration above, we have specified PATH and PERL5LIB environment variables to work with the modules build for local::lib on each host (the actual configuration is probably a little more complicated), so we can use modules that we have installed in local::lib.

 % clad webservers -- perl -Mojo -E 'say g("")->dom->at("title")->text'
 [www1 out ]
 [www1 out ]       Mojolicious - Perl real-time web framework
 [www1 out ]
 [www2 out ]
 [www2 out ]       Mojolicious - Perl real-time web framework
 [www2 out ]
 [www3 out ]
 [www3 out ]       Mojolicious - Perl real-time web framework
 [www3 out ]

pulling remote configuration using git

Clustericious servers and client use configuration files that are usually stored in ~/etc. We usually manage these configurations on a cluster by cluster basis using git, and deploy them using clad.

For example, to initialize the configuration directory using the <config_init> alias:

   config_init: git clone git1:/cm/config-$CLUSTER.git ~/etc

and run:

 % clad webservers config_init

...we can update using the config_update alias:

   config_update: cd ~/etc && git pull

and run:

 % clad webservers config_update

...and if the configuration becomes hosed, we can remove it and start over. Since the master configuration is stored in git this may not be disaster.

   config_destory: rm -rf ~/etc

and run:

 % clad webservers config_destroy

using shell

clad runs the command on the remote end using the same exact arguments as you pass it on the client side. That means that it uses either the single argument or list version of system depending on input. That means that if you want to use shell logic, pipes or redirection, you need to use the single argument version! For example:

 % clad webservers cd ~/etc && git pull    # WRONG !
 % clad webservers 'cd ~/etc && git pull'  # RIGHT !

Sometimes if you don't want to worry about the escaping of meta characters the list version will be more appropriate

 % clad webservers perl -E 'say "hi there"'



This environment variable is set to the cluster name from the configuration file on each node that the command is run. The deprecated CLUSTER is also set, though that may be removed in a future version.


You can override the default values for the --fat, --server_command and --fat_server_command at install time using options to Build.PL.

 perl Build.PL --clad_fat \
               --clad_server_command /usr/local/bin/perl \
               --server_command /usr/local/bin/perl /usr/local/bin/clad --server

In this example, we specify fully qualified pathnames for Perl and clad, which may be what you want in environments where the system Perl (usually installed in /usr/bin/perl) comes before the Perl that you want to use.


Clustericious::Admin and clad require an AnyEvent event loop that allows entering the event loop by calling recv on a condition variable. This is not supported by all AnyEvent event loops and is discouraged by the AnyEvent documentation for CPAN modules, though most of the important event loops, such as EV and the pure perl implementation that comes with AnyEvent DO support this behavior.


Graham Ollis <>


This software is copyright (c) 2015 by Graham Ollis.

This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.