22. Best practice

Practices used in real environments

22.1. Passwordless connections

It is recommended to run cdist with public key authentication. This requires a private/public key pair and the entry "PermitRootLogin without-password" in the sshd server. See sshd_config(5) and ssh-keygen(1).

22.2. Speeding up ssh connections

When connecting to a new host, the initial delay with ssh connections is pretty big. As cdist makes many connections to each host successive connections can be sped up by "sharing of multiple sessions over a single network connection" (quote from ssh_config(5)). This is also called "connection multiplexing".

Cdist implements this since v4.0.0 by executing ssh with the appropriate options (-o ControlMaster=auto -o ControlPath=/tmp/<tmpdir>/s -o ControlPersist=2h).

Note that the sshd_config on the server can configure the maximum number of parallel multiplexed connections this with MaxSessions N (N defaults to 10 for OpenSSH v7.4).

22.3. Speeding up shell execution

On the source host, ensure that /bin/sh is not bash: bash is quite slow for script execution. Instead, you could use dash after installing it:

ln -sf /bin/dash /bin/sh

22.4. Multi master or environment setups

If you plan to distribute cdist among servers or use different environments, you can do so easily with the included version control git. For instance if you plan to use the typical three environments production, integration and development, you can realise this with git branches:

# Go to cdist checkout
cd /path/to/cdist

# Create branches
git branch development
git branch integration
git branch production

# Make use of a branch, for instance production
git checkout production

Similar if you want to have cdist checked out at multiple machines, you can clone it multiple times:

machine-a % git clone git://your-git-server/cdist
machine-b % git clone git://your-git-server/cdist

22.5. Separating work by groups

If you are working with different groups on one cdist-configuration, you can delegate to other manifests and have the groups edit only their manifests. You can use the following snippet in conf/manifests/init:

# Include other groups
sh -e "$__manifest/systems"

sh -e "$__manifest/cbrg"

22.6. Maintaining multiple configurations

When you need to manage multiple sites with cdist, like company_a, company_b and private for instance, you can easily use git for this purpose. Including a possible common base that is reused across the different sites:

# create branches
git branch company_a company_b common private

# make stuff for company a
git checkout company_a
# work, commit, etc.

# make stuff for company b
git checkout company_b
# work, commit, etc.

# make stuff relevant for all sites
git checkout common
# work, commit, etc.

# change to private and include latest common stuff
git checkout private
git merge common

The following .git/config is taken from a real world scenario:

# Track upstream, merge from time to time
[remote "upstream"]
   url = git://git.schottelius.org/cdist
   fetch = +refs/heads/*:refs/remotes/upstream/*

# Same as upstream, but works when being offline
[remote "local"]
   fetch = +refs/heads/*:refs/remotes/local/*
   url = /home/users/nico/p/cdist

# Remote containing various ETH internal branches
[remote "eth"]
   url = sans.ethz.ch:/home/services/sans/git/cdist-eth
   fetch = +refs/heads/*:refs/remotes/eth/*

# Public remote that contains my private changes to cdist upstream
[remote "nico"]
   url = git.schottelius.org:/home/services/git/cdist-nico
   fetch = +refs/heads/*:refs/remotes/nico/*

# The "nico" branch will be synced with the remote nico, branch master
[branch "nico"]
   remote = nico
   merge = refs/heads/master

# ETH stable contains rock solid configurations used in various places
[branch "eth-stable"]
   remote = eth
   merge = refs/heads/stable

Have a look at git-remote(1) to adjust the remote configuration, which allows

22.7. Multiple developers with different trust

If you are working in an environment that requires different people to work on the same configuration, but having different privileges, you can implement this scenario with a gateway host and sudo:

  • Create a dedicated user (for instance cdist)

  • Setup the ssh-pubkey for this user that has the right to configure all hosts

  • Create a wrapper to update the cdist configuration in ~cdist/cdist

  • Allow every developer to execute this script via sudo as the user cdist

  • Allow run of cdist as user cdist on specific hosts on a per user/group basis.

    • f.i. nico ALL=(ALL) NOPASSWD: /home/cdist/bin/cdist config hostabc

For more details consult sudoers(5)

22.8. Templating

  • create directory files/ in your type (convention)
  • create the template as an executable file like files/basic.conf.sh, it will output text using shell variables for the values
# in the template, use cat << eof (here document) to output the text
# and use standard shell variables in the template
# output everything in the template script to stdout
cat << EOF
server {
  listen                          80;
  server_name                     $SERVERNAME;
  root                            $ROOT;

  access_log /var/log/nginx/$SERVERNAME_access.log
  error_log /var/log/nginx/$SERVERNAME_error.log
  • in the manifest, export the relevant variables and add the following lines to your manifest:
# export variables needed for the template
  export SERVERNAME='test"
  export ROOT='/var/www/test'
# render the template
  mkdir -p "$__object/files"
  "$__type/files/basic.conf.sh" > "$__object/files/basic.conf"
# send the rendered template
  __file /etc/nginx/sites-available/test.conf  \
    --state present
    --source "$__object/files/basic.conf"

22.9. Testing a new type

If you want to test a new type on a node, you can tell cdist to only use an object of this type: Use the '--initial-manifest' parameter with - (stdin) as argument and feed object into stdin of cdist:

# Singleton type without parameter
echo __ungleich_munin_server | cdist --initial-manifest - munin.panter.ch

# Singleton type with parameter
echo __ungleich_munin_node --allow | \
    cdist --initial-manifest - rails-19.panter.ch

# Normal type
echo __file /tmp/stdintest --mode 0644 | \
    cdist --initial-manifest - cdist-dev-01.ungleich.ch

22.10. Other content in cdist repository

Usually the cdist repository contains all configuration items. Sometimes you may have additional resources that you would like to store in your central configuration repository (like password files from KeepassX, Libreoffice diagrams, etc.).

It is recommended to use a subfolder named "non-cdist" in the repository for such content: It allows you to easily distinguish what is used by cdist and what is not and also to store all important files in one repository.


With CDIST_ORDER_DEPENDENCY all types are executed in the order in which they are created in the manifest. The current created object automatically depends on the previously created object.

It essentially helps you to build up blocks of code that build upon each other (like first creating the directory xyz than the file below the directory).

This can be helpful, but it can also be the source of evil.

22.11.1. CDIST_ORDER_DEPENDENCY easily causes unobvious dependency cycles

Let's see an example. Suppose you have special init manifest where among other things you are assuring that remote host has packages sudo and curl installed.



for p in sudo curl
   __package "${p}"

Then you have some other special init manifest where among other things you are assuring sudo package is installed.



__package sudo

Then you have third init manifest where you combine those two init manifests, by including them:


sh -e "$__manifest/init1"
sh -e "$__manifest/init2"

The resulting init manifest is then equal to:


for p in sudo curl
   __package "${p}"


__package sudo

In the end you get the following dependencies:

  • __package/curl depends on __package/sudo
  • __package/sudo depends on __package/curl

And here you have a circular dependency!

In the real world manifest can be quite complex, dependencies can become complicated and circual dependencies are not so obvious. Resolving it can become cumbersome.

Practical solution?

Instead of managing complex init manifests you can write custom types. Each custom type can do one thing, it has well defined dependencies that will not leak into init manifest. In custom type you can also add special explorers and gencode.

Then, in init manifest you combine your complex types. It is:

  • cleaner
  • easier to follow
  • easier to maintain
  • easier to debug.

22.11.2. CDIST_ORDER_DEPENDENCY kills parallelization

Suppose you have defined CDIST_ORDER_DEPENDENCY and then, among other things, you specify creation of three, by nature independent, files.



__file /tmp/file1
__file /tmp/file2
__file /tmp/file3

Due to defined CDIST_ORDER_DEPENDENCY cdist will execute them in specified order. It is better to use CDIST_ORDER_DEPENDENCY in well defined blocks:



__file /tmp/file1
__file /tmp/file2
__file /tmp/file3