def start_worker(self, image, tenant, actor_id, worker_id, client_id, client_access_token, client_refresh_token, ch, api_server, client_secret): # start an actor executor container and wait for a confirmation that image was pulled. attempts = 0 # worker = get_worker(worker_id) # worker['status'] = PULLING_IMAGE Worker.update_worker_status(actor_id, worker_id, PULLING_IMAGE) try: logger.debug("Worker pulling image {}...".format(image)) pull_image(image) except DockerError as e: # return a message to the spawner that there was an error pulling image and abort # this is not necessarily an error state: the user simply could have provided an # image name that does not exist in the registry. This is the first time we would # find that out. logger.info("worker got a DockerError trying to pull image. Error: {}.".format(e)) raise e logger.info("Image {} pulled successfully.".format(image)) # Done pulling image # Run Worker Container while True: try: Worker.update_worker_status(actor_id, worker_id, CREATING_CONTAINER) logger.debug('spawner creating worker container') worker_dict = run_worker( image, actor_id, worker_id, client_id, client_access_token, client_refresh_token, tenant, api_server, client_secret ) logger.debug(f'finished run worker; worker dict: {worker_dict}') except DockerError as e: logger.error("Spawner got a docker exception from run_worker; Exception: {}".format(e)) if 'read timeout' in e.message: logger.info("Exception was a read timeout; trying run_worker again..") time.sleep(5) attempts = attempts + 1 if attempts > 20: msg = "Spawner continued to get DockerError for 20 attempts. Exception: {}".format(e) logger.critical(msg) # todo - should we be calling kill_worker here? (it is called in the exception block of the else below) raise SpawnerException(msg) continue else: logger.info("Exception was NOT a read timeout; quiting on this worker.") # delete this worker from the workers store: try: self.kill_worker(actor_id, worker_id) except WorkerException as e: logger.info("Got WorkerException from delete_worker(). " "worker_id: {}" "Exception: {}".format(worker_id, e)) raise SpawnerException(message="Unable to start worker; error: {}".format(e)) break logger.debug('finished loop') worker_dict['ch_name'] = WorkerChannel.get_name(worker_id) # if the actor is not already in READY status, set actor status to READY before worker status has been # set to READY. # it is possible the actor status is already READY because this request is the autoscaler starting a new worker # for an existing actor. actor = Actor.from_db(actors_store[actor_id]) if not actor.status == READY: try: Actor.set_status(actor_id, READY, status_message=" ") except KeyError: # it is possible the actor was already deleted during worker start up; if # so, the worker should have a stop message waiting for it. starting subscribe # as usual should allow this process to work as expected. pass # finalize worker with READY status worker = Worker(tenant=tenant, **worker_dict) logger.info("calling add_worker for worker: {}.".format(worker)) Worker.add_worker(actor_id, worker) ch.put('READY') # step 4 logger.info('sent message through channel')
def process(self, cmd): print("Processing cmd:{}".format(str(cmd))) actor_id = cmd['actor_id'] worker_ids = cmd['worker_ids'] image = cmd['image'] tenant = cmd['tenant'] stop_existing = cmd.get('stop_existing', True) num_workers = cmd.get('num', self.num_workers) print("Actor id:{}".format(actor_id)) try: new_channels, anon_channels, new_workers = self.start_workers( actor_id, worker_ids, image, tenant, num_workers) except SpawnerException as e: # for now, start_workers will do clean up for a SpawnerException, so we just need # to return back to the run loop. return print("Created new workers: {}".format(str(new_workers))) # stop any existing workers: if stop_existing: self.stop_workers(actor_id, worker_ids) # add workers to store first so that the records will be there when the workers go # to update their status if not stop_existing: for _, worker in new_workers.items(): Worker.add_worker(actor_id, worker) else: workers_store[actor_id] = new_workers # Tell new worker to subscribe to the actor channel. # If abaco is configured to generate clients for the workers, generate them now # and send new workers their clients. generate_clients = Config.get('workers', 'generate_clients').lower() for idx, channel in enumerate(anon_channels): if generate_clients == 'true': print("Getting client for worker {}".format(idx)) client_ch = ClientsChannel() client_msg = client_ch.request_client( tenant=tenant, actor_id=actor_id, # new_workers is a dictionary of dictionaries; list(d) creates a # list of keys for a dictionary d. hence, the idx^th entry # of list(ner_workers) should be the key. worker_id=new_workers[list(new_workers)[idx]]['id'], secret=self.secret) # we need to ignore errors when generating clients because it's possible it is not set up for a specific # tenant. we log it instead. if client_msg.get('status') == 'error': print("Error generating client: {}".format( client_msg.get('message'))) channel.put({ 'status': 'ok', 'actor_id': actor_id, 'tenant': tenant, 'client': 'no' }) # else, client was generated successfully: else: print("Got a client: {}, {}, {}".format( client_msg['client_id'], client_msg['access_token'], client_msg['refresh_token'])) channel.put({ 'status': 'ok', 'actor_id': actor_id, 'tenant': tenant, 'client': 'yes', 'client_id': client_msg['client_id'], 'client_secret': client_msg['client_secret'], 'access_token': client_msg['access_token'], 'refresh_token': client_msg['refresh_token'], 'api_server': client_msg['api_server'], }) else: print("Not generating clients. Config value was: {}".format( generate_clients)) channel.put({ 'status': 'ok', 'actor_id': actor_id, 'tenant': tenant, 'client': 'no' }) print("Done processing command.")
def process(self, cmd): """Main spawner method for processing a command from the CommandChannel.""" logger.info("Spawner processing new command:{}".format(cmd)) actor_id = cmd['actor_id'] worker_ids = cmd['worker_ids'] image = cmd['image'] tenant = cmd['tenant'] stop_existing = cmd.get('stop_existing', True) num_workers = cmd.get('num', self.num_workers) logger.info("command params: actor_id: {} worker_ids: {} image: {} stop_existing: {} mum_workers: {}".format( actor_id, worker_ids, image, tenant, stop_existing, num_workers)) try: new_channels, anon_channels, new_workers = self.start_workers(actor_id, worker_ids, image, tenant, num_workers) except SpawnerException as e: # for now, start_workers will do clean up for a SpawnerException, so we just need # to return back to the run loop. logger.info("Spawner returning to main run loop.") return logger.info("Created new workers: {}".format(new_workers)) # stop any existing workers: if stop_existing: logger.info("Stopping existing workers: {}".format(worker_ids)) self.stop_workers(actor_id, worker_ids) # add workers to store first so that the records will be there when the workers go # to update their status if not stop_existing: # if we're not stopping the existing workers, we need to add each worker to the # actor's collection. for _, worker in new_workers.items(): logger.info("calling add_worker for worker: {}.".format(worker)) Worker.add_worker(actor_id, worker) else: # since we're stopping the existing workers, the actor's collection should just # be equal to the new_workers. workers_store[actor_id] = new_workers logger.info("workers_store set to new_workers: {}.".format(new_workers)) # Tell new worker to subscribe to the actor channel. # If abaco is configured to generate clients for the workers, generate them now # and send new workers their clients. generate_clients = Config.get('workers', 'generate_clients').lower() logger.info("Sending messages to new workers over anonymous channels to subscribe to inbox.") for idx, channel in enumerate(anon_channels): if generate_clients == 'true': worker_id = new_workers[list(new_workers)[idx]]['id'] logger.info("Getting client for worker number {}, id: {}".format(idx, worker_id)) client_ch = ClientsChannel() try: client_msg = client_ch.request_client(tenant=tenant, actor_id=actor_id, # new_workers is a dictionary of dictionaries; list(d) creates a # list of keys for a dictionary d. hence, the idx^th entry # of list(ner_workers) should be the key. worker_id=worker_id, secret=self.secret) except ChannelTimeoutException as e: logger.error("Got a ChannelTimeoutException trying to generate a client for " "actor_id: {}; worker_id: {}; exception: {}".format(actor_id, worker_id, e)) # put actor in an error state and return self.error_out_actor(actor_id, worker_id, "Abaco was unable to generate an OAuth client for a new " "worker for this actor. System administrators have been notified.") client_ch.close() return client_ch.close() # we need to ignore errors when generating clients because it's possible it is not set up for a specific # tenant. we log it instead. if client_msg.get('status') == 'error': logger.error("Error generating client: {}".format(client_msg.get('message'))) channel.put({'status': 'ok', 'actor_id': actor_id, 'tenant': tenant, 'client': 'no'}) logger.debug("Sent OK message over anonymous worker channel.") # else, client was generated successfully: else: logger.info("Got a client: {}, {}, {}".format(client_msg['client_id'], client_msg['access_token'], client_msg['refresh_token'])) channel.put({'status': 'ok', 'actor_id': actor_id, 'tenant': tenant, 'client': 'yes', 'client_id': client_msg['client_id'], 'client_secret': client_msg['client_secret'], 'access_token': client_msg['access_token'], 'refresh_token': client_msg['refresh_token'], 'api_server': client_msg['api_server'], }) logger.debug("Sent OK message AND client over anonymous worker channel.") else: logger.info("Not generating clients. Config value was: {}".format(generate_clients)) channel.put({'status': 'ok', 'actor_id': actor_id, 'tenant': tenant, 'client': 'no'}) logger.debug("Sent OK message over anonymous worker channel.") # @TODO - # delete the anonymous channel from this thread but sleep first to avoid the race condition. time.sleep(1.5) channel.delete() # due to the race condition deleting channels (potentially before all workers have received all messages) # we put a sleep here. time.sleep(1) for ch in new_channels: try: # the new_channels are the spawnerworker channels so they can be deleted. ch.delete() except Exception as e: logger.error("Got exception trying to delete spawnerworker channel: {}".format(e)) logger.info("Done processing command.")
def process(self, cmd): print("Processing cmd:{}".format(str(cmd))) actor_id = cmd['actor_id'] image = cmd['image'] tenant = cmd['tenant'] stop_existing = cmd.get('stop_existing', True) num_workers = cmd.get('num', self.num_workers) print("Actor id:{}".format(actor_id)) try: new_channels, anon_channels, new_workers = self.start_workers(actor_id, image, tenant, num_workers) except SpawnerException as e: # for now, start_workers will do clean up for a SpawnerException, so we just need # to return back to the run loop. return print("Created new workers: {}".format(str(new_workers))) # stop any existing workers: if stop_existing: self.stop_workers(actor_id) # add workers to store first so that the records will be there when the workers go # to update their status if not stop_existing: for _, worker in new_workers.items(): Worker.add_worker(actor_id, worker) else: workers_store[actor_id] = new_workers # send new workers their clients and tell them to subscribe to the actor channel. for idx, channel in enumerate(anon_channels): print("Getting client for worker {}".format(idx)) client_ch = ClientsChannel() client_msg = client_ch.request_client(tenant=tenant, actor_id=actor_id, # new_workers is a dictionary of dictionaries; list(d) creates a # list of keys for a dictionary d. hence, the idx^th entry # of list(ner_workers) should be the key. worker_id=new_workers[list(new_workers)[idx]]['ch_name'], secret=self.secret) # we need to ignore errors when generating clients because it's possible it is not set up for a specific # tenant. we log it instead. if client_msg.get('status') == 'error': print("Error generating client: {}".format(client_msg.get('message'))) channel.put({'status': 'ok', 'actor_id': actor_id, 'tenant': tenant, 'client': 'no'}) # else, client was generated successfully: else: print("Got a client: {}, {}, {}".format(client_msg['client_id'], client_msg['access_token'], client_msg['refresh_token'])) channel.put({'status': 'ok', 'actor_id': actor_id, 'tenant': tenant, 'client': 'yes', 'client_id': client_msg['client_id'], 'client_secret': client_msg['client_secret'], 'access_token': client_msg['access_token'], 'refresh_token': client_msg['refresh_token'], 'api_server': client_msg['api_server'], }) print("Done processing command.")