def get(self, request, *args, **kwargs):
auditor.record(event_type=EXPERIMENT_LOGS_VIEWED,
instance=self.experiment,
actor_id=request.user.id,
actor_name=request.user.username)
experiment_name = self.experiment.unique_name
if self.experiment.is_done:
log_path = stores.get_experiment_logs_path(
experiment_name=experiment_name, temp=False)
log_path = archive_logs_file(log_path=log_path,
namepath=experiment_name)
elif self.experiment.run_env and self.experiment.run_env.get(
'in_cluster'):
process_logs(experiment=self.experiment, temp=True)
log_path = stores.get_experiment_logs_path(
experiment_name=experiment_name, temp=True)
else:
return Response(status=status.HTTP_404_NOT_FOUND,
data='Experiment is still running, no logs.')
filename = os.path.basename(log_path)
chunk_size = 8192
try:
wrapped_file = FileWrapper(open(log_path, 'rb'), chunk_size)
response = StreamingHttpResponse(
wrapped_file, content_type=mimetypes.guess_type(log_path)[0])
response['Content-Length'] = os.path.getsize(log_path)
response['Content-Disposition'] = "attachment; filename={}".format(
filename)
return response
except FileNotFoundError:
_logger.warning('Log file not found: log_path=%s', log_path)
return Response(
status=status.HTTP_404_NOT_FOUND,
data='Log file not found: log_path={}'.format(log_path))
def test_experiment_logs_path_creation_deletion(self):
stores.create_experiment_logs_path(
experiment_name=self.experiment.unique_name, temp=False)
experiment_logs_path = stores.get_experiment_logs_path(
experiment_name=self.experiment.unique_name, temp=False)
filepath = stores.get_experiment_logs_path(
experiment_name=self.experiment.unique_name, temp=False)
open(filepath, '+w')
assert os.path.exists(experiment_logs_path) is True
assert os.path.exists(filepath) is True
stores_schedule_logs_deletion(persistence=None,
subpath=self.experiment.subpath)
assert os.path.exists(filepath) is False
def get_experiment_logs_path(experiment: Experiment) -> Optional[str]:
experiment_name = experiment.unique_name
if experiment.is_done:
log_path = stores.get_experiment_logs_path(experiment_name=experiment_name, temp=False)
logs_path = archive_logs_file(
log_path=log_path,
namepath=experiment_name)
elif experiment.is_managed:
process_logs(experiment=experiment, temp=True)
logs_path = stores.get_experiment_logs_path(experiment_name=experiment_name, temp=True)
else:
return None
return logs_path
def get_experiment_logs_path(experiment: Experiment) -> Optional[str]:
experiment_name = experiment.unique_name
if experiment.is_done:
log_path = stores.get_experiment_logs_path(experiment_name=experiment_name, temp=False)
logs_path = archive_logs_file(
log_path=log_path,
namepath=experiment_name)
elif experiment.is_managed:
process_logs(experiment=experiment, temp=True)
logs_path = stores.get_experiment_logs_path(experiment_name=experiment_name, temp=True)
else:
# Normally it should load from the tmp file where we append the logs
return None
return logs_path
def get_config_map(namespace, project_name, experiment_group_name,
experiment_name, project_uuid, experiment_group_uuid,
experiment_uuid, original_name, cloning_strategy,
cluster_def, persistence_outputs, persistence_data, params,
log_level):
name = constants.CONFIG_MAP_NAME.format(uuid=experiment_uuid)
labels = get_map_labels(project_name, experiment_group_name,
experiment_name, project_uuid,
experiment_group_uuid, experiment_uuid)
metadata = client.V1ObjectMeta(name=name,
labels=labels,
namespace=namespace)
experiment_outputs_path = stores.get_experiment_outputs_path(
persistence=persistence_outputs,
experiment_name=experiment_name,
original_name=original_name,
cloning_strategy=cloning_strategy)
experiment_logs_path = stores.get_experiment_logs_path(
experiment_name=experiment_name, temp=False)
data = {
constants.CONFIG_MAP_CLUSTER_KEY_NAME: json.dumps(cluster_def),
constants.CONFIG_MAP_PARAMS_KEY_NAME: json.dumps(params) or '{}',
constants.CONFIG_MAP_EXPERIMENT_INFO_KEY_NAME: json.dumps(labels),
constants.CONFIG_MAP_LOG_LEVEL_KEY_NAME: log_level,
constants.CONFIG_MAP_RUN_OUTPUTS_PATH_KEY_NAME:
experiment_outputs_path,
constants.CONFIG_MAP_RUN_LOGS_PATH_KEY_NAME: experiment_logs_path,
constants.CONFIG_MAP_RUN_DATA_PATHS_KEY_NAME: persistence_data,
API_HTTP_URL: get_settings_http_api_url(),
API_WS_HOST: get_settings_ws_api_url(),
}
return client.V1ConfigMap(api_version=k8s_constants.K8S_API_VERSION_V1,
kind=k8s_constants.K8S_CONFIG_MAP_KIND,
metadata=metadata,
data=data)
def safe_log_experiment(experiment_name, log_lines, temp, append=False):
log_path = stores.get_experiment_logs_path(experiment_name=experiment_name,
temp=temp)
try:
stores.create_experiment_logs_path(experiment_name=experiment_name,
temp=temp)
_lock_log(log_path, log_lines, append=append)
except OSError:
# Retry
stores.create_experiment_logs_path(experiment_name=experiment_name,
temp=temp)
_lock_log(log_path, log_lines, append=append)
def test_experiment_group_logs_path_creation_deletion(self):
experiment = ExperimentFactory(user=self.project.user,
project=self.project,
experiment_group=self.experiment_group)
experiment_logs_path = stores.get_experiment_logs_path(
experiment_name=experiment.unique_name,
temp=False)
stores.create_experiment_logs_path(experiment_name=experiment.unique_name, temp=False)
open(experiment_logs_path, '+w')
experiment_group_logs_path = stores.get_experiment_group_logs_path(
experiment_group_name=self.experiment_group.unique_name)
# Should be true, created by the signal
assert os.path.exists(experiment_logs_path) is True
assert os.path.exists(experiment_group_logs_path) is True
stores_schedule_logs_deletion(persistence=None, subpath=self.experiment_group.subpath)
assert os.path.exists(experiment_logs_path) is False
assert os.path.exists(experiment_group_logs_path) is False
def test_project_logs_path_creation_deletion(self):
with patch('scheduler.tasks.experiments.experiments_build.apply_async'
) as _: # noqa
experiment = ExperimentFactory(user=self.project.user,
project=self.project)
experiment_logs_path = stores.get_experiment_logs_path(
experiment_name=experiment.unique_name, temp=False)
stores.create_experiment_logs_path(
experiment_name=experiment.unique_name, temp=False)
open(experiment_logs_path, '+w')
project_logs_path = stores.get_project_logs_path(
project_name=self.project.unique_name)
project_repos_path = get_project_repos_path(self.project.unique_name)
# Should be true, created by the signal
assert os.path.exists(experiment_logs_path) is True
assert os.path.exists(project_logs_path) is True
assert os.path.exists(project_repos_path) is True
stores_schedule_logs_deletion(persistence=None,
subpath=self.project.subpath)
delete_project_repos(self.project.unique_name)
assert os.path.exists(experiment_logs_path) is False
assert os.path.exists(project_logs_path) is False
assert os.path.exists(project_repos_path) is False
def _get_logs_path(self, persistence_logs='default'):
return stores.get_experiment_logs_path(
persistence=persistence_logs,
experiment_name=self.experiment_name,
temp=False)
def get_log_path(instance):
return stores.get_experiment_logs_path(
experiment_name=instance.unique_name, temp=False)
def get_pod_container(self,
volume_mounts,
env_vars=None,
command=None,
args=None,
persistence_outputs=None,
persistence_data=None,
outputs_refs_jobs=None,
outputs_refs_experiments=None,
secret_refs=None,
configmap_refs=None,
resources=None,
ephemeral_token=None):
"""Pod job container for task."""
assert self.cluster_def is not None
# Env vars preparations
env_vars = to_list(env_vars, check_none=True)
logs_path = stores.get_experiment_logs_path(
experiment_name=self.experiment_name, temp=False)
outputs_path = stores.get_experiment_outputs_path(
persistence=persistence_outputs,
experiment_name=self.experiment_name,
original_name=self.original_name,
cloning_strategy=self.cloning_strategy)
env_vars += get_job_env_vars(
persistence_outputs=persistence_outputs,
outputs_path=outputs_path,
persistence_data=persistence_data,
log_level=self.log_level,
logs_path=logs_path,
outputs_refs_jobs=outputs_refs_jobs,
outputs_refs_experiments=outputs_refs_experiments,
ephemeral_token=ephemeral_token,
)
env_vars += [
get_env_var(name=constants.CONFIG_MAP_CLUSTER_KEY_NAME,
value=json.dumps(self.cluster_def)),
get_env_var(name=constants.CONFIG_MAP_DECLARATIONS_KEY_NAME,
value=self.declarations),
get_env_var(name=constants.CONFIG_MAP_EXPERIMENT_INFO_KEY_NAME,
value=json.dumps(self.experiment_labels)),
]
env_vars += get_resources_env_vars(resources=resources)
# Env from configmap and secret refs
env_from = get_pod_env_from(secret_refs=secret_refs,
configmap_refs=configmap_refs)
ports = [
client.V1ContainerPort(container_port=port) for port in self.ports
]
return client.V1Container(name=self.job_container_name,
image=self.job_docker_image,
command=command,
args=args,
ports=ports,
env=env_vars,
env_from=env_from,
resources=get_resources(resources),
volume_mounts=volume_mounts)
