1
0
Fork 0
mirror of https://github.com/ansible-collections/community.general.git synced 2024-09-14 20:13:21 +02:00
community.general/plugins/modules/cloud/rackspace/rax.py
Felix Fontein d119905bd5
Fix non-matching defaults. (#5452) (#5454)
(cherry picked from commit f84a9bf932)
2022-11-01 20:09:31 +01:00

896 lines
32 KiB
Python

#!/usr/bin/python
# -*- coding: utf-8 -*-
# Copyright: Ansible Project
# GNU General Public License v3.0+ (see COPYING or https://www.gnu.org/licenses/gpl-3.0.txt)
from __future__ import absolute_import, division, print_function
__metaclass__ = type
DOCUMENTATION = '''
---
module: rax
short_description: create / delete an instance in Rackspace Public Cloud
description:
- creates / deletes a Rackspace Public Cloud instance and optionally
waits for it to be 'running'.
options:
auto_increment:
description:
- Whether or not to increment a single number with the name of the
created servers. Only applicable when used with the I(group) attribute
or meta key.
type: bool
default: 'yes'
boot_from_volume:
description:
- Whether or not to boot the instance from a Cloud Block Storage volume.
If C(yes) and I(image) is specified a new volume will be created at
boot time. I(boot_volume_size) is required with I(image) to create a
new volume at boot time.
type: bool
default: 'no'
boot_volume:
type: str
description:
- Cloud Block Storage ID or Name to use as the boot volume of the
instance
boot_volume_size:
type: int
description:
- Size of the volume to create in Gigabytes. This is only required with
I(image) and I(boot_from_volume).
default: 100
boot_volume_terminate:
description:
- Whether the I(boot_volume) or newly created volume from I(image) will
be terminated when the server is terminated
type: bool
default: 'no'
config_drive:
description:
- Attach read-only configuration drive to server as label config-2
type: bool
default: 'no'
count:
type: int
description:
- number of instances to launch
default: 1
count_offset:
type: int
description:
- number count to start at
default: 1
disk_config:
type: str
description:
- Disk partitioning strategy
- If not specified it will assume the value C(auto).
choices:
- auto
- manual
exact_count:
description:
- Explicitly ensure an exact count of instances, used with
state=active/present. If specified as C(yes) and I(count) is less than
the servers matched, servers will be deleted to match the count. If
the number of matched servers is fewer than specified in I(count)
additional servers will be added.
type: bool
default: 'no'
extra_client_args:
type: dict
default: {}
description:
- A hash of key/value pairs to be used when creating the cloudservers
client. This is considered an advanced option, use it wisely and
with caution.
extra_create_args:
type: dict
default: {}
description:
- A hash of key/value pairs to be used when creating a new server.
This is considered an advanced option, use it wisely and with caution.
files:
type: dict
default: {}
description:
- Files to insert into the instance. remotefilename:localcontent
flavor:
type: str
description:
- flavor to use for the instance
group:
type: str
description:
- host group to assign to server, is also used for idempotent operations
to ensure a specific number of instances
image:
type: str
description:
- image to use for the instance. Can be an C(id), C(human_id) or C(name).
With I(boot_from_volume), a Cloud Block Storage volume will be created
with this image
instance_ids:
type: list
elements: str
description:
- list of instance ids, currently only used when state='absent' to
remove instances
key_name:
type: str
description:
- key pair to use on the instance
aliases:
- keypair
meta:
type: dict
default: {}
description:
- A hash of metadata to associate with the instance
name:
type: str
description:
- Name to give the instance
networks:
type: list
elements: str
description:
- The network to attach to the instances. If specified, you must include
ALL networks including the public and private interfaces. Can be C(id)
or C(label).
default:
- public
- private
state:
type: str
description:
- Indicate desired state of the resource
choices:
- present
- absent
default: present
user_data:
type: str
description:
- Data to be uploaded to the servers config drive. This option implies
I(config_drive). Can be a file path or a string
wait:
description:
- wait for the instance to be in state 'running' before returning
type: bool
default: 'no'
wait_timeout:
type: int
description:
- how long before wait gives up, in seconds
default: 300
author:
- "Jesse Keating (@omgjlk)"
- "Matt Martz (@sivel)"
notes:
- I(exact_count) can be "destructive" if the number of running servers in
the I(group) is larger than that specified in I(count). In such a case, the
I(state) is effectively set to C(absent) and the extra servers are deleted.
In the case of deletion, the returned data structure will have C(action)
set to C(delete), and the oldest servers in the group will be deleted.
extends_documentation_fragment:
- community.general.rackspace.openstack
'''
EXAMPLES = '''
- name: Build a Cloud Server
gather_facts: False
tasks:
- name: Server build request
local_action:
module: rax
credentials: ~/.raxpub
name: rax-test1
flavor: 5
image: b11d9567-e412-4255-96b9-bd63ab23bcfe
key_name: my_rackspace_key
files:
/root/test.txt: /home/localuser/test.txt
wait: yes
state: present
networks:
- private
- public
register: rax
- name: Build an exact count of cloud servers with incremented names
hosts: local
gather_facts: False
tasks:
- name: Server build requests
local_action:
module: rax
credentials: ~/.raxpub
name: test%03d.example.org
flavor: performance1-1
image: ubuntu-1204-lts-precise-pangolin
state: present
count: 10
count_offset: 10
exact_count: yes
group: test
wait: yes
register: rax
'''
import json
import os
import re
import time
try:
import pyrax
HAS_PYRAX = True
except ImportError:
HAS_PYRAX = False
from ansible.module_utils.basic import AnsibleModule
from ansible_collections.community.general.plugins.module_utils.rax import (FINAL_STATUSES, rax_argument_spec, rax_find_bootable_volume,
rax_find_image, rax_find_network, rax_find_volume,
rax_required_together, rax_to_dict, setup_rax_module)
from ansible.module_utils.six.moves import xrange
from ansible.module_utils.six import string_types
def rax_find_server_image(module, server, image, boot_volume):
if not image and boot_volume:
vol = rax_find_bootable_volume(module, pyrax, server,
exit=False)
if not vol:
return None
volume_image_metadata = vol.volume_image_metadata
vol_image_id = volume_image_metadata.get('image_id')
if vol_image_id:
server_image = rax_find_image(module, pyrax,
vol_image_id, exit=False)
if server_image:
server.image = dict(id=server_image)
# Match image IDs taking care of boot from volume
if image and not server.image:
vol = rax_find_bootable_volume(module, pyrax, server)
volume_image_metadata = vol.volume_image_metadata
vol_image_id = volume_image_metadata.get('image_id')
if not vol_image_id:
return None
server_image = rax_find_image(module, pyrax,
vol_image_id, exit=False)
if image != server_image:
return None
server.image = dict(id=server_image)
elif image and server.image['id'] != image:
return None
return server.image
def create(module, names=None, flavor=None, image=None, meta=None, key_name=None,
files=None, wait=True, wait_timeout=300, disk_config=None,
group=None, nics=None, extra_create_args=None, user_data=None,
config_drive=False, existing=None, block_device_mapping_v2=None):
names = [] if names is None else names
meta = {} if meta is None else meta
files = {} if files is None else files
nics = [] if nics is None else nics
extra_create_args = {} if extra_create_args is None else extra_create_args
existing = [] if existing is None else existing
block_device_mapping_v2 = [] if block_device_mapping_v2 is None else block_device_mapping_v2
cs = pyrax.cloudservers
changed = False
if user_data:
config_drive = True
if user_data and os.path.isfile(os.path.expanduser(user_data)):
try:
user_data = os.path.expanduser(user_data)
f = open(user_data)
user_data = f.read()
f.close()
except Exception as e:
module.fail_json(msg='Failed to load %s' % user_data)
# Handle the file contents
for rpath in files.keys():
lpath = os.path.expanduser(files[rpath])
try:
fileobj = open(lpath, 'r')
files[rpath] = fileobj.read()
fileobj.close()
except Exception as e:
module.fail_json(msg='Failed to load %s' % lpath)
try:
servers = []
bdmv2 = block_device_mapping_v2
for name in names:
servers.append(cs.servers.create(name=name, image=image,
flavor=flavor, meta=meta,
key_name=key_name,
files=files, nics=nics,
disk_config=disk_config,
config_drive=config_drive,
userdata=user_data,
block_device_mapping_v2=bdmv2,
**extra_create_args))
except Exception as e:
if e.message:
msg = str(e.message)
else:
msg = repr(e)
module.fail_json(msg=msg)
else:
changed = True
if wait:
end_time = time.time() + wait_timeout
infinite = wait_timeout == 0
while infinite or time.time() < end_time:
for server in servers:
try:
server.get()
except Exception:
server.status = 'ERROR'
if not filter(lambda s: s.status not in FINAL_STATUSES,
servers):
break
time.sleep(5)
success = []
error = []
timeout = []
for server in servers:
try:
server.get()
except Exception:
server.status = 'ERROR'
instance = rax_to_dict(server, 'server')
if server.status == 'ACTIVE' or not wait:
success.append(instance)
elif server.status == 'ERROR':
error.append(instance)
elif wait:
timeout.append(instance)
untouched = [rax_to_dict(s, 'server') for s in existing]
instances = success + untouched
results = {
'changed': changed,
'action': 'create',
'instances': instances,
'success': success,
'error': error,
'timeout': timeout,
'instance_ids': {
'instances': [i['id'] for i in instances],
'success': [i['id'] for i in success],
'error': [i['id'] for i in error],
'timeout': [i['id'] for i in timeout]
}
}
if timeout:
results['msg'] = 'Timeout waiting for all servers to build'
elif error:
results['msg'] = 'Failed to build all servers'
if 'msg' in results:
module.fail_json(**results)
else:
module.exit_json(**results)
def delete(module, instance_ids=None, wait=True, wait_timeout=300, kept=None):
instance_ids = [] if instance_ids is None else instance_ids
kept = [] if kept is None else kept
cs = pyrax.cloudservers
changed = False
instances = {}
servers = []
for instance_id in instance_ids:
servers.append(cs.servers.get(instance_id))
for server in servers:
try:
server.delete()
except Exception as e:
module.fail_json(msg=e.message)
else:
changed = True
instance = rax_to_dict(server, 'server')
instances[instance['id']] = instance
# If requested, wait for server deletion
if wait:
end_time = time.time() + wait_timeout
infinite = wait_timeout == 0
while infinite or time.time() < end_time:
for server in servers:
instance_id = server.id
try:
server.get()
except Exception:
instances[instance_id]['status'] = 'DELETED'
instances[instance_id]['rax_status'] = 'DELETED'
if not filter(lambda s: s['status'] not in ('', 'DELETED',
'ERROR'),
instances.values()):
break
time.sleep(5)
timeout = filter(lambda s: s['status'] not in ('', 'DELETED', 'ERROR'),
instances.values())
error = filter(lambda s: s['status'] in ('ERROR'),
instances.values())
success = filter(lambda s: s['status'] in ('', 'DELETED'),
instances.values())
instances = [rax_to_dict(s, 'server') for s in kept]
results = {
'changed': changed,
'action': 'delete',
'instances': instances,
'success': success,
'error': error,
'timeout': timeout,
'instance_ids': {
'instances': [i['id'] for i in instances],
'success': [i['id'] for i in success],
'error': [i['id'] for i in error],
'timeout': [i['id'] for i in timeout]
}
}
if timeout:
results['msg'] = 'Timeout waiting for all servers to delete'
elif error:
results['msg'] = 'Failed to delete all servers'
if 'msg' in results:
module.fail_json(**results)
else:
module.exit_json(**results)
def cloudservers(module, state=None, name=None, flavor=None, image=None,
meta=None, key_name=None, files=None, wait=True, wait_timeout=300,
disk_config=None, count=1, group=None, instance_ids=None,
exact_count=False, networks=None, count_offset=0,
auto_increment=False, extra_create_args=None, user_data=None,
config_drive=False, boot_from_volume=False,
boot_volume=None, boot_volume_size=None,
boot_volume_terminate=False):
meta = {} if meta is None else meta
files = {} if files is None else files
instance_ids = [] if instance_ids is None else instance_ids
networks = [] if networks is None else networks
extra_create_args = {} if extra_create_args is None else extra_create_args
cs = pyrax.cloudservers
cnw = pyrax.cloud_networks
if not cnw:
module.fail_json(msg='Failed to instantiate client. This '
'typically indicates an invalid region or an '
'incorrectly capitalized region name.')
if state == 'present' or (state == 'absent' and instance_ids is None):
if not boot_from_volume and not boot_volume and not image:
module.fail_json(msg='image is required for the "rax" module')
for arg, value in dict(name=name, flavor=flavor).items():
if not value:
module.fail_json(msg='%s is required for the "rax" module' %
arg)
if boot_from_volume and not image and not boot_volume:
module.fail_json(msg='image or boot_volume are required for the '
'"rax" with boot_from_volume')
if boot_from_volume and image and not boot_volume_size:
module.fail_json(msg='boot_volume_size is required for the "rax" '
'module with boot_from_volume and image')
if boot_from_volume and image and boot_volume:
image = None
servers = []
# Add the group meta key
if group and 'group' not in meta:
meta['group'] = group
elif 'group' in meta and group is None:
group = meta['group']
# Normalize and ensure all metadata values are strings
for k, v in meta.items():
if isinstance(v, list):
meta[k] = ','.join(['%s' % i for i in v])
elif isinstance(v, dict):
meta[k] = json.dumps(v)
elif not isinstance(v, string_types):
meta[k] = '%s' % v
# When using state=absent with group, the absent block won't match the
# names properly. Use the exact_count functionality to decrease the count
# to the desired level
was_absent = False
if group is not None and state == 'absent':
exact_count = True
state = 'present'
was_absent = True
if image:
image = rax_find_image(module, pyrax, image)
nics = []
if networks:
for network in networks:
nics.extend(rax_find_network(module, pyrax, network))
# act on the state
if state == 'present':
# Idempotent ensurance of a specific count of servers
if exact_count is not False:
# See if we can find servers that match our options
if group is None:
module.fail_json(msg='"group" must be provided when using '
'"exact_count"')
if auto_increment:
numbers = set()
# See if the name is a printf like string, if not append
# %d to the end
try:
name % 0
except TypeError as e:
if e.message.startswith('not all'):
name = '%s%%d' % name
else:
module.fail_json(msg=e.message)
# regex pattern to match printf formatting
pattern = re.sub(r'%\d*[sd]', r'(\d+)', name)
for server in cs.servers.list():
# Ignore DELETED servers
if server.status == 'DELETED':
continue
if server.metadata.get('group') == group:
servers.append(server)
match = re.search(pattern, server.name)
if match:
number = int(match.group(1))
numbers.add(number)
number_range = xrange(count_offset, count_offset + count)
available_numbers = list(set(number_range)
.difference(numbers))
else: # Not auto incrementing
for server in cs.servers.list():
# Ignore DELETED servers
if server.status == 'DELETED':
continue
if server.metadata.get('group') == group:
servers.append(server)
# available_numbers not needed here, we inspect auto_increment
# again later
# If state was absent but the count was changed,
# assume we only wanted to remove that number of instances
if was_absent:
diff = len(servers) - count
if diff < 0:
count = 0
else:
count = diff
if len(servers) > count:
# We have more servers than we need, set state='absent'
# and delete the extras, this should delete the oldest
state = 'absent'
kept = servers[:count]
del servers[:count]
instance_ids = []
for server in servers:
instance_ids.append(server.id)
delete(module, instance_ids=instance_ids, wait=wait,
wait_timeout=wait_timeout, kept=kept)
elif len(servers) < count:
# we have fewer servers than we need
if auto_increment:
# auto incrementing server numbers
names = []
name_slice = count - len(servers)
numbers_to_use = available_numbers[:name_slice]
for number in numbers_to_use:
names.append(name % number)
else:
# We are not auto incrementing server numbers,
# create a list of 'name' that matches how many we need
names = [name] * (count - len(servers))
else:
# we have the right number of servers, just return info
# about all of the matched servers
instances = []
instance_ids = []
for server in servers:
instances.append(rax_to_dict(server, 'server'))
instance_ids.append(server.id)
module.exit_json(changed=False, action=None,
instances=instances,
success=[], error=[], timeout=[],
instance_ids={'instances': instance_ids,
'success': [], 'error': [],
'timeout': []})
else: # not called with exact_count=True
if group is not None:
if auto_increment:
# we are auto incrementing server numbers, but not with
# exact_count
numbers = set()
# See if the name is a printf like string, if not append
# %d to the end
try:
name % 0
except TypeError as e:
if e.message.startswith('not all'):
name = '%s%%d' % name
else:
module.fail_json(msg=e.message)
# regex pattern to match printf formatting
pattern = re.sub(r'%\d*[sd]', r'(\d+)', name)
for server in cs.servers.list():
# Ignore DELETED servers
if server.status == 'DELETED':
continue
if server.metadata.get('group') == group:
servers.append(server)
match = re.search(pattern, server.name)
if match:
number = int(match.group(1))
numbers.add(number)
number_range = xrange(count_offset,
count_offset + count + len(numbers))
available_numbers = list(set(number_range)
.difference(numbers))
names = []
numbers_to_use = available_numbers[:count]
for number in numbers_to_use:
names.append(name % number)
else:
# Not auto incrementing
names = [name] * count
else:
# No group was specified, and not using exact_count
# Perform more simplistic matching
search_opts = {
'name': '^%s$' % name,
'flavor': flavor
}
servers = []
for server in cs.servers.list(search_opts=search_opts):
# Ignore DELETED servers
if server.status == 'DELETED':
continue
if not rax_find_server_image(module, server, image,
boot_volume):
continue
# Ignore servers with non matching metadata
if server.metadata != meta:
continue
servers.append(server)
if len(servers) >= count:
# We have more servers than were requested, don't do
# anything. Not running with exact_count=True, so we assume
# more is OK
instances = []
for server in servers:
instances.append(rax_to_dict(server, 'server'))
instance_ids = [i['id'] for i in instances]
module.exit_json(changed=False, action=None,
instances=instances, success=[], error=[],
timeout=[],
instance_ids={'instances': instance_ids,
'success': [], 'error': [],
'timeout': []})
# We need more servers to reach out target, create names for
# them, we aren't performing auto_increment here
names = [name] * (count - len(servers))
block_device_mapping_v2 = []
if boot_from_volume:
mapping = {
'boot_index': '0',
'delete_on_termination': boot_volume_terminate,
'destination_type': 'volume',
}
if image:
mapping.update({
'uuid': image,
'source_type': 'image',
'volume_size': boot_volume_size,
})
image = None
elif boot_volume:
volume = rax_find_volume(module, pyrax, boot_volume)
mapping.update({
'uuid': pyrax.utils.get_id(volume),
'source_type': 'volume',
})
block_device_mapping_v2.append(mapping)
create(module, names=names, flavor=flavor, image=image,
meta=meta, key_name=key_name, files=files, wait=wait,
wait_timeout=wait_timeout, disk_config=disk_config, group=group,
nics=nics, extra_create_args=extra_create_args,
user_data=user_data, config_drive=config_drive,
existing=servers,
block_device_mapping_v2=block_device_mapping_v2)
elif state == 'absent':
if instance_ids is None:
# We weren't given an explicit list of server IDs to delete
# Let's match instead
search_opts = {
'name': '^%s$' % name,
'flavor': flavor
}
for server in cs.servers.list(search_opts=search_opts):
# Ignore DELETED servers
if server.status == 'DELETED':
continue
if not rax_find_server_image(module, server, image,
boot_volume):
continue
# Ignore servers with non matching metadata
if meta != server.metadata:
continue
servers.append(server)
# Build a list of server IDs to delete
instance_ids = []
for server in servers:
if len(instance_ids) < count:
instance_ids.append(server.id)
else:
break
if not instance_ids:
# No server IDs were matched for deletion, or no IDs were
# explicitly provided, just exit and don't do anything
module.exit_json(changed=False, action=None, instances=[],
success=[], error=[], timeout=[],
instance_ids={'instances': [],
'success': [], 'error': [],
'timeout': []})
delete(module, instance_ids=instance_ids, wait=wait,
wait_timeout=wait_timeout)
def main():
argument_spec = rax_argument_spec()
argument_spec.update(
dict(
auto_increment=dict(default=True, type='bool'),
boot_from_volume=dict(default=False, type='bool'),
boot_volume=dict(type='str'),
boot_volume_size=dict(type='int', default=100),
boot_volume_terminate=dict(type='bool', default=False),
config_drive=dict(default=False, type='bool'),
count=dict(default=1, type='int'),
count_offset=dict(default=1, type='int'),
disk_config=dict(choices=['auto', 'manual']),
exact_count=dict(default=False, type='bool'),
extra_client_args=dict(type='dict', default={}),
extra_create_args=dict(type='dict', default={}),
files=dict(type='dict', default={}),
flavor=dict(),
group=dict(),
image=dict(),
instance_ids=dict(type='list', elements='str'),
key_name=dict(aliases=['keypair']),
meta=dict(type='dict', default={}),
name=dict(),
networks=dict(type='list', elements='str', default=['public', 'private']),
state=dict(default='present', choices=['present', 'absent']),
user_data=dict(no_log=True),
wait=dict(default=False, type='bool'),
wait_timeout=dict(default=300, type='int'),
)
)
module = AnsibleModule(
argument_spec=argument_spec,
required_together=rax_required_together(),
)
if not HAS_PYRAX:
module.fail_json(msg='pyrax is required for this module')
auto_increment = module.params.get('auto_increment')
boot_from_volume = module.params.get('boot_from_volume')
boot_volume = module.params.get('boot_volume')
boot_volume_size = module.params.get('boot_volume_size')
boot_volume_terminate = module.params.get('boot_volume_terminate')
config_drive = module.params.get('config_drive')
count = module.params.get('count')
count_offset = module.params.get('count_offset')
disk_config = module.params.get('disk_config')
if disk_config:
disk_config = disk_config.upper()
exact_count = module.params.get('exact_count', False)
extra_client_args = module.params.get('extra_client_args')
extra_create_args = module.params.get('extra_create_args')
files = module.params.get('files')
flavor = module.params.get('flavor')
group = module.params.get('group')
image = module.params.get('image')
instance_ids = module.params.get('instance_ids')
key_name = module.params.get('key_name')
meta = module.params.get('meta')
name = module.params.get('name')
networks = module.params.get('networks')
state = module.params.get('state')
user_data = module.params.get('user_data')
wait = module.params.get('wait')
wait_timeout = int(module.params.get('wait_timeout'))
setup_rax_module(module, pyrax)
if extra_client_args:
pyrax.cloudservers = pyrax.connect_to_cloudservers(
region=pyrax.cloudservers.client.region_name,
**extra_client_args)
client = pyrax.cloudservers.client
if 'bypass_url' in extra_client_args:
client.management_url = extra_client_args['bypass_url']
if pyrax.cloudservers is None:
module.fail_json(msg='Failed to instantiate client. This '
'typically indicates an invalid region or an '
'incorrectly capitalized region name.')
cloudservers(module, state=state, name=name, flavor=flavor,
image=image, meta=meta, key_name=key_name, files=files,
wait=wait, wait_timeout=wait_timeout, disk_config=disk_config,
count=count, group=group, instance_ids=instance_ids,
exact_count=exact_count, networks=networks,
count_offset=count_offset, auto_increment=auto_increment,
extra_create_args=extra_create_args, user_data=user_data,
config_drive=config_drive, boot_from_volume=boot_from_volume,
boot_volume=boot_volume, boot_volume_size=boot_volume_size,
boot_volume_terminate=boot_volume_terminate)
if __name__ == '__main__':
main()