Michael Trojanek (relativkreativ) — Bootstrapper and creator of things

This article was published on September 29th 2014 and takes about 5 minutes to read.

Use it with caution — it is probably still valid, but it has not been updated for over a year.

5 bash tips useful for scripted provisioning

Setting up servers with a shell script is a great way to reduce errors, spare yourself from having to carry out the same steps over and over again and provision a machine in minutes.

You don't have to necessarily use a provisioning tool like Puppet or Chef to provision a server. To get started, a shell script is enough to promote your machine from a base installation to a full-fledged box.

But you have to know how to carry out a few common tasks that are crucial for doing system administration tasks from inside a script.

Although this article is written with bash in mind, these tips will also work with other shells with small modifications.

Modifying a configuration file

Modifying a configuration file is one of the most common tasks needed in a provisioning script. It can easily be done using the sed command.

As an example, you may want to prohibit logins as root via SSH:

sed -i 's/^#\?PermitRootLogin \(no\|yes\)/PermitRootLogin no/' /etc/ssh/sshd_config
/etc/init.d/sshd restart

Keep in mind that while sed works with regular expressions, you have to escape special characters (in this example ?, | and the braces) with a backslash. It's also worth noting that in order to replace a whole line, you have to adjust your regular expression accordingly.

In any case you should test your command carefully before using it in a script (if you run sed without the -i flag, it prints the modified output to the screen which can be of help).

There are quite a few different versions of sed and depending on your environment, the above command may or may not work without modifications (its syntax works on CentOS 6.5 but does not on Mac OS X for example).

sed is also a very powerful command capable of a whole lot more than just small modifications like this one. People have written books about it but as always, the man page is a good starting point to dig deeper.

Reloading the environment

If you want to reload the environment for the user which runs the current script, you can do so by sourcing the corresponding configuration file (source ~/.bash_profile or source ~/.bashrc).

Maybe you have installed rbenv and want to reload your environment so that rbenv is initialized:

git clone https://github.com/sstephenson/rbenv.git ~/.rbenv
echo 'export PATH="$HOME/.rbenv/bin:$PATH"' >> ~/.bash_profile
echo 'eval "$(rbenv init -)"' >> ~/.bash_profile
source ~/.bash_profile

If you are not sure which file to load (or even which file to put your configuration directives in), take a look at the chapter Bash startup files in the Bash reference manual.

Running a series of commands as another user

Using a combination of the su command and bash's heredoc syntax, you can easily run a series of commands as another user:

su - app <<END
  git clone https://github.com/sstephenson/rbenv.git ~/.rbenv
  git clone https://github.com/sstephenson/ruby-build.git ~/.rbenv/plugins/ruby-build
  echo 'export PATH="$HOME/.rbenv/bin:$PATH"' >> ~/.bash_profile
  echo 'eval "$(rbenv init -)"' >> ~/.bash_profile
  source ~/.bash_profile
  rbenv install 2.1.2
  rbenv global 2.1.2
  gem install bundler

If you want to print a literal $ inside a heredoc, you have to escape it with a backslash (\$) to prevent variable expansion.

In combination with the cat command, heredocs can also be used to write whole configuration files at once:

cat > /etc/profile.d/rails.sh <<END
# Set a system-wide Rails environment
export RAILS_ENV=production

However, using this technique things can get messy when your files are big or if you need to write a lot of them. In this case it may be worth considering to create the files beforehand and simply copy them over their original using cp -f.

Printing each command in a script before execution

If your script takes some time to execute or you use a lot of commands which do not generate output (cp and mv for example), then it is useful to have each command printed to screen (or wherever you redirect the output to) before it is executed.

You can achieve this by setting set -x at the start of your script. If you want this detailed output just for a section of your script, you can turn this behaviour off with set +x.

The set command has a bunch of other useful switches which are explained in its man page.

Writing a log file while still having output on screen

To redirect all output of a script (including errors printed to STDERR) to a file, you have to first redirect STDERR to STDOUT:

sh script.sh 2>&1 > script.log

This way, all of the script's output is captured in the file script.log. However, when you do this, you will get no output on screen.

You can use the tee command to capture ouput in a file while still having output sent to STDOUT:

sh script.sh 2>&1 | tee script.log

A pipe only reads from STDOUT in any case, so leave the redirection of STDERR in place. If you want to append to the file instead of replacing it, use tee -a script.log.

Get in the loop

Join my email list to get new articles delivered straight to your inbox and discounts on my products.

No spam — guaranteed.

You can unsubscribe at any time.

Got it, thanks a lot!

Please check your emails for the confirmation request I just sent you. Once you clicked the link therein, you will no longer see these signup forms.