summaryrefslogtreecommitdiffstats
path: root/inventory
diff options
context:
space:
mode:
authorRussell Teague <rteague@redhat.com>2017-08-29 08:06:20 -0400
committerGitHub <noreply@github.com>2017-08-29 08:06:20 -0400
commitab7b515d8ed8576b23147e64e4da73e6dea41df7 (patch)
tree01517c01840882c637a4fabf62512830fb6888c3 /inventory
parent60d544846d049b85539e7b2ea91a7aee00eb15a0 (diff)
parent7da3f730a2e2525190dc2c42661275f62891650c (diff)
downloadopenshift-ab7b515d8ed8576b23147e64e4da73e6dea41df7.tar.gz
openshift-ab7b515d8ed8576b23147e64e4da73e6dea41df7.tar.bz2
openshift-ab7b515d8ed8576b23147e64e4da73e6dea41df7.tar.xz
openshift-ab7b515d8ed8576b23147e64e4da73e6dea41df7.zip
Merge pull request #5201 from mtnbikenc/remove-unsupported
Remove unsupported playbooks and utilities
Diffstat (limited to 'inventory')
-rw-r--r--inventory/README.md6
-rw-r--r--inventory/aws/hosts/ec2.ini189
-rwxr-xr-xinventory/aws/hosts/ec2.py1511
-rw-r--r--inventory/aws/hosts/hosts1
-rwxr-xr-xinventory/gce/hosts/gce.py477
-rw-r--r--inventory/gce/hosts/hosts1
-rw-r--r--inventory/libvirt/hosts/hosts1
-rw-r--r--inventory/libvirt/hosts/libvirt.ini20
-rwxr-xr-xinventory/libvirt/hosts/libvirt_generic.py191
-rw-r--r--inventory/openstack/hosts/hosts1
-rwxr-xr-xinventory/openstack/hosts/openstack.py247
11 files changed, 1 insertions, 2644 deletions
diff --git a/inventory/README.md b/inventory/README.md
index b61bfff18..5e26e3c32 100644
--- a/inventory/README.md
+++ b/inventory/README.md
@@ -2,8 +2,4 @@
You can install OpenShift on:
-* [Amazon Web Services](aws/hosts/)
-* [BYO](byo/) (Bring your own), use this inventory config file to install OpenShift on your bare metal servers
-* [GCE](gce/) (Google Compute Engine)
-* [libvirt](libvirt/hosts/)
-* [OpenStack](openstack/hosts/)
+* [BYO](byo/) (Bring your own), use this inventory config file to install OpenShift on your pre-existing hosts
diff --git a/inventory/aws/hosts/ec2.ini b/inventory/aws/hosts/ec2.ini
deleted file mode 100644
index 64c097d47..000000000
--- a/inventory/aws/hosts/ec2.ini
+++ /dev/null
@@ -1,189 +0,0 @@
-# Ansible EC2 external inventory script settings
-#
-
-[ec2]
-
-# to talk to a private eucalyptus instance uncomment these lines
-# and edit edit eucalyptus_host to be the host name of your cloud controller
-#eucalyptus = True
-#eucalyptus_host = clc.cloud.domain.org
-
-# AWS regions to make calls to. Set this to 'all' to make request to all regions
-# in AWS and merge the results together. Alternatively, set this to a comma
-# separated list of regions. E.g. 'us-east-1,us-west-1,us-west-2'
-regions = all
-regions_exclude = us-gov-west-1,cn-north-1
-
-# When generating inventory, Ansible needs to know how to address a server.
-# Each EC2 instance has a lot of variables associated with it. Here is the list:
-# http://docs.pythonboto.org/en/latest/ref/ec2.html#module-boto.ec2.instance
-# Below are 2 variables that are used as the address of a server:
-# - destination_variable
-# - vpc_destination_variable
-
-# This is the normal destination variable to use. If you are running Ansible
-# from outside EC2, then 'public_dns_name' makes the most sense. If you are
-# running Ansible from within EC2, then perhaps you want to use the internal
-# address, and should set this to 'private_dns_name'. The key of an EC2 tag
-# may optionally be used; however the boto instance variables hold precedence
-# in the event of a collision.
-destination_variable = public_dns_name
-
-# This allows you to override the inventory_name with an ec2 variable, instead
-# of using the destination_variable above. Addressing (aka ansible_ssh_host)
-# will still use destination_variable. Tags should be written as 'tag_TAGNAME'.
-hostname_variable = tag_Name
-
-# For server inside a VPC, using DNS names may not make sense. When an instance
-# has 'subnet_id' set, this variable is used. If the subnet is public, setting
-# this to 'ip_address' will return the public IP address. For instances in a
-# private subnet, this should be set to 'private_ip_address', and Ansible must
-# be run from within EC2. The key of an EC2 tag may optionally be used; however
-# the boto instance variables hold precedence in the event of a collision.
-# WARNING: - instances that are in the private vpc, _without_ public ip address
-# will not be listed in the inventory until You set:
-# vpc_destination_variable = private_ip_address
-vpc_destination_variable = ip_address
-
-# The following two settings allow flexible ansible host naming based on a
-# python format string and a comma-separated list of ec2 tags. Note that:
-#
-# 1) If the tags referenced are not present for some instances, empty strings
-# will be substituted in the format string.
-# 2) This overrides both destination_variable and vpc_destination_variable.
-#
-#destination_format = {0}.{1}.example.com
-#destination_format_tags = Name,environment
-
-# To tag instances on EC2 with the resource records that point to them from
-# Route53, uncomment and set 'route53' to True.
-route53 = False
-
-# To exclude RDS instances from the inventory, uncomment and set to False.
-rds = False
-
-# To exclude ElastiCache instances from the inventory, uncomment and set to False.
-elasticache = False
-
-# Additionally, you can specify the list of zones to exclude looking up in
-# 'route53_excluded_zones' as a comma-separated list.
-# route53_excluded_zones = samplezone1.com, samplezone2.com
-
-# By default, only EC2 instances in the 'running' state are returned. Set
-# 'all_instances' to True to return all instances regardless of state.
-all_instances = False
-
-# By default, only EC2 instances in the 'running' state are returned. Specify
-# EC2 instance states to return as a comma-separated list. This
-# option is overridden when 'all_instances' is True.
-# instance_states = pending, running, shutting-down, terminated, stopping, stopped
-
-# By default, only RDS instances in the 'available' state are returned. Set
-# 'all_rds_instances' to True return all RDS instances regardless of state.
-all_rds_instances = False
-
-# Include RDS cluster information (Aurora etc.)
-include_rds_clusters = False
-
-# By default, only ElastiCache clusters and nodes in the 'available' state
-# are returned. Set 'all_elasticache_clusters' and/or 'all_elastic_nodes'
-# to True return all ElastiCache clusters and nodes, regardless of state.
-#
-# Note that all_elasticache_nodes only applies to listed clusters. That means
-# if you set all_elastic_clusters to false, no node will be return from
-# unavailable clusters, regardless of the state and to what you set for
-# all_elasticache_nodes.
-all_elasticache_replication_groups = False
-all_elasticache_clusters = False
-all_elasticache_nodes = False
-
-# API calls to EC2 are slow. For this reason, we cache the results of an API
-# call. Set this to the path you want cache files to be written to. Two files
-# will be written to this directory:
-# - ansible-ec2.cache
-# - ansible-ec2.index
-cache_path = ~/.ansible/tmp
-
-# The number of seconds a cache file is considered valid. After this many
-# seconds, a new API call will be made, and the cache file will be updated.
-# To disable the cache, set this value to 0
-cache_max_age = 300
-
-# Organize groups into a nested/hierarchy instead of a flat namespace.
-nested_groups = False
-
-# Replace - tags when creating groups to avoid issues with ansible
-replace_dash_in_groups = False
-
-# If set to true, any tag of the form "a,b,c" is expanded into a list
-# and the results are used to create additional tag_* inventory groups.
-expand_csv_tags = False
-
-# The EC2 inventory output can become very large. To manage its size,
-# configure which groups should be created.
-group_by_instance_id = True
-group_by_region = True
-group_by_availability_zone = True
-group_by_ami_id = True
-group_by_instance_type = True
-group_by_key_pair = True
-group_by_vpc_id = True
-group_by_security_group = True
-group_by_tag_keys = True
-group_by_tag_none = True
-group_by_route53_names = True
-group_by_rds_engine = True
-group_by_rds_parameter_group = True
-group_by_elasticache_engine = True
-group_by_elasticache_cluster = True
-group_by_elasticache_parameter_group = True
-group_by_elasticache_replication_group = True
-
-# If you only want to include hosts that match a certain regular expression
-# pattern_include = staging-*
-
-# If you want to exclude any hosts that match a certain regular expression
-# pattern_exclude = staging-*
-
-# Instance filters can be used to control which instances are retrieved for
-# inventory. For the full list of possible filters, please read the EC2 API
-# docs: http://docs.aws.amazon.com/AWSEC2/latest/APIReference/ApiReference-query-DescribeInstances.html#query-DescribeInstances-filters
-# Filters are key/value pairs separated by '=', to list multiple filters use
-# a list separated by commas. See examples below.
-
-# Retrieve only instances with (key=value) env=staging tag
-# instance_filters = tag:env=staging
-
-# Retrieve only instances with role=webservers OR role=dbservers tag
-# instance_filters = tag:role=webservers,tag:role=dbservers
-
-# Retrieve only t1.micro instances OR instances with tag env=staging
-# instance_filters = instance-type=t1.micro,tag:env=staging
-
-# You can use wildcards in filter values also. Below will list instances which
-# tag Name value matches webservers1*
-# (ex. webservers15, webservers1a, webservers123 etc)
-# instance_filters = tag:Name=webservers1*
-
-# A boto configuration profile may be used to separate out credentials
-# see http://boto.readthedocs.org/en/latest/boto_config_tut.html
-# boto_profile = some-boto-profile-name
-
-
-[credentials]
-
-# The AWS credentials can optionally be specified here. Credentials specified
-# here are ignored if the environment variable AWS_ACCESS_KEY_ID or
-# AWS_PROFILE is set, or if the boto_profile property above is set.
-#
-# Supplying AWS credentials here is not recommended, as it introduces
-# non-trivial security concerns. When going down this route, please make sure
-# to set access permissions for this file correctly, e.g. handle it the same
-# way as you would a private SSH key.
-#
-# Unlike the boto and AWS configure files, this section does not support
-# profiles.
-#
-# aws_access_key_id = AXXXXXXXXXXXXXX
-# aws_secret_access_key = XXXXXXXXXXXXXXXXXXX
-# aws_security_token = XXXXXXXXXXXXXXXXXXXXXXXXXXXX
diff --git a/inventory/aws/hosts/ec2.py b/inventory/aws/hosts/ec2.py
deleted file mode 100755
index b71458a29..000000000
--- a/inventory/aws/hosts/ec2.py
+++ /dev/null
@@ -1,1511 +0,0 @@
-#!/usr/bin/env python2
-# pylint: skip-file
-
-'''
-EC2 external inventory script
-=================================
-
-Generates inventory that Ansible can understand by making API request to
-AWS EC2 using the Boto library.
-
-NOTE: This script assumes Ansible is being executed where the environment
-variables needed for Boto have already been set:
- export AWS_ACCESS_KEY_ID='AK123'
- export AWS_SECRET_ACCESS_KEY='abc123'
-
-This script also assumes there is an ec2.ini file alongside it. To specify a
-different path to ec2.ini, define the EC2_INI_PATH environment variable:
-
- export EC2_INI_PATH=/path/to/my_ec2.ini
-
-If you're using eucalyptus you need to set the above variables and
-you need to define:
-
- export EC2_URL=http://hostname_of_your_cc:port/services/Eucalyptus
-
-If you're using boto profiles (requires boto>=2.24.0) you can choose a profile
-using the --boto-profile command line argument (e.g. ec2.py --boto-profile prod) or using
-the AWS_PROFILE variable:
-
- AWS_PROFILE=prod ansible-playbook -i ec2.py myplaybook.yml
-
-For more details, see: http://docs.pythonboto.org/en/latest/boto_config_tut.html
-
-When run against a specific host, this script returns the following variables:
- - ec2_ami_launch_index
- - ec2_architecture
- - ec2_association
- - ec2_attachTime
- - ec2_attachment
- - ec2_attachmentId
- - ec2_block_devices
- - ec2_client_token
- - ec2_deleteOnTermination
- - ec2_description
- - ec2_deviceIndex
- - ec2_dns_name
- - ec2_eventsSet
- - ec2_group_name
- - ec2_hypervisor
- - ec2_id
- - ec2_image_id
- - ec2_instanceState
- - ec2_instance_type
- - ec2_ipOwnerId
- - ec2_ip_address
- - ec2_item
- - ec2_kernel
- - ec2_key_name
- - ec2_launch_time
- - ec2_monitored
- - ec2_monitoring
- - ec2_networkInterfaceId
- - ec2_ownerId
- - ec2_persistent
- - ec2_placement
- - ec2_platform
- - ec2_previous_state
- - ec2_private_dns_name
- - ec2_private_ip_address
- - ec2_publicIp
- - ec2_public_dns_name
- - ec2_ramdisk
- - ec2_reason
- - ec2_region
- - ec2_requester_id
- - ec2_root_device_name
- - ec2_root_device_type
- - ec2_security_group_ids
- - ec2_security_group_names
- - ec2_shutdown_state
- - ec2_sourceDestCheck
- - ec2_spot_instance_request_id
- - ec2_state
- - ec2_state_code
- - ec2_state_reason
- - ec2_status
- - ec2_subnet_id
- - ec2_tenancy
- - ec2_virtualization_type
- - ec2_vpc_id
-
-These variables are pulled out of a boto.ec2.instance object. There is a lack of
-consistency with variable spellings (camelCase and underscores) since this
-just loops through all variables the object exposes. It is preferred to use the
-ones with underscores when multiple exist.
-
-In addition, if an instance has AWS Tags associated with it, each tag is a new
-variable named:
- - ec2_tag_[Key] = [Value]
-
-Security groups are comma-separated in 'ec2_security_group_ids' and
-'ec2_security_group_names'.
-'''
-
-# (c) 2012, Peter Sankauskas
-#
-# This file is part of Ansible,
-#
-# Ansible is free software: you can redistribute it and/or modify
-# it under the terms of the GNU General Public License as published by
-# the Free Software Foundation, either version 3 of the License, or
-# (at your option) any later version.
-#
-# Ansible is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with Ansible. If not, see <http://www.gnu.org/licenses/>.
-
-######################################################################
-
-import sys
-import os
-import argparse
-import re
-from time import time
-import boto
-from boto import ec2
-from boto import rds
-from boto import elasticache
-from boto import route53
-import six
-
-from ansible.module_utils import ec2 as ec2_utils
-
-HAS_BOTO3 = False
-try:
- import boto3
- HAS_BOTO3 = True
-except ImportError:
- pass
-
-from six.moves import configparser
-from collections import defaultdict
-
-try:
- import json
-except ImportError:
- import simplejson as json
-
-
-class Ec2Inventory(object):
-
- def _empty_inventory(self):
- return {"_meta" : {"hostvars" : {}}}
-
- def __init__(self):
- ''' Main execution path '''
-
- # Inventory grouped by instance IDs, tags, security groups, regions,
- # and availability zones
- self.inventory = self._empty_inventory()
-
- # Index of hostname (address) to instance ID
- self.index = {}
-
- # Boto profile to use (if any)
- self.boto_profile = None
-
- # AWS credentials.
- self.credentials = {}
-
- # Read settings and parse CLI arguments
- self.parse_cli_args()
- self.read_settings()
-
- # Make sure that profile_name is not passed at all if not set
- # as pre 2.24 boto will fall over otherwise
- if self.boto_profile:
- if not hasattr(boto.ec2.EC2Connection, 'profile_name'):
- self.fail_with_error("boto version must be >= 2.24 to use profile")
-
- # Cache
- if self.args.refresh_cache:
- self.do_api_calls_update_cache()
- elif not self.is_cache_valid():
- self.do_api_calls_update_cache()
-
- # Data to print
- if self.args.host:
- data_to_print = self.get_host_info()
-
- elif self.args.list:
- # Display list of instances for inventory
- if self.inventory == self._empty_inventory():
- data_to_print = self.get_inventory_from_cache()
- else:
- data_to_print = self.json_format_dict(self.inventory, True)
-
- print(data_to_print)
-
-
- def is_cache_valid(self):
- ''' Determines if the cache files have expired, or if it is still valid '''
-
- if os.path.isfile(self.cache_path_cache):
- mod_time = os.path.getmtime(self.cache_path_cache)
- current_time = time()
- if (mod_time + self.cache_max_age) > current_time:
- if os.path.isfile(self.cache_path_index):
- return True
-
- return False
-
-
- def read_settings(self):
- ''' Reads the settings from the ec2.ini file '''
- if six.PY3:
- config = configparser.ConfigParser()
- else:
- config = configparser.SafeConfigParser()
- ec2_default_ini_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'ec2.ini')
- ec2_ini_path = os.path.expanduser(os.path.expandvars(os.environ.get('EC2_INI_PATH', ec2_default_ini_path)))
- config.read(ec2_ini_path)
-
- # is eucalyptus?
- self.eucalyptus_host = None
- self.eucalyptus = False
- if config.has_option('ec2', 'eucalyptus'):
- self.eucalyptus = config.getboolean('ec2', 'eucalyptus')
- if self.eucalyptus and config.has_option('ec2', 'eucalyptus_host'):
- self.eucalyptus_host = config.get('ec2', 'eucalyptus_host')
-
- # Regions
- self.regions = []
- configRegions = config.get('ec2', 'regions')
- configRegions_exclude = config.get('ec2', 'regions_exclude')
- if (configRegions == 'all'):
- if self.eucalyptus_host:
- self.regions.append(boto.connect_euca(host=self.eucalyptus_host).region.name, **self.credentials)
- else:
- for regionInfo in ec2.regions():
- if regionInfo.name not in configRegions_exclude:
- self.regions.append(regionInfo.name)
- else:
- self.regions = configRegions.split(",")
-
- # Destination addresses
- self.destination_variable = config.get('ec2', 'destination_variable')
- self.vpc_destination_variable = config.get('ec2', 'vpc_destination_variable')
-
- if config.has_option('ec2', 'hostname_variable'):
- self.hostname_variable = config.get('ec2', 'hostname_variable')
- else:
- self.hostname_variable = None
-
- if config.has_option('ec2', 'destination_format') and \
- config.has_option('ec2', 'destination_format_tags'):
- self.destination_format = config.get('ec2', 'destination_format')
- self.destination_format_tags = config.get('ec2', 'destination_format_tags').split(',')
- else:
- self.destination_format = None
- self.destination_format_tags = None
-
- # Route53
- self.route53_enabled = config.getboolean('ec2', 'route53')
- self.route53_excluded_zones = []
- if config.has_option('ec2', 'route53_excluded_zones'):
- self.route53_excluded_zones.extend(
- config.get('ec2', 'route53_excluded_zones', '').split(','))
-
- # Include RDS instances?
- self.rds_enabled = True
- if config.has_option('ec2', 'rds'):
- self.rds_enabled = config.getboolean('ec2', 'rds')
-
- # Include RDS cluster instances?
- if config.has_option('ec2', 'include_rds_clusters'):
- self.include_rds_clusters = config.getboolean('ec2', 'include_rds_clusters')
- else:
- self.include_rds_clusters = False
-
- # Include ElastiCache instances?
- self.elasticache_enabled = True
- if config.has_option('ec2', 'elasticache'):
- self.elasticache_enabled = config.getboolean('ec2', 'elasticache')
-
- # Return all EC2 instances?
- if config.has_option('ec2', 'all_instances'):
- self.all_instances = config.getboolean('ec2', 'all_instances')
- else:
- self.all_instances = False
-
- # Instance states to be gathered in inventory. Default is 'running'.
- # Setting 'all_instances' to 'yes' overrides this option.
- ec2_valid_instance_states = [
- 'pending',
- 'running',
- 'shutting-down',
- 'terminated',
- 'stopping',
- 'stopped'
- ]
- self.ec2_instance_states = []
- if self.all_instances:
- self.ec2_instance_states = ec2_valid_instance_states
- elif config.has_option('ec2', 'instance_states'):
- for instance_state in config.get('ec2', 'instance_states').split(','):
- instance_state = instance_state.strip()
- if instance_state not in ec2_valid_instance_states:
- continue
- self.ec2_instance_states.append(instance_state)
- else:
- self.ec2_instance_states = ['running']
-
- # Return all RDS instances? (if RDS is enabled)
- if config.has_option('ec2', 'all_rds_instances') and self.rds_enabled:
- self.all_rds_instances = config.getboolean('ec2', 'all_rds_instances')
- else:
- self.all_rds_instances = False
-
- # Return all ElastiCache replication groups? (if ElastiCache is enabled)
- if config.has_option('ec2', 'all_elasticache_replication_groups') and self.elasticache_enabled:
- self.all_elasticache_replication_groups = config.getboolean('ec2', 'all_elasticache_replication_groups')
- else:
- self.all_elasticache_replication_groups = False
-
- # Return all ElastiCache clusters? (if ElastiCache is enabled)
- if config.has_option('ec2', 'all_elasticache_clusters') and self.elasticache_enabled:
- self.all_elasticache_clusters = config.getboolean('ec2', 'all_elasticache_clusters')
- else:
- self.all_elasticache_clusters = False
-
- # Return all ElastiCache nodes? (if ElastiCache is enabled)
- if config.has_option('ec2', 'all_elasticache_nodes') and self.elasticache_enabled:
- self.all_elasticache_nodes = config.getboolean('ec2', 'all_elasticache_nodes')
- else:
- self.all_elasticache_nodes = False
-
- # boto configuration profile (prefer CLI argument)
- self.boto_profile = self.args.boto_profile
- if config.has_option('ec2', 'boto_profile') and not self.boto_profile:
- self.boto_profile = config.get('ec2', 'boto_profile')
-
- # AWS credentials (prefer environment variables)
- if not (self.boto_profile or os.environ.get('AWS_ACCESS_KEY_ID') or
- os.environ.get('AWS_PROFILE')):
- if config.has_option('credentials', 'aws_access_key_id'):
- aws_access_key_id = config.get('credentials', 'aws_access_key_id')
- else:
- aws_access_key_id = None
- if config.has_option('credentials', 'aws_secret_access_key'):
- aws_secret_access_key = config.get('credentials', 'aws_secret_access_key')
- else:
- aws_secret_access_key = None
- if config.has_option('credentials', 'aws_security_token'):
- aws_security_token = config.get('credentials', 'aws_security_token')
- else:
- aws_security_token = None
- if aws_access_key_id:
- self.credentials = {
- 'aws_access_key_id': aws_access_key_id,
- 'aws_secret_access_key': aws_secret_access_key
- }
- if aws_security_token:
- self.credentials['security_token'] = aws_security_token
-
- # Cache related
- cache_dir = os.path.expanduser(config.get('ec2', 'cache_path'))
- if self.boto_profile:
- cache_dir = os.path.join(cache_dir, 'profile_' + self.boto_profile)
- if not os.path.exists(cache_dir):
- os.makedirs(cache_dir)
-
- cache_name = 'ansible-ec2'
- aws_profile = lambda: (self.boto_profile or
- os.environ.get('AWS_PROFILE') or
- os.environ.get('AWS_ACCESS_KEY_ID') or
- self.credentials.get('aws_access_key_id', None))
- if aws_profile():
- cache_name = '%s-%s' % (cache_name, aws_profile())
- self.cache_path_cache = cache_dir + "/%s.cache" % cache_name
- self.cache_path_index = cache_dir + "/%s.index" % cache_name
- self.cache_max_age = config.getint('ec2', 'cache_max_age')
-
- if config.has_option('ec2', 'expand_csv_tags'):
- self.expand_csv_tags = config.getboolean('ec2', 'expand_csv_tags')
- else:
- self.expand_csv_tags = False
-
- # Configure nested groups instead of flat namespace.
- if config.has_option('ec2', 'nested_groups'):
- self.nested_groups = config.getboolean('ec2', 'nested_groups')
- else:
- self.nested_groups = False
-
- # Replace dash or not in group names
- if config.has_option('ec2', 'replace_dash_in_groups'):
- self.replace_dash_in_groups = config.getboolean('ec2', 'replace_dash_in_groups')
- else:
- self.replace_dash_in_groups = True
-
- # Configure which groups should be created.
- group_by_options = [
- 'group_by_instance_id',
- 'group_by_region',
- 'group_by_availability_zone',
- 'group_by_ami_id',
- 'group_by_instance_type',
- 'group_by_key_pair',
- 'group_by_vpc_id',
- 'group_by_security_group',
- 'group_by_tag_keys',
- 'group_by_tag_none',
- 'group_by_route53_names',
- 'group_by_rds_engine',
- 'group_by_rds_parameter_group',
- 'group_by_elasticache_engine',
- 'group_by_elasticache_cluster',
- 'group_by_elasticache_parameter_group',
- 'group_by_elasticache_replication_group',
- ]
- for option in group_by_options:
- if config.has_option('ec2', option):
- setattr(self, option, config.getboolean('ec2', option))
- else:
- setattr(self, option, True)
-
- # Do we need to just include hosts that match a pattern?
- try:
- pattern_include = config.get('ec2', 'pattern_include')
- if pattern_include and len(pattern_include) > 0:
- self.pattern_include = re.compile(pattern_include)
- else:
- self.pattern_include = None
- except configparser.NoOptionError:
- self.pattern_include = None
-
- # Do we need to exclude hosts that match a pattern?
- try:
- pattern_exclude = config.get('ec2', 'pattern_exclude');
- if pattern_exclude and len(pattern_exclude) > 0:
- self.pattern_exclude = re.compile(pattern_exclude)
- else:
- self.pattern_exclude = None
- except configparser.NoOptionError:
- self.pattern_exclude = None
-
- # Instance filters (see boto and EC2 API docs). Ignore invalid filters.
- self.ec2_instance_filters = defaultdict(list)
- if config.has_option('ec2', 'instance_filters'):
-
- filters = [f for f in config.get('ec2', 'instance_filters').split(',') if f]
-
- for instance_filter in filters:
- instance_filter = instance_filter.strip()
- if not instance_filter or '=' not in instance_filter:
- continue
- filter_key, filter_value = [x.strip() for x in instance_filter.split('=', 1)]
- if not filter_key:
- continue
- self.ec2_instance_filters[filter_key].append(filter_value)
-
- def parse_cli_args(self):
- ''' Command line argument processing '''
-
- parser = argparse.ArgumentParser(description='Produce an Ansible Inventory file based on EC2')
- parser.add_argument('--list', action='store_true', default=True,
- help='List instances (default: True)')
- parser.add_argument('--host', action='store',
- help='Get all the variables about a specific instance')
- parser.add_argument('--refresh-cache', action='store_true', default=False,
- help='Force refresh of cache by making API requests to EC2 (default: False - use cache files)')
- parser.add_argument('--profile', '--boto-profile', action='store', dest='boto_profile',
- help='Use boto profile for connections to EC2')
- self.args = parser.parse_args()
-
-
- def do_api_calls_update_cache(self):
- ''' Do API calls to each region, and save data in cache files '''
-
- if self.route53_enabled:
- self.get_route53_records()
-
- for region in self.regions:
- self.get_instances_by_region(region)
- if self.rds_enabled:
- self.get_rds_instances_by_region(region)
- if self.elasticache_enabled:
- self.get_elasticache_clusters_by_region(region)
- self.get_elasticache_replication_groups_by_region(region)
- if self.include_rds_clusters:
- self.include_rds_clusters_by_region(region)
-
- self.write_to_cache(self.inventory, self.cache_path_cache)
- self.write_to_cache(self.index, self.cache_path_index)
-
- def connect(self, region):
- ''' create connection to api server'''
- if self.eucalyptus:
- conn = boto.connect_euca(host=self.eucalyptus_host, **self.credentials)
- conn.APIVersion = '2010-08-31'
- else:
- conn = self.connect_to_aws(ec2, region)
- return conn
-
- def boto_fix_security_token_in_profile(self, connect_args):
- ''' monkey patch for boto issue boto/boto#2100 '''
- profile = 'profile ' + self.boto_profile
- if boto.config.has_option(profile, 'aws_security_token'):
- connect_args['security_token'] = boto.config.get(profile, 'aws_security_token')
- return connect_args
-
- def connect_to_aws(self, module, region):
- connect_args = self.credentials
-
- # only pass the profile name if it's set (as it is not supported by older boto versions)
- if self.boto_profile:
- connect_args['profile_name'] = self.boto_profile
- self.boto_fix_security_token_in_profile(connect_args)
-
- conn = module.connect_to_region(region, **connect_args)
- # connect_to_region will fail "silently" by returning None if the region name is wrong or not supported
- if conn is None:
- self.fail_with_error("region name: %s likely not supported, or AWS is down. connection to region failed." % region)
- return conn
-
- def get_instances_by_region(self, region):
- ''' Makes an AWS EC2 API call to the list of instances in a particular
- region '''
-
- try:
- conn = self.connect(region)
- reservations = []
- if self.ec2_instance_filters:
- for filter_key, filter_values in self.ec2_instance_filters.items():
- reservations.extend(conn.get_all_instances(filters = { filter_key : filter_values }))
- else:
- reservations = conn.get_all_instances()
-
- # Pull the tags back in a second step
- # AWS are on record as saying that the tags fetched in the first `get_all_instances` request are not
- # reliable and may be missing, and the only way to guarantee they are there is by calling `get_all_tags`
- instance_ids = []
- for reservation in reservations:
- instance_ids.extend([instance.id for instance in reservation.instances])
-
- max_filter_value = 199
- tags = []
- for i in range(0, len(instance_ids), max_filter_value):
- tags.extend(conn.get_all_tags(filters={'resource-type': 'instance', 'resource-id': instance_ids[i:i+max_filter_value]}))
-
- tags_by_instance_id = defaultdict(dict)
- for tag in tags:
- tags_by_instance_id[tag.res_id][tag.name] = tag.value
-
- for reservation in reservations:
- for instance in reservation.instances:
- instance.tags = tags_by_instance_id[instance.id]
- self.add_instance(instance, region)
-
- except boto.exception.BotoServerError as e:
- if e.error_code == 'AuthFailure':
- error = self.get_auth_error_message()
- else:
- backend = 'Eucalyptus' if self.eucalyptus else 'AWS'
- error = "Error connecting to %s backend.\n%s" % (backend, e.message)
- self.fail_with_error(error, 'getting EC2 instances')
-
- def get_rds_instances_by_region(self, region):
- ''' Makes an AWS API call to the list of RDS instances in a particular
- region '''
-
- try:
- conn = self.connect_to_aws(rds, region)
- if conn:
- marker = None
- while True:
- instances = conn.get_all_dbinstances(marker=marker)
- marker = instances.marker
- for instance in instances:
- self.add_rds_instance(instance, region)
- if not marker:
- break
- except boto.exception.BotoServerError as e:
- error = e.reason
-
- if e.error_code == 'AuthFailure':
- error = self.get_auth_error_message()
- if not e.reason == "Forbidden":
- error = "Looks like AWS RDS is down:\n%s" % e.message
- self.fail_with_error(error, 'getting RDS instances')
-
- def include_rds_clusters_by_region(self, region):
- if not HAS_BOTO3:
- self.fail_with_error("Working with RDS clusters requires boto3 - please install boto3 and try again",
- "getting RDS clusters")
-
- client = ec2_utils.boto3_inventory_conn('client', 'rds', region, **self.credentials)
-
- marker, clusters = '', []
- while marker is not None:
- resp = client.describe_db_clusters(Marker=marker)
- clusters.extend(resp["DBClusters"])
- marker = resp.get('Marker', None)
-
- account_id = boto.connect_iam().get_user().arn.split(':')[4]
- c_dict = {}
- for c in clusters:
- # remove these datetime objects as there is no serialisation to json
- # currently in place and we don't need the data yet
- if 'EarliestRestorableTime' in c:
- del c['EarliestRestorableTime']
- if 'LatestRestorableTime' in c:
- del c['LatestRestorableTime']
-
- if self.ec2_instance_filters == {}:
- matches_filter = True
- else:
- matches_filter = False
-
- try:
- # arn:aws:rds:<region>:<account number>:<resourcetype>:<name>
- tags = client.list_tags_for_resource(
- ResourceName='arn:aws:rds:' + region + ':' + account_id + ':cluster:' + c['DBClusterIdentifier'])
- c['Tags'] = tags['TagList']
-
- if self.ec2_instance_filters:
- for filter_key, filter_values in self.ec2_instance_filters.items():
- # get AWS tag key e.g. tag:env will be 'env'
- tag_name = filter_key.split(":", 1)[1]
- # Filter values is a list (if you put multiple values for the same tag name)
- matches_filter = any(d['Key'] == tag_name and d['Value'] in filter_values for d in c['Tags'])
-
- if matches_filter:
- # it matches a filter, so stop looking for further matches
- break
-
- except Exception as e:
- if e.message.find('DBInstanceNotFound') >= 0:
- # AWS RDS bug (2016-01-06) means deletion does not fully complete and leave an 'empty' cluster.
- # Ignore errors when trying to find tags for these
- pass
-
- # ignore empty clusters caused by AWS bug
- if len(c['DBClusterMembers']) == 0:
- continue
- elif matches_filter:
- c_dict[c['DBClusterIdentifier']] = c
-
- self.inventory['db_clusters'] = c_dict
-
- def get_elasticache_clusters_by_region(self, region):
- ''' Makes an AWS API call to the list of ElastiCache clusters (with
- nodes' info) in a particular region.'''
-
- # ElastiCache boto module doesn't provide a get_all_intances method,
- # that's why we need to call describe directly (it would be called by
- # the shorthand method anyway...)
- try:
- conn = self.connect_to_aws(elasticache, region)
- if conn:
- # show_cache_node_info = True
- # because we also want nodes' information
- response = conn.describe_cache_clusters(None, None, None, True)
-
- except boto.exception.BotoServerError as e:
- error = e.reason
-
- if e.error_code == 'AuthFailure':
- error = self.get_auth_error_message()
- if not e.reason == "Forbidden":
- error = "Looks like AWS ElastiCache is down:\n%s" % e.message
- self.fail_with_error(error, 'getting ElastiCache clusters')
-
- try:
- # Boto also doesn't provide wrapper classes to CacheClusters or
- # CacheNodes. Because of that we can't make use of the get_list
- # method in the AWSQueryConnection. Let's do the work manually
- clusters = response['DescribeCacheClustersResponse']['DescribeCacheClustersResult']['CacheClusters']
-
- except KeyError as e:
- error = "ElastiCache query to AWS failed (unexpected format)."
- self.fail_with_error(error, 'getting ElastiCache clusters')
-
- for cluster in clusters:
- self.add_elasticache_cluster(cluster, region)
-
- def get_elasticache_replication_groups_by_region(self, region):
- ''' Makes an AWS API call to the list of ElastiCache replication groups
- in a particular region.'''
-
- # ElastiCache boto module doesn't provide a get_all_intances method,
- # that's why we need to call describe directly (it would be called by
- # the shorthand method anyway...)
- try:
- conn = self.connect_to_aws(elasticache, region)
- if conn:
- response = conn.describe_replication_groups()
-
- except boto.exception.BotoServerError as e:
- error = e.reason
-
- if e.error_code == 'AuthFailure':
- error = self.get_auth_error_message()
- if not e.reason == "Forbidden":
- error = "Looks like AWS ElastiCache [Replication Groups] is down:\n%s" % e.message
- self.fail_with_error(error, 'getting ElastiCache clusters')
-
- try:
- # Boto also doesn't provide wrapper classes to ReplicationGroups
- # Because of that we can't make use of the get_list method in the
- # AWSQueryConnection. Let's do the work manually
- replication_groups = response['DescribeReplicationGroupsResponse']['DescribeReplicationGroupsResult']['ReplicationGroups']
-
- except KeyError as e:
- error = "ElastiCache [Replication Groups] query to AWS failed (unexpected format)."
- self.fail_with_error(error, 'getting ElastiCache clusters')
-
- for replication_group in replication_groups:
- self.add_elasticache_replication_group(replication_group, region)
-
- def get_auth_error_message(self):
- ''' create an informative error message if there is an issue authenticating'''
- errors = ["Authentication error retrieving ec2 inventory."]
- if None in [os.environ.get('AWS_ACCESS_KEY_ID'), os.environ.get('AWS_SECRET_ACCESS_KEY')]:
- errors.append(' - No AWS_ACCESS_KEY_ID or AWS_SECRET_ACCESS_KEY environment vars found')
- else:
- errors.append(' - AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment vars found but may not be correct')
-
- boto_paths = ['/etc/boto.cfg', '~/.boto', '~/.aws/credentials']
- boto_config_found = list(p for p in boto_paths if os.path.isfile(os.path.expanduser(p)))
- if len(boto_config_found) > 0:
- errors.append(" - Boto configs found at '%s', but the credentials contained may not be correct" % ', '.join(boto_config_found))
- else:
- errors.append(" - No Boto config found at any expected location '%s'" % ', '.join(boto_paths))
-
- return '\n'.join(errors)
-
- def fail_with_error(self, err_msg, err_operation=None):
- '''log an error to std err for ansible-playbook to consume and exit'''
- if err_operation:
- err_msg = 'ERROR: "{err_msg}", while: {err_operation}'.format(
- err_msg=err_msg, err_operation=err_operation)
- sys.stderr.write(err_msg)
- sys.exit(1)
-
- def get_instance(self, region, instance_id):
- conn = self.connect(region)
-
- reservations = conn.get_all_instances([instance_id])
- for reservation in reservations:
- for instance in reservation.instances:
- return instance
-
- def add_instance(self, instance, region):
- ''' Adds an instance to the inventory and index, as long as it is
- addressable '''
-
- # Only return instances with desired instance states
- if instance.state not in self.ec2_instance_states:
- return
-
- # Select the best destination address
- if self.destination_format and self.destination_format_tags:
- dest = self.destination_format.format(*[ getattr(instance, 'tags').get(tag, '') for tag in self.destination_format_tags ])
- elif instance.subnet_id:
- dest = getattr(instance, self.vpc_destination_variable, None)
- if dest is None:
- dest = getattr(instance, 'tags').get(self.vpc_destination_variable, None)
- else:
- dest = getattr(instance, self.destination_variable, None)
- if dest is None:
- dest = getattr(instance, 'tags').get(self.destination_variable, None)
-
- if not dest:
- # Skip instances we cannot address (e.g. private VPC subnet)
- return
-
- # Set the inventory name
- hostname = None
- if self.hostname_variable:
- if self.hostname_variable.startswith('tag_'):
- hostname = instance.tags.get(self.hostname_variable[4:], None)
- else:
- hostname = getattr(instance, self.hostname_variable)
-
- # If we can't get a nice hostname, use the destination address
- if not hostname:
- hostname = dest
- else:
- hostname = self.to_safe(hostname).lower()
-
- # if we only want to include hosts that match a pattern, skip those that don't
- if self.pattern_include and not self.pattern_include.match(hostname):
- return
-
- # if we need to exclude hosts that match a pattern, skip those
- if self.pattern_exclude and self.pattern_exclude.match(hostname):
- return
-
- # Add to index
- self.index[hostname] = [region, instance.id]
-
- # Inventory: Group by instance ID (always a group of 1)
- if self.group_by_instance_id:
- self.inventory[instance.id] = [hostname]
- if self.nested_groups:
- self.push_group(self.inventory, 'instances', instance.id)
-
- # Inventory: Group by region
- if self.group_by_region:
- self.push(self.inventory, region, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'regions', region)
-
- # Inventory: Group by availability zone
- if self.group_by_availability_zone:
- self.push(self.inventory, instance.placement, hostname)
- if self.nested_groups:
- if self.group_by_region:
- self.push_group(self.inventory, region, instance.placement)
- self.push_group(self.inventory, 'zones', instance.placement)
-
- # Inventory: Group by Amazon Machine Image (AMI) ID
- if self.group_by_ami_id:
- ami_id = self.to_safe(instance.image_id)
- self.push(self.inventory, ami_id, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'images', ami_id)
-
- # Inventory: Group by instance type
- if self.group_by_instance_type:
- type_name = self.to_safe('type_' + instance.instance_type)
- self.push(self.inventory, type_name, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'types', type_name)
-
- # Inventory: Group by key pair
- if self.group_by_key_pair and instance.key_name:
- key_name = self.to_safe('key_' + instance.key_name)
- self.push(self.inventory, key_name, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'keys', key_name)
-
- # Inventory: Group by VPC
- if self.group_by_vpc_id and instance.vpc_id:
- vpc_id_name = self.to_safe('vpc_id_' + instance.vpc_id)
- self.push(self.inventory, vpc_id_name, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'vpcs', vpc_id_name)
-
- # Inventory: Group by security group
- if self.group_by_security_group:
- try:
- for group in instance.groups:
- key = self.to_safe("security_group_" + group.name)
- self.push(self.inventory, key, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'security_groups', key)
- except AttributeError:
- self.fail_with_error('\n'.join(['Package boto seems a bit older.',
- 'Please upgrade boto >= 2.3.0.']))
-
- # Inventory: Group by tag keys
- if self.group_by_tag_keys:
- for k, v in instance.tags.items():
- if self.expand_csv_tags and v and ',' in v:
- values = map(lambda x: x.strip(), v.split(','))
- else:
- values = [v]
-
- for v in values:
- if v:
- key = self.to_safe("tag_" + k + "=" + v)
- else:
- key = self.to_safe("tag_" + k)
- self.push(self.inventory, key, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'tags', self.to_safe("tag_" + k))
- if v:
- self.push_group(self.inventory, self.to_safe("tag_" + k), key)
-
- # Inventory: Group by Route53 domain names if enabled
- if self.route53_enabled and self.group_by_route53_names:
- route53_names = self.get_instance_route53_names(instance)
- for name in route53_names:
- self.push(self.inventory, name, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'route53', name)
-
- # Global Tag: instances without tags
- if self.group_by_tag_none and len(instance.tags) == 0:
- self.push(self.inventory, 'tag_none', hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'tags', 'tag_none')
-
- # Global Tag: tag all EC2 instances
- self.push(self.inventory, 'ec2', hostname)
-
- self.inventory["_meta"]["hostvars"][hostname] = self.get_host_info_dict_from_instance(instance)
- self.inventory["_meta"]["hostvars"][hostname]['ansible_ssh_host'] = dest
-
-
- def add_rds_instance(self, instance, region):
- ''' Adds an RDS instance to the inventory and index, as long as it is
- addressable '''
-
- # Only want available instances unless all_rds_instances is True
- if not self.all_rds_instances and instance.status != 'available':
- return
-
- # Select the best destination address
- dest = instance.endpoint[0]
-
- if not dest:
- # Skip instances we cannot address (e.g. private VPC subnet)
- return
-
- # Set the inventory name
- hostname = None
- if self.hostname_variable:
- if self.hostname_variable.startswith('tag_'):
- hostname = instance.tags.get(self.hostname_variable[4:], None)
- else:
- hostname = getattr(instance, self.hostname_variable)
-
- # If we can't get a nice hostname, use the destination address
- if not hostname:
- hostname = dest
-
- hostname = self.to_safe(hostname).lower()
-
- # Add to index
- self.index[hostname] = [region, instance.id]
-
- # Inventory: Group by instance ID (always a group of 1)
- if self.group_by_instance_id:
- self.inventory[instance.id] = [hostname]
- if self.nested_groups:
- self.push_group(self.inventory, 'instances', instance.id)
-
- # Inventory: Group by region
- if self.group_by_region:
- self.push(self.inventory, region, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'regions', region)
-
- # Inventory: Group by availability zone
- if self.group_by_availability_zone:
- self.push(self.inventory, instance.availability_zone, hostname)
- if self.nested_groups:
- if self.group_by_region:
- self.push_group(self.inventory, region, instance.availability_zone)
- self.push_group(self.inventory, 'zones', instance.availability_zone)
-
- # Inventory: Group by instance type
- if self.group_by_instance_type:
- type_name = self.to_safe('type_' + instance.instance_class)
- self.push(self.inventory, type_name, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'types', type_name)
-
- # Inventory: Group by VPC
- if self.group_by_vpc_id and instance.subnet_group and instance.subnet_group.vpc_id:
- vpc_id_name = self.to_safe('vpc_id_' + instance.subnet_group.vpc_id)
- self.push(self.inventory, vpc_id_name, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'vpcs', vpc_id_name)
-
- # Inventory: Group by security group
- if self.group_by_security_group:
- try:
- if instance.security_group:
- key = self.to_safe("security_group_" + instance.security_group.name)
- self.push(self.inventory, key, hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'security_groups', key)
-
- except AttributeError:
- self.fail_with_error('\n'.join(['Package boto seems a bit older.',
- 'Please upgrade boto >= 2.3.0.']))
-
-
- # Inventory: Group by engine
- if self.group_by_rds_engine:
- self.push(self.inventory, self.to_safe("rds_" + instance.engine), hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'rds_engines', self.to_safe("rds_" + instance.engine))
-
- # Inventory: Group by parameter group
- if self.group_by_rds_parameter_group:
- self.push(self.inventory, self.to_safe("rds_parameter_group_" + instance.parameter_group.name), hostname)
- if self.nested_groups:
- self.push_group(self.inventory, 'rds_parameter_groups', self.to_safe("rds_parameter_group_" + instance.parameter_group.name))
-
- # Global Tag: all RDS instances
- self.push(self.inventory, 'rds', hostname)
-
- self.inventory["_meta"]["hostvars"][hostname] = self.get_host_info_dict_from_instance(instance)
- self.inventory["_meta"]["hostvars"][hostname]['ansible_ssh_host'] = dest
-
- def add_elasticache_cluster(self, cluster, region):
- ''' Adds an ElastiCache cluster to the inventory and index, as long as
- it's nodes are addressable '''
-
- # Only want available clusters unless all_elasticache_clusters is True
- if not self.all_elasticache_clusters and cluster['CacheClusterStatus'] != 'available':
- return
-
- # Select the best destination address
- if 'ConfigurationEndpoint' in cluster and cluster['ConfigurationEndpoint']:
- # Memcached cluster
- dest = cluster['ConfigurationEndpoint']['Address']
- is_redis = False
- else:
- # Redis sigle node cluster
- # Because all Redis clusters are single nodes, we'll merge the
- # info from the cluster with info about the node
- dest = cluster['CacheNodes'][0]['Endpoint']['Address']
- is_redis = True
-
- if not dest:
- # Skip clusters we cannot address (e.g. private VPC subnet)
- return
-
- # Add to index
- self.index[dest] = [region, cluster['CacheClusterId']]
-
- # Inventory: Group by instance ID (always a group of 1)
- if self.group_by_instance_id:
- self.inventory[cluster['CacheClusterId']] = [dest]
- if self.nested_groups:
- self.push_group(self.inventory, 'instances', cluster['CacheClusterId'])
-
- # Inventory: Group by region
- if self.group_by_region and not is_redis:
- self.push(self.inventory, region, dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'regions', region)
-
- # Inventory: Group by availability zone
- if self.group_by_availability_zone and not is_redis:
- self.push(self.inventory, cluster['PreferredAvailabilityZone'], dest)
- if self.nested_groups:
- if self.group_by_region:
- self.push_group(self.inventory, region, cluster['PreferredAvailabilityZone'])
- self.push_group(self.inventory, 'zones', cluster['PreferredAvailabilityZone'])
-
- # Inventory: Group by node type
- if self.group_by_instance_type and not is_redis:
- type_name = self.to_safe('type_' + cluster['CacheNodeType'])
- self.push(self.inventory, type_name, dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'types', type_name)
-
- # Inventory: Group by VPC (information not available in the current
- # AWS API version for ElastiCache)
-
- # Inventory: Group by security group
- if self.group_by_security_group and not is_redis:
-
- # Check for the existence of the 'SecurityGroups' key and also if
- # this key has some value. When the cluster is not placed in a SG
- # the query can return None here and cause an error.
- if 'SecurityGroups' in cluster and cluster['SecurityGroups'] is not None:
- for security_group in cluster['SecurityGroups']:
- key = self.to_safe("security_group_" + security_group['SecurityGroupId'])
- self.push(self.inventory, key, dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'security_groups', key)
-
- # Inventory: Group by engine
- if self.group_by_elasticache_engine and not is_redis:
- self.push(self.inventory, self.to_safe("elasticache_" + cluster['Engine']), dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'elasticache_engines', self.to_safe(cluster['Engine']))
-
- # Inventory: Group by parameter group
- if self.group_by_elasticache_parameter_group:
- self.push(self.inventory, self.to_safe("elasticache_parameter_group_" + cluster['CacheParameterGroup']['CacheParameterGroupName']), dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'elasticache_parameter_groups', self.to_safe(cluster['CacheParameterGroup']['CacheParameterGroupName']))
-
- # Inventory: Group by replication group
- if self.group_by_elasticache_replication_group and 'ReplicationGroupId' in cluster and cluster['ReplicationGroupId']:
- self.push(self.inventory, self.to_safe("elasticache_replication_group_" + cluster['ReplicationGroupId']), dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'elasticache_replication_groups', self.to_safe(cluster['ReplicationGroupId']))
-
- # Global Tag: all ElastiCache clusters
- self.push(self.inventory, 'elasticache_clusters', cluster['CacheClusterId'])
-
- host_info = self.get_host_info_dict_from_describe_dict(cluster)
-
- self.inventory["_meta"]["hostvars"][dest] = host_info
-
- # Add the nodes
- for node in cluster['CacheNodes']:
- self.add_elasticache_node(node, cluster, region)
-
- def add_elasticache_node(self, node, cluster, region):
- ''' Adds an ElastiCache node to the inventory and index, as long as
- it is addressable '''
-
- # Only want available nodes unless all_elasticache_nodes is True
- if not self.all_elasticache_nodes and node['CacheNodeStatus'] != 'available':
- return
-
- # Select the best destination address
- dest = node['Endpoint']['Address']
-
- if not dest:
- # Skip nodes we cannot address (e.g. private VPC subnet)
- return
-
- node_id = self.to_safe(cluster['CacheClusterId'] + '_' + node['CacheNodeId'])
-
- # Add to index
- self.index[dest] = [region, node_id]
-
- # Inventory: Group by node ID (always a group of 1)
- if self.group_by_instance_id:
- self.inventory[node_id] = [dest]
- if self.nested_groups:
- self.push_group(self.inventory, 'instances', node_id)
-
- # Inventory: Group by region
- if self.group_by_region:
- self.push(self.inventory, region, dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'regions', region)
-
- # Inventory: Group by availability zone
- if self.group_by_availability_zone:
- self.push(self.inventory, cluster['PreferredAvailabilityZone'], dest)
- if self.nested_groups:
- if self.group_by_region:
- self.push_group(self.inventory, region, cluster['PreferredAvailabilityZone'])
- self.push_group(self.inventory, 'zones', cluster['PreferredAvailabilityZone'])
-
- # Inventory: Group by node type
- if self.group_by_instance_type:
- type_name = self.to_safe('type_' + cluster['CacheNodeType'])
- self.push(self.inventory, type_name, dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'types', type_name)
-
- # Inventory: Group by VPC (information not available in the current
- # AWS API version for ElastiCache)
-
- # Inventory: Group by security group
- if self.group_by_security_group:
-
- # Check for the existence of the 'SecurityGroups' key and also if
- # this key has some value. When the cluster is not placed in a SG
- # the query can return None here and cause an error.
- if 'SecurityGroups' in cluster and cluster['SecurityGroups'] is not None:
- for security_group in cluster['SecurityGroups']:
- key = self.to_safe("security_group_" + security_group['SecurityGroupId'])
- self.push(self.inventory, key, dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'security_groups', key)
-
- # Inventory: Group by engine
- if self.group_by_elasticache_engine:
- self.push(self.inventory, self.to_safe("elasticache_" + cluster['Engine']), dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'elasticache_engines', self.to_safe("elasticache_" + cluster['Engine']))
-
- # Inventory: Group by parameter group (done at cluster level)
-
- # Inventory: Group by replication group (done at cluster level)
-
- # Inventory: Group by ElastiCache Cluster
- if self.group_by_elasticache_cluster:
- self.push(self.inventory, self.to_safe("elasticache_cluster_" + cluster['CacheClusterId']), dest)
-
- # Global Tag: all ElastiCache nodes
- self.push(self.inventory, 'elasticache_nodes', dest)
-
- host_info = self.get_host_info_dict_from_describe_dict(node)
-
- if dest in self.inventory["_meta"]["hostvars"]:
- self.inventory["_meta"]["hostvars"][dest].update(host_info)
- else:
- self.inventory["_meta"]["hostvars"][dest] = host_info
-
- def add_elasticache_replication_group(self, replication_group, region):
- ''' Adds an ElastiCache replication group to the inventory and index '''
-
- # Only want available clusters unless all_elasticache_replication_groups is True
- if not self.all_elasticache_replication_groups and replication_group['Status'] != 'available':
- return
-
- # Select the best destination address (PrimaryEndpoint)
- dest = replication_group['NodeGroups'][0]['PrimaryEndpoint']['Address']
-
- if not dest:
- # Skip clusters we cannot address (e.g. private VPC subnet)
- return
-
- # Add to index
- self.index[dest] = [region, replication_group['ReplicationGroupId']]
-
- # Inventory: Group by ID (always a group of 1)
- if self.group_by_instance_id:
- self.inventory[replication_group['ReplicationGroupId']] = [dest]
- if self.nested_groups:
- self.push_group(self.inventory, 'instances', replication_group['ReplicationGroupId'])
-
- # Inventory: Group by region
- if self.group_by_region:
- self.push(self.inventory, region, dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'regions', region)
-
- # Inventory: Group by availability zone (doesn't apply to replication groups)
-
- # Inventory: Group by node type (doesn't apply to replication groups)
-
- # Inventory: Group by VPC (information not available in the current
- # AWS API version for replication groups
-
- # Inventory: Group by security group (doesn't apply to replication groups)
- # Check this value in cluster level
-
- # Inventory: Group by engine (replication groups are always Redis)
- if self.group_by_elasticache_engine:
- self.push(self.inventory, 'elasticache_redis', dest)
- if self.nested_groups:
- self.push_group(self.inventory, 'elasticache_engines', 'redis')
-
- # Global Tag: all ElastiCache clusters
- self.push(self.inventory, 'elasticache_replication_groups', replication_group['ReplicationGroupId'])
-
- host_info = self.get_host_info_dict_from_describe_dict(replication_group)
-
- self.inventory["_meta"]["hostvars"][dest] = host_info
-
- def get_route53_records(self):
- ''' Get and store the map of resource records to domain names that
- point to them. '''
-
- r53_conn = route53.Route53Connection()
- all_zones = r53_conn.get_zones()
-
- route53_zones = [ zone for zone in all_zones if zone.name[:-1]
- not in self.route53_excluded_zones ]
-
- self.route53_records = {}
-
- for zone in route53_zones:
- rrsets = r53_conn.get_all_rrsets(zone.id)
-
- for record_set in rrsets:
- record_name = record_set.name
-
- if record_name.endswith('.'):
- record_name = record_name[:-1]
-
- for resource in record_set.resource_records:
- self.route53_records.setdefault(resource, set())
- self.route53_records[resource].add(record_name)
-
-
- def get_instance_route53_names(self, instance):
- ''' Check if an instance is referenced in the records we have from
- Route53. If it is, return the list of domain names pointing to said
- instance. If nothing points to it, return an empty list. '''
-
- instance_attributes = [ 'public_dns_name', 'private_dns_name',
- 'ip_address', 'private_ip_address' ]
-
- name_list = set()
-
- for attrib in instance_attributes:
- try:
- value = getattr(instance, attrib)
- except AttributeError:
- continue
-
- if value in self.route53_records:
- name_list.update(self.route53_records[value])
-
- return list(name_list)
-
- def get_host_info_dict_from_instance(self, instance):
- instance_vars = {}
- for key in vars(instance):
- value = getattr(instance, key)
- key = self.to_safe('ec2_' + key)
-
- # Handle complex types
- # state/previous_state changed to properties in boto in https://github.com/boto/boto/commit/a23c379837f698212252720d2af8dec0325c9518
- if key == 'ec2__state':
- instance_vars['ec2_state'] = instance.state or ''
- instance_vars['ec2_state_code'] = instance.state_code
- elif key == 'ec2__previous_state':
- instance_vars['ec2_previous_state'] = instance.previous_state or ''
- instance_vars['ec2_previous_state_code'] = instance.previous_state_code
- elif type(value) in [int, bool]:
- instance_vars[key] = value
- elif isinstance(value, six.string_types):
- instance_vars[key] = value.strip()
- elif type(value) == type(None):
- instance_vars[key] = ''
- elif key == 'ec2_region':
- instance_vars[key] = value.name
- elif key == 'ec2__placement':
- instance_vars['ec2_placement'] = value.zone
- elif key == 'ec2_tags':
- for k, v in value.items():
- if self.expand_csv_tags and ',' in v:
- v = list(map(lambda x: x.strip(), v.split(',')))
- key = self.to_safe('ec2_tag_' + k)
- instance_vars[key] = v
- elif key == 'ec2_groups':
- group_ids = []
- group_names = []
- for group in value:
- group_ids.append(group.id)
- group_names.append(group.name)
- instance_vars["ec2_security_group_ids"] = ','.join([str(i) for i in group_ids])
- instance_vars["ec2_security_group_names"] = ','.join([str(i) for i in group_names])
- elif key == 'ec2_block_device_mapping':
- instance_vars["ec2_block_devices"] = {}
- for k, v in value.items():
- instance_vars["ec2_block_devices"][ os.path.basename(k) ] = v.volume_id
- else:
- pass
- # TODO Product codes if someone finds them useful
- #print key
- #print type(value)
- #print value
-
- return instance_vars
-
- def get_host_info_dict_from_describe_dict(self, describe_dict):
- ''' Parses the dictionary returned by the API call into a flat list
- of parameters. This method should be used only when 'describe' is
- used directly because Boto doesn't provide specific classes. '''
-
- # I really don't agree with prefixing everything with 'ec2'
- # because EC2, RDS and ElastiCache are different services.
- # I'm just following the pattern used until now to not break any
- # compatibility.
-
- host_info = {}
- for key in describe_dict:
- value = describe_dict[key]
- key = self.to_safe('ec2_' + self.uncammelize(key))
-
- # Handle complex types
-
- # Target: Memcached Cache Clusters
- if key == 'ec2_configuration_endpoint' and value:
- host_info['ec2_configuration_endpoint_address'] = value['Address']
- host_info['ec2_configuration_endpoint_port'] = value['Port']
-
- # Target: Cache Nodes and Redis Cache Clusters (single node)
- if key == 'ec2_endpoint' and value:
- host_info['ec2_endpoint_address'] = value['Address']
- host_info['ec2_endpoint_port'] = value['Port']
-
- # Target: Redis Replication Groups
- if key == 'ec2_node_groups' and value:
- host_info['ec2_endpoint_address'] = value[0]['PrimaryEndpoint']['Address']
- host_info['ec2_endpoint_port'] = value[0]['PrimaryEndpoint']['Port']
- replica_count = 0
- for node in value[0]['NodeGroupMembers']:
- if node['CurrentRole'] == 'primary':
- host_info['ec2_primary_cluster_address'] = node['ReadEndpoint']['Address']
- host_info['ec2_primary_cluster_port'] = node['ReadEndpoint']['Port']
- host_info['ec2_primary_cluster_id'] = node['CacheClusterId']
- elif node['CurrentRole'] == 'replica':
- host_info['ec2_replica_cluster_address_'+ str(replica_count)] = node['ReadEndpoint']['Address']
- host_info['ec2_replica_cluster_port_'+ str(replica_count)] = node['ReadEndpoint']['Port']
- host_info['ec2_replica_cluster_id_'+ str(replica_count)] = node['CacheClusterId']
- replica_count += 1
-
- # Target: Redis Replication Groups
- if key == 'ec2_member_clusters' and value:
- host_info['ec2_member_clusters'] = ','.join([str(i) for i in value])
-
- # Target: All Cache Clusters
- elif key == 'ec2_cache_parameter_group':
- host_info["ec2_cache_node_ids_to_reboot"] = ','.join([str(i) for i in value['CacheNodeIdsToReboot']])
- host_info['ec2_cache_parameter_group_name'] = value['CacheParameterGroupName']
- host_info['ec2_cache_parameter_apply_status'] = value['ParameterApplyStatus']
-
- # Target: Almost everything
- elif key == 'ec2_security_groups':
-
- # Skip if SecurityGroups is None
- # (it is possible to have the key defined but no value in it).
- if value is not None:
- sg_ids = []
- for sg in value:
- sg_ids.append(sg['SecurityGroupId'])
- host_info["ec2_security_group_ids"] = ','.join([str(i) for i in sg_ids])
-
- # Target: Everything
- # Preserve booleans and integers
- elif type(value) in [int, bool]:
- host_info[key] = value
-
- # Target: Everything
- # Sanitize string values
- elif isinstance(value, six.string_types):
- host_info[key] = value.strip()
-
- # Target: Everything
- # Replace None by an empty string
- elif type(value) == type(None):
- host_info[key] = ''
-
- else:
- # Remove non-processed complex types
- pass
-
- return host_info
-
- def get_host_info(self):
- ''' Get variables about a specific host '''
-
- if len(self.index) == 0:
- # Need to load index from cache
- self.load_index_from_cache()
-
- if not self.args.host in self.index:
- # try updating the cache
- self.do_api_calls_update_cache()
- if not self.args.host in self.index:
- # host might not exist anymore
- return self.json_format_dict({}, True)
-
- (region, instance_id) = self.index[self.args.host]
-
- instance = self.get_instance(region, instance_id)
- return self.json_format_dict(self.get_host_info_dict_from_instance(instance), True)
-
- def push(self, my_dict, key, element):
- ''' Push an element onto an array that may not have been defined in
- the dict '''
- group_info = my_dict.setdefault(key, [])
- if isinstance(group_info, dict):
- host_list = group_info.setdefault('hosts', [])
- host_list.append(element)
- else:
- group_info.append(element)
-
- def push_group(self, my_dict, key, element):
- ''' Push a group as a child of another group. '''
- parent_group = my_dict.setdefault(key, {})
- if not isinstance(parent_group, dict):
- parent_group = my_dict[key] = {'hosts': parent_group}
- child_groups = parent_group.setdefault('children', [])
- if element not in child_groups:
- child_groups.append(element)
-
- def get_inventory_from_cache(self):
- ''' Reads the inventory from the cache file and returns it as a JSON
- object '''
-
- cache = open(self.cache_path_cache, 'r')
- json_inventory = cache.read()
- return json_inventory
-
-
- def load_index_from_cache(self):
- ''' Reads the index from the cache file sets self.index '''
-
- cache = open(self.cache_path_index, 'r')
- json_index = cache.read()
- self.index = json.loads(json_index)
-
-
- def write_to_cache(self, data, filename):
- ''' Writes data in JSON format to a file '''
-
- json_data = self.json_format_dict(data, True)
- cache = open(filename, 'w')
- cache.write(json_data)
- cache.close()
-
- def uncammelize(self, key):
- temp = re.sub('(.)([A-Z][a-z]+)', r'\1_\2', key)
- return re.sub('([a-z0-9])([A-Z])', r'\1_\2', temp).lower()
-
- def to_safe(self, word):
- ''' Converts 'bad' characters in a string to underscores so they can be used as Ansible groups '''
- regex = "[^A-Za-z0-9\_"
- if not self.replace_dash_in_groups:
- regex += "\-"
- return re.sub(regex + "]", "_", word)
-
- def json_format_dict(self, data, pretty=False):
- ''' Converts a dict to a JSON object and dumps it as a formatted
- string '''
-
- if pretty:
- return json.dumps(data, sort_keys=True, indent=2)
- else:
- return json.dumps(data)
-
-
-# Run the script
-Ec2Inventory()
diff --git a/inventory/aws/hosts/hosts b/inventory/aws/hosts/hosts
deleted file mode 100644
index 3996e577e..000000000
--- a/inventory/aws/hosts/hosts
+++ /dev/null
@@ -1 +0,0 @@
-localhost ansible_connection=local ansible_become=no ansible_python_interpreter='/usr/bin/env python2'
diff --git a/inventory/gce/hosts/gce.py b/inventory/gce/hosts/gce.py
deleted file mode 100755
index 2be46a58c..000000000
--- a/inventory/gce/hosts/gce.py
+++ /dev/null
@@ -1,477 +0,0 @@
-#!/usr/bin/env python2
-# pylint: skip-file
-# Copyright 2013 Google Inc.
-#
-# This file is part of Ansible
-#
-# Ansible is free software: you can redistribute it and/or modify
-# it under the terms of the GNU General Public License as published by
-# the Free Software Foundation, either version 3 of the License, or
-# (at your option) any later version.
-#
-# Ansible is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with Ansible. If not, see <http://www.gnu.org/licenses/>.
-
-'''
-GCE external inventory script
-=================================
-
-Generates inventory that Ansible can understand by making API requests
-Google Compute Engine via the libcloud library. Full install/configuration
-instructions for the gce* modules can be found in the comments of
-ansible/test/gce_tests.py.
-
-When run against a specific host, this script returns the following variables
-based on the data obtained from the libcloud Node object:
- - gce_uuid
- - gce_id
- - gce_image
- - gce_machine_type
- - gce_private_ip
- - gce_public_ip
- - gce_name
- - gce_description
- - gce_status
- - gce_zone
- - gce_tags
- - gce_metadata
- - gce_network
-
-When run in --list mode, instances are grouped by the following categories:
- - zone:
- zone group name examples are us-central1-b, europe-west1-a, etc.
- - instance tags:
- An entry is created for each tag. For example, if you have two instances
- with a common tag called 'foo', they will both be grouped together under
- the 'tag_foo' name.
- - network name:
- the name of the network is appended to 'network_' (e.g. the 'default'
- network will result in a group named 'network_default')
- - machine type
- types follow a pattern like n1-standard-4, g1-small, etc.
- - running status:
- group name prefixed with 'status_' (e.g. status_running, status_stopped,..)
- - image:
- when using an ephemeral/scratch disk, this will be set to the image name
- used when creating the instance (e.g. debian-7-wheezy-v20130816). when
- your instance was created with a root persistent disk it will be set to
- 'persistent_disk' since there is no current way to determine the image.
-
-Examples:
- Execute uname on all instances in the us-central1-a zone
- $ ansible -i gce.py us-central1-a -m shell -a "/bin/uname -a"
-
- Use the GCE inventory script to print out instance specific information
- $ contrib/inventory/gce.py --host my_instance
-
-Author: Eric Johnson <erjohnso@google.com>
-Contributors: Matt Hite <mhite@hotmail.com>, Tom Melendez <supertom@google.com>
-Version: 0.0.3
-'''
-
-__requires__ = ['pycrypto>=2.6']
-try:
- import pkg_resources
-except ImportError:
- # Use pkg_resources to find the correct versions of libraries and set
- # sys.path appropriately when there are multiversion installs. We don't
- # fail here as there is code that better expresses the errors where the
- # library is used.
- pass
-
-USER_AGENT_PRODUCT="Ansible-gce_inventory_plugin"
-USER_AGENT_VERSION="v2"
-
-import sys
-import os
-import argparse
-
-from time import time
-
-import ConfigParser
-
-import logging
-logging.getLogger('libcloud.common.google').addHandler(logging.NullHandler())
-
-try:
- import json
-except ImportError:
- import simplejson as json
-
-try:
- from libcloud.compute.types import Provider
- from libcloud.compute.providers import get_driver
- _ = Provider.GCE
-except:
- sys.exit("GCE inventory script requires libcloud >= 0.13")
-
-
-class CloudInventoryCache(object):
- def __init__(self, cache_name='ansible-cloud-cache', cache_path='/tmp',
- cache_max_age=300):
- cache_dir = os.path.expanduser(cache_path)
- if not os.path.exists(cache_dir):
- os.makedirs(cache_dir)
- self.cache_path_cache = os.path.join(cache_dir, cache_name)
-
- self.cache_max_age = cache_max_age
-
- def is_valid(self, max_age=None):
- ''' Determines if the cache files have expired, or if it is still valid '''
-
- if max_age is None:
- max_age = self.cache_max_age
-
- if os.path.isfile(self.cache_path_cache):
- mod_time = os.path.getmtime(self.cache_path_cache)
- current_time = time()
- if (mod_time + max_age) > current_time:
- return True
-
- return False
-
- def get_all_data_from_cache(self, filename=''):
- ''' Reads the JSON inventory from the cache file. Returns Python dictionary. '''
-
- data = ''
- if not filename:
- filename = self.cache_path_cache
- with open(filename, 'r') as cache:
- data = cache.read()
- return json.loads(data)
-
- def write_to_cache(self, data, filename=''):
- ''' Writes data to file as JSON. Returns True. '''
- if not filename:
- filename = self.cache_path_cache
- json_data = json.dumps(data)
- with open(filename, 'w') as cache:
- cache.write(json_data)
- return True
-
-
-class GceInventory(object):
- def __init__(self):
- # Cache object
- self.cache = None
- # dictionary containing inventory read from disk
- self.inventory = {}
-
- # Read settings and parse CLI arguments
- self.parse_cli_args()
- self.config = self.get_config()
- self.driver = self.get_gce_driver()
- self.ip_type = self.get_inventory_options()
- if self.ip_type:
- self.ip_type = self.ip_type.lower()
-
- # Cache management
- start_inventory_time = time()
- cache_used = False
- if self.args.refresh_cache or not self.cache.is_valid():
- self.do_api_calls_update_cache()
- else:
- self.load_inventory_from_cache()
- cache_used = True
- self.inventory['_meta']['stats'] = {'use_cache': True}
- self.inventory['_meta']['stats'] = {
- 'inventory_load_time': time() - start_inventory_time,
- 'cache_used': cache_used
- }
-
- # Just display data for specific host
- if self.args.host:
- print(self.json_format_dict(
- self.inventory['_meta']['hostvars'][self.args.host],
- pretty=self.args.pretty))
- else:
- # Otherwise, assume user wants all instances grouped
- zones = self.parse_env_zones()
- print(self.json_format_dict(self.inventory,
- pretty=self.args.pretty))
- sys.exit(0)
-
- def get_config(self):
- """
- Reads the settings from the gce.ini file.
-
- Populates a SafeConfigParser object with defaults and
- attempts to read an .ini-style configuration from the filename
- specified in GCE_INI_PATH. If the environment variable is
- not present, the filename defaults to gce.ini in the current
- working directory.
- """
- gce_ini_default_path = os.path.join(
- os.path.dirname(os.path.realpath(__file__)), "gce.ini")
- gce_ini_path = os.environ.get('GCE_INI_PATH', gce_ini_default_path)
-
- # Create a ConfigParser.
- # This provides empty defaults to each key, so that environment
- # variable configuration (as opposed to INI configuration) is able
- # to work.
- config = ConfigParser.SafeConfigParser(defaults={
- 'gce_service_account_email_address': '',
- 'gce_service_account_pem_file_path': '',
- 'gce_project_id': '',
- 'libcloud_secrets': '',
- 'inventory_ip_type': '',
- 'cache_path': '~/.ansible/tmp',
- 'cache_max_age': '300'
- })
- if 'gce' not in config.sections():
- config.add_section('gce')
- if 'inventory' not in config.sections():
- config.add_section('inventory')
- if 'cache' not in config.sections():
- config.add_section('cache')
-
- config.read(gce_ini_path)
-
- #########
- # Section added for processing ini settings
- #########
-
- # Set the instance_states filter based on config file options
- self.instance_states = []
- if config.has_option('gce', 'instance_states'):
- states = config.get('gce', 'instance_states')
- # Ignore if instance_states is an empty string.
- if states:
- self.instance_states = states.split(',')
-
- # Caching
- cache_path = config.get('cache', 'cache_path')
- cache_max_age = config.getint('cache', 'cache_max_age')
- # TOOD(supertom): support project-specific caches
- cache_name = 'ansible-gce.cache'
- self.cache = CloudInventoryCache(cache_path=cache_path,
- cache_max_age=cache_max_age,
- cache_name=cache_name)
- return config
-
- def get_inventory_options(self):
- """Determine inventory options. Environment variables always
- take precedence over configuration files."""
- ip_type = self.config.get('inventory', 'inventory_ip_type')
- # If the appropriate environment variables are set, they override
- # other configuration
- ip_type = os.environ.get('INVENTORY_IP_TYPE', ip_type)
- return ip_type
-
- def get_gce_driver(self):
- """Determine the GCE authorization settings and return a
- libcloud driver.
- """
- # Attempt to get GCE params from a configuration file, if one
- # exists.
- secrets_path = self.config.get('gce', 'libcloud_secrets')
- secrets_found = False
- try:
- import secrets
- args = list(getattr(secrets, 'GCE_PARAMS', []))
- kwargs = getattr(secrets, 'GCE_KEYWORD_PARAMS', {})
- secrets_found = True
- except:
- pass
-
- if not secrets_found and secrets_path:
- if not secrets_path.endswith('secrets.py'):
- err = "Must specify libcloud secrets file as "
- err += "/absolute/path/to/secrets.py"
- sys.exit(err)
- sys.path.append(os.path.dirname(secrets_path))
- try:
- import secrets
- args = list(getattr(secrets, 'GCE_PARAMS', []))
- kwargs = getattr(secrets, 'GCE_KEYWORD_PARAMS', {})
- secrets_found = True
- except:
- pass
- if not secrets_found:
- args = [
- self.config.get('gce','gce_service_account_email_address'),
- self.config.get('gce','gce_service_account_pem_file_path')
- ]
- kwargs = {'project': self.config.get('gce', 'gce_project_id')}
-
- # If the appropriate environment variables are set, they override
- # other configuration; process those into our args and kwargs.
- args[0] = os.environ.get('GCE_EMAIL', args[0])
- args[1] = os.environ.get('GCE_PEM_FILE_PATH', args[1])
- kwargs['project'] = os.environ.get('GCE_PROJECT', kwargs['project'])
-
- # Retrieve and return the GCE driver.
- gce = get_driver(Provider.GCE)(*args, **kwargs)
- gce.connection.user_agent_append(
- '%s/%s' % (USER_AGENT_PRODUCT, USER_AGENT_VERSION),
- )
- return gce
-
- def parse_env_zones(self):
- '''returns a list of comma separated zones parsed from the GCE_ZONE environment variable.
- If provided, this will be used to filter the results of the grouped_instances call'''
- import csv
- reader = csv.reader([os.environ.get('GCE_ZONE',"")], skipinitialspace=True)
- zones = [r for r in reader]
- return [z for z in zones[0]]
-
- def parse_cli_args(self):
- ''' Command line argument processing '''
-
- parser = argparse.ArgumentParser(
- description='Produce an Ansible Inventory file based on GCE')
- parser.add_argument('--list', action='store_true', default=True,
- help='List instances (default: True)')
- parser.add_argument('--host', action='store',
- help='Get all information about an instance')
- parser.add_argument('--pretty', action='store_true', default=False,
- help='Pretty format (default: False)')
- parser.add_argument(
- '--refresh-cache', action='store_true', default=False,
- help='Force refresh of cache by making API requests (default: False - use cache files)')
- self.args = parser.parse_args()
-
-
- def node_to_dict(self, inst):
- md = {}
-
- if inst is None:
- return {}
-
- if 'items' in inst.extra['metadata']:
- for entry in inst.extra['metadata']['items']:
- md[entry['key']] = entry['value']
-
- net = inst.extra['networkInterfaces'][0]['network'].split('/')[-1]
- # default to exernal IP unless user has specified they prefer internal
- if self.ip_type == 'internal':
- ssh_host = inst.private_ips[0]
- else:
- ssh_host = inst.public_ips[0] if len(inst.public_ips) >= 1 else inst.private_ips[0]
-
- return {
- 'gce_uuid': inst.uuid,
- 'gce_id': inst.id,
- 'gce_image': inst.image,
- 'gce_machine_type': inst.size,
- 'gce_private_ip': inst.private_ips[0],
- 'gce_public_ip': inst.public_ips[0] if len(inst.public_ips) >= 1 else None,
- 'gce_name': inst.name,
- 'gce_description': inst.extra['description'],
- 'gce_status': inst.extra['status'],
- 'gce_zone': inst.extra['zone'].name,
- 'gce_tags': inst.extra['tags'],
- 'gce_metadata': md,
- 'gce_network': net,
- # Hosts don't have a public name, so we add an IP
- 'ansible_ssh_host': ssh_host
- }
-
- def load_inventory_from_cache(self):
- ''' Loads inventory from JSON on disk. '''
-
- try:
- self.inventory = self.cache.get_all_data_from_cache()
- hosts = self.inventory['_meta']['hostvars']
- except Exception as e:
- print(
- "Invalid inventory file %s. Please rebuild with -refresh-cache option."
- % (self.cache.cache_path_cache))
- raise
-
- def do_api_calls_update_cache(self):
- ''' Do API calls and save data in cache. '''
- zones = self.parse_env_zones()
- data = self.group_instances(zones)
- self.cache.write_to_cache(data)
- self.inventory = data
-
- def list_nodes(self):
- all_nodes = []
- params, more_results = {'maxResults': 500}, True
- while more_results:
- self.driver.connection.gce_params=params
- all_nodes.extend(self.driver.list_nodes())
- more_results = 'pageToken' in params
- return all_nodes
-
- def group_instances(self, zones=None):
- '''Group all instances'''
- groups = {}
- meta = {}
- meta["hostvars"] = {}
-
- for node in self.list_nodes():
-
- # This check filters on the desired instance states defined in the
- # config file with the instance_states config option.
- #
- # If the instance_states list is _empty_ then _ALL_ states are returned.
- #
- # If the instance_states list is _populated_ then check the current
- # state against the instance_states list
- if self.instance_states and not node.extra['status'] in self.instance_states:
- continue
-
- name = node.name
-
- meta["hostvars"][name] = self.node_to_dict(node)
-
- zone = node.extra['zone'].name
-
- # To avoid making multiple requests per zone
- # we list all nodes and then filter the results
- if zones and zone not in zones:
- continue
-
- if zone in groups: groups[zone].append(name)
- else: groups[zone] = [name]
-
- tags = node.extra['tags']
- for t in tags:
- if t.startswith('group-'):
- tag = t[6:]
- else:
- tag = 'tag_%s' % t
- if tag in groups: groups[tag].append(name)
- else: groups[tag] = [name]
-
- net = node.extra['networkInterfaces'][0]['network'].split('/')[-1]
- net = 'network_%s' % net
- if net in groups: groups[net].append(name)
- else: groups[net] = [name]
-
- machine_type = node.size
- if machine_type in groups: groups[machine_type].append(name)
- else: groups[machine_type] = [name]
-
- image = node.image and node.image or 'persistent_disk'
- if image in groups: groups[image].append(name)
- else: groups[image] = [name]
-
- status = node.extra['status']
- stat = 'status_%s' % status.lower()
- if stat in groups: groups[stat].append(name)
- else: groups[stat] = [name]
-
- groups["_meta"] = meta
-
- return groups
-
- def json_format_dict(self, data, pretty=False):
- ''' Converts a dict to a JSON object and dumps it as a formatted
- string '''
-
- if pretty:
- return json.dumps(data, sort_keys=True, indent=2)
- else:
- return json.dumps(data)
-
-# Run the script
-if __name__ == '__main__':
- GceInventory()
diff --git a/inventory/gce/hosts/hosts b/inventory/gce/hosts/hosts
deleted file mode 100644
index 3996e577e..000000000
--- a/inventory/gce/hosts/hosts
+++ /dev/null
@@ -1 +0,0 @@
-localhost ansible_connection=local ansible_become=no ansible_python_interpreter='/usr/bin/env python2'
diff --git a/inventory/libvirt/hosts/hosts b/inventory/libvirt/hosts/hosts
deleted file mode 100644
index 3996e577e..000000000
--- a/inventory/libvirt/hosts/hosts
+++ /dev/null
@@ -1 +0,0 @@
-localhost ansible_connection=local ansible_become=no ansible_python_interpreter='/usr/bin/env python2'
diff --git a/inventory/libvirt/hosts/libvirt.ini b/inventory/libvirt/hosts/libvirt.ini
deleted file mode 100644
index 62ff204dd..000000000
--- a/inventory/libvirt/hosts/libvirt.ini
+++ /dev/null
@@ -1,20 +0,0 @@
-# Ansible libvirt external inventory script settings
-#
-
-[libvirt]
-
-uri = qemu:///system
-
-# API calls to libvirt can be slow. For this reason, we cache the results of an API
-# call. Set this to the path you want cache files to be written to. Two files
-# will be written to this directory:
-# - ansible-libvirt.cache
-# - ansible-libvirt.index
-cache_path = /tmp
-
-# The number of seconds a cache file is considered valid. After this many
-# seconds, a new API call will be made, and the cache file will be updated.
-cache_max_age = 900
-
-
-
diff --git a/inventory/libvirt/hosts/libvirt_generic.py b/inventory/libvirt/hosts/libvirt_generic.py
deleted file mode 100755
index d63e07b64..000000000
--- a/inventory/libvirt/hosts/libvirt_generic.py
+++ /dev/null
@@ -1,191 +0,0 @@
-#!/usr/bin/env python2
-# pylint: skip-file
-
-'''
-libvirt external inventory script
-=================================
-
-Ansible has a feature where instead of reading from /etc/ansible/hosts
-as a text file, it can query external programs to obtain the list
-of hosts, groups the hosts are in, and even variables to assign to each host.
-
-To use this, copy this file over /etc/ansible/hosts and chmod +x the file.
-This, more or less, allows you to keep one central database containing
-info about all of your managed instances.
-
-'''
-
-# (c) 2015, Jason DeTiberus <jdetiber@redhat.com>
-#
-# This file is part of Ansible,
-#
-# Ansible is free software: you can redistribute it and/or modify
-# it under the terms of the GNU General Public License as published by
-# the Free Software Foundation, either version 3 of the License, or
-# (at your option) any later version.
-#
-# Ansible is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with Ansible. If not, see <http://www.gnu.org/licenses/>.
-
-######################################################################
-
-import argparse
-import ConfigParser
-import os
-import sys
-import libvirt
-import xml.etree.ElementTree as ET
-
-try:
- import json
-except ImportError:
- import simplejson as json
-
-
-class LibvirtInventory(object):
- ''' libvirt dynamic inventory '''
-
- def __init__(self):
- ''' Main execution path '''
-
- self.inventory = dict() # A list of groups and the hosts in that group
- self.cache = dict() # Details about hosts in the inventory
-
- # Read settings and parse CLI arguments
- self.read_settings()
- self.parse_cli_args()
-
- if self.args.host:
- print(_json_format_dict(self.get_host_info(), self.args.pretty))
- elif self.args.list:
- print(_json_format_dict(self.get_inventory(), self.args.pretty))
- else: # default action with no options
- print(_json_format_dict(self.get_inventory(), self.args.pretty))
-
- def read_settings(self):
- ''' Reads the settings from the libvirt.ini file '''
-
- config = ConfigParser.SafeConfigParser()
- config.read(
- os.path.dirname(os.path.realpath(__file__)) + '/libvirt.ini'
- )
- self.libvirt_uri = config.get('libvirt', 'uri')
-
- def parse_cli_args(self):
- ''' Command line argument processing '''
-
- parser = argparse.ArgumentParser(
- description='Produce an Ansible Inventory file based on libvirt'
- )
- parser.add_argument(
- '--list',
- action='store_true',
- default=True,
- help='List instances (default: True)'
- )
- parser.add_argument(
- '--host',
- action='store',
- help='Get all the variables about a specific instance'
- )
- parser.add_argument(
- '--pretty',
- action='store_true',
- default=False,
- help='Pretty format (default: False)'
- )
- self.args = parser.parse_args()
-
- def get_host_info(self):
- ''' Get variables about a specific host '''
-
- inventory = self.get_inventory()
- if self.args.host in inventory['_meta']['hostvars']:
- return inventory['_meta']['hostvars'][self.args.host]
-
- def get_inventory(self):
- ''' Construct the inventory '''
-
- inventory = dict(_meta=dict(hostvars=dict()))
-
- conn = libvirt.openReadOnly(self.libvirt_uri)
- if conn is None:
- print("Failed to open connection to %s" % self.libvirt_uri)
- sys.exit(1)
-
- domains = conn.listAllDomains()
- if domains is None:
- print("Failed to list domains for connection %s" % self.libvirt_uri)
- sys.exit(1)
-
- for domain in domains:
- hostvars = dict(libvirt_name=domain.name(),
- libvirt_id=domain.ID(),
- libvirt_uuid=domain.UUIDString())
- domain_name = domain.name()
-
- # TODO: add support for guests that are not in a running state
- state, _ = domain.state()
- # 2 is the state for a running guest
- if state != 1:
- continue
-
- hostvars['libvirt_status'] = 'running'
-
- root = ET.fromstring(domain.XMLDesc())
- ansible_ns = {'ansible': 'https://github.com/ansible/ansible'}
- for tag_elem in root.findall('./metadata/ansible:tags/ansible:tag', ansible_ns):
- tag = tag_elem.text
- _push(inventory, "tag_%s" % tag, domain_name)
- _push(hostvars, 'libvirt_tags', tag)
-
- # TODO: support more than one network interface, also support
- # interface types other than 'network'
- interface = root.find("./devices/interface[@type='network']")
- if interface is not None:
- source_elem = interface.find('source')
- mac_elem = interface.find('mac')
- if source_elem is not None and \
- mac_elem is not None:
- # Adding this to disable pylint check specifically
- # ignoring libvirt-python versions that
- # do not include DHCPLeases
- # This is needed until we upgrade the build bot to
- # RHEL7 (>= 1.2.6 libvirt)
- # pylint: disable=no-member
- dhcp_leases = conn.networkLookupByName(source_elem.get('network')) \
- .DHCPLeases(mac_elem.get('address'))
- if len(dhcp_leases) > 0:
- ip_address = dhcp_leases[0]['ipaddr']
- hostvars['ansible_ssh_host'] = ip_address
- hostvars['libvirt_ip_address'] = ip_address
-
- inventory['_meta']['hostvars'][domain_name] = hostvars
-
- return inventory
-
-def _push(my_dict, key, element):
- '''
- Push element to the my_dict[key] list.
- After having initialized my_dict[key] if it dosn't exist.
- '''
-
- if key in my_dict:
- my_dict[key].append(element)
- else:
- my_dict[key] = [element]
-
-def _json_format_dict(data, pretty=False):
- ''' Serialize data to a JSON formated str '''
-
- if pretty:
- return json.dumps(data, sort_keys=True, indent=2)
- else:
- return json.dumps(data)
-
-LibvirtInventory()
diff --git a/inventory/openstack/hosts/hosts b/inventory/openstack/hosts/hosts
deleted file mode 100644
index 9b63e98f4..000000000
--- a/inventory/openstack/hosts/hosts
+++ /dev/null
@@ -1 +0,0 @@
-localhost ansible_become=no ansible_python_interpreter='/usr/bin/env python2' connection=local
diff --git a/inventory/openstack/hosts/openstack.py b/inventory/openstack/hosts/openstack.py
deleted file mode 100755
index deefd3b5d..000000000
--- a/inventory/openstack/hosts/openstack.py
+++ /dev/null
@@ -1,247 +0,0 @@
-#!/usr/bin/env python
-# pylint: skip-file
-
-# Copyright (c) 2012, Marco Vito Moscaritolo <marco@agavee.com>
-# Copyright (c) 2013, Jesse Keating <jesse.keating@rackspace.com>
-# Copyright (c) 2015, Hewlett-Packard Development Company, L.P.
-# Copyright (c) 2016, Rackspace Australia
-#
-# This module is free software: you can redistribute it and/or modify
-# it under the terms of the GNU General Public License as published by
-# the Free Software Foundation, either version 3 of the License, or
-# (at your option) any later version.
-#
-# This software is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this software. If not, see <http://www.gnu.org/licenses/>.
-
-# The OpenStack Inventory module uses os-client-config for configuration.
-# https://github.com/stackforge/os-client-config
-# This means it will either:
-# - Respect normal OS_* environment variables like other OpenStack tools
-# - Read values from a clouds.yaml file.
-# If you want to configure via clouds.yaml, you can put the file in:
-# - Current directory
-# - ~/.config/openstack/clouds.yaml
-# - /etc/openstack/clouds.yaml
-# - /etc/ansible/openstack.yml
-# The clouds.yaml file can contain entries for multiple clouds and multiple
-# regions of those clouds. If it does, this inventory module will connect to
-# all of them and present them as one contiguous inventory.
-#
-# See the adjacent openstack.yml file for an example config file
-# There are two ansible inventory specific options that can be set in
-# the inventory section.
-# expand_hostvars controls whether or not the inventory will make extra API
-# calls to fill out additional information about each server
-# use_hostnames changes the behavior from registering every host with its UUID
-# and making a group of its hostname to only doing this if the
-# hostname in question has more than one server
-# fail_on_errors causes the inventory to fail and return no hosts if one cloud
-# has failed (for example, bad credentials or being offline).
-# When set to False, the inventory will return hosts from
-# whichever other clouds it can contact. (Default: True)
-
-import argparse
-import collections
-import os
-import sys
-import time
-from distutils.version import StrictVersion
-
-try:
- import json
-except:
- import simplejson as json
-
-import os_client_config
-import shade
-import shade.inventory
-
-CONFIG_FILES = ['/etc/ansible/openstack.yaml', '/etc/ansible/openstack.yml']
-
-
-def get_groups_from_server(server_vars, namegroup=True):
- groups = []
-
- region = server_vars['region']
- cloud = server_vars['cloud']
- metadata = server_vars.get('metadata', {})
-
- # Create a group for the cloud
- groups.append(cloud)
-
- # Create a group on region
- groups.append(region)
-
- # And one by cloud_region
- groups.append("%s_%s" % (cloud, region))
-
- # Check if group metadata key in servers' metadata
- if 'group' in metadata:
- groups.append(metadata['group'])
-
- for extra_group in metadata.get('groups', '').split(','):
- if extra_group:
- groups.append(extra_group.strip())
-
- groups.append('instance-%s' % server_vars['id'])
- if namegroup:
- groups.append(server_vars['name'])
-
- for key in ('flavor', 'image'):
- if 'name' in server_vars[key]:
- groups.append('%s-%s' % (key, server_vars[key]['name']))
-
- for key, value in iter(metadata.items()):
- groups.append('meta-%s_%s' % (key, value))
-
- az = server_vars.get('az', None)
- if az:
- # Make groups for az, region_az and cloud_region_az
- groups.append(az)
- groups.append('%s_%s' % (region, az))
- groups.append('%s_%s_%s' % (cloud, region, az))
- return groups
-
-
-def get_host_groups(inventory, refresh=False):
- (cache_file, cache_expiration_time) = get_cache_settings()
- if is_cache_stale(cache_file, cache_expiration_time, refresh=refresh):
- groups = to_json(get_host_groups_from_cloud(inventory))
- open(cache_file, 'w').write(groups)
- else:
- groups = open(cache_file, 'r').read()
- return groups
-
-
-def append_hostvars(hostvars, groups, key, server, namegroup=False):
- hostvars[key] = dict(
- ansible_ssh_host=server['interface_ip'],
- openstack=server)
- for group in get_groups_from_server(server, namegroup=namegroup):
- groups[group].append(key)
-
-
-def get_host_groups_from_cloud(inventory):
- groups = collections.defaultdict(list)
- firstpass = collections.defaultdict(list)
- hostvars = {}
- list_args = {}
- if hasattr(inventory, 'extra_config'):
- use_hostnames = inventory.extra_config['use_hostnames']
- list_args['expand'] = inventory.extra_config['expand_hostvars']
- if StrictVersion(shade.__version__) >= StrictVersion("1.6.0"):
- list_args['fail_on_cloud_config'] = \
- inventory.extra_config['fail_on_errors']
- else:
- use_hostnames = False
-
- for server in inventory.list_hosts(**list_args):
-
- if 'interface_ip' not in server:
- continue
- firstpass[server['name']].append(server)
- for name, servers in firstpass.items():
- if len(servers) == 1 and use_hostnames:
- append_hostvars(hostvars, groups, name, servers[0])
- else:
- server_ids = set()
- # Trap for duplicate results
- for server in servers:
- server_ids.add(server['id'])
- if len(server_ids) == 1 and use_hostnames:
- append_hostvars(hostvars, groups, name, servers[0])
- else:
- for server in servers:
- append_hostvars(
- hostvars, groups, server['id'], server,
- namegroup=True)
- groups['_meta'] = {'hostvars': hostvars}
- return groups
-
-
-def is_cache_stale(cache_file, cache_expiration_time, refresh=False):
- ''' Determines if cache file has expired, or if it is still valid '''
- if refresh:
- return True
- if os.path.isfile(cache_file) and os.path.getsize(cache_file) > 0:
- mod_time = os.path.getmtime(cache_file)
- current_time = time.time()
- if (mod_time + cache_expiration_time) > current_time:
- return False
- return True
-
-
-def get_cache_settings():
- config = os_client_config.config.OpenStackConfig(
- config_files=os_client_config.config.CONFIG_FILES + CONFIG_FILES)
- # For inventory-wide caching
- cache_expiration_time = config.get_cache_expiration_time()
- cache_path = config.get_cache_path()
- if not os.path.exists(cache_path):
- os.makedirs(cache_path)
- cache_file = os.path.join(cache_path, 'ansible-inventory.cache')
- return (cache_file, cache_expiration_time)
-
-
-def to_json(in_dict):
- return json.dumps(in_dict, sort_keys=True, indent=2)
-
-
-def parse_args():
- parser = argparse.ArgumentParser(description='OpenStack Inventory Module')
- parser.add_argument('--private',
- action='store_true',
- help='Use private address for ansible host')
- parser.add_argument('--refresh', action='store_true',
- help='Refresh cached information')
- parser.add_argument('--debug', action='store_true', default=False,
- help='Enable debug output')
- group = parser.add_mutually_exclusive_group(required=True)
- group.add_argument('--list', action='store_true',
- help='List active servers')
- group.add_argument('--host', help='List details about the specific host')
-
- return parser.parse_args()
-
-
-def main():
- args = parse_args()
- try:
- config_files = os_client_config.config.CONFIG_FILES + CONFIG_FILES
- shade.simple_logging(debug=args.debug)
- inventory_args = dict(
- refresh=args.refresh,
- config_files=config_files,
- private=args.private,
- )
- if hasattr(shade.inventory.OpenStackInventory, 'extra_config'):
- inventory_args.update(dict(
- config_key='ansible',
- config_defaults={
- 'use_hostnames': False,
- 'expand_hostvars': True,
- 'fail_on_errors': True,
- }
- ))
-
- inventory = shade.inventory.OpenStackInventory(**inventory_args)
-
- if args.list:
- output = get_host_groups(inventory, refresh=args.refresh)
- elif args.host:
- output = to_json(inventory.get_host(args.host))
- print(output)
- except shade.OpenStackCloudException as e:
- sys.stderr.write('%s\n' % e.message)
- sys.exit(1)
- sys.exit(0)
-
-
-if __name__ == '__main__':
- main()