in src/nccl_healthcheck/nccl_startup.py [0:0]
def main() -> None:
"""Main function."""
ensure_env_variables()
node_name = os.environ["NODE_NAME"]
remove_nccl_labels(node_name)
apply_namespace_resolution()
configure_ssh()
nhosts = int(os.environ["NHOSTS"])
nr = os.environ["nr"]
with open("/host.name", "w") as f:
f.write(node_name)
hosts = get_host_list(nhosts)
create_hostfile(hosts, nr)
run_nccl_test(hosts)
cleanup(hosts)
print("my job is done")