view README.txt @ 275:22bb1ac50eb7

remove obselete project
author Jeff Hammel <jhammel@mozilla.com>
date Tue, 31 Jul 2012 14:20:40 -0700
parents c77b946e4496
children 268349e6c941
line wrap: on
line source

autobot
=======

buildbot for the A*Team


What is autobot?
----------------

autobot is a continuous integration solution for the Automation and
Tools Team.  We have a lot of software.  We're really talented, so
usually it doesn't break.  But we're not infalliable.  Our robot ally,
autobot, is there to test things for us.  Let's meet autobot!


Installing autobot
------------------

autobot may be installed using the install script::

 curl http://k0s.org/mozilla/hg/autobot/raw-file/tip/INSTALL.sh | bash

This will create a virtualenv and install autobot for development
($VIRTUAL_ENV/src/autobot).  You can also use the script as
instructions and install it in the same manner as any other python package.


Setting up a buildmaster and slave
----------------------------------

Once you have autobot installed and the virtualenv activated, you'll
want to create a buildmaster and a buildslave.

You can create a master-slave pair by running ``create-autobot`` after
activating the virtualenv. This is mostly useful for autobot
development. The scripts ``create-autobot-master`` and
``create-autobot-slave`` are also available and are more useful for
production, as normally you will not want to keep the master and the
slave on the same machine except for development and testing.  The
scripts will prompt you for a factory or you can specify one or more
from the commandline using ``--factory`` or ``-f``. The factories are
from ``autobot.projects`` and its subdirectories.  You can list the
available factories with ``create-autobot --list-factories`` (or
``create-autobot-master --list-factories``).

Running ``create-autobot --help`` will give the variables you can set
when it makes a new bot for you (``create-autobot-master`` and
``create-autobot-slave`` also have a ``--help``, the variables in
``create-autobot`` being a superset of these).

If you don't specify a variable, the default will be used to create
your particular bot.  You can change these in the resultant
configuration later.


Using autobot
-------------

The buildmaster and buildslave are started with
``buildbot start master`` and ``buildslave stop slave`` from the bot's
directories. Respective ``stop`` commands also exist.  If you used the
``create-autobot`` command the generated bot will have a
``restart_buildbot.py`` script that will stop and start both the
master and slave and (if debug is set) remove the log as well.

The generated ``master.cfg`` file reads values from a ``master.ini``
file in the same directory.  The master.ini contains a number of
different sections:

* ``[:master:]`` contains the top level configuration for the master
  as well as default settings for slaves
* sections like ``[slave:slavename]`` indicate a slave of name
  ``slavename``
* all other sections are construction parameters for factories

The format of the ``.ini`` file is detailed in the subsequent
section.  You may change which .ini file is used by editing the
``master.cfg`` file, or, if for whatever reason you don't want to use
an .ini file as a driver you can construct the appropriate
configuration therein yourself.

An overview of the build status is detailed at the waterfall display,
by default at http://localhost:8010/waterfall . To force a build,
click on the desired builder and there will be a force build button
towards the bottom.

It is important to remember that continuous integration is a safety
net, not a first line of defense.  In other words, continue to check
your code *before* sending it to autobot.


.ini file format
----------------

As stated above, a ``master.ini`` file contains three different types
of sections:

* ``[:master:]`` contains the top level configuration for the master
  as well as default settings for slaves
* sections like ``[slave:slavename]`` indicate a slave of name
  ``slavename``
* other sections are construction parameters for factories

What goes in each of these?

``:master:`` The master section contains parameters appropriate to the
buildmaster:

* slaveport: which port to listen for the slaves on
* htmlport: which port to serve the waterfall display on
* pollInterval: how often to poll the repositories, in seconds [DEFAULT: 60]
* treeStable: how long to wait before the tree settles down before
  building [DEFAULT: 60]

The other defaults may be seen by running ``create-autobot-master --help``.

Other parameters given in the ``[:master:]`` section are used as the
baseline defaults for each slave.  They may be overridden in each
``slave:`` section

----

``slave:`` parameters for each slave:

* password: each slave *must* have a password.  Unless there's a
  reason to have a different password per slave, its probably better
  to put this in the ``:master:`` section

* factories: space-separated list of factories to run on that slave.
  If the special value ``*`` is used, all factories will be run.  You
  can view the factories available with
  ``create-autobot --list-factories``. The factory name corresponds to
  the directory (or module) name in ``autobot.projects``.  If
  every slave is meant to run the same factories, you can put this in
  the ``:master:`` section

* os: the operating system of the slave. Should be one of ``linux``,
  ``win``, or ``mac`` (though see TODO about future use of MozInfo
  making this obselescent). You probably *shouldn't* have a default
  key for this in the ``:master:`` section (though autobot won't try
  to stop you!).

----

factory sections:  All other sections are used to build the factories.
Each parameter in a factory section is used as a constructor
(``__init__``) keyword argument when they are created in the
``master.cfg``. So a factory section like::

 [foo]
 bar = fleem
 baz = another parameter

will invoke the foo factory (lets say it maps to ``MyFooFactory``) on
each slave like::

 MyFooFactory(**dict(bar='fleem', baz='another parameter'))

In addition, if a factory has a special key, ``platform``, the slave
will pass its platform information when instantiating a factory.
Currently, this is a dict with a single key, ``os``, but more may be
added as needed.  As noted in the TODO below, ideally this would be
deprecated entirely by MozInfo but such is the interim solution.


Projects
--------

What does autobot test?

* mozbase        [WORKING]
* logparser      [WORKING]
* profilemanager [IN FLIGHT]
* mozmill        [IN FLIGHT]
* devicemanager  [IN FLIGHT]
* firebug        [TODO]
* jetpack        [TODO]
* talos          [TODO]

The projects are obtained from any factories in subdirectories of
``autobot.projects`` (use
``python -c 'from autobot import projects; print projects.__file__'``
to see this location).


Sources
-------

``autobot.process.factory:SourceFactory`` is an abstract base class
for specifying sources.  Sources live as a member on the instance, named
(oddly) ``sources``.  This is a dict with the key being the source
type and the value being the source to use.  Source types are
currently "git" and "hg", though this is easily extensible.

Internally, an individual source is stored as a 2-tuple with the first
item being the URL of the source and the second item being the
branch.  If the branch is ``None`` or not specified, this is assumed
to equate to ``master`` for git or ``default`` for hg.  However, you
may also use a string in the form of ``URL#branch``. In addition, you
may specify a whitespace separated source, which ``SourceFactory``
will split into a list of sources.  So all of these translate into the
same source::

 sources = {'hg': 'http://hg.mozilla.org/automation/logparser'}

 sources = {'hg': ['http://hg.mozilla.org/automation/logparser#default']}

 sources = {'hg': [('http://hg.mozilla.org/automation/logparser', None)]

The generated ``master.cfg`` polls for changes on the given source,
then uses ``buildbot.schedulers.filter:Changefilter`` in conjunction
with a ``Scheduler`` to trigger the appropriate builds.  This is done
by ``GitPoller`` and ``HgPoller`` in
``autobot.changes.poller``. Again, more pollers can be added as
needed or the provided buildbot change sources may be utilized.

While, in general, the canonical sources should be specified at the
class level, if an appropriate argument (e.g. *hg* or *git*) is passed
in to the ``SourceFactory`` constructor, this will override that
source type.  This is useful for testing changes on non-canonical URLs
and/or branches.


Adding a New Project
--------------------

Occassionally, you'll need to add a new project to test.  You can add
a ``__init__.py`` file to a directory under ``autobot.projects`` and,
if you have a factory therein, autobot will find it and add it to the
list of factories it knows about.

There is also a script, ``create-autobot-project``, that can create
this stub for you::

 Usage: create-autobot-project [options] project <output>

``project`` is the name of the project and ``output``, if specified,
is the path of the file to create.  If output is not specified, it
will find its way to a directory named for the project under
``autobot.projects``.

Several factories (``BuildFactory`` descendents) are in
``autobot.process`` to make building a new project easier:

* ``SourceFactory``: as described above, this processes the sources
  and gives a method (``checkout()``) for downloading them.
  ``master.cfg`` looks in each factory and gleems its needed
  repository from its ``sources`` attribute (if any) in the internal
  storage mechanism of ``SourceFactory`` and sets up schedulers
  accordingly.

* ``VirtualenvFactory``: inherits from ``SourceFactory``. creates a
  python virtualenv and provides build properties ``%(virtualenv)s``,
  ``%(python)s``, and ``%(scripts)s`` (location of ``bin`` on unix or
  ``Scripts`` on windows with respect to the virtualenv.  It also has
  a ``findScript()`` method which accepts the unix-style script name
  (no extension) and returns the path to the script on the platform.

* ``PythonSourceFactory``: inherits from ``VirtualenvFactory``. It
  treats *all* (for now, can be changed) sources as sources of python
  packages.  In addition to checking them out in
  ``%(virtualenv)s/src``, it will also install them (using
  ``%(python)s setup.py install``)

* ``FirefoxDownloaderFactory``:  downloads and unpacks a Firefox
  bundle, using
  `GetLatestTinderbox <http://hg.mozilla.org/automation/getlatest-tinderbox>`_ .
  It also provides three properties: ``%(firefox_url)s``,
  ``%(firefox_bundle)s``, and ``%(firefox)s`` -- the last, and most
  interesting, being the absolute path to the Firefox executable. You
  (currently) *must* pass a ``platform`` argument with
  ``platform['os']`` set to "linux", "win", or "mac" in order for
  ``FirefoxDownloaderFactory`` to figure out what to do with the thing
  that it downloads (though see TODO, in the future it should/could
  use MozInfo or guess the OS from the bundle extension...maybe).
  Currently, mac is *NOT* supported, though its on my list of things
  to do (I'd love advice/help!).

See the ``autobot.process.factory`` file for more details.  These
classes are intended as mix-ins (not my favorite pattern, but it seems
to basically be what buildbot wants you to do).  More work needs to be
done to provide for the breadth of use-cases, but what exists now is
considered a "good start" versus a "final form".  More factories may
be added per necessity.


Is your autobot being feisty?
-----------------------------

Let me know! I'd like to make autobot a solution that works for all
stake-holders, and if you're reading this, that means you!

jhammel@mozilla.com


TODO
----

No software of any size is ever finished.  Here are a few things I
would like to add:

* post to autolog:
  http://hg.mozilla.org/users/jgriffin_mozilla.com/mozautolog/
  Example code can be found at
  http://hg.mozilla.org/services/services-central/file/9f29daaecbcc/testing/tps/tps/testrunner.py#l491
  add a 'server' keyword to the AutologTestGroup constructor, set to
  10.2.76.100:9200, that will cause the data to be written to
  brasstacks test server. you can omit the restserver keyword, it will
  use the correct default if you skip it

* email notifications

* get FirefoxDownloaderFactory working on Mac with .dmg files!

* singular checkout of repos on slaves: the slaves should have a
  singular master repo that is checked out once for each repo URL,
  branch pair.  It is then updated as the slaves need and (e.g.)
  cloned from there.  This should effectively minimize fetch time.