def tolerations(self):
cluster, is_distributed, = self.spec.cluster_def
worker_tolerations = HorovodSpecification.get_worker_tolerations(
environment=self.spec.config.horovod,
cluster=cluster,
is_distributed=is_distributed
)
return {
TaskType.MASTER: {0: self.spec.master_affinity},
TaskType.WORKER: worker_tolerations,
}
def handle_horovod_experiment(experiment, spawner, response):
# Get the number of jobs this experiment started
master = response[TaskType.MASTER]
job_uuid = master['pod']['metadata']['labels']['job_uuid']
job_uuid = uuid.UUID(job_uuid)
create_job(job_uuid=job_uuid,
experiment=experiment,
definition=get_job_definition(master),
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
worker_resources = HorovodSpecification.get_worker_resources(
environment=spawner.spec.environment,
cluster=cluster,
is_distributed=is_distributed
)
worker_node_selectors = HorovodSpecification.get_worker_node_selectors(
environment=spawner.spec.environment,
cluster=cluster,
is_distributed=is_distributed
)
worker_affinities = HorovodSpecification.get_worker_affinities(
environment=spawner.spec.environment,
cluster=cluster,
is_distributed=is_distributed
)
worker_tolerations = HorovodSpecification.get_worker_tolerations(
environment=spawner.spec.environment,
cluster=cluster,
is_distributed=is_distributed
)
for i, worker in enumerate(response[TaskType.WORKER]):
job_uuid = worker['pod']['metadata']['labels']['job_uuid']
job_uuid = uuid.UUID(job_uuid)
create_job(job_uuid=job_uuid,
experiment=experiment,
definition=get_job_definition(worker),
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))
def create_horovod_experiment_jobs(experiment, spawner):
master_job_uuid = spawner.job_uuids[TaskType.MASTER][0]
k8s_replica = None
create_job(job_uuid=master_job_uuid,
experiment=experiment,
k8s_replica=k8s_replica,
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.horovod
worker_resources = HorovodSpecification.get_worker_resources(
environment=environment,
cluster=cluster,
is_distributed=is_distributed
)
worker_node_selectors = HorovodSpecification.get_worker_node_selectors(
environment=environment,
cluster=cluster,
is_distributed=is_distributed
)
worker_affinities = HorovodSpecification.get_worker_affinities(
environment=environment,
cluster=cluster,
is_distributed=is_distributed
)
worker_tolerations = HorovodSpecification.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))