Pdsh

The pdsh command is a parallel version of the dsh (Distributed Shell). It is used to run commands on multiple remote hosts in parallel and supports a number of remote shell programs including SSH.

Cheat sheet

Depending on your pdsh install, it may default to rsh rather than ssh. Pass in -R ssh to change the default module.

Action	Command
List all installed modules	pdsh -L
Use a specific module. Eg. SSH	pdsh -R ssh -w node[0-10] uptime
Connect to a certain nodes with a specific module	pdsh -w ssh:node[0-5],rsh:node[6-7] uptime
Run a command on all hosts	pdsh -a uptime
Run a command on all hosts EXCEPT one	pdsh -a -x exclude-node uptime
Run a command on a set of hosts by hostname Or by a specific gender (ie. attribute), see below.	pdsh -w node[0-10] uptimepdsh -w cpus=8 uptime
Collect the OS version on all hosts and group it	pdsh -a cat /etc/redhat-release | dshbak -c

Installation

You can install pdsh from the epel repo.

# yum -y install epel-release
# yum -y install pdsh pdsh-rcmd-ssh

Alternatively, install from source from the GitHub repo at: https://github.com/chaos/pdsh

Modules

Run pdsh -L to list all remote command modules available.

You can see the default module that will be used when running pdsh --help.

SSH remote module

For SSH support, you will need to install the pdsh-rcmd-ssh package. Additional SSH arguments can be passed by passing in the PDSH_SSH_ARGS_APPEND environment variable. Eg:

# PDSH_SSH_ARGS_APPEND=”-o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -o LogLevel=error” pdsh -a uptime

Configuration

Genders file

The genders file can be used to give attributes to nodes in a cluster. It can be thought of as groups that can be used as a target when running pdsh commands. While the genders file isn't required, it's still a good idea to have set up because it allows the -a option to function properly.

The genders file is usually located at /etc/genders. Each line can contain one or more hosts and attributes are given as a comma delimited list. Here is an example genders file:

ca-controller management
ca-login cluster-a,login
ca[001-168] cluster-a,cpus=4
cb-login cluster-b,login
cb[091-320] cluster-b,cpus=8

Using this example, you can target:

• Only the control node with pdsh -w management
• Only cluster A nodes with pdsh -w cluster-a
• Only nodes with 8 CPUs with pdsh -w cpus=8

See also

• Libgenders man page talks about using genders and target grouping - https://linux.die.net/man/3/libgenders