-
Committer:
Scott Moser
-
Date:
2012-07-06 21:19:37 UTC
-
mfrom:
(559.2.436 cloud-init)
-
Revision ID:
smoser@ubuntu.com-20120706211937-i4bhe6ncje8vg0m7
Merge rework branch in [Joshua Harlow]
- unified binary that activates the various stages
- Now using argparse + subcommands to specify the various CLI options
- a stage module that clearly separates the stages of the different
components (also described how they are used and in what order in the
new unified binary)
- user_data is now a module that just does user data processing while the
actual activation and 'handling' of the processed user data is done via
a separate set of files (and modules) with the main 'init' stage being the
controller of this
- creation of boot_hook, cloud_config, shell_script, upstart_job version 2
modules (with classes that perform there functionality) instead of those
having functionality that is attached to the cloudinit object (which
reduces reuse and limits future functionality, and makes testing harder)
- removal of global config that defined paths, shared config, now this is
via objects making unit testing testing and global side-effects a non issue
- creation of a 'helpers.py'
- this contains an abstraction for the 'lock' like objects that the various
module/handler running stages use to avoid re-running a given
module/handler for a given frequency. this makes it separated from
the actual usage of that object (thus helpful for testing and clear lines
usage and how the actual job is accomplished)
- a common 'runner' class is the main entrypoint using these locks to
run function objects passed in (along with there arguments) and there
frequency
- add in a 'paths' object that provides access to the previously global
and/or config based paths (thus providing a single entrypoint object/type
that provides path information)
- this also adds in the ability to change the path when constructing
that path 'object' and adding in additional config that can be used to
alter the root paths of 'joins' (useful for testing or possibly useful
in chroots?)
- config options now avaiable that can alter the 'write_root' and the
'read_root' when backing code uses the paths join() function
- add a config parser subclass that will automatically add unknown sections
and return default values (instead of throwing exceptions for these cases)
- a new config merging class that will be the central object that knows
how to do the common configuration merging from the various configuration
sources. The order is the following:
- cli config files override environment config files
which override instance configs which override datasource
configs which override base configuration which overrides
default configuration.
- remove the passing around of the 'cloudinit' object as a 'cloud' variable
and instead pass around an 'interface' object that can be given to modules
and handlers as there cloud access layer while the backing of that
object can be varied (good for abstraction and testing)
- use a single set of functions to do importing of modules
- add a function in which will search for a given set of module names with
a given set of attributes and return those which are found
- refactor logging so that instead of using a single top level 'log' that
instead each component/module can use its own logger (if desired), this
should be backwards compatible with handlers and config modules that used
the passed in logger (its still passed in)
- ensure that all places where exception are caught and where applicable
that the util logexc() is called, so that no exceptions that may occur
are dropped without first being logged (where it makes sense for this
to happen)
- add a 'requires' file that lists cloud-init dependencies
- applying it in package creation (bdeb and brpm) as well as using it
in the modified setup.py to ensure dependencies are installed when
using that method of packaging
- add a 'version.py' that lists the active version (in code) so that code
inside cloud-init can report the version in messaging and other config files
- cleanup of subprocess usage so that all subprocess calls go through the
subp() utility method, which now has an exception type that will provide
detailed information on python 2.6 and 2.7
- forced all code loading, moving, chmod, writing files and other system
level actions to go through standard set of util functions, this greatly
helps in debugging and determining exactly which system actions cloud-init is
performing
- switching out the templating engine cheetah for tempita since tempita has
no external dependencies (minus python) while cheetah has many dependencies
which makes it more difficult to adopt cloud-init in distros that may not
have those dependencies
- adjust url fetching and url trying to go through a single function that
reads urls in the new 'url helper' file, this helps in tracing, debugging
and knowing which urls are being called and/or posted to from with-in
cloud-init code
- add in the sending of a 'User-Agent' header for all urls fetched that
do not provide there own header mapping, derive this user-agent from
the following template, 'Cloud-Init/{version}' where the version is the
cloud-init version number
- using prettytable for netinfo 'debug' printing since it provides a standard
and defined output that should be easier to parse than a custom format
- add a set of distro specific classes, that handle distro specific actions
that modules and or handler code can use as needed, this is organized into
a base abstract class with child classes that implement the shared
functionality. config determines exactly which subclass to load, so it can
be easily extended as needed.
- current functionality
- network interface config file writing
- hostname setting/updating
- locale/timezone/ setting
- updating of /etc/hosts (with templates or generically)
- package commands (ie installing, removing)/mirror finding
- interface up/down activating
- implemented a debian + ubuntu subclass
- implemented a redhat + fedora subclass
- adjust the root 'cloud.cfg' file to now have distrobution/path specific
configuration values in it. these special configs are merged as the normal
config is, but the system level config is not passed into modules/handlers
- modules/handlers must go through the path and distro object instead
- have the cloudstack datasource test the url before calling into boto to
avoid the long wait for boto to finish retrying and finally fail when
the gateway meta-data address is unavailable
- add a simple mock ec2 meta-data python based http server that can serve a
very simple set of ec2 meta-data back to callers
- useful for testing or for understanding what the ec2 meta-data
service can provide in terms of data or functionality
- for ssh key and authorized key file parsing add in classes and util functions
that maintain the state of individual lines, allowing for a clearer
separation of parsing and modification (useful for testing and tracing)
- add a set of 'base' init.d scripts that can be used on systems that do
not have full upstart or systemd support (or support that does not match
the standard fedora/ubuntu implementation)
- currently these are being tested on RHEL 6.2
- separate the datasources into there own subdirectory (instead of being
a top-level item), this matches how config 'modules' and user-data 'handlers'
are also in there own subdirectory (thus helping new developers and others
understand the code layout in a quicker manner)
- add the building of rpms based off a new cli tool and template 'spec' file
that will templatize and perform the necessary commands to create a source
and binary package to be used with a cloud-init install on a 'rpm' supporting
system
- uses the new standard set of requires and converts those pypi requirements
into a local set of package requirments (that are known to exist on RHEL
systems but should also exist on fedora systems)
- adjust the bdeb builder to be a python script (instead of a shell script) and
make its 'control' file a template that takes in the standard set of pypi
dependencies and uses a local mapping (known to work on ubuntu) to create the
packages set of dependencies (that should also work on ubuntu-like systems)
- pythonify a large set of various pieces of code
- remove wrapping return statements with () when it has no effect
- upper case all constants used
- correctly 'case' class and method names (where applicable)
- use os.path.join (and similar commands) instead of custom path creation
- use 'is None' instead of the frowned upon '== None' which picks up a large
set of 'true' cases than is typically desired (ie for objects that have
there own equality)
- use context managers on locks, tempdir, chdir, file, selinux, umask,
unmounting commands so that these actions do not have to be closed and/or
cleaned up manually in finally blocks, which is typically not done and will
eventually be a bug in the future
- use the 'abc' module for abstract classes base where possible
- applied in the datasource root class, the distro root class, and the
user-data v2 root class
- when loading yaml, check that the 'root' type matches a predefined set of
valid types (typically just 'dict') and throw a type error if a mismatch
occurs, this seems to be a good idea to do when loading user config files
- when forking a long running task (ie resizing a filesytem) use a new util
function that will fork and then call a callback, instead of having to
implement all that code in a non-shared location (thus allowing it to be
used by others in the future)
- when writing out filenames, go through a util function that will attempt to
ensure that the given filename is 'filesystem' safe by replacing '/' with
'_' and removing characters which do not match a given whitelist of allowed
filename characters
- for the varying usages of the 'blkid' command make a function in the util
module that can be used as the single point of entry for interaction with
that command (and its results) instead of having X separate implementations
- place the rfc 8222 time formatting and uptime repeated pieces of code in the
util module as a set of function with the name 'time_rfc2822'/'uptime'
- separate the pylint+pep8 calling from one tool into two indivudal tools so
that they can be called independently, add make file sections that can be
used to call these independently
- remove the support for the old style config that was previously located in
'/etc/ec2-init/ec2-config.cfg', no longer supported!
- instead of using a altered config parser that added its own 'dummy' section
on in the 'mcollective' module, use configobj which handles the parsing of
config without sections better (and it also maintains comments instead of
removing them)
- use the new defaulting config parser (that will not raise errors on sections
that do not exist or return errors when values are fetched that do not exist)
in the 'puppet' module
- for config 'modules' add in the ability for the module to provide a list of
distro names which it is known to work with, if when ran and the distro being
used name does not match one of those in this list, a warning will be written
out saying that this module may not work correctly on this distrobution
- for all dynamically imported modules ensure that they are fixed up before
they are used by ensuring that they have certain attributes, if they do not
have those attributes they will be set to a sensible set of defaults instead
- adjust all 'config' modules and handlers to use the adjusted util functions
and the new distro objects where applicable so that those pieces of code can
benefit from the unified and enhanced functionality being provided in that
util module
- fix a potential bug whereby when a #includeonce was encountered it would
enable checking of urls against a cache, if later a #include was encountered
it would continue checking against that cache, instead of refetching (which
would likely be the expected case)
- add a openstack/nova based pep8 extension utility ('hacking.py') that allows
for custom checks (along with the standard pep8 checks) to occur when running
'make pep8' and its derivatives