Bug 1651 fix
[osm/N2VC.git] / n2vc / libjuju.py
1 # Copyright 2020 Canonical Ltd.
2 #
3 # Licensed under the Apache License, Version 2.0 (the "License");
4 # you may not use this file except in compliance with the License.
5 # You may obtain a copy of the License at
6 #
7 # http://www.apache.org/licenses/LICENSE-2.0
8 #
9 # Unless required by applicable law or agreed to in writing, software
10 # distributed under the License is distributed on an "AS IS" BASIS,
11 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 # See the License for the specific language governing permissions and
13 # limitations under the License.
14
15 import asyncio
16 import logging
17 import typing
18
19 import time
20
21 import juju.errors
22 from juju.model import Model
23 from juju.machine import Machine
24 from juju.application import Application
25 from juju.unit import Unit
26 from juju.client._definitions import (
27 FullStatus,
28 QueryApplicationOffersResults,
29 Cloud,
30 CloudCredential,
31 )
32 from juju.controller import Controller
33 from juju.client import client
34 from juju import tag
35
36 from n2vc.juju_watcher import JujuModelWatcher
37 from n2vc.provisioner import AsyncSSHProvisioner
38 from n2vc.n2vc_conn import N2VCConnector
39 from n2vc.exceptions import (
40 JujuMachineNotFound,
41 JujuApplicationNotFound,
42 JujuLeaderUnitNotFound,
43 JujuActionNotFound,
44 JujuControllerFailedConnecting,
45 JujuApplicationExists,
46 JujuInvalidK8sConfiguration,
47 JujuError,
48 )
49 from n2vc.vca.cloud import Cloud as VcaCloud
50 from n2vc.vca.connection import Connection
51 from kubernetes.client.configuration import Configuration
52 from retrying_async import retry
53
54
55 RBAC_LABEL_KEY_NAME = "rbac-id"
56
57
58 class Libjuju:
59 def __init__(
60 self,
61 vca_connection: Connection,
62 loop: asyncio.AbstractEventLoop = None,
63 log: logging.Logger = None,
64 n2vc: N2VCConnector = None,
65 ):
66 """
67 Constructor
68
69 :param: vca_connection: n2vc.vca.connection object
70 :param: loop: Asyncio loop
71 :param: log: Logger
72 :param: n2vc: N2VC object
73 """
74
75 self.log = log or logging.getLogger("Libjuju")
76 self.n2vc = n2vc
77 self.vca_connection = vca_connection
78
79 self.loop = loop or asyncio.get_event_loop()
80 self.loop.set_exception_handler(self.handle_exception)
81 self.creating_model = asyncio.Lock(loop=self.loop)
82
83 if self.vca_connection.is_default:
84 self.health_check_task = self._create_health_check_task()
85
86 def _create_health_check_task(self):
87 return self.loop.create_task(self.health_check())
88
89 async def get_controller(self, timeout: float = 60.0) -> Controller:
90 """
91 Get controller
92
93 :param: timeout: Time in seconds to wait for controller to connect
94 """
95 controller = None
96 try:
97 controller = Controller(loop=self.loop)
98 await asyncio.wait_for(
99 controller.connect(
100 endpoint=self.vca_connection.data.endpoints,
101 username=self.vca_connection.data.user,
102 password=self.vca_connection.data.secret,
103 cacert=self.vca_connection.data.cacert,
104 ),
105 timeout=timeout,
106 )
107 if self.vca_connection.is_default:
108 endpoints = await controller.api_endpoints
109 if not all(
110 endpoint in self.vca_connection.endpoints for endpoint in endpoints
111 ):
112 await self.vca_connection.update_endpoints(endpoints)
113 return controller
114 except asyncio.CancelledError as e:
115 raise e
116 except Exception as e:
117 self.log.error(
118 "Failed connecting to controller: {}... {}".format(
119 self.vca_connection.data.endpoints, e
120 )
121 )
122 if controller:
123 await self.disconnect_controller(controller)
124 raise JujuControllerFailedConnecting(e)
125
126 async def disconnect(self):
127 """Disconnect"""
128 # Cancel health check task
129 self.health_check_task.cancel()
130 self.log.debug("Libjuju disconnected!")
131
132 async def disconnect_model(self, model: Model):
133 """
134 Disconnect model
135
136 :param: model: Model that will be disconnected
137 """
138 await model.disconnect()
139
140 async def disconnect_controller(self, controller: Controller):
141 """
142 Disconnect controller
143
144 :param: controller: Controller that will be disconnected
145 """
146 if controller:
147 await controller.disconnect()
148
149 @retry(attempts=3, delay=5, timeout=None)
150 async def add_model(self, model_name: str, cloud: VcaCloud):
151 """
152 Create model
153
154 :param: model_name: Model name
155 :param: cloud: Cloud object
156 """
157
158 # Get controller
159 controller = await self.get_controller()
160 model = None
161 try:
162 # Block until other workers have finished model creation
163 while self.creating_model.locked():
164 await asyncio.sleep(0.1)
165
166 # Create the model
167 async with self.creating_model:
168 if await self.model_exists(model_name, controller=controller):
169 return
170 self.log.debug("Creating model {}".format(model_name))
171 model = await controller.add_model(
172 model_name,
173 config=self.vca_connection.data.model_config,
174 cloud_name=cloud.name,
175 credential_name=cloud.credential_name,
176 )
177 except juju.errors.JujuAPIError as e:
178 if "already exists" in e.message:
179 pass
180 else:
181 raise e
182 finally:
183 if model:
184 await self.disconnect_model(model)
185 await self.disconnect_controller(controller)
186
187 async def get_executed_actions(self, model_name: str) -> list:
188 """
189 Get executed/history of actions for a model.
190
191 :param: model_name: Model name, str.
192 :return: List of executed actions for a model.
193 """
194 model = None
195 executed_actions = []
196 controller = await self.get_controller()
197 try:
198 model = await self.get_model(controller, model_name)
199 # Get all unique action names
200 actions = {}
201 for application in model.applications:
202 application_actions = await self.get_actions(application, model_name)
203 actions.update(application_actions)
204 # Get status of all actions
205 for application_action in actions:
206 app_action_status_list = await model.get_action_status(
207 name=application_action
208 )
209 for action_id, action_status in app_action_status_list.items():
210 executed_action = {
211 "id": action_id,
212 "action": application_action,
213 "status": action_status,
214 }
215 # Get action output by id
216 action_status = await model.get_action_output(executed_action["id"])
217 for k, v in action_status.items():
218 executed_action[k] = v
219 executed_actions.append(executed_action)
220 except Exception as e:
221 raise JujuError(
222 "Error in getting executed actions for model: {}. Error: {}".format(
223 model_name, str(e)
224 )
225 )
226 finally:
227 if model:
228 await self.disconnect_model(model)
229 await self.disconnect_controller(controller)
230 return executed_actions
231
232 async def get_application_configs(
233 self, model_name: str, application_name: str
234 ) -> dict:
235 """
236 Get available configs for an application.
237
238 :param: model_name: Model name, str.
239 :param: application_name: Application name, str.
240
241 :return: A dict which has key - action name, value - action description
242 """
243 model = None
244 application_configs = {}
245 controller = await self.get_controller()
246 try:
247 model = await self.get_model(controller, model_name)
248 application = self._get_application(
249 model, application_name=application_name
250 )
251 application_configs = await application.get_config()
252 except Exception as e:
253 raise JujuError(
254 "Error in getting configs for application: {} in model: {}. Error: {}".format(
255 application_name, model_name, str(e)
256 )
257 )
258 finally:
259 if model:
260 await self.disconnect_model(model)
261 await self.disconnect_controller(controller)
262 return application_configs
263
264 @retry(attempts=3, delay=5)
265 async def get_model(self, controller: Controller, model_name: str) -> Model:
266 """
267 Get model from controller
268
269 :param: controller: Controller
270 :param: model_name: Model name
271
272 :return: Model: The created Juju model object
273 """
274 return await controller.get_model(model_name)
275
276 async def model_exists(
277 self, model_name: str, controller: Controller = None
278 ) -> bool:
279 """
280 Check if model exists
281
282 :param: controller: Controller
283 :param: model_name: Model name
284
285 :return bool
286 """
287 need_to_disconnect = False
288
289 # Get controller if not passed
290 if not controller:
291 controller = await self.get_controller()
292 need_to_disconnect = True
293
294 # Check if model exists
295 try:
296 return model_name in await controller.list_models()
297 finally:
298 if need_to_disconnect:
299 await self.disconnect_controller(controller)
300
301 async def models_exist(self, model_names: [str]) -> (bool, list):
302 """
303 Check if models exists
304
305 :param: model_names: List of strings with model names
306
307 :return (bool, list[str]): (True if all models exists, List of model names that don't exist)
308 """
309 if not model_names:
310 raise Exception(
311 "model_names must be a non-empty array. Given value: {}".format(
312 model_names
313 )
314 )
315 non_existing_models = []
316 models = await self.list_models()
317 existing_models = list(set(models).intersection(model_names))
318 non_existing_models = list(set(model_names) - set(existing_models))
319
320 return (
321 len(non_existing_models) == 0,
322 non_existing_models,
323 )
324
325 async def get_model_status(self, model_name: str) -> FullStatus:
326 """
327 Get model status
328
329 :param: model_name: Model name
330
331 :return: Full status object
332 """
333 controller = await self.get_controller()
334 model = await self.get_model(controller, model_name)
335 try:
336 return await model.get_status()
337 finally:
338 await self.disconnect_model(model)
339 await self.disconnect_controller(controller)
340
341 async def create_machine(
342 self,
343 model_name: str,
344 machine_id: str = None,
345 db_dict: dict = None,
346 progress_timeout: float = None,
347 total_timeout: float = None,
348 series: str = "bionic",
349 wait: bool = True,
350 ) -> (Machine, bool):
351 """
352 Create machine
353
354 :param: model_name: Model name
355 :param: machine_id: Machine id
356 :param: db_dict: Dictionary with data of the DB to write the updates
357 :param: progress_timeout: Maximum time between two updates in the model
358 :param: total_timeout: Timeout for the entity to be active
359 :param: series: Series of the machine (xenial, bionic, focal, ...)
360 :param: wait: Wait until machine is ready
361
362 :return: (juju.machine.Machine, bool): Machine object and a boolean saying
363 if the machine is new or it already existed
364 """
365 new = False
366 machine = None
367
368 self.log.debug(
369 "Creating machine (id={}) in model: {}".format(machine_id, model_name)
370 )
371
372 # Get controller
373 controller = await self.get_controller()
374
375 # Get model
376 model = await self.get_model(controller, model_name)
377 try:
378 if machine_id is not None:
379 self.log.debug(
380 "Searching machine (id={}) in model {}".format(
381 machine_id, model_name
382 )
383 )
384
385 # Get machines from model and get the machine with machine_id if exists
386 machines = await model.get_machines()
387 if machine_id in machines:
388 self.log.debug(
389 "Machine (id={}) found in model {}".format(
390 machine_id, model_name
391 )
392 )
393 machine = machines[machine_id]
394 else:
395 raise JujuMachineNotFound("Machine {} not found".format(machine_id))
396
397 if machine is None:
398 self.log.debug("Creating a new machine in model {}".format(model_name))
399
400 # Create machine
401 machine = await model.add_machine(
402 spec=None, constraints=None, disks=None, series=series
403 )
404 new = True
405
406 # Wait until the machine is ready
407 self.log.debug(
408 "Wait until machine {} is ready in model {}".format(
409 machine.entity_id, model_name
410 )
411 )
412 if wait:
413 await JujuModelWatcher.wait_for(
414 model=model,
415 entity=machine,
416 progress_timeout=progress_timeout,
417 total_timeout=total_timeout,
418 db_dict=db_dict,
419 n2vc=self.n2vc,
420 vca_id=self.vca_connection._vca_id,
421 )
422 finally:
423 await self.disconnect_model(model)
424 await self.disconnect_controller(controller)
425
426 self.log.debug(
427 "Machine {} ready at {} in model {}".format(
428 machine.entity_id, machine.dns_name, model_name
429 )
430 )
431 return machine, new
432
433 async def provision_machine(
434 self,
435 model_name: str,
436 hostname: str,
437 username: str,
438 private_key_path: str,
439 db_dict: dict = None,
440 progress_timeout: float = None,
441 total_timeout: float = None,
442 ) -> str:
443 """
444 Manually provisioning of a machine
445
446 :param: model_name: Model name
447 :param: hostname: IP to access the machine
448 :param: username: Username to login to the machine
449 :param: private_key_path: Local path for the private key
450 :param: db_dict: Dictionary with data of the DB to write the updates
451 :param: progress_timeout: Maximum time between two updates in the model
452 :param: total_timeout: Timeout for the entity to be active
453
454 :return: (Entity): Machine id
455 """
456 self.log.debug(
457 "Provisioning machine. model: {}, hostname: {}, username: {}".format(
458 model_name, hostname, username
459 )
460 )
461
462 # Get controller
463 controller = await self.get_controller()
464
465 # Get model
466 model = await self.get_model(controller, model_name)
467
468 try:
469 # Get provisioner
470 provisioner = AsyncSSHProvisioner(
471 host=hostname,
472 user=username,
473 private_key_path=private_key_path,
474 log=self.log,
475 )
476
477 # Provision machine
478 params = await provisioner.provision_machine()
479
480 params.jobs = ["JobHostUnits"]
481
482 self.log.debug("Adding machine to model")
483 connection = model.connection()
484 client_facade = client.ClientFacade.from_connection(connection)
485
486 results = await client_facade.AddMachines(params=[params])
487 error = results.machines[0].error
488
489 if error:
490 msg = "Error adding machine: {}".format(error.message)
491 self.log.error(msg=msg)
492 raise ValueError(msg)
493
494 machine_id = results.machines[0].machine
495
496 self.log.debug("Installing Juju agent into machine {}".format(machine_id))
497 asyncio.ensure_future(
498 provisioner.install_agent(
499 connection=connection,
500 nonce=params.nonce,
501 machine_id=machine_id,
502 proxy=self.vca_connection.data.api_proxy,
503 series=params.series,
504 )
505 )
506
507 machine = None
508 for _ in range(10):
509 machine_list = await model.get_machines()
510 if machine_id in machine_list:
511 self.log.debug("Machine {} found in model!".format(machine_id))
512 machine = model.machines.get(machine_id)
513 break
514 await asyncio.sleep(2)
515
516 if machine is None:
517 msg = "Machine {} not found in model".format(machine_id)
518 self.log.error(msg=msg)
519 raise JujuMachineNotFound(msg)
520
521 self.log.debug(
522 "Wait until machine {} is ready in model {}".format(
523 machine.entity_id, model_name
524 )
525 )
526 await JujuModelWatcher.wait_for(
527 model=model,
528 entity=machine,
529 progress_timeout=progress_timeout,
530 total_timeout=total_timeout,
531 db_dict=db_dict,
532 n2vc=self.n2vc,
533 vca_id=self.vca_connection._vca_id,
534 )
535 except Exception as e:
536 raise e
537 finally:
538 await self.disconnect_model(model)
539 await self.disconnect_controller(controller)
540
541 self.log.debug(
542 "Machine provisioned {} in model {}".format(machine_id, model_name)
543 )
544
545 return machine_id
546
547 async def deploy(
548 self, uri: str, model_name: str, wait: bool = True, timeout: float = 3600
549 ):
550 """
551 Deploy bundle or charm: Similar to the juju CLI command `juju deploy`
552
553 :param: uri: Path or Charm Store uri in which the charm or bundle can be found
554 :param: model_name: Model name
555 :param: wait: Indicates whether to wait or not until all applications are active
556 :param: timeout: Time in seconds to wait until all applications are active
557 """
558 controller = await self.get_controller()
559 model = await self.get_model(controller, model_name)
560 try:
561 await model.deploy(uri, trust=True)
562 if wait:
563 await JujuModelWatcher.wait_for_model(model, timeout=timeout)
564 self.log.debug("All units active in model {}".format(model_name))
565 finally:
566 await self.disconnect_model(model)
567 await self.disconnect_controller(controller)
568
569 async def add_unit(
570 self,
571 application_name: str,
572 model_name: str,
573 machine_id: str,
574 db_dict: dict = None,
575 progress_timeout: float = None,
576 total_timeout: float = None,
577 ):
578 """Add unit
579
580 :param: application_name: Application name
581 :param: model_name: Model name
582 :param: machine_id Machine id
583 :param: db_dict: Dictionary with data of the DB to write the updates
584 :param: progress_timeout: Maximum time between two updates in the model
585 :param: total_timeout: Timeout for the entity to be active
586
587 :return: None
588 """
589
590 model = None
591 controller = await self.get_controller()
592 try:
593 model = await self.get_model(controller, model_name)
594 application = self._get_application(model, application_name)
595
596 if application is not None:
597
598 # Checks if the given machine id in the model,
599 # otherwise function raises an error
600 _machine, _series = self._get_machine_info(model, machine_id)
601
602 self.log.debug(
603 "Adding unit (machine {}) to application {} in model ~{}".format(
604 machine_id, application_name, model_name
605 )
606 )
607
608 await application.add_unit(to=machine_id)
609
610 await JujuModelWatcher.wait_for(
611 model=model,
612 entity=application,
613 progress_timeout=progress_timeout,
614 total_timeout=total_timeout,
615 db_dict=db_dict,
616 n2vc=self.n2vc,
617 vca_id=self.vca_connection._vca_id,
618 )
619 self.log.debug(
620 "Unit is added to application {} in model {}".format(
621 application_name, model_name
622 )
623 )
624 else:
625 raise JujuApplicationNotFound(
626 "Application {} not exists".format(application_name)
627 )
628 finally:
629 if model:
630 await self.disconnect_model(model)
631 await self.disconnect_controller(controller)
632
633 async def destroy_unit(
634 self,
635 application_name: str,
636 model_name: str,
637 machine_id: str,
638 total_timeout: float = None,
639 ):
640 """Destroy unit
641
642 :param: application_name: Application name
643 :param: model_name: Model name
644 :param: machine_id Machine id
645 :param: total_timeout: Timeout for the entity to be active
646
647 :return: None
648 """
649
650 model = None
651 controller = await self.get_controller()
652 try:
653 model = await self.get_model(controller, model_name)
654 application = self._get_application(model, application_name)
655
656 if application is None:
657 raise JujuApplicationNotFound(
658 "Application not found: {} (model={})".format(
659 application_name, model_name
660 )
661 )
662
663 unit = self._get_unit(application, machine_id)
664 if not unit:
665 raise JujuError(
666 "A unit with machine id {} not in available units".format(
667 machine_id
668 )
669 )
670
671 unit_name = unit.name
672
673 self.log.debug(
674 "Destroying unit {} from application {} in model {}".format(
675 unit_name, application_name, model_name
676 )
677 )
678 await application.destroy_unit(unit_name)
679
680 self.log.debug(
681 "Waiting for unit {} to be destroyed in application {} (model={})...".format(
682 unit_name, application_name, model_name
683 )
684 )
685
686 # TODO: Add functionality in the Juju watcher to replace this kind of blocks
687 if total_timeout is None:
688 total_timeout = 3600
689 end = time.time() + total_timeout
690 while time.time() < end:
691 if not self._get_unit(application, machine_id):
692 self.log.debug(
693 "The unit {} was destroyed in application {} (model={}) ".format(
694 unit_name, application_name, model_name
695 )
696 )
697 return
698 await asyncio.sleep(5)
699 self.log.debug(
700 "Unit {} is destroyed from application {} in model {}".format(
701 unit_name, application_name, model_name
702 )
703 )
704 finally:
705 if model:
706 await self.disconnect_model(model)
707 await self.disconnect_controller(controller)
708
709 async def deploy_charm(
710 self,
711 application_name: str,
712 path: str,
713 model_name: str,
714 machine_id: str,
715 db_dict: dict = None,
716 progress_timeout: float = None,
717 total_timeout: float = None,
718 config: dict = None,
719 series: str = None,
720 num_units: int = 1,
721 ):
722 """Deploy charm
723
724 :param: application_name: Application name
725 :param: path: Local path to the charm
726 :param: model_name: Model name
727 :param: machine_id ID of the machine
728 :param: db_dict: Dictionary with data of the DB to write the updates
729 :param: progress_timeout: Maximum time between two updates in the model
730 :param: total_timeout: Timeout for the entity to be active
731 :param: config: Config for the charm
732 :param: series: Series of the charm
733 :param: num_units: Number of units
734
735 :return: (juju.application.Application): Juju application
736 """
737 self.log.debug(
738 "Deploying charm {} to machine {} in model ~{}".format(
739 application_name, machine_id, model_name
740 )
741 )
742 self.log.debug("charm: {}".format(path))
743
744 # Get controller
745 controller = await self.get_controller()
746
747 # Get model
748 model = await self.get_model(controller, model_name)
749
750 try:
751 if application_name not in model.applications:
752
753 if machine_id is not None:
754 machine, series = self._get_machine_info(model, machine_id)
755
756 application = await model.deploy(
757 entity_url=path,
758 application_name=application_name,
759 channel="stable",
760 num_units=1,
761 series=series,
762 to=machine_id,
763 config=config,
764 )
765
766 self.log.debug(
767 "Wait until application {} is ready in model {}".format(
768 application_name, model_name
769 )
770 )
771 if num_units > 1:
772 for _ in range(num_units - 1):
773 m, _ = await self.create_machine(model_name, wait=False)
774 await application.add_unit(to=m.entity_id)
775
776 await JujuModelWatcher.wait_for(
777 model=model,
778 entity=application,
779 progress_timeout=progress_timeout,
780 total_timeout=total_timeout,
781 db_dict=db_dict,
782 n2vc=self.n2vc,
783 vca_id=self.vca_connection._vca_id,
784 )
785 self.log.debug(
786 "Application {} is ready in model {}".format(
787 application_name, model_name
788 )
789 )
790 else:
791 raise JujuApplicationExists(
792 "Application {} exists".format(application_name)
793 )
794 finally:
795 await self.disconnect_model(model)
796 await self.disconnect_controller(controller)
797
798 return application
799
800 async def scale_application(
801 self,
802 model_name: str,
803 application_name: str,
804 scale: int = 1,
805 total_timeout: float = None,
806 ):
807 """
808 Scale application (K8s)
809
810 :param: model_name: Model name
811 :param: application_name: Application name
812 :param: scale: Scale to which to set this application
813 :param: total_timeout: Timeout for the entity to be active
814 """
815
816 model = None
817 controller = await self.get_controller()
818 try:
819 model = await self.get_model(controller, model_name)
820
821 self.log.debug(
822 "Scaling application {} in model {}".format(
823 application_name, model_name
824 )
825 )
826 application = self._get_application(model, application_name)
827 if application is None:
828 raise JujuApplicationNotFound("Cannot scale application")
829 await application.scale(scale=scale)
830 # Wait until application is scaled in model
831 self.log.debug(
832 "Waiting for application {} to be scaled in model {}...".format(
833 application_name, model_name
834 )
835 )
836 if total_timeout is None:
837 total_timeout = 1800
838 end = time.time() + total_timeout
839 while time.time() < end:
840 application_scale = self._get_application_count(model, application_name)
841 # Before calling wait_for_model function,
842 # wait until application unit count and scale count are equal.
843 # Because there is a delay before scaling triggers in Juju model.
844 if application_scale == scale:
845 await JujuModelWatcher.wait_for_model(
846 model=model, timeout=total_timeout
847 )
848 self.log.debug(
849 "Application {} is scaled in model {}".format(
850 application_name, model_name
851 )
852 )
853 return
854 await asyncio.sleep(5)
855 raise Exception(
856 "Timeout waiting for application {} in model {} to be scaled".format(
857 application_name, model_name
858 )
859 )
860 finally:
861 if model:
862 await self.disconnect_model(model)
863 await self.disconnect_controller(controller)
864
865 def _get_application_count(self, model: Model, application_name: str) -> int:
866 """Get number of units of the application
867
868 :param: model: Model object
869 :param: application_name: Application name
870
871 :return: int (or None if application doesn't exist)
872 """
873 application = self._get_application(model, application_name)
874 if application is not None:
875 return len(application.units)
876
877 def _get_application(self, model: Model, application_name: str) -> Application:
878 """Get application
879
880 :param: model: Model object
881 :param: application_name: Application name
882
883 :return: juju.application.Application (or None if it doesn't exist)
884 """
885 if model.applications and application_name in model.applications:
886 return model.applications[application_name]
887
888 def _get_unit(self, application: Application, machine_id: str) -> Unit:
889 """Get unit
890
891 :param: application: Application object
892 :param: machine_id: Machine id
893
894 :return: Unit
895 """
896 unit = None
897 for u in application.units:
898 if u.machine_id == machine_id:
899 unit = u
900 break
901 return unit
902
903 def _get_machine_info(
904 self,
905 model,
906 machine_id: str,
907 ) -> (str, str):
908 """Get machine info
909
910 :param: model: Model object
911 :param: machine_id: Machine id
912
913 :return: (str, str): (machine, series)
914 """
915 if machine_id not in model.machines:
916 msg = "Machine {} not found in model".format(machine_id)
917 self.log.error(msg=msg)
918 raise JujuMachineNotFound(msg)
919 machine = model.machines[machine_id]
920 return machine, machine.series
921
922 async def execute_action(
923 self,
924 application_name: str,
925 model_name: str,
926 action_name: str,
927 db_dict: dict = None,
928 machine_id: str = None,
929 progress_timeout: float = None,
930 total_timeout: float = None,
931 **kwargs,
932 ):
933 """Execute action
934
935 :param: application_name: Application name
936 :param: model_name: Model name
937 :param: action_name: Name of the action
938 :param: db_dict: Dictionary with data of the DB to write the updates
939 :param: machine_id Machine id
940 :param: progress_timeout: Maximum time between two updates in the model
941 :param: total_timeout: Timeout for the entity to be active
942
943 :return: (str, str): (output and status)
944 """
945 self.log.debug(
946 "Executing action {} using params {}".format(action_name, kwargs)
947 )
948 # Get controller
949 controller = await self.get_controller()
950
951 # Get model
952 model = await self.get_model(controller, model_name)
953
954 try:
955 # Get application
956 application = self._get_application(
957 model,
958 application_name=application_name,
959 )
960 if application is None:
961 raise JujuApplicationNotFound("Cannot execute action")
962 # Racing condition:
963 # Ocassionally, self._get_leader_unit() will return None
964 # because the leader elected hook has not been triggered yet.
965 # Therefore, we are doing some retries. If it happens again,
966 # re-open bug 1236
967 if machine_id is None:
968 unit = await self._get_leader_unit(application)
969 self.log.debug(
970 "Action {} is being executed on the leader unit {}".format(
971 action_name, unit.name
972 )
973 )
974 else:
975 unit = self._get_unit(application, machine_id)
976 if not unit:
977 raise JujuError(
978 "A unit with machine id {} not in available units".format(
979 machine_id
980 )
981 )
982 self.log.debug(
983 "Action {} is being executed on {} unit".format(
984 action_name, unit.name
985 )
986 )
987
988 actions = await application.get_actions()
989
990 if action_name not in actions:
991 raise JujuActionNotFound(
992 "Action {} not in available actions".format(action_name)
993 )
994
995 action = await unit.run_action(action_name, **kwargs)
996
997 self.log.debug(
998 "Wait until action {} is completed in application {} (model={})".format(
999 action_name, application_name, model_name
1000 )
1001 )
1002 await JujuModelWatcher.wait_for(
1003 model=model,
1004 entity=action,
1005 progress_timeout=progress_timeout,
1006 total_timeout=total_timeout,
1007 db_dict=db_dict,
1008 n2vc=self.n2vc,
1009 vca_id=self.vca_connection._vca_id,
1010 )
1011
1012 output = await model.get_action_output(action_uuid=action.entity_id)
1013 status = await model.get_action_status(uuid_or_prefix=action.entity_id)
1014 status = (
1015 status[action.entity_id] if action.entity_id in status else "failed"
1016 )
1017
1018 self.log.debug(
1019 "Action {} completed with status {} in application {} (model={})".format(
1020 action_name, action.status, application_name, model_name
1021 )
1022 )
1023 finally:
1024 await self.disconnect_model(model)
1025 await self.disconnect_controller(controller)
1026
1027 return output, status
1028
1029 async def get_actions(self, application_name: str, model_name: str) -> dict:
1030 """Get list of actions
1031
1032 :param: application_name: Application name
1033 :param: model_name: Model name
1034
1035 :return: Dict with this format
1036 {
1037 "action_name": "Description of the action",
1038 ...
1039 }
1040 """
1041 self.log.debug(
1042 "Getting list of actions for application {}".format(application_name)
1043 )
1044
1045 # Get controller
1046 controller = await self.get_controller()
1047
1048 # Get model
1049 model = await self.get_model(controller, model_name)
1050
1051 try:
1052 # Get application
1053 application = self._get_application(
1054 model,
1055 application_name=application_name,
1056 )
1057
1058 # Return list of actions
1059 return await application.get_actions()
1060
1061 finally:
1062 # Disconnect from model and controller
1063 await self.disconnect_model(model)
1064 await self.disconnect_controller(controller)
1065
1066 async def get_metrics(self, model_name: str, application_name: str) -> dict:
1067 """Get the metrics collected by the VCA.
1068
1069 :param model_name The name or unique id of the network service
1070 :param application_name The name of the application
1071 """
1072 if not model_name or not application_name:
1073 raise Exception("model_name and application_name must be non-empty strings")
1074 metrics = {}
1075 controller = await self.get_controller()
1076 model = await self.get_model(controller, model_name)
1077 try:
1078 application = self._get_application(model, application_name)
1079 if application is not None:
1080 metrics = await application.get_metrics()
1081 finally:
1082 self.disconnect_model(model)
1083 self.disconnect_controller(controller)
1084 return metrics
1085
1086 async def add_relation(
1087 self,
1088 model_name: str,
1089 endpoint_1: str,
1090 endpoint_2: str,
1091 ):
1092 """Add relation
1093
1094 :param: model_name: Model name
1095 :param: endpoint_1 First endpoint name
1096 ("app:endpoint" format or directly the saas name)
1097 :param: endpoint_2: Second endpoint name (^ same format)
1098 """
1099
1100 self.log.debug("Adding relation: {} -> {}".format(endpoint_1, endpoint_2))
1101
1102 # Get controller
1103 controller = await self.get_controller()
1104
1105 # Get model
1106 model = await self.get_model(controller, model_name)
1107
1108 # Add relation
1109 try:
1110 await model.add_relation(endpoint_1, endpoint_2)
1111 except juju.errors.JujuAPIError as e:
1112 if "not found" in e.message:
1113 self.log.warning("Relation not found: {}".format(e.message))
1114 return
1115 if "already exists" in e.message:
1116 self.log.warning("Relation already exists: {}".format(e.message))
1117 return
1118 # another exception, raise it
1119 raise e
1120 finally:
1121 await self.disconnect_model(model)
1122 await self.disconnect_controller(controller)
1123
1124 async def consume(
1125 self,
1126 offer_url: str,
1127 model_name: str,
1128 ):
1129 """
1130 Adds a remote offer to the model. Relations can be created later using "juju relate".
1131
1132 :param: offer_url: Offer Url
1133 :param: model_name: Model name
1134
1135 :raises ParseError if there's a problem parsing the offer_url
1136 :raises JujuError if remote offer includes and endpoint
1137 :raises JujuAPIError if the operation is not successful
1138 """
1139 controller = await self.get_controller()
1140 model = await controller.get_model(model_name)
1141
1142 try:
1143 await model.consume(offer_url)
1144 finally:
1145 await self.disconnect_model(model)
1146 await self.disconnect_controller(controller)
1147
1148 async def destroy_model(self, model_name: str, total_timeout: float = 1800):
1149 """
1150 Destroy model
1151
1152 :param: model_name: Model name
1153 :param: total_timeout: Timeout
1154 """
1155
1156 controller = await self.get_controller()
1157 model = None
1158 try:
1159 if not await self.model_exists(model_name, controller=controller):
1160 return
1161
1162 self.log.debug("Destroying model {}".format(model_name))
1163
1164 model = await self.get_model(controller, model_name)
1165 # Destroy machines that are manually provisioned
1166 # and still are in pending state
1167 await self._destroy_pending_machines(model, only_manual=True)
1168 await self.disconnect_model(model)
1169
1170 await self._destroy_model(
1171 model_name,
1172 controller,
1173 timeout=total_timeout,
1174 )
1175 finally:
1176 if model:
1177 await self.disconnect_model(model)
1178 await self.disconnect_controller(controller)
1179
1180 async def _destroy_model(
1181 self, model_name: str, controller: Controller, timeout: float = 1800
1182 ):
1183 """
1184 Destroy model from controller
1185
1186 :param: model: Model name to be removed
1187 :param: controller: Controller object
1188 :param: timeout: Timeout in seconds
1189 """
1190
1191 async def _destroy_model_loop(model_name: str, controller: Controller):
1192 while await self.model_exists(model_name, controller=controller):
1193 await controller.destroy_model(
1194 model_name, destroy_storage=True, force=True, max_wait=0
1195 )
1196 await asyncio.sleep(5)
1197
1198 try:
1199 await asyncio.wait_for(
1200 _destroy_model_loop(model_name, controller), timeout=timeout
1201 )
1202 except asyncio.TimeoutError:
1203 raise Exception(
1204 "Timeout waiting for model {} to be destroyed".format(model_name)
1205 )
1206
1207 async def destroy_application(
1208 self, model_name: str, application_name: str, total_timeout: float
1209 ):
1210 """
1211 Destroy application
1212
1213 :param: model_name: Model name
1214 :param: application_name: Application name
1215 :param: total_timeout: Timeout
1216 """
1217
1218 controller = await self.get_controller()
1219 model = None
1220
1221 try:
1222 model = await self.get_model(controller, model_name)
1223 self.log.debug(
1224 "Destroying application {} in model {}".format(
1225 application_name, model_name
1226 )
1227 )
1228 application = self._get_application(model, application_name)
1229 if application:
1230 await application.destroy()
1231 else:
1232 self.log.warning("Application not found: {}".format(application_name))
1233
1234 self.log.debug(
1235 "Waiting for application {} to be destroyed in model {}...".format(
1236 application_name, model_name
1237 )
1238 )
1239 if total_timeout is None:
1240 total_timeout = 3600
1241 end = time.time() + total_timeout
1242 while time.time() < end:
1243 if not self._get_application(model, application_name):
1244 self.log.debug(
1245 "The application {} was destroyed in model {} ".format(
1246 application_name, model_name
1247 )
1248 )
1249 return
1250 await asyncio.sleep(5)
1251 raise Exception(
1252 "Timeout waiting for application {} to be destroyed in model {}".format(
1253 application_name, model_name
1254 )
1255 )
1256 finally:
1257 if model is not None:
1258 await self.disconnect_model(model)
1259 await self.disconnect_controller(controller)
1260
1261 async def _destroy_pending_machines(self, model: Model, only_manual: bool = False):
1262 """
1263 Destroy pending machines in a given model
1264
1265 :param: only_manual: Bool that indicates only manually provisioned
1266 machines should be destroyed (if True), or that
1267 all pending machines should be destroyed
1268 """
1269 status = await model.get_status()
1270 for machine_id in status.machines:
1271 machine_status = status.machines[machine_id]
1272 if machine_status.agent_status.status == "pending":
1273 if only_manual and not machine_status.instance_id.startswith("manual:"):
1274 break
1275 machine = model.machines[machine_id]
1276 await machine.destroy(force=True)
1277
1278 async def configure_application(
1279 self, model_name: str, application_name: str, config: dict = None
1280 ):
1281 """Configure application
1282
1283 :param: model_name: Model name
1284 :param: application_name: Application name
1285 :param: config: Config to apply to the charm
1286 """
1287 self.log.debug("Configuring application {}".format(application_name))
1288
1289 if config:
1290 controller = await self.get_controller()
1291 model = None
1292 try:
1293 model = await self.get_model(controller, model_name)
1294 application = self._get_application(
1295 model,
1296 application_name=application_name,
1297 )
1298 await application.set_config(config)
1299 finally:
1300 if model:
1301 await self.disconnect_model(model)
1302 await self.disconnect_controller(controller)
1303
1304 def handle_exception(self, loop, context):
1305 # All unhandled exceptions by libjuju are handled here.
1306 pass
1307
1308 async def health_check(self, interval: float = 300.0):
1309 """
1310 Health check to make sure controller and controller_model connections are OK
1311
1312 :param: interval: Time in seconds between checks
1313 """
1314 controller = None
1315 while True:
1316 try:
1317 controller = await self.get_controller()
1318 # self.log.debug("VCA is alive")
1319 except Exception as e:
1320 self.log.error("Health check to VCA failed: {}".format(e))
1321 finally:
1322 await self.disconnect_controller(controller)
1323 await asyncio.sleep(interval)
1324
1325 async def list_models(self, contains: str = None) -> [str]:
1326 """List models with certain names
1327
1328 :param: contains: String that is contained in model name
1329
1330 :retur: [models] Returns list of model names
1331 """
1332
1333 controller = await self.get_controller()
1334 try:
1335 models = await controller.list_models()
1336 if contains:
1337 models = [model for model in models if contains in model]
1338 return models
1339 finally:
1340 await self.disconnect_controller(controller)
1341
1342 async def list_offers(self, model_name: str) -> QueryApplicationOffersResults:
1343 """List models with certain names
1344
1345 :param: model_name: Model name
1346
1347 :return: Returns list of offers
1348 """
1349
1350 controller = await self.get_controller()
1351 try:
1352 return await controller.list_offers(model_name)
1353 finally:
1354 await self.disconnect_controller(controller)
1355
1356 async def add_k8s(
1357 self,
1358 name: str,
1359 rbac_id: str,
1360 token: str,
1361 client_cert_data: str,
1362 configuration: Configuration,
1363 storage_class: str,
1364 credential_name: str = None,
1365 ):
1366 """
1367 Add a Kubernetes cloud to the controller
1368
1369 Similar to the `juju add-k8s` command in the CLI
1370
1371 :param: name: Name for the K8s cloud
1372 :param: configuration: Kubernetes configuration object
1373 :param: storage_class: Storage Class to use in the cloud
1374 :param: credential_name: Storage Class to use in the cloud
1375 """
1376
1377 if not storage_class:
1378 raise Exception("storage_class must be a non-empty string")
1379 if not name:
1380 raise Exception("name must be a non-empty string")
1381 if not configuration:
1382 raise Exception("configuration must be provided")
1383
1384 endpoint = configuration.host
1385 credential = self.get_k8s_cloud_credential(
1386 configuration,
1387 client_cert_data,
1388 token,
1389 )
1390 credential.attrs[RBAC_LABEL_KEY_NAME] = rbac_id
1391 cloud = client.Cloud(
1392 type_="kubernetes",
1393 auth_types=[credential.auth_type],
1394 endpoint=endpoint,
1395 ca_certificates=[client_cert_data],
1396 config={
1397 "operator-storage": storage_class,
1398 "workload-storage": storage_class,
1399 },
1400 )
1401
1402 return await self.add_cloud(
1403 name, cloud, credential, credential_name=credential_name
1404 )
1405
1406 def get_k8s_cloud_credential(
1407 self,
1408 configuration: Configuration,
1409 client_cert_data: str,
1410 token: str = None,
1411 ) -> client.CloudCredential:
1412 attrs = {}
1413 # TODO: Test with AKS
1414 key = None # open(configuration.key_file, "r").read()
1415 username = configuration.username
1416 password = configuration.password
1417
1418 if client_cert_data:
1419 attrs["ClientCertificateData"] = client_cert_data
1420 if key:
1421 attrs["ClientKeyData"] = key
1422 if token:
1423 if username or password:
1424 raise JujuInvalidK8sConfiguration("Cannot set both token and user/pass")
1425 attrs["Token"] = token
1426
1427 auth_type = None
1428 if key:
1429 auth_type = "oauth2"
1430 if client_cert_data:
1431 auth_type = "oauth2withcert"
1432 if not token:
1433 raise JujuInvalidK8sConfiguration(
1434 "missing token for auth type {}".format(auth_type)
1435 )
1436 elif username:
1437 if not password:
1438 self.log.debug(
1439 "credential for user {} has empty password".format(username)
1440 )
1441 attrs["username"] = username
1442 attrs["password"] = password
1443 if client_cert_data:
1444 auth_type = "userpasswithcert"
1445 else:
1446 auth_type = "userpass"
1447 elif client_cert_data and token:
1448 auth_type = "certificate"
1449 else:
1450 raise JujuInvalidK8sConfiguration("authentication method not supported")
1451 return client.CloudCredential(auth_type=auth_type, attrs=attrs)
1452
1453 async def add_cloud(
1454 self,
1455 name: str,
1456 cloud: Cloud,
1457 credential: CloudCredential = None,
1458 credential_name: str = None,
1459 ) -> Cloud:
1460 """
1461 Add cloud to the controller
1462
1463 :param: name: Name of the cloud to be added
1464 :param: cloud: Cloud object
1465 :param: credential: CloudCredentials object for the cloud
1466 :param: credential_name: Credential name.
1467 If not defined, cloud of the name will be used.
1468 """
1469 controller = await self.get_controller()
1470 try:
1471 _ = await controller.add_cloud(name, cloud)
1472 if credential:
1473 await controller.add_credential(
1474 credential_name or name, credential=credential, cloud=name
1475 )
1476 # Need to return the object returned by the controller.add_cloud() function
1477 # I'm returning the original value now until this bug is fixed:
1478 # https://github.com/juju/python-libjuju/issues/443
1479 return cloud
1480 finally:
1481 await self.disconnect_controller(controller)
1482
1483 async def remove_cloud(self, name: str):
1484 """
1485 Remove cloud
1486
1487 :param: name: Name of the cloud to be removed
1488 """
1489 controller = await self.get_controller()
1490 try:
1491 await controller.remove_cloud(name)
1492 except juju.errors.JujuError as e:
1493 if len(e.errors) == 1 and f'cloud "{name}" not found' == e.errors[0]:
1494 self.log.warning(f"Cloud {name} not found, so it could not be deleted.")
1495 else:
1496 raise e
1497 finally:
1498 await self.disconnect_controller(controller)
1499
1500 @retry(attempts=20, delay=5, fallback=JujuLeaderUnitNotFound())
1501 async def _get_leader_unit(self, application: Application) -> Unit:
1502 unit = None
1503 for u in application.units:
1504 if await u.is_leader_from_status():
1505 unit = u
1506 break
1507 if not unit:
1508 raise Exception()
1509 return unit
1510
1511 async def get_cloud_credentials(self, cloud: Cloud) -> typing.List:
1512 """
1513 Get cloud credentials
1514
1515 :param: cloud: Cloud object. The returned credentials will be from this cloud.
1516
1517 :return: List of credentials object associated to the specified cloud
1518
1519 """
1520 controller = await self.get_controller()
1521 try:
1522 facade = client.CloudFacade.from_connection(controller.connection())
1523 cloud_cred_tag = tag.credential(
1524 cloud.name, self.vca_connection.data.user, cloud.credential_name
1525 )
1526 params = [client.Entity(cloud_cred_tag)]
1527 return (await facade.Credential(params)).results
1528 finally:
1529 await self.disconnect_controller(controller)
1530
1531 async def check_application_exists(self, model_name, application_name) -> bool:
1532 """Check application exists
1533
1534 :param: model_name: Model Name
1535 :param: application_name: Application Name
1536
1537 :return: Boolean
1538 """
1539
1540 model = None
1541 controller = await self.get_controller()
1542 try:
1543 model = await self.get_model(controller, model_name)
1544 self.log.debug(
1545 "Checking if application {} exists in model {}".format(
1546 application_name, model_name
1547 )
1548 )
1549 return self._get_application(model, application_name) is not None
1550 finally:
1551 if model:
1552 await self.disconnect_model(model)
1553 await self.disconnect_controller(controller)