pave 0.69

Simple push-based configuration and deployment tool, leveraging fabric.  No servers, few dependencies.

                                            H   |P""""`.
                                          __H._ || o_  |
- - - - - - - - - - - - - - - - - - - -   |=====\||((,/ '_______  - - -
                                          |  __  YHaHHHaP ___   #:
      ______________ ___   ______        / ,'   `.YHHHHP.'   `. #:
      ___  __ \  __ `/_ | / /  _ \      | :       :HHH(/       . \
      __  /_/ / /_/ /_  |/ //  __/      `-|  (\)  |:=:"|  (\)  | _>
      _  ____/\__,_/ _____/ \___/         \       /(=) `       '
      /_/                                  `-._.-       `-._.-'

tl;dr

Simple push-based deployment and remote host management, leveraging fabric. No servers, few dependencies.

Intro

There are also slides on this topic, located here.

Another one?

Been looking for the “perfect” config management tool to deploy software to networked boxes… and still looking. I’ve tried but didn’t choose these:

bcfg, cfengine, puppet, chef, saltstack, ansible, etc.

I found them powerful and complex. They required daemons running, had their own languages, unique terminology, XML, Ruby, or Jinja dependencies. Despite reading the word “simple” multiple times, each had a varingly steep learning curve, and dozens if not hundreds of pages of abstract or “enterprisey” docs.

For my modest needs, ansible seemed the best fit (ssh-based and no servers), but there were a few things bugging me. Finally realized I was telling it how to do things, though I really wanted the tool to figure it out itself.

Likewise, I stumbled across fabric early on, and like most loved it—at first. Unfortunately as a script gets longer it dawns on you why people say, “don’t use it for configuration management!” Intentions, implementation details, and code all get mixed up in the same script and things get ugly fast. Cuisine (a fabric addon) appeared promising, but it looked mainly to shorten the fabfile. I wanted to go further and completely separate my setup from general implementation code.

Maybe I could whip one up myself? I started looking into how to build such a system. Thankfully after a day or so I realized the building blocks are already in fabric. Why not just commandeer it for my own evil purposes? </grin>

So here is pave.

How does it work?

Pave logs you into your networked hosts using SSH and configures them to your liking “automatically” using the command line. Just tell it what to do. It’ll check first and won’t make changes unless it needs to.

Don’t have to write any code for common tasks. Don’t have to know anything about Python or the OS you’re using (in theory—You’ll get the most out of it if you know how to use the default shell installed on your hosts, such as bash.) No gnarly/obscure runtimes to install, no root servers taking up memory/listening/waiting to be cracked, no new languages, nor weird lingo to learn.

Just a config file, a command, … and space for a new parking lot.

Use

Requirements

An OS from the last 5 years or so, with Python 2.6+ (locally) and sshd (remotely) ready to go.

Development is primarily on Ubuntu clients with Ubuntu Server 12.04 (Precise LTS) as the target (with some recent attention to Fedora, CentOS, Gentoo, Rasbian, and OSX), but aims to work everywhere practical, Unix and Windows.

Installation & Setup

Need pip?

sudo apt-get install python-pip   # Or yum, brew, etc.

“Express” install:

sudo pip install pave

Custom install options:

OSX: (Not yet tested, alpha quality) Install pip and the command-line developer tools if they are not already (type easy_install pip and gcc at a terminal), since pave needs paramiko/pycrypto and the latter needs to be built.

Windows: (Not yet tested, alpha quality) Install pip and pycrypto. Type pip install pave[win] in an Admin prompt.

Debian Linux: Note if pycrypto (ssh) needs to be built, or you’d like to build libyaml (for speed), do this first:

sudo apt-get install build-essential python-dev libyaml-dev

For the dev version:

sudo pip install \
https://bitbucket.org/mixmastamyk/pave/get/default.zip

Also, note that pip is not required; You could also download the package manually from PyPI or bitbucket. Unpack the archive, cd into the folder created, then:

sudo ./setup.py install    # manual install

Like typing in passwords over and over?  No?  ;)

Do you have an ssh key-pair set up?  No?

ssh-keygen -t rsa
ssh-copy-id [user@]machine

Pave can do the steps above if you’d like, however for “piece” of mind it’s better to do it yourself.

You may need to paste the matching line from ~/.ssh/id_rsa.pub into your EC2 or other cloud console… if you’re into that sort of thing.

Creating El Pavefile

The pavefile is what tells pave what to do. This file is named ./pave.yml by default, and “hello world” might look like this:

main:
    user: ubuntu
    sudo: True
    targets:
        - 192.168.2.1             # a list/str of hosts/groups

# passwords:                      # use only if not using keys
#   - ask password SSH Password   # default passwd for fabric

tasks:
    - packages:
        if-platform: Debian
        install: sysvbanner

    - packages:
        if-platform: Redhat
        install: banner

    - banner "Hello!"             # world

To create this file now, you may use the “skeleton” shortcut:

pave -S

Each section above declares what you’d like pave to do, and the details under each must be indented with spaces. Comments are written following the # character. The only required section is main which contains options for pave. Lists and mappings (dicts/hashes) are welcome in appropriate places. The format is called yaml, and syntax details can be found here. You don’t need to know much about it for now, just keep to the form above.

Common tasks already have their own modules, such as packages: above. These modules handle mundane details so you don’t need to be concerned. Don’t reinvent the wheel by putting platform-dependent commands into an ad-hoc tasklist—That’s what we’re trying to avoid.

Reference docs can be found on the next page and examples here. How to target groups of hosts and do other necessary things like declare variables, render templates, create users, configure the OS, etc, etc. is demonstrated.

Running it

pave

Now, that’s my definition of simple. ;) Of course pave does less than the big-guns but handles quite a lot already and will only grow. Consult the references and pave -h for many options.

You should see output similar to this (if you use pave -V). (Verbose logs from the last five runs are kept in ~/.cache/pave/logs/.)

✓ INFO    main: Version 0.60, Python: 2.7.4, fabric: 1.6.1, ./pave.yml
✓ INFO    transport: Connected (version 2.0, client OpenSSH_5.9p1)
✓ INFO    transport: Authentication (publickey) successful!
✓ INFO    main: found a(n) Ubuntu system.
✓ INFO    packages: install: sysvbanner

[snip install-text]

[192.168.2.1] out: sudo password:
[192.168.2.1] out: #     #                                    ###
[192.168.2.1] out: #     #  ######  #       #        ####     ###
[192.168.2.1] out: #     #  #       #       #       #    #    ###
[192.168.2.1] out: #######  #####   #       #       #    #     #
[192.168.2.1] out: #     #  #       #       #       #    #
[192.168.2.1] out: #     #  #       #       #       #    #    ###
[192.168.2.1] out: #     #  ######  ######  ######   ####     ###

✓ INFO    main: {u'192.168.2.1': (2, 0)}

The numbers at the end mean pave made two changes and had zero errors on the remote host.

“Exxxxcellent, Smithers.”

License

Licensed under the GPL, version 3+. A commercial license may also be available in the future.

Next »»

(Looking for a very simple build tool to complement pave? BuildIt was designed for that purpose.)

This concludes the introductory readme. Kindly continue on to the next chapter of the full docs. You may also go to the table of contents instead.