Unified config handling for client libraries and programs
Go to file
Doug Hellmann a71511468e Update reno for stable/mitaka
Fix a few page titles at the same time

Change-Id: I68d082f1cad51bbe58deed6a7e4b0de122c22fc7
2016-03-11 15:06:47 -05:00
doc/source Remove HP and RunAbove from vendor profiles 2016-02-10 11:24:10 -06:00
os_client_config Merge "Fix formulation" 2016-02-25 21:59:10 +00:00
releasenotes Update reno for stable/mitaka 2016-03-11 15:06:47 -05:00
.coveragerc Change ignore-errors to ignore_errors 2015-09-21 14:41:36 +00:00
.gitignore add .venv to gitignore 2015-04-13 12:45:50 -06:00
.gitreview Update .gitreview for git section rename 2015-03-28 11:36:44 -04:00
.mailmap Initial Cookiecutter Commit. 2014-09-20 16:16:13 -07:00
.testr.conf Initial Cookiecutter Commit. 2014-09-20 16:16:13 -07:00
CONTRIBUTING.rst Workflow documentation is now in infra-manual 2014-12-05 03:30:46 +00:00
HACKING.rst Initial Cookiecutter Commit. 2014-09-20 16:16:13 -07:00
LICENSE Initial Cookiecutter Commit. 2014-09-20 16:16:13 -07:00
MANIFEST.in Initial Cookiecutter Commit. 2014-09-20 16:16:13 -07:00
README.rst Update the README a bit 2016-02-22 09:29:22 -08:00
requirements.txt Add method for registering argparse options 2015-12-06 21:39:49 -05:00
setup.cfg remove python 2.6 os-client-config classifier 2015-12-24 01:22:07 +00:00
setup.py Port in config reading from shade 2014-09-21 12:17:09 -07:00
test-requirements.txt Clean up removed hacking rule from [flake8] ignore lists 2016-01-19 21:54:17 +08:00
tox.ini Go ahead and remove final excludes 2016-01-22 19:52:45 +00:00

os-client-config

os-client-config is a library for collecting client configuration for using an OpenStack cloud in a consistent and comprehensive manner. It will find cloud config for as few as 1 cloud and as many as you want to put in a config file. It will read environment variables and config files, and it also contains some vendor specific default values so that you don't have to know extra info to use OpenStack

  • If you have a config file, you will get the clouds listed in it
  • If you have environment variables, you will get a cloud named envvars
  • If you have neither, you will get a cloud named defaults with base defaults

Environment Variables

os-client-config honors all of the normal OS_* variables. It does not provide backwards compatibility to service-specific variables such as NOVA_USERNAME.

If you have OpenStack environment variables set, os-client-config will produce a cloud config object named envvars containing your values from the environment. If you don't like the name envvars, that's ok, you can override it by setting OS_CLOUD_NAME.

Service specific settings, like the nova service type, are set with the default service type as a prefix. For instance, to set a special service_type for trove set

export OS_DATABASE_SERVICE_TYPE=rax:database

Config Files

os-client-config will look for a file called clouds.yaml in the following locations:

  • Current Directory
  • ~/.config/openstack
  • /etc/openstack

The first file found wins.

You can also set the environment variable OS_CLIENT_CONFIG_FILE to an absolute path of a file to look for and that location will be inserted at the front of the file search list.

The keys are all of the keys you'd expect from OS_* - except lower case and without the OS prefix. So, region name is set with region_name.

Service specific settings, like the nova service type, are set with the default service type as a prefix. For instance, to set a special service_type for trove (because you're using Rackspace) set:

database_service_type: 'rax:database'

Site Specific File Locations

In addition to ~/.config/openstack and /etc/openstack - some platforms have other locations they like to put things. os-client-config will also look in an OS specific config dir

  • USER_CONFIG_DIR
  • SITE_CONFIG_DIR

USER_CONFIG_DIR is different on Linux, OSX and Windows.

  • Linux: ~/.config/openstack
  • OSX: ~/Library/Application Support/openstack
  • Windows: C:UsersUSERNAMEAppDataLocalOpenStackopenstack

SITE_CONFIG_DIR is different on Linux, OSX and Windows.

  • Linux: /etc/openstack
  • OSX: /Library/Application Support/openstack
  • Windows: C:ProgramDataOpenStackopenstack

An example config file is probably helpful:

clouds:
  mtvexx:
    profile: vexxhost
    auth:
      username: mordred@inaugust.com
      password: XXXXXXXXX
      project_name: mordred@inaugust.com
    region_name: ca-ymq-1
    dns_api_version: 1
  mordred:
    region_name: RegionOne
    auth:
      username: 'mordred'
      password: XXXXXXX
      project_name: 'shade'
      auth_url: 'https://montytaylor-sjc.openstack.blueboxgrid.com:5001/v2.0'
  infra:
    profile: rackspace
    auth:
      username: openstackci
      password: XXXXXXXX
      project_id: 610275
    regions:
    - DFW
    - ORD
    - IAD

You may note a few things. First, since auth_url settings are silly and embarrassingly ugly, known cloud vendor profile information is included and may be referenced by name. One of the benefits of that is that auth_url isn't the only thing the vendor defaults contain. For instance, since Rackspace lists rax:database as the service type for trove, os-client-config knows that so that you don't have to. In case the cloud vendor profile is not available, you can provide one called clouds-public.yaml, following the same location rules previously mentioned for the config files.

regions can be a list of regions. When you call get_all_clouds, you'll get a cloud config object for each cloud/region combo.

As seen with dns_service_type, any setting that makes sense to be per-service, like service_type or endpoint or api_version can be set by prefixing the setting with the default service type. That might strike you funny when setting service_type and it does me too - but that's just the world we live in.

Auth Settings

Keystone has auth plugins - which means it's not possible to know ahead of time which auth settings are needed. os-client-config sets the default plugin type to password, which is what things all were before plugins came about. In order to facilitate validation of values, all of the parameters that exist as a result of a chosen plugin need to go into the auth dict. For password auth, this includes auth_url, username and password as well as anything related to domains, projects and trusts.

Splitting Secrets

In some scenarios, such as configuration management controlled environments, it might be easier to have secrets in one file and non-secrets in another. This is fully supported via an optional file secure.yaml which follows all the same location rules as clouds.yaml. It can contain anything you put in clouds.yaml and will take precedence over anything in the clouds.yaml file.

# clouds.yaml
clouds:
  internap:
    profile: internap
    auth:
      username: api-55f9a00fb2619
      project_name: inap-17037
    regions:
    - ams01
    - nyj01
# secure.yaml
clouds:
  internap:
    auth:
      password: XXXXXXXXXXXXXXXXX

SSL Settings

When the access to a cloud is done via a secure connection, os-client-config will always verify the SSL cert by default. This can be disabled by setting verify to False. In case the cert is signed by an unknown CA, a specific cacert can be provided via cacert. WARNING: verify will always have precedence over cacert, so when setting a CA cert but disabling verify, the cloud cert will never be validated.

Client certs are also configurable. cert will be the client cert file location. In case the cert key is not included within the client cert file, its file location needs to be set via key.

Cache Settings

Accessing a cloud is often expensive, so it's quite common to want to do some client-side caching of those operations. To facilitate that, os-client-config understands passing through cache settings to dogpile.cache, with the following behaviors:

  • Listing no config settings means you get a null cache.
  • cache.expiration_time and nothing else gets you memory cache.
  • Otherwise, cache.class and cache.arguments are passed in

Different cloud behaviors are also differently expensive to deal with. If you want to get really crazy and tweak stuff, you can specify different expiration times on a per-resource basis by passing values, in seconds to an expiration mapping keyed on the singular name of the resource. A value of -1 indicates that the resource should never expire.

os-client-config does not actually cache anything itself, but it collects and presents the cache information so that your various applications that are connecting to OpenStack can share a cache should you desire.

cache:
  class: dogpile.cache.pylibmc
  expiration_time: 3600
  arguments:
    url:
      - 127.0.0.1
  expiration:
    server: 5
    flavor: -1
clouds:
  mtvexx:
    profile: vexxhost
    auth:
      username: mordred@inaugust.com
      password: XXXXXXXXX
      project_name: mordred@inaugust.com
    region_name: ca-ymq-1
    dns_api_version: 1

IPv6

IPv6 is the future, and you should always use it if your cloud supports it and if your local network supports it. Both of those are easily detectable and all friendly software should do the right thing. However, sometimes you might exist in a location where you have an IPv6 stack, but something evil has caused it to not actually function. In that case, there is a config option you can set to unbreak you force_ipv4, or OS_FORCE_IPV4 boolean environment variable.

client:
  force_ipv4: true
clouds:
  mtvexx:
    profile: vexxhost
    auth:
      username: mordred@inaugust.com
      password: XXXXXXXXX
      project_name: mordred@inaugust.com
    region_name: ca-ymq-1
    dns_api_version: 1
  monty:
    profile: rax
    auth:
      username: mordred@inaugust.com
      password: XXXXXXXXX
      project_name: mordred@inaugust.com
    region_name: DFW

The above snippet will tell client programs to prefer returning an IPv4 address.

Per-region settings

Sometimes you have a cloud provider that has config that is common to the cloud, but also with some things you might want to express on a per-region basis. For instance, Internap provides a public and private network specific to the user in each region, and putting the values of those networks into config can make consuming programs more efficient.

To support this, the region list can actually be a list of dicts, and any setting that can be set at the cloud level can be overridden for that region.

clouds:
  internap:
    profile: internap
    auth:
      password: XXXXXXXXXXXXXXXXX
      username: api-55f9a00fb2619
      project_name: inap-17037
    regions:
    - name: ams01
      values:
        external_network: inap-17037-WAN1654
        internal_network: inap-17037-LAN4820
    - name: nyj01
      values:
        external_network: inap-17037-WAN7752
        internal_network: inap-17037-LAN6745

Usage

The simplest and least useful thing you can do is:

python -m os_client_config.config

Which will print out whatever if finds for your config. If you want to use it from python, which is much more likely what you want to do, things like:

Get a named cloud.

import os_client_config

cloud_config = os_client_config.OpenStackConfig().get_one_cloud(
    'internap', region_name='ams01')
print(cloud_config.name, cloud_config.region, cloud_config.config)

Or, get all of the clouds.

import os_client_config

cloud_config = os_client_config.OpenStackConfig().get_all_clouds()
for cloud in cloud_config:
    print(cloud.name, cloud.region, cloud.config)

argparse

If you're using os-client-config from a program that wants to process command line options, there is a registration function to register the arguments that both os-client-config and keystoneauth know how to deal with - as well as a consumption argument.

import argparse
import sys

import os_client_config

cloud_config = os_client_config.OpenStackConfig()
parser = argparse.ArgumentParser()
cloud_config.register_argparse_arguments(parser, sys.argv)

options = parser.parse_args()

cloud = cloud_config.get_one_cloud(argparse=options)

Constructing Legacy Client objects

If all you want to do is get a Client object from a python-*client library, and you want it to do all the normal things related to clouds.yaml, OS_ environment variables, a helper function is provided. The following will get you a fully configured novaclient instance.

import os_client_config

nova = os_client_config.make_client('compute')

If you want to do the same thing but on a named cloud.

import os_client_config

nova = os_client_config.make_client('compute', cloud='mtvexx')

If you want to do the same thing but also support command line parsing.

import argparse

import os_client_config

nova = os_client_config.make_client(
    'compute', options=argparse.ArgumentParser())

If you want to get fancier than that in your python, then the rest of the API is available to you. But often times, you just want to do the one thing.

Constructing Mounted Session Objects

What if you want to make direct REST calls via a Session interface? You're in luck. The same interface for make_client is supported for session_client and will return you a keystoneauth Session object that is mounted on the endpoint for the service you're looking for.

import os_client_config

session = os_client_config.session_client('compute', cloud='vexxhost')

response = session.get('/servers') server_list = response.json()['servers']

Source