def install_addon():
"""Install Benchmark Addons."""
logging.basicConfig(level=logging.INFO,
format=('%(asctime)s %(name)-12s %(levelname)-8s %(message)s'),
datefmt='%Y-%m-%dT%H:%M:%S',
)
logging.getLogger().setLevel(logging.INFO)
args = parse_args()
namespace = args.namespace
base_dir = args.base_dir
kubeconfig_path = str(os.environ['KUBECONFIG'])
api_client = deploy_utils.create_k8s_client(kubeconfig_path)
# Setup GPU Device Plugin
install_gpu_drivers(api_client)
wait_for_gpu_driver_install(api_client)
# Setup ksonnet application
app_dir = deploy_utils.setup_ks_app(base_dir, namespace, api_client, args.kubeflow_registry, args.kubebench_registry)
# Deploy Kubeflow
install_kubeflow(api_client, app_dir, namespace)
wait_for_kubeflow_install(api_client, namespace)
# change the namespace to default to set up nfs-volume and nfs-server
namespace = "default"
# Deploy NFS for kubebench
install_kubebench_nfs(api_client, app_dir, namespace)
# Deploy Github Secret
github_token = str(os.environ['GITHUB_TOKEN'])
install_github_secret(api_client, namespace, args.github_secret_name, github_token)
def install_addon():
"""Install Benchmark Addons."""
logging.basicConfig(
level=logging.INFO,
format=('%(asctime)s %(name)-12s %(levelname)-8s %(message)s'),
datefmt='%Y-%m-%dT%H:%M:%S',
)
logging.getLogger().setLevel(logging.INFO)
args = parse_args()
namespace = args.namespace
base_dir = args.base_dir
kubeconfig_path = str(os.environ['KUBECONFIG'])
api_client = deploy_utils.create_k8s_client(kubeconfig_path)
kubeflow_app_dir = os.path.join(base_dir, "ks-app")
if os.path.exists(kubeflow_app_dir):
logging.info("kubeflow app dir already exists. skip install")
return
# Setup ksonnet application
app_dir = deploy_utils.setup_ks_app(base_dir, namespace, api_client,
args.kubeflow_registry,
args.kubebench_registry)
# Deploy Kubeflow
install_kubeflow(api_client, app_dir, namespace)
wait_for_kubeflow_install(api_client, namespace)
# change the namespace to default to set up nfs-volume and nfs-server
namespace = "default"
# Deploy NFS for kubebench
install_kubebench_nfs(api_client, app_dir, namespace)
def copy_dataset():
"""Install Benchmark Addons."""
logging.basicConfig(
level=logging.INFO,
format=('%(asctime)s %(name)-12s %(levelname)-8s %(message)s'),
datefmt='%Y-%m-%dT%H:%M:%S',
)
logging.getLogger().setLevel(logging.INFO)
args = parse_args()
namespace = args.namespace
kubeconfig_path = str(os.environ['KUBECONFIG'])
api_client = deploy_utils.create_k8s_client(kubeconfig_path)
batchv1_api = k8s_client.BatchV1Api(api_client)
# need pvc, s3 bucket dataset name
deployment = create_job_object(args.runner_image, args.region,
args.s3_import_path, args.pvc)
batchv1_api.create_namespaced_job(namespace, deployment)
# describe
logging.info("Wait for data copy finish.")
wait_for_job(api_client, namespace, "copy-dataset-worker")
logging.info("Finish copy data from %s to pvc %s", args.s3_import_path,
args.pvc)
def install_addon():
"""Install Benchmark Addons."""
logging.basicConfig(level=logging.INFO,
format=('%(asctime)s %(name)-12s %(levelname)-8s %(message)s'),
datefmt='%Y-%m-%dT%H:%M:%S',
)
logging.getLogger().setLevel(logging.INFO)
args = parse_args()
namespace = args.namespace
base_dir = args.base_dir
kubeconfig_path = str(os.environ['KUBECONFIG'])
api_client = deploy_utils.create_k8s_client(kubeconfig_path)
# Setup GPU Device Plugin
install_gpu_drivers(api_client)
wait_for_gpu_driver_install(api_client)
def install_addon():
"""Install Benchmark Addons."""
logging.basicConfig(
level=logging.INFO,
format=('%(asctime)s %(name)-12s %(levelname)-8s %(message)s'),
datefmt='%Y-%m-%dT%H:%M:%S',
)
logging.getLogger().setLevel(logging.INFO)
args = parse_args()
namespace = args.namespace
kubeconfig_path = str(os.environ['KUBECONFIG'])
api_client = deploy_utils.create_k8s_client(kubeconfig_path)
# Deploy Github Secret. Can be passed from user's parameter
github_token = str(os.environ['GITHUB_TOKEN'])
install_github_secret(api_client, namespace, args.github_secret_name,
base64.b64encode(github_token))
def install_addon():
"""Install Benchmark Addons."""
logging.basicConfig(
level=logging.INFO,
format=('%(asctime)s %(name)-12s %(levelname)-8s %(message)s'),
datefmt='%Y-%m-%dT%H:%M:%S',
)
logging.getLogger().setLevel(logging.INFO)
args = parse_args()
namespace = args.namespace
kubeconfig_path = str(os.environ['KUBECONFIG'])
api_client = deploy_utils.create_k8s_client(kubeconfig_path)
# Deploy AWS Secret. Get from environment.
access_key_id = str(os.environ['AWS_ACCESS_KEY_ID'])
access_key = str(os.environ['AWS_SECRET_ACCESS_KEY'])
install_aws_secret(api_client, namespace, args.aws_secret_name,
base64.b64encode(access_key_id),
base64.b64encode(access_key))
def run_benchmark_job():
"""Submit benchmark jobs to remote kubernetes cluster."""
args = parse_args()
app_dir = os.path.join(str(os.environ['BENCHMARK_DIR']), "ks-app")
kubeconfig_path = str(os.environ['KUBECONFIG'])
api_client = deploy_utils.create_k8s_client(kubeconfig_path)
namespace = args.namespace
job_name = args.experiment_name
# Set the namespace of kb job to default
namespace = "default"
# Deploy Kubebench
util.run(
["ks", "generate", "kubebench-job", job_name, "--name=" + job_name],
cwd=app_dir)
job_config_prefix = "ks param set " + job_name + " "
cmd = job_config_prefix + "mainJobKsRegistry " + args.training_job_registry
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "mainJobKsPackage " + args.training_job_pkg
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "mainJobKsPrototype " + args.training_job_prototype
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "mainJobConfig " + args.training_job_config
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "awsCredentialsSecret " + args.aws_secret
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "awsCredentialsSecretAccessKeyId " + args.aws_access_key_id
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "awsCredentialsSecretAccessKey " + args.aws_secret_access_key
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "awsRegion " + args.aws_region
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "githubTokenSecret " + args.github_secret_name
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "githubTokenSecretKey GITHUB_TOKEN"
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "controllerImage seedjeffwan/configurator:20190415"
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "postJobImage seedjeffwan/mpi-post-processor:logs"
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "postJobArgs null"
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "reporterType null"
util.run(cmd.split(), cwd=app_dir)
cmd = job_config_prefix + "experimentDataPvc " + args.data_pvc
util.run(cmd.split(), cwd=app_dir)
# cmd = "ks param set " + job_name + " config_args -- --config-file=" + pvc_mount + \
# "/config/" + config_name + ".yaml"
# util.run(cmd.split(), cwd=app_dir)
# cmd = "ks param set " + job_name + " report_args -- --output-file=" + pvc_mount + \
# "/output/results.csv"
# util.run(cmd.split(), cwd=app_dir)
apply_command = ["ks", "apply", "default", "-c", job_name]
util.run(apply_command, cwd=app_dir)
# TODO: expose timeout setting here.
deploy_utils.wait_for_benchmark_job(job_name, namespace)
deploy_utils.cleanup_benchmark_job(app_dir, job_name)