A Fabric/Puppet Build/Deploy System

Adrian
Nye, Dimensio
nal Fund
Advisors
A PUPPET/FABRIC
BUILD/DEPLOY SYSTEM

• Python Software engineer, not a dev-ops guy
• Long-time Fabric user, just learned puppet
• Developed this system with Gary Wilson, another python dev
who also just learned puppet
WHO BUILT THIS?

 Start from bare RHEL 6 VMs, with only basic services pre-
installed (puppet, ntp, networking/firewall rules)
 Provide tools to build, configure, and deploy:
 15 existing websites in various technologies:
python, perl, php, ruby, & combinations
 Mysql & Mongo databases
 Memcache servers
 Proxy servers
 Search servers
 Dev/Stage/Prod copies of all this
 Automate everything
 Never touch any server by hand
THE TASK

 RHEL 6 is stable but very old versions of most software. For
example puppet hiera just became available as RPM.
 Stage & Prod servers won’t have internet access
 Deployment to Stage/Prod will be done by operations
people, not apps people.
 Need rollback
 Must have GUI or be simple
SOME CHALLENGES

 RPM or Source Installs?
 Git or Tar-based Deployment?
 Chef/Puppet/Ansible/SaltStack?
 Puppet preferred by our infrastructure group
 We’re python devs, so Fabric seemed obvious, it’s not going
away
SOME CHOICES

 Executes commands either local or remote (via ssh)
 Has functions for many common tasks
 Easy to script
 Anything you can do manually by ssh to a server, you can
script fabric to do.
 Goal is a repeatable, idempotent sequence of steps.
SO WHAT IS FABRIC?

 Useful stuff it can do:
 Confirm before doing things if you want
 Run stuff in parallel on multiple machines, or serially
 run stuff as if run from a directory
 Get & put files, append to files, comment or uncomment lines
 Upload templates and fill in variables
 Run sudo commands
 Connect to one host then to another within same function
BRIEF INTRO TO FABRIC

 Many ways to specify
 Most common is to use the Env variable, and set env.hosts
 Can specify on command line
 Can hardcode it (build tasks always happen on build server)
 Can make lists of hosts
 Functions on fab command line are executed in order, so first
function can set host, and subsequence functions can use
setting
FABRIC HOSTS

def tail_log(logname):
"""tail a log file.
fab hostname tail_log:access
logname is filename of log (without .log)
"""
log = env.logdir + „/‟ + logname + „.log‟
if file_exists(log):
run(„tail –f %s‟ % log, pty=True)
else:
print “Logfile not found in %s” % env.log_dir
EXAMPLE FABRIC TASK

def dev(service_name=None):
"""Sets server as appropriate for service_name for the dev
environment.
Also sets environ, server, and service_name in env so it is inherited
by later fab commands. Some fab commands need an environment but
are not
specific to a service (such as mysql commands), so service_name is
optional.
"""
_set_host_for_environment(service_name, Environment.DEV)
The above function is just a clever way of setting env.host to a hostname, so that later
commands in the same fab command line know what system to work on.
SETTING HOST CLEVERLY

 We wrote classes using fabric to do all the tasks needed in
the build and deployment process for our sites (and invoke
puppet, which does the rest).
 All of it is data-driven. There is a file that defines the needs
of each of our services and one that defines all of our servers.
HOW WE USED FABRIC

 Your responsibility to make things idempotent.
 For example, running “mkdir dirname” is not idempotent, because it
will fail the second time. In this case use a routine that tests
whether the dir exists, and if not then create it.
 Output control
 Normal output is everything (very verbose). Good for
debugging, although it can hide problems in sheer volume of
information.
 You can turn down the verbosity.
 Really want two levels simultaneously: less verbose output displayed
to the terminal, and fully verbose output logged to a file. But fabric
doesn’t support that yet.
CHALLENGES WITH FABRIC

 Puppet and Fabric capabilities overlap
 both can do most tasks
 Puppet is naturally idempotent
 Fabric is naturally step by step
 Puppet: use to enforce STATE
 RPM installation
 Creation of upstart scripts from templates
 User accounts
 Files, Directories, and Permissions
 Fabric: use to enforce WORKFLOW
 Build software environment (python virtualenv/perl modules etc)
 Protect from simultaneous deploys
 Testing of support services
 Sync of software environment from build server to deploy server
 Checkout of git repos, switching branches
 Media syncing
 Run puppet
 Graceful server restart
 Smoke testing (is site actually working after deployment)
DIVISION OF RESPONSIBILITIES

 All our custom puppet and fabric code is in a single git repo
called “sysconfig”
 Enables everything to be run from anywhere with network
access to build server.
PUPPET/FABRIC GIT REPO

 Graphic of our dev servers and networking
DEV ENVIRONMENT

Utility modules used by
multiple sites
Each site/service has
module
4 internal sites run on
intweb01. Mongo &
Mysql run on DB1Host
Intweb01(d,s,p) is internal
web server, db01 is db
server , etc
Nodes
IntWeb1Host
IntWebsite1
Nginx Proxy
IntWebsite2
Website
Layout
DB1Host
MySQL MongoDB
PUPPET MODULES

 Nodes manifest connects hostnames with the type of host it
will be, for all servers in all environments.
 Hosts manifest for each type of host (4 types of web
servers, db server, cache server, proxy server, etc). This
assigns sites to hosts.
 Site manifests for each type of service (each
website, proxy, database). Does rpm installation, site-specific
files & dirs, upstart scripts to start and stop service.
 Utility manifests for stuff needed by multiple sites, to
minimize duplication. For example nginx module supports 3
different uses of nginx: fake dev load balancer proxy, ssl
offloading proxy, local proxy.
PUPPET MODULES/MANIFESTS

 Use Virtual Packages to enable every manifest to install its
dependencies without regard to whether some other manifest
has already installed it on the same server.
 Should use Hiera to enable Puppet/Fabric to pull from the
same YAML database, but we haven’t done this yet since Hiera
just became available on RHEL 6.
PUPPET FEATURES

 1. Build step builds software environment for site on build
server
 2. Deploy step copies the software environment from build
server to the destination server, then deploys app code from
scratch.
 Two-step process does several things:
 Speeds up deployment, since build step is needed less often and
takes a long time.
 Speeds parallel deployment if you have redundant servers
 Keeps compiling tools off destination servers. The less you
install, the more secure they are.
FABRIC WORKFLOW

 Pip/Virtualenv used for Python packages, requirements file in
git repo
 Cpanm used for Perl modules
 Rbenv used for ruby modules
 All packages, modules, and rpms mirrored locally
 Improves reliability and speed
 Simplifies version control
 Everything (except rpms) installed in
/opt/comms/servicename, not system-wide. This simplifies
copying the environment to the deployed server, and
simplifies recreating a clean build.
FABRIC BUILD WORKFLOW

 Example service definition
 Name (for fab commands)
 Server type it should be installed on (not hostname)
 Domain (without dev/stg/com)
 Ssl or not
 How to smoke test it
 Init scripts it needs
 Git repos to check out, including branch, and any media
 Log dirs
 Languages needed
 Prerequisite services to check (memcache, db)
 Related services to reload (nginx, memcache)
 dirs containing built software environment (virtualenv, cpanm etc)
FABRIC BUILD SERVICE DEFINITION

 1. Mark deployment as in progress (using lock file)
 2. Check support services
 If db needed, is it running?
 If memcache needed, is it running?
 If critical support service not running, ask whether to continue.
FABRIC DEPLOYMENT WORKFLOW

 3. clone/pull the git repo(s) needed for the site. Checkout the
specified branch.

 4. Move previous software environment for fast rollback
 5. Rsync software environment for site
 6. Rsync media for site

 7. Run puppet. For convenience we support two modes:
 Use puppet master
 Copy developer’s sysconfig repo and run puppet using those modules.
This makes development a lot faster.

 8. Zero Downtime Restart
 9. Smoke Test
 For web servers, check that site is up, login works, and run selenium
tests.
 For memcache etc, use nc to test basic operation.
 Note that if deployment fails, there would be some downtime until
previous version reinstated.

 Setting up SSH keys on all servers
 Log viewers
 Database backups
 Database copy from one environment to another (i.e. copy
production db back to dev)
 Determine hostname from service name and environment
 Status/Start/stop/reload any remote service/site
 Media syncing from environment to environment
 Proxy server config generation
 Running puppet, using puppet master and without
 Smoke testers for different types of sites & services
 Tools to make local mirrors of internet software
FABRIC UTILITIES WE WROTE

 Support for replicated (redundant) servers.
 GUI for common tasks, using Rundeck or Jenkins or TeamCity
 Network logging (Splunk)
FUTURE WORK

A Fabric/Puppet Build/Deploy System

More Related Content

What's hot (20)

Viewers also liked (9)

Similar to A Fabric/Puppet Build/Deploy System (20)

Recently uploaded (20)

A Fabric/Puppet Build/Deploy System