def annotations(self): cluster, is_distributed, = self.spec.cluster_def worker_annotations = MXNetSpecification.get_worker_annotations( environment=self.spec.config.mxnet, cluster=cluster, is_distributed=is_distributed) ps_annotations = MXNetSpecification.get_ps_annotations( environment=self.spec.config.mxnet, cluster=cluster, is_distributed=is_distributed) return { TaskType.MASTER: { 0: self.spec.master_annotations }, TaskType.WORKER: worker_annotations, TaskType.SERVER: ps_annotations, }
def create_mxnet_experiment_jobs(experiment, spawner): master_job_uuid = spawner.job_uuids[TaskType.MASTER][0] create_job(job_uuid=master_job_uuid, experiment=experiment, resources=spawner.spec.master_resources, node_selector=spawner.spec.master_node_selector, affinity=spawner.spec.master_affinity, tolerations=spawner.spec.master_tolerations) cluster, is_distributed = spawner.spec.cluster_def environment = spawner.spec.config.mxnet worker_resources = MXNetSpecification.get_worker_resources( environment=environment, cluster=cluster, is_distributed=is_distributed) worker_node_selectors = MXNetSpecification.get_worker_node_selectors( environment=environment, cluster=cluster, is_distributed=is_distributed) worker_affinities = MXNetSpecification.get_worker_affinities( environment=environment, cluster=cluster, is_distributed=is_distributed) worker_tolerations = MXNetSpecification.get_worker_tolerations( environment=environment, cluster=cluster, is_distributed=is_distributed) for i, worker_job_uuid in enumerate(spawner.job_uuids[TaskType.WORKER]): create_job(job_uuid=worker_job_uuid, experiment=experiment, role=TaskType.WORKER, sequence=i, resources=worker_resources.get(i), node_selector=worker_node_selectors.get(i), affinity=worker_affinities.get(i), tolerations=worker_tolerations.get(i)) server_resources = MXNetSpecification.get_ps_resources( environment=environment, cluster=cluster, is_distributed=is_distributed) server_node_selectors = MXNetSpecification.get_ps_node_selectors( environment=environment, cluster=cluster, is_distributed=is_distributed) server_affinities = MXNetSpecification.get_ps_affinities( environment=environment, cluster=cluster, is_distributed=is_distributed) server_tolerations = MXNetSpecification.get_ps_tolerations( environment=environment, cluster=cluster, is_distributed=is_distributed) for i, server_job_uuid in enumerate(spawner.job_uuids[TaskType.SERVER]): create_job(job_uuid=server_job_uuid, experiment=experiment, role=TaskType.SERVER, sequence=i, resources=server_resources.get(i), node_selector=server_node_selectors, affinity=server_affinities, tolerations=server_tolerations)