mirror of
https://github.com/trailofbits/algo
synced 2024-11-13 19:12:06 +00:00
6b33d09d9f
* Scaleway modules * Update docs
620 lines
20 KiB
Python
620 lines
20 KiB
Python
#!/usr/bin/python
|
|
#
|
|
# Scaleway Compute management module
|
|
#
|
|
# Copyright (C) 2018 Online SAS.
|
|
# https://www.scaleway.com
|
|
#
|
|
# 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
|
|
|
|
ANSIBLE_METADATA = {
|
|
'metadata_version': '1.1',
|
|
'status': ['preview'],
|
|
'supported_by': 'community'
|
|
}
|
|
|
|
DOCUMENTATION = '''
|
|
---
|
|
module: scaleway_compute
|
|
short_description: Scaleway compute management module
|
|
version_added: "2.6"
|
|
author: Remy Leone (@sieben)
|
|
description:
|
|
- "This module manages compute instances on Scaleway."
|
|
extends_documentation_fragment: scaleway
|
|
|
|
options:
|
|
|
|
enable_ipv6:
|
|
description:
|
|
- Enable public IPv6 connectivity on the instance
|
|
default: false
|
|
type: bool
|
|
|
|
boot_type:
|
|
description:
|
|
- Boot method
|
|
default: bootscript
|
|
choices:
|
|
- bootscript
|
|
- local
|
|
|
|
image:
|
|
description:
|
|
- Image identifier used to start the instance with
|
|
required: true
|
|
|
|
name:
|
|
description:
|
|
- Name of the instance
|
|
|
|
organization:
|
|
description:
|
|
- Organization identifier
|
|
required: true
|
|
|
|
state:
|
|
description:
|
|
- Indicate desired state of the instance.
|
|
default: present
|
|
choices:
|
|
- present
|
|
- absent
|
|
- running
|
|
- restarted
|
|
- stopped
|
|
|
|
tags:
|
|
description:
|
|
- List of tags to apply to the instance (5 max)
|
|
required: false
|
|
default: []
|
|
|
|
region:
|
|
description:
|
|
- Scaleway compute zone
|
|
required: true
|
|
choices:
|
|
- ams1
|
|
- EMEA-NL-EVS
|
|
- par1
|
|
- EMEA-FR-PAR1
|
|
|
|
commercial_type:
|
|
description:
|
|
- Commercial name of the compute node
|
|
required: true
|
|
choices:
|
|
- ARM64-2GB
|
|
- ARM64-4GB
|
|
- ARM64-8GB
|
|
- ARM64-16GB
|
|
- ARM64-32GB
|
|
- ARM64-64GB
|
|
- ARM64-128GB
|
|
- C1
|
|
- C2S
|
|
- C2M
|
|
- C2L
|
|
- START1-XS
|
|
- START1-S
|
|
- START1-M
|
|
- START1-L
|
|
- X64-15GB
|
|
- X64-30GB
|
|
- X64-60GB
|
|
- X64-120GB
|
|
|
|
wait:
|
|
description:
|
|
- Wait for the instance to reach its desired state before returning.
|
|
type: bool
|
|
default: 'no'
|
|
|
|
wait_timeout:
|
|
description:
|
|
- Time to wait for the server to reach the expected state
|
|
required: false
|
|
default: 300
|
|
|
|
wait_sleep_time:
|
|
description:
|
|
- Time to wait before every attempt to check the state of the server
|
|
required: false
|
|
default: 3
|
|
'''
|
|
|
|
EXAMPLES = '''
|
|
- name: Create a server
|
|
scaleway_compute:
|
|
name: foobar
|
|
state: present
|
|
image: 89ee4018-f8c3-4dc4-a6b5-bca14f985ebe
|
|
organization: 951df375-e094-4d26-97c1-ba548eeb9c42
|
|
region: ams1
|
|
commercial_type: VC1S
|
|
tags:
|
|
- test
|
|
- www
|
|
|
|
- name: Destroy it right after
|
|
scaleway_compute:
|
|
name: foobar
|
|
state: absent
|
|
image: 89ee4018-f8c3-4dc4-a6b5-bca14f985ebe
|
|
organization: 951df375-e094-4d26-97c1-ba548eeb9c42
|
|
region: ams1
|
|
commercial_type: VC1S
|
|
'''
|
|
|
|
RETURN = '''
|
|
'''
|
|
|
|
import datetime
|
|
import time
|
|
|
|
from ansible.module_utils.basic import AnsibleModule
|
|
from ansible.module_utils.six.moves.urllib.parse import quote as urlquote
|
|
from ansible.module_utils.scaleway import SCALEWAY_LOCATION, scaleway_argument_spec, Scaleway
|
|
|
|
SCALEWAY_COMMERCIAL_TYPES = [
|
|
|
|
# Virtual ARM64 compute instance
|
|
'ARM64-2GB',
|
|
'ARM64-4GB',
|
|
'ARM64-8GB',
|
|
'ARM64-16GB',
|
|
'ARM64-32GB',
|
|
'ARM64-64GB',
|
|
'ARM64-128GB',
|
|
|
|
# Baremetal
|
|
'C1', # ARM64 (4 cores) - 2GB
|
|
'C2S', # X86-64 (4 cores) - 8GB
|
|
'C2M', # X86-64 (8 cores) - 16GB
|
|
'C2L', # x86-64 (8 cores) - 32 GB
|
|
|
|
# Virtual X86-64 compute instance
|
|
'START1-XS', # Starter X86-64 (1 core) - 1GB - 25 GB NVMe
|
|
'START1-S', # Starter X86-64 (2 cores) - 2GB - 50 GB NVMe
|
|
'START1-M', # Starter X86-64 (4 cores) - 4GB - 100 GB NVMe
|
|
'START1-L', # Starter X86-64 (8 cores) - 8GB - 200 GB NVMe
|
|
'X64-15GB',
|
|
'X64-30GB',
|
|
'X64-60GB',
|
|
'X64-120GB',
|
|
]
|
|
|
|
SCALEWAY_SERVER_STATES = (
|
|
'stopped',
|
|
'stopping',
|
|
'starting',
|
|
'running',
|
|
'locked'
|
|
)
|
|
|
|
SCALEWAY_TRANSITIONS_STATES = (
|
|
"stopping",
|
|
"starting",
|
|
"pending"
|
|
)
|
|
|
|
|
|
def fetch_state(compute_api, server):
|
|
compute_api.module.debug("fetch_state of server: %s" % server["id"])
|
|
response = compute_api.get(path="servers/%s" % server["id"])
|
|
|
|
if response.status_code == 404:
|
|
return "absent"
|
|
|
|
if not response.ok:
|
|
msg = 'Error during state fetching: (%s) %s' % (response.status_code, response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
try:
|
|
compute_api.module.debug("Server %s in state: %s" % (server["id"], response.json["server"]["state"]))
|
|
return response.json["server"]["state"]
|
|
except KeyError:
|
|
compute_api.module.fail_json(msg="Could not fetch state in %s" % response.json)
|
|
|
|
|
|
def wait_to_complete_state_transition(compute_api, server):
|
|
wait = compute_api.module.params["wait"]
|
|
if not wait:
|
|
return
|
|
wait_timeout = compute_api.module.params["wait_timeout"]
|
|
wait_sleep_time = compute_api.module.params["wait_sleep_time"]
|
|
|
|
start = datetime.datetime.utcnow()
|
|
end = start + datetime.timedelta(seconds=wait_timeout)
|
|
while datetime.datetime.utcnow() < end:
|
|
compute_api.module.debug("We are going to wait for the server to finish its transition")
|
|
if fetch_state(compute_api, server) not in SCALEWAY_TRANSITIONS_STATES:
|
|
compute_api.module.debug("It seems that the server is not in transition anymore.")
|
|
compute_api.module.debug("Server in state: %s" % fetch_state(compute_api, server))
|
|
break
|
|
time.sleep(wait_sleep_time)
|
|
else:
|
|
compute_api.module.fail_json(msg="Server takes too long to finish its transition")
|
|
|
|
|
|
def create_server(compute_api, server):
|
|
compute_api.module.debug("Starting a create_server")
|
|
target_server = None
|
|
response = compute_api.post(path="servers",
|
|
data={"enable_ipv6": server["enable_ipv6"],
|
|
"boot_type": server["boot_type"],
|
|
"tags": server["tags"],
|
|
"commercial_type": server["commercial_type"],
|
|
"image": server["image"],
|
|
"name": server["name"],
|
|
"organization": server["organization"]})
|
|
|
|
if not response.ok:
|
|
msg = 'Error during server creation: (%s) %s' % (response.status_code, response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
try:
|
|
target_server = response.json["server"]
|
|
except KeyError:
|
|
compute_api.module.fail_json(msg="Error in getting the server information from: %s" % response.json)
|
|
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=target_server)
|
|
|
|
return target_server
|
|
|
|
|
|
def restart_server(compute_api, server):
|
|
return perform_action(compute_api=compute_api, server=server, action="reboot")
|
|
|
|
|
|
def stop_server(compute_api, server):
|
|
return perform_action(compute_api=compute_api, server=server, action="poweroff")
|
|
|
|
|
|
def start_server(compute_api, server):
|
|
return perform_action(compute_api=compute_api, server=server, action="poweron")
|
|
|
|
|
|
def perform_action(compute_api, server, action):
|
|
response = compute_api.post(path="servers/%s/action" % server["id"],
|
|
data={"action": action})
|
|
if not response.ok:
|
|
msg = 'Error during server %s: (%s) %s' % (action, response.status_code, response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=server)
|
|
|
|
return response
|
|
|
|
|
|
def remove_server(compute_api, server):
|
|
compute_api.module.debug("Starting remove server strategy")
|
|
response = compute_api.delete(path="servers/%s" % server["id"])
|
|
if not response.ok:
|
|
msg = 'Error during server deletion: (%s) %s' % (response.status_code, response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=server)
|
|
|
|
return response
|
|
|
|
|
|
def present_strategy(compute_api, wished_server):
|
|
compute_api.module.debug("Starting present strategy")
|
|
changed = False
|
|
query_results = find(compute_api=compute_api, wished_server=wished_server, per_page=1)
|
|
|
|
if not query_results:
|
|
changed = True
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "A server would be created."}
|
|
|
|
target_server = create_server(compute_api=compute_api, server=wished_server)
|
|
else:
|
|
target_server = query_results[0]
|
|
|
|
if server_attributes_should_be_changed(compute_api=compute_api, target_server=target_server,
|
|
wished_server=wished_server):
|
|
changed = True
|
|
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "Server %s attributes would be changed." % target_server["id"]}
|
|
|
|
server_change_attributes(compute_api=compute_api, target_server=target_server, wished_server=wished_server)
|
|
|
|
return changed, target_server
|
|
|
|
|
|
def absent_strategy(compute_api, wished_server):
|
|
compute_api.module.debug("Starting absent strategy")
|
|
changed = False
|
|
target_server = None
|
|
query_results = find(compute_api=compute_api, wished_server=wished_server, per_page=1)
|
|
|
|
if not query_results:
|
|
return changed, {"status": "Server already absent."}
|
|
else:
|
|
target_server = query_results[0]
|
|
|
|
changed = True
|
|
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "Server %s would be made absent." % target_server["id"]}
|
|
|
|
# A server MUST be stopped to be deleted.
|
|
while not fetch_state(compute_api=compute_api, server=target_server) == "stopped":
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=target_server)
|
|
response = stop_server(compute_api=compute_api, server=target_server)
|
|
|
|
if not response.ok:
|
|
err_msg = 'Error while stopping a server before removing it [{0}: {1}]'.format(response.status_code,
|
|
response.json)
|
|
compute_api.module.fail_json(msg=err_msg)
|
|
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=target_server)
|
|
|
|
response = remove_server(compute_api=compute_api, server=target_server)
|
|
|
|
if not response.ok:
|
|
err_msg = 'Error while removing server [{0}: {1}]'.format(response.status_code, response.json)
|
|
compute_api.module.fail_json(msg=err_msg)
|
|
|
|
return changed, {"status": "Server %s deleted" % target_server["id"]}
|
|
|
|
|
|
def running_strategy(compute_api, wished_server):
|
|
compute_api.module.debug("Starting running strategy")
|
|
changed = False
|
|
query_results = find(compute_api=compute_api, wished_server=wished_server, per_page=1)
|
|
|
|
if not query_results:
|
|
changed = True
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "A server would be created before being run."}
|
|
|
|
target_server = create_server(compute_api=compute_api, server=wished_server)
|
|
else:
|
|
target_server = query_results[0]
|
|
|
|
if server_attributes_should_be_changed(compute_api=compute_api, target_server=target_server,
|
|
wished_server=wished_server):
|
|
changed = True
|
|
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "Server %s attributes would be changed before running it." % target_server["id"]}
|
|
|
|
server_change_attributes(compute_api=compute_api, target_server=target_server, wished_server=wished_server)
|
|
|
|
current_state = fetch_state(compute_api=compute_api, server=target_server)
|
|
if current_state not in ("running", "starting"):
|
|
compute_api.module.debug("running_strategy: Server in state: %s" % current_state)
|
|
changed = True
|
|
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "Server %s attributes would be changed." % target_server["id"]}
|
|
|
|
response = start_server(compute_api=compute_api, server=target_server)
|
|
if not response.ok:
|
|
msg = 'Error while running server [{0}: {1}]'.format(response.status_code, response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
return changed, target_server
|
|
|
|
|
|
def stop_strategy(compute_api, wished_server):
|
|
compute_api.module.debug("Starting stop strategy")
|
|
query_results = find(compute_api=compute_api, wished_server=wished_server, per_page=1)
|
|
|
|
changed = False
|
|
|
|
if not query_results:
|
|
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "A server would be created before being stopped."}
|
|
|
|
target_server = create_server(compute_api=compute_api, server=wished_server)
|
|
changed = True
|
|
else:
|
|
target_server = query_results[0]
|
|
|
|
compute_api.module.debug("stop_strategy: Servers are found.")
|
|
|
|
if server_attributes_should_be_changed(compute_api=compute_api, target_server=target_server,
|
|
wished_server=wished_server):
|
|
changed = True
|
|
|
|
if compute_api.module.check_mode:
|
|
return changed, {
|
|
"status": "Server %s attributes would be changed before stopping it." % target_server["id"]}
|
|
|
|
server_change_attributes(compute_api=compute_api, target_server=target_server, wished_server=wished_server)
|
|
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=target_server)
|
|
|
|
current_state = fetch_state(compute_api=compute_api, server=target_server)
|
|
if current_state not in ("stopped",):
|
|
compute_api.module.debug("stop_strategy: Server in state: %s" % current_state)
|
|
|
|
changed = True
|
|
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "Server %s would be stopped." % target_server["id"]}
|
|
|
|
response = stop_server(compute_api=compute_api, server=target_server)
|
|
compute_api.module.debug(response.json)
|
|
compute_api.module.debug(response.ok)
|
|
|
|
if not response.ok:
|
|
msg = 'Error while stopping server [{0}: {1}]'.format(response.status_code, response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
return changed, target_server
|
|
|
|
|
|
def restart_strategy(compute_api, wished_server):
|
|
compute_api.module.debug("Starting restart strategy")
|
|
changed = False
|
|
query_results = find(compute_api=compute_api, wished_server=wished_server, per_page=1)
|
|
|
|
if not query_results:
|
|
changed = True
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "A server would be created before being rebooted."}
|
|
|
|
target_server = create_server(compute_api=compute_api, server=wished_server)
|
|
else:
|
|
target_server = query_results[0]
|
|
|
|
if server_attributes_should_be_changed(compute_api=compute_api,
|
|
target_server=target_server,
|
|
wished_server=wished_server):
|
|
changed = True
|
|
|
|
if compute_api.module.check_mode:
|
|
return changed, {
|
|
"status": "Server %s attributes would be changed before rebooting it." % target_server["id"]}
|
|
|
|
server_change_attributes(compute_api=compute_api, target_server=target_server, wished_server=wished_server)
|
|
|
|
changed = True
|
|
if compute_api.module.check_mode:
|
|
return changed, {"status": "Server %s would be rebooted." % target_server["id"]}
|
|
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=target_server)
|
|
|
|
if fetch_state(compute_api=compute_api, server=target_server) in ("running",):
|
|
response = restart_server(compute_api=compute_api, server=target_server)
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=target_server)
|
|
if not response.ok:
|
|
msg = 'Error while restarting server that was running [{0}: {1}].'.format(response.status_code,
|
|
response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
if fetch_state(compute_api=compute_api, server=target_server) in ("stopped",):
|
|
response = restart_server(compute_api=compute_api, server=target_server)
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=target_server)
|
|
if not response.ok:
|
|
msg = 'Error while restarting server that was stopped [{0}: {1}].'.format(response.status_code,
|
|
response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
return changed, target_server
|
|
|
|
|
|
state_strategy = {
|
|
"present": present_strategy,
|
|
"restarted": restart_strategy,
|
|
"stopped": stop_strategy,
|
|
"running": running_strategy,
|
|
"absent": absent_strategy
|
|
}
|
|
|
|
|
|
def find(compute_api, wished_server, per_page=1):
|
|
compute_api.module.debug("Getting inside find")
|
|
# Only the name attribute is accepted in the Compute query API
|
|
url = 'servers?name=%s&per_page=%d' % (urlquote(wished_server["name"]), per_page)
|
|
response = compute_api.get(url)
|
|
|
|
if not response.ok:
|
|
msg = 'Error during server search: (%s) %s' % (response.status_code, response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
search_results = response.json["servers"]
|
|
|
|
return search_results
|
|
|
|
|
|
PATCH_MUTABLE_SERVER_ATTRIBUTES = (
|
|
"ipv6",
|
|
"tags",
|
|
"name",
|
|
"dynamic_ip_required",
|
|
)
|
|
|
|
|
|
def server_attributes_should_be_changed(compute_api, target_server, wished_server):
|
|
compute_api.module.debug("Checking if server attributes should be changed")
|
|
compute_api.module.debug("Current Server: %s" % target_server)
|
|
compute_api.module.debug("Wished Server: %s" % wished_server)
|
|
debug_dict = dict((x, (target_server[x], wished_server[x]))
|
|
for x in PATCH_MUTABLE_SERVER_ATTRIBUTES
|
|
if x in target_server and x in wished_server)
|
|
compute_api.module.debug("Debug dict %s" % debug_dict)
|
|
|
|
try:
|
|
return any([target_server[x] != wished_server[x]
|
|
for x in PATCH_MUTABLE_SERVER_ATTRIBUTES
|
|
if x in target_server and x in wished_server])
|
|
except AttributeError:
|
|
compute_api.module.fail_json(msg="Error while checking if attributes should be changed")
|
|
|
|
|
|
def server_change_attributes(compute_api, target_server, wished_server):
|
|
compute_api.module.debug("Starting patching server attributes")
|
|
patch_payload = dict((x, wished_server[x])
|
|
for x in PATCH_MUTABLE_SERVER_ATTRIBUTES
|
|
if x in wished_server and x in target_server)
|
|
response = compute_api.patch(path="servers/%s" % target_server["id"],
|
|
data=patch_payload)
|
|
if not response.ok:
|
|
msg = 'Error during server attributes patching: (%s) %s' % (response.status_code, response.json)
|
|
compute_api.module.fail_json(msg=msg)
|
|
|
|
wait_to_complete_state_transition(compute_api=compute_api, server=target_server)
|
|
|
|
return response
|
|
|
|
|
|
def core(module):
|
|
region = module.params["region"]
|
|
wished_server = {
|
|
"state": module.params["state"],
|
|
"image": module.params["image"],
|
|
"name": module.params["name"],
|
|
"commercial_type": module.params["commercial_type"],
|
|
"enable_ipv6": module.params["enable_ipv6"],
|
|
"boot_type": module.params["boot_type"],
|
|
"tags": module.params["tags"],
|
|
"organization": module.params["organization"]
|
|
}
|
|
module.params['api_url'] = SCALEWAY_LOCATION[region]["api_endpoint"]
|
|
|
|
compute_api = Scaleway(module=module)
|
|
|
|
changed, summary = state_strategy[wished_server["state"]](compute_api=compute_api, wished_server=wished_server)
|
|
module.exit_json(changed=changed, msg=summary)
|
|
|
|
|
|
def main():
|
|
argument_spec = scaleway_argument_spec()
|
|
argument_spec.update(dict(
|
|
image=dict(required=True),
|
|
name=dict(),
|
|
region=dict(required=True, choices=SCALEWAY_LOCATION.keys()),
|
|
commercial_type=dict(required=True, choices=SCALEWAY_COMMERCIAL_TYPES),
|
|
enable_ipv6=dict(default=False, type="bool"),
|
|
boot_type=dict(default="bootscript"),
|
|
state=dict(choices=state_strategy.keys(), default='present'),
|
|
tags=dict(type="list", default=[]),
|
|
organization=dict(required=True),
|
|
wait=dict(type="bool", default=False),
|
|
wait_timeout=dict(type="int", default=300),
|
|
wait_sleep_time=dict(type="int", default=3),
|
|
))
|
|
module = AnsibleModule(
|
|
argument_spec=argument_spec,
|
|
supports_check_mode=True,
|
|
)
|
|
|
|
core(module)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|