Overview

A tool to compare two Puppet catalogs.

While upgrading versions of Puppet or refactoring Puppet code you want to ensure that no unexpected changes will be made prior to doing the upgrade.

This tool will allow you to diff catalogs created by different versions of Puppet. This will let you guage the impact of a Puppet upgrade before actually touching any of your nodes.

This tool is delivered as a Puppet Face. It thus requires a Puppet installation to properly run.

The diff tool recognizes catalogs in yaml, marshall, or pson format. Currently automatic generation of the catalogs is done in the pson format.

The tool can automatically compile the catalogs for both your new and older servers. It can ask the master to use the yaml cache to compile the catalog for the last known environment with the last known facts. It can then validate against the rest terminus ( or by proxy puppetdb ) that the node is still active. This filtered list should contain only machines that have not been decommissioned in puppetdb (important as compiling their catalogs would also reactive them and their exports otherwise).

Usage

Before starting you need to copy or mount the contents of your current master's yamldir on the diff node, new master and old master.If you have multiple masters then combine the yamldirs of all nodes to give the fullest picture of all catalogs

You can retrieve the current yamldir location with the following command: puppet master --configprint yamldir. If you are using Puppet Enterprise this directory is '/var/opt/lib/pe-puppet/yaml'. It is not required to use a specific "diff" node , as you could use the "new" puppet server.

Once the yamldir is in place you need to allow access to the "diff" node to compile the catalogs for all nodes on both your old and new masters. In your confdir modify auth.conf to add allows for both /facts and /catalog. If there is an existing reference i.e. the $1 back reference for machines to compile their own catalog then simply add another line with the certificate name of the diff machine. As mentioned this can be the new master as required.

# allow the diff server to query facts
path  /facts
method find, search
auth any
allow diff.example.com

# allow the diff server to retrieve any catalog
path ~ ^/catalog/([^/]+)$
method find
allow $1
allow diff.example.com

The /facts ACL is optional onlly if you are using puppetdb and running the query from the master. You can pass --use_puppetdb to query the puppetdb server directly rather than via the rest terminus. However the rest terminus should be the same list of nodes if they are both backed by puppetdb and so this option is only for environments where puppetdb is more authoritative in some way they the facts rest query. This option is provided for future compatibility.

You can run this face without root access if you run the script as the puppet user on the system. The following is an example script doing so. You can alternatively install this in your module path on Puppet 3 and higher without the need of exporting RUBYLIB. One example of when you would need to run this as non root would be if you mounted the yamldir via NFS and had root squash enabled.

#!/bin/bash -x
if !-d "${HOME}/puppet-catalog-diff" ; then
  git clone https://github.com/acidprime/puppet-catalog-diff.git
fi

# These should be changed for Puppet Enterprise
export RUBYLIB="${HOME}/puppet-catalog-diff/lib/"
export YAMLDIR='/var/lib/puppet/yaml'
export SSLDIR='/var/lib/puppet/ssl'
export PUSER='puppet'
[ $USER == $PUSER ] || exit 1
time puppet catalog diff puppet2.example.com puppet3.example.com \
--show_resource_diff \
--content_diff \
--yamldir $YAMLDIR \
--ssldir $SSLDIR \
--changed_depth 1000 \
--configtimeout 1000 \
--output_report "${HOME}/lastrun-$$.json" \
--debug \
\ #--fact_search kernel='Darwin' \
--threads 50 | tee -a lastrun-$$.log

Multi threaded compile requests

You can change the number of concurrent connections to the masters by passing an interger to the --threads option. This will balence the catalogs evenly on the old and new masters. This option defaults to 10 and in testing 50 threads seemed correct for 4 masters with two load balancers.

Note: When using catalog diff to compare directories, one thread per catalog comparison will be created. However, since Ruby cannot take advantage of multiple CPUs this may be of limited use comparing local catalogs. If the 'parallel' gem is installed, then one process will be forked off per CPU on the system, allowing use of all CPUs.

Fact search

You can pass --fact_search to filter the list of nodes based on a single fact value. This currently defaults to kernel=Linux if you do not pass it. The yaml cache will be queried to find all nodes whose fact matches that value. Once a list of nodes with known facts is compiled a rest or puppetdb connection ( as mentioned above) filters the list to only nodes who are "active" based typically on puppetdb. For more information on deactiving nodes in puppetdb see this article

Changed depth

Once each catalog is compiled , it is saved to the /tmp directory on the system and the face will then automatically calculate the differences between the catalogs. Once this is complete a summary of number of nodes with changes as well as nodes whose catalog would not compile are listed. You can modify the number of nodes shown here using --changed_depth option.

Output Report

You can save the last report as json to a specific location using "--output_report" This report will contain the structured data in the format of running this command with --render-as json. An example Rakefile is provided with a docs task for converting this report to (GitHub flavored) markdown. The script above also will save the output with escaped color. If you want to view that text report run less -r lastrun-$$.log

Output description

During the transition of 0.24.x to 0.25.x there was a serialization bug that resulted in unexpected file content changes. I've recreated this bug in a tiny 2 resource catalog, the output below shows how this tool would have highlighted this bug prior to upgrading any nodes.

Resource counts:
  Old: 2
  New: 2

Catalogs contain the same resources by resource title


Individual Resource differences:
Old Resource:
  file{"/tmp/foo":
    content => d3b07384d113edec49eaa6238ad5ff00
  }

New Resource:
  file{"/tmp/foo":
    content => dbb53f3699703c028483658773628452
  }

Had resources simply gone missing - not the case here - you would have seen a list of that.

This code only validates the catalogs, it cannot tell you if the behavior of the providers that interpret the catalog has changed so testing is still recommended, this is just one tool to take away some of the uncertainty.

You can get some inline help with:

puppet man catalog

The reports generated by the this tool can be rendered as json as well as viewed in markdown using the Rakefile in this directory. A web viewer is also being developed at https://github.com/camptocamp/puppet-catalog-diff-viewer

Authors

R.I.Pienaar rip@devco.net / www.devco.net / @ripienaar
Zack Smith zack@puppetlabs.com / @acidprime
See change log