converge
Resolve Data from Abstract Hierarchies and Templates
Managing configuration is hard. More often than not you have high key/values duplication and storage. DRY your tears, it's time for some hierarchical magic, so you can get back to the important stuff.
This is where converge comes in. There are a few basic concepts when using or extend converge:
- Readers: get data from your backends
- Filters: filter data before or after resolution (example: search & replace values, inject data post resolution)
- Resolvers: convert abstract data to resolved data
- Writers: output your data to the format and backend you need
Abstract hierarchies of data chewed up and spit out to your liking.
Getting started
install pyconverge, this will add the converge
command to your classpath using setup.py/PyPi
pip install pyconverge
converge version
converge --help
Create a converge.yaml.template file in your project working directory, modify and move it to converge.yaml
# create the converge.yaml.template file
converge init conf
# modify your converge parameters
vim converge.yaml.template
# activate converge
mv converge.yaml.template converge.yaml
# verify the integrity of your configuration file
converge check
Try it out! You chould have a bunch more options!
converge --help
Example: Simple testing
create a converge.yaml as described above
converge init conf
mv converge.yaml.template converge.yaml
converge check
Create a sample repository structure
converge init repository target_directory
# you should now have the following structure
find target_directory/ -type d
# this is where the application centric data goes
./data
./data/default
./data/default/shared
./data/default/app
# the hierarchy file is situated here
./hierarchy
# target (or host for most) centric data resides here
./targets
./targets/hosts
./targets/mapping
Configuration
converge.yaml
This file is a bit peculiar, it allows you to add programs, options, configurations on the fly. Let's try it out before adding the converge.yaml:
converge --help
# positional arguments:
# {init,check,version}
# init initialize configuration or repository
# check run sanity check on configuration
# version get converge version and build information
Put the following in a converge.yaml:
conf:
default:
logging_level: "INFO"
programs:
# user command
amazingcommand:
# arguments expected by the command
args:
- "argument_one"
- "argument_two"
# description for the python help
description: "application description"
modes:
amazing_mode_one:
- "com.insane.class.path.Class1"
- "com.insane.class.path.Class2"
amazing_mode_two:
- "com.insane.class.path.Class1"
- "com.insane.class.path.Class3"
Now you should see a new option:
converge --help
# positional arguments:
# {init,check,version,amazingcommand}
# init initialize configuration or repository
# check run sanity check on configuration
# version get converge version and build information
# amazingcommand application description <-- MAGIC STUFF with description!!
And even more so there are sub-options available too:
converge amazingcommand --help
# usage: converge-runner.py amazingcommand [-h]
# argument_one argument_two
# {amazing_mode_two,amazing_mode_one}
#
# positional arguments:
# argument_one
# argument_two
# {amazing_mode_two,amazing_mode_one}
Classpath Execution Explained
Those classpaths you listed under the modes, will be executed expecting a method matching the following:
def run(self, data, conf, **kwargs):
"""
Args:
data (object): The data object that is passes and returned from all class runs
conf (dict): Dictionary of all configurations found in converge.yaml (directories, logging etc)
kwargs (object): magical python variable variable variables.
Returns:
dict: the data object that will be passed to all following class-runs.
"""
return data
Example: Configuration for Java property files
A rough overview
A general example in diagram form:
A Brief History of Pain
You may have hit some (or all) of these stages in the pursuit of configurability:
In short: from the file, to the GUI, back to the file you idiot. #wevecomefullcircle
- Externalize configuration from your applications, to avoid re-releases due to simple conf tuning
- Realizing that you're now managing a million de-centralized files with no similar structure
- Create or use a centralized, GUI/DB based configuration management system (woohoo! configuration liberation!)
- Realizing that you are missing flexibility, automation is complex, added abstraction layers are painful.
The better model is to accept any data format, process it and output it as you wish.
Files are better because:
- you can use time tested versioning systems like git or mercurial to branch, release, rollback, check history
- you can automate the modification of files with any tool you want
- doing migrations/deployment/modifications on DB values/REST endpoints sucks (unnecessarily complex)