User Manual¶
- Fulll Documentation
- https://config-resolver.readthedocs.org/en/latest/
- Repository
- https://github.com/exhuma/config_resolver
- PyPI
- https://pypi.python.org/pypi/config_resolver
Rationale¶
Many of the larger frameworks (not only web frameworks) offer their own
configuration management. But it looks different everywhere. Both in code and
in usage later on. Additionally, the operating system usually has some default,
predictable place to look for configuration values. On Linux, this is /etc
and the XDG Base Dir Spec.
The code for finding these config files is always the same. But finding config files can be more interesting than that:
- If config files contain passwords, the application should issue appropriate warnings if it encounters an insecure file and refuse to load it.
- The expected structure in the config file can be versioned (think: schema). If an application is upgraded and expects new values to exist in an old version file, it should notify the user.
- It should be possible to override the configuration per installed instance, even per execution.
config_resolver
tackles all these challenges in a simple-to-use drop-in
module. The module uses no additional external modules (no additional
dependencies, pure Python) so it can be used in any application without adding
unnecessary bloat.
One last thing that config_resolver
provides, is a better handling of
default values than instances of SafeConfigParser
of the standard library.
The stdlib config parser can only specify defaults for options without
associating them to a section! This means that you cannot have two options with
the same name in multiple sections with different default values.
config_resolver
handles default values at the time you call .get()
,
which makes it independent of the section.
Description / Usage¶
The module provides two main classes:
Config
: This is the default class.SecuredConfig
: This is a subclass ofConfig
which refuses to load files which a readable by other people than the owner.
The simple usage for both is identical. The only difference is the above mentioned decision to load files or not:
from config_resolver imoprt Config
cfg = Config('acmecorp', 'bird_feeder')
This will look for config files in (in that order):
/etc/acmecorp/bird_feeder/app.ini
/etc/xdg/acmecorp/bird_feeder/app.ini
~/.acmecorp/bird_feeder/app.ini
– This will be deprecated (no longer loaded) inconfig_resolver 5.0
~/.config/acmecorp/bird_feeder/app.ini
./.acmecorp/bird_feeder/app.ini
If all files exist, one which is loaded later, will override the values of an
earlier file. No values will be removed, this means you can put system-wide
defaults in /etc
and specialise/override from there.
The Freedesktop XDG standard¶
freedesktop.org standardises the location of configuration files in the XDG
specification Since version 4.1.0, config_resolver
reads these paths as
well, and honors the defined environment variables. To ensure backwards
compatibility, those paths have only been added to the resolution order. They
have a higher precedence than the old locations though. So the following
applies:
XDG item | overrides |
---|---|
/etc/xdg/<group>/<app> |
/etc/<group>/<app> |
~/.config/<group>/</app> |
~/.<group>/<app> |
$XDG_DATA_HOME |
$GROUP_APP_PATH |
$XDG_CONFIG_DIRS |
$GROUP_APP_PATH |
Tip
If a config file is found at ~/.<group>/<app>
, a log message with
a warning is issued since config_resolver 4.1.0 encouraging the
end-user to move the config file to ~/.config/<group>/<app>
.
Files are parsed using the default Python configparser.ConfigParser
(i.e. ini
files).
Advanced Usage¶
Versioning¶
It is pretty much always useful to keep track of the expected “schema” of a config file. If in a later version of your application, you decide to change a configuration value’s name, remove a variable, or require a new one the end-user needs to be notified.
For this use-case, you can create versioned config_resolver.Config
instances in your application:
cfg = Config('group', 'app', version='2.1')
Config file example:
[meta]
version=2.1
[database]
dsn=foobar
If you don’t specify a version number in the construcor versioning will trigger automatically on the first file encountered which has a version number. The reason this triggers is to prevent accidentally loading files which incompatible version.
Only “major” and “minor” numbers are supported. If the application encounters a
file with a different “major” value, it will emit a log message with severity
ERROR
and the file will be skipped. Differences in minor numbers are only
logged with a “warning” level but the file will be loaded.
Rule of thumb: If your application accepts a new config value, but can function just fine with previous and default values, increment the minor number. If on the other hand, something has changed, and the user needs to change the config file, increment the major number.
Requiring files (bail out if no config is found)¶
Since version 3.3.0, you have a bit more control about how files are loaded.
The config_resolver.Config
class takes a new argument:
require_load
. If this is set to True
, an OSError
is raised
if no config file was loaded. Alternatively, and, purely a matter of taste, you
can leave this on it’s default False
value and inspect the loaded_files
attribute on the config_resolver.Config
instance. If it’s empty,
nothing has been loaded.
Overriding internal defaults¶
Both the search path and the basename of the file (app.ini
) can be
overridden by the application developer via the API and by the end-user via
environment variables.
By the application developer¶
Apart from the “group name” and “application name”, the
config_resolver.Config
class accepts search_path
and
filename
as arguments. search_path
controls to what folders are
searched for config files, filename
controls the basename of the config
file. filename
is especially useful if you want to separate different
concepts into different files:
app_cfg = Config('acmecorp', 'bird_feeder')
db_cfg = Config('acmecorp', 'bird_feeder', filename='db.ini')
By the end-user¶
The end-user has access to two environment variables:
<GROUP_NAME>_<APP_NAME>_PATH
overrides the default search path.XDG_CONFIG_HOME
overrides the path considered as “home” locations for config files (default=``~/.config``)XDG_CONFIG_DIRS
overrides additional path elements as recommended by the freedesktop.org XDG basedir spec. Paths are separated by:
and are sorted with descending precedence (leftmost is the most important one).<GROUP_NAME>_<APP_NAME>_FILENAME
overrides the default basename of the config file (default=``app.ini``).
Logging¶
All operations are logged using the default logging
package with a
logger with the name config_resolver
. All operational logs (opening/reading
file) are logged with the INFO
level. The log messages include the absolute
names of the loaded files. If a file is not loadable, a WARNING
message is
emitted. It also contains a couple of DEBUG
messages. If you want to see
those messages on-screen you could do the following:
import logging
from config_resolver import Config
logging.basicConfig(level=logging.DEBUG)
conf = Config('mycompany', 'myapplication')
If you want to use the INFO
level in your application, but silence only the
config_resolver logs, add the following to your code:
logging.getLogger('config_resolver').setLevel(logging.WARNING)
As of version 4.2.0, all log messages are prefixed with the group and
application name. This helps identifying log messages if multiple packages in
your application use config_resolver
. The prefix filter can be accessed via
the instance member _prefix_filter
if you want to change or remove it:
from config_resolver import Config
conf = Config('mycompany', 'myapplication')
print conf._prefix_filter
More detailed information about logging is out of the scope of this document. Consider reading the logging tutorial of the official Python docs.
Environment Variables¶
The resolver can also be manipulated using environment variables to allow different values for different running instances. The variable names are all upper-case and are prefixed with both group- and application-name.
<group_name>_<app_name>_PATH
The search path for config files. You can specify multiple paths by separating it by the system’s path separator default (
:
on Linux).If the path is prefixed with
+
, then the path elements are appended to the default search path.<group_name>_<app_name>_FILENAME
- The file name of the config file. Note that this should not be given with
leading path elements. It should simply be a file basename (f.ex.:
my_config.ini
) XDG_CONFIG_HOME
andXDG_CONFIG_DIRS
- See the XDG specification
Difference to ConfigParser¶
There is one major difference to the default Python
ConfigParser
: the
get()
method accepts a “default” parameter. If
specified, that value is returned in case
ConfigParser
does not return a value. Remember that
the ConfigParser
instance supports defaults as well if specified in the
constructor.
Using the default
parameter on get()
, you
can now have two options with the same name in two sections with different
values. Imagine the following:
[database1]
dsn=sqlite:///tmp/db.sqlite3
[database2]
dsn=sqlite:///tmp/db2.sqlite3
In the core ConfigParser
you could not specify two
different default values! The default
parameter makes this possible.
Note
AGAIN: The core ConfigParser
default mechanism
still takes precedence!
Debugging¶
Creating an instance of Config
will not raise an
error (except if explicitly asked to do so). Instead it will always return a
valid, (but possibly empty) instance. So errors can be hard to see sometimes.
The idea behind this, is to encourage you to have sensible default values, so
that the application can run, even without configuration. For
“development-time” exceptions, consider calling
get()
without a default value.
Your first stop should be to configure logging and look at the emitted messages.
In order to determine whether any config file was loaded, you can look into the
loaded_files
instance variable. It contains a list of all the loaded files,
in the order of loading. If that list is empty, no config has been found. Also
remember that the order is important. Later elements will override values from
earlier elements.
Additionally, another instance variable named active_path
represents the
search path after processing of environment variables and runtime parameters.
This may also be useful to display informtation to the end-user.
Examples¶
A simple config instance (with logging):
import logging
from config_resolver import Config
logging.basicConfig(level=logging.DEBUG)
cfg = Config("acmecorp", "bird_feeder")
print cfg.get('section', 'var')
An instance which will not load unsecured files:
import logging
from config_resolver import SecuredConfig
logging.basicConfig(level=logging.DEBUG)
cfg = SecuredConfig("acmecorp", "bird_feeder")
print cfg.get('section', 'var')
Loading a versioned config file:
import logging
from config_resolver import Config
logging.basicConfig(level=logging.DEBUG)
cfg = Config("acmecorp", "bird_feeder", version="1.0")
print cfg.get('section', 'var')
Default values:
import logging
from config_resolver import Config
logging.basicConfig(level=logging.DEBUG)
cfg = Config("acmecorp", "bird_feeder", version="1.0")
# This will not raise an error (but emit a DEBUG log entry).
print cfg.get('section', 'example_non_existing_option_name', default=10)
# this may raise a "NoOptionError"
print cfg.get('section', 'example_non_existing_option_name')
# this may raise a "NoSectionError"
print cfg.get('example_non_existing_section_name', 'varname')