slides

space, →	next slide
←	previous slide
d	debug mode
## <ret>	go to slide #
c	table of contents (vi)
f	toggle footer
r	reload slides
z	toggle help (this)

loading presentation...

Dependency management with Puppet

http://rcrowley.org/talks/pycon-2011/

Hi, I’m Richard Crowley

Equal opportunity technology hater.
DevStructure’s operator and UNIX hacker.

Install Puppet

sudo apt-get -y install ruby rubygems
sudo gem install --no-rdoc --no-ri puppet

You want 2.6 or better.
Ubuntu Lucid and pretty much
every Red Hat should use RubyGems.

Configuration management

Express all dependencies in a
legible and iterable format.

`pip freeze`

Not good enough.

Resources

The smallest unit of configuration.
Have a type, a name, and parameters.

Resources

package { "python":
    ensure => installed,
    provider => apt,
}

Resources

package { "python":
    ensure => "2.6.6-2ubuntu1",
    provider => apt,
}

Resources

package { "python":
    ensure => latest,
    provider => apt,
}

Resources

package {
    "build-essential":
        ensure => installed,
        provider => apt;
    "python":
        ensure => installed,
        provider => apt;
    "python-dev":
        ensure => installed,
        provider => apt;
    "python-setuptools":
        ensure => installed,
        provider => apt;
}

Resources

package {
    "build-essential": ensure => installed;
    "python": ensure => installed;
    "python-dev": ensure => installed;
    "python-setuptools": ensure => installed;
}

Resource types

There’s more to life than
package management.

Users and groups

group { "rcrowley":
    ensure => present,
    gid => 1000,
}
user { "rcrowley":
    ensure => present,
    gid => 1000,
    uid => 1000,
}

Authorized keys

ssh_authorized_key { "Richard's VM":
    ensure => present,
    key => "AAAAB3NzaC1yc2EAAAABIwAAAQEA6NF8iallvQVp22WDkTkyrtvp9eWW6A8YVr+kz4TjGYe7gHzIw+niNltGEFHzD8+v1I2YJ6oXevct1YeS0o9HZyN1Q9qgCgzUFtdOKLv6IedplqoPkcmF0aYet2PkEDo3MlTBckFXPITAMzF8dJSIFo9D8HfdOV0IAdx4O7PtixWKn5y2hMNG0zQPyUecp4pzC6kivAIhyfHilFR61RGL+GPXQ2MWZWFYbAGjyiYJnAmCP3NOTd0jMZEnDkbUvxhMmBYSdETk1rRgm+R4LOzFUGaHqHDLKLX+FIPKcF96hrucXzcWyLbIbEgE98OHlnVYCzRdK8jlqm8tehUc9c9WhQ",
    type => "ssh-rsa",
    user => "rcrowley",
}

Services

service { "apache2":
    enable => true,
    ensure => running,
    hasstatus => true,
}

Generic resource types

`exec` and `file`

Arbitrary commands

exec { "easy_install pip":
    path => "/usr/local/bin:/usr/bin:/bin",
    unless => "which pip",
}

Arbitrary files

file {
    "/root/.pip":
        ensure => directory,
        group => "root",
        mode => 0755,
        owner => "root";
    "/root/.pip/pip.conf":
        content => template("python/pip.conf.erb"),
        ensure => file,
        group => "root",
        mode => 0644,
        owner => "root";
}

`template` function

template("python/pip.conf.erb")

Argument: special pathname.
Result: string rendered from ERB.

`pip.conf.erb` template

[install]
user-mirrors = true
mirrors =
    http://pypi.<%= domainname %>
    http://pypi.python.org

Facts

facter | less
facter factname
Normalized names for
dozens of system parameters.

Running Puppet manifests

sudo puppet apply pathname
--noop and --verbose are handy.

Composing resources

Classes

Singleton collections of resources.

Classes

class python {
    package {
        "build-essential": ensure => installed;
        "python": ensure => installed;
        "python-dev": ensure => installed;
        "python-setuptools": ensure => installed;
    }
    exec { "easy_install pip":
        path => "/usr/local/bin:/usr/bin:/bin",
        unless => "which pip",
    }
}

Using classes

include python

include is a shorthand function.

Using classes

class { "python": }

Class instances are just resources.

Parameterized classes

class foo($bar = "baz") {}

class { "foo": bar => "quux" }

Like any resource,
classes may accept parameters.

Composing resources

Definitions

Resource types expressed as Puppet code.

Definitions

define pip($ensure = installed) {
    case $ensure {
        installed: {
            exec { "pip install $name":
                path => "/usr/local/bin:/usr/bin:/bin",
            }
        }
        latest: {
            exec { "pip install --upgrade $name":
                path => "/usr/local/bin:/usr/bin:/bin",
            }
        }
        default: {
            exec { "pip install $name==$ensure":
                path => "/usr/local/bin:/usr/bin:/bin",
            }
        }
    }
}

Using definitions

pip { "mysql-python": ensure => installed }

Variables and expressions

Ruby plugins

For when definitions aren’t good enough.
Implementation details are
beyond the scope of this talk.

`gem install puppet-pip`

Ruby implementation of a provider
for the package resource type.
Submitted to be included in
the next release of Puppet.

Using `puppet-pip`

package { "mysql-python":
    ensure => installed,
    provider => pip,
}

Modules

/etc/puppet/
    manifests/
        nodes.pp
        site.pp
    modules/
        python/
            manifests/
                init.pp
                foo.pp

Code organization pays dividends.

`modules/python/ manifests/init.pp`

class python {
    # ...
}

`modules/python/ manifests/foo.pp`

class python::foo {
    # ...
}

Module autoloading

include python will expect the python class in modules/python/manifests/init.pp.
Templates referenced as template("python/foo.erb") will render modules/python/templates/foo.erb.

Running Puppet modules

sudo puppet apply -e "include classname"
--noop and --verbose are still handy.

Which modules are applied to which servers?

Nodes

node default {
    include python
}

node /^www\d+\./ inherits default {
    include apache
    include django
}

node "mail.example.com" inherits default {
    include lamson
}

Conventionally in manifests/nodes.pp.

`manifests/site.pp`

Exec {
    path => "/usr/local/bin:/usr/bin:/bin",
}

import "nodes"

Default entrypoint.

Running Puppet catalog

sudo puppet apply /etc/puppet/manifests/site.pp
--noop and --verbose are still handy.

Django module

class django {
    package {
        "django":
            ensure => "1.2.5",
            provider => pip;
        "mysql-python":
            ensure => installed,
            provider => pip;
    }
}

A bug!

class django {
    package {
        "django":
            ensure => "1.2.5",
            provider => pip;
        "mysql-python":
            ensure => installed,
            provider => pip;
    }
}

Declare all dependencies

class django {
    package {
        "django":
            ensure => "1.2.5",
            provider => pip;
        "libmysqlclient-dev":
            ensure => installed;
        "mysql-python":
            ensure => installed,
            provider => pip;
    }
}

Nondeterminism!

class django {
    package {
        "django":
            ensure => "1.2.5",
            provider => pip;
        "libmysqlclient-dev":
            ensure => installed;
        "mysql-python":
            ensure => installed,
            provider => pip;
    }
}

Declare interdependencies

class django {
    package {
        "django":
            ensure => "1.2.5",
            provider => pip;
        "libmysqlclient-dev":
            ensure => installed;
        "mysql-python":
            ensure => installed,
            provider => pip,
            require => Package["libmysqlclient-dev"];
    }
}

Declare interdependencies

class python {
    package {
        "build-essential": ensure => installed;
        "python": ensure => installed;
        "python-dev": ensure => installed;
        "python-setuptools": ensure => installed;
    }
    exec { "easy_install pip":
        path => "/usr/local/bin:/usr/bin:/bin",
        require => Package["python-setuptools"],
        unless => "which pip",
    }
}

Coarser dependencies

class python {
    package {
        "build-essential": ensure => installed;
        "python": ensure => installed;
        "python-dev": ensure => installed;
        "python-setuptools": ensure => installed;
    }
    exec { "easy_install pip":
        alias => "pip",
        path => "/usr/local/bin:/usr/bin:/bin",
        require => Package["python-setuptools"],
        unless => "which pip",
    }
    Class["python"] -> Package<| provider == pip |>
}

All pip packages require pip to be setup.

`before` and `require`

One resource can require another.
One resource can come before another.

Same difference

class django {
    package {
        "django":
            ensure => "1.2.5",
            provider => pip;
        "libmysqlclient-dev":
            before => Package["mysql-python"],
            ensure => installed;
        "mysql-python":
            ensure => installed,
            provider => pip;
    }
}

Why all the fuss?

Python packages are ignorant of
lower-level dependencies.
For example, libmysqlclient-dev.
Explicit is better than implicit.

Apache module

class apache {
    package {
        "apache2-mpm-worker": ensure => installed;
        "libapache2-mod-wsgi": ensure => installed;
    }
    file {
        "/etc/apache2/sites-available/mysite":
            content => template("apache/mysite.erb"),
            ensure => file,
            require => Package["apache2-mpm-worker"];
        "/etc/apache2/sites-enabled/001-mysite":
            ensure =>
                "/etc/apache2/sites-available/mysite",
            require => Package["apache2-mpm-worker"];
    }
    # Next slide goes here.
}

Apache service

service { "apache2":
    enable => true,
    ensure => running,
    hasstatus => true,
    require => Package["apache2-mpm-worker"],
    subscribe => [
        Package[
            "apache2-mpm-worker",
            "libapache2-mod-wsgi"],
        File[
            "/etc/apache2/sites-available/mysite",
            "/etc/apache2/sites-enabled/001-mysite"]],
}

`notify` and `subscribe`

Complementary just like before and require.
Notified exec resources run their command.
Notified service resources
restart the service.

`modules/apache/templates/mysite.erb`

<VirtualHost *:80>
    DocumentRoot /usr/local/share/wsgi/mysite/media
    Alias /media /usr/local/share/wsgi/mysite/media
    WSGIScriptAlias / /usr/local/share/wsgi/mysite/mysite.wsgi
    WSGIDaemonProcess mysite processes=2
    WSGIProcessGroup mysite
</VirtualHost>

`modules/apache/templates/mysite.erb` with one process per core

<VirtualHost *:80>
    DocumentRoot /usr/local/share/wsgi/mysite/media
    Alias /media /usr/local/share/wsgi/mysite/media
    WSGIScriptAlias / /usr/local/share/wsgi/mysite/mysite.wsgi
    WSGIDaemonProcess mysite processes=<%= processorcount %>
    WSGIProcessGroup mysite
</VirtualHost>

Deploy

Deploy with Fabric

from fabric.api import *
from fabric.contrib.project import rsync_project

env.hosts = ['example.com']

def deploy():
    rsync_project(remote_dir='/etc/puppet')
    sudo('RUBYLIB=/var/lib/gems/1.8/gems/puppet-pip-0.0.5/lib'
         'puppet apply /etc/puppet/manifests/site.pp')

Puppet master

Deploy your Puppet code only to the master.
Agents phone home periodically
to fetch their configuration.

Puppet master

sudo puppet master
Drops privileges after starting.
Can be run by Apache or Nginx if needed.

Puppet agent

Daemon or cron job that phones home.

An agent’s `/etc/puppet/puppet.conf`

[main]
    server = puppet.example.com

Accept defaults otherwise.
See http://bit.ly/puppet-config-ref.

Puppet agent

sudo puppet agent --no-daemonize --onetime
First time will fail in SSL client verify.

Sign SSL certificates

Tell the master the agent is cool

sudo puppet cert -l
sudo puppet cert -s hostname

Puppet agent

sudo puppet agent --no-daemonize --onetime
If certificates verify, it receives
and executes a catalog.
--noop and --verbose are still handy.

Why should I bother with Puppet master?

Exported resources

[main]
    server = puppet.example.com

    dbadapter = mysql
    dbname = puppet
    dbpassword = puppet
    dbserver = localhost
    dbuser = puppet

[master]
    thin_storeconfigs = true

Run MySQL on the master.

Exported resources

node /^www\d+\./ inherits default {
    include apache
    include django
    @@nagios_host { "$hostname":
        address => "$ipaddress",
        hostgroups => ["www"],
        use => "generic-host",
        # ...
    }
}

node "ops.example.com" inherits default {
    include nagios
    Nagios_host<<||>>
}

Blueprint

If all that sounds like a lot of trouble...
blueprint(1) reverse engineers servers to generate Puppet code.
https://github.com/devstructure/blueprint
http://devstructure.com/

Thank you

richard@devstructure.com or @rcrowley
P.S. I’m available for Puppet
and operations consulting.

Dependency management with Puppet

Hi, I’m Richard Crowley

Install Puppet

Configuration management

pip freeze

Resources

Resources

Resources

Resources

Resources

Resources

Resource types

Users and groups

Authorized keys

Services

Generic resource types

exec and file

Arbitrary commands

Arbitrary files

template function

pip.conf.erb template

Facts

Running Puppet manifests

Composing resources

Classes

Classes

Using classes

Using classes

Parameterized classes

Composing resources

Definitions

Definitions

Using definitions

Variables and expressions

Further reading

Ruby plugins

gem install puppet-pip

Using puppet-pip

Modules

modules/python/ manifests/init.pp

modules/python/ manifests/foo.pp

Module autoloading

Running Puppet modules

Which modules are applied to which servers?

Nodes

manifests/site.pp

Running Puppet catalog

Django module

A bug!

Declare all dependencies

Nondeterminism!

Declare interdependencies

Declare interdependencies

Coarser dependencies

before and require

Same difference

Why all the fuss?

Apache module

Apache service

notify and subscribe

modules/apache/templates/mysite.erb

modules/apache/templates/mysite.erb with one process per core

Deploy

Deploy with Fabric

Puppet master

Puppet master

Puppet agent

An agent’s /etc/puppet/puppet.conf

Puppet agent

Sign SSL certificates

Tell the master the agent is cool

Puppet agent

Why should I bother with Puppet master?

Exported resources

Exported resources

Blueprint

Thank you

`pip freeze`

`exec` and `file`

`template` function

`pip.conf.erb` template

`gem install puppet-pip`

Using `puppet-pip`

`modules/python/ manifests/init.pp`

`modules/python/ manifests/foo.pp`

`manifests/site.pp`

`before` and `require`

`notify` and `subscribe`

`modules/apache/templates/mysite.erb`

`modules/apache/templates/mysite.erb` with one process per core

An agent’s `/etc/puppet/puppet.conf`