def test_mpi_worker_run_no_wait(popen, ssh_client):
worker = _mpi.WorkerRunner(user_entry_point='train.sh',
args=['-v', '--lr', '35'],
env_vars={'LD_CONFIG_PATH': '/etc/ld'},
master_hostname='algo-1')
worker.run(wait=False)
ssh_client.assert_not_called()
popen.assert_called_with(['/usr/sbin/sshd', '-D'])
def test_mpi_worker_run_no_wait(popen, ssh_client):
worker = _mpi.WorkerRunner(
user_entry_point="train.sh",
args=["-v", "--lr", "35"],
env_vars={"LD_CONFIG_PATH": "/etc/ld"},
master_hostname="algo-1",
)
worker.run(wait=False)
ssh_client.assert_not_called()
popen.assert_called_with(["/usr/sbin/sshd", "-D"])
def test_mpi_worker_run(popen, policy, process_iter, wait_procs, ssh_client):
process = MagicMock(info={'name': 'orted'})
process_iter.side_effect = lambda attrs: [process]
worker = _mpi.WorkerRunner(user_entry_point='train.sh',
args=['-v', '--lr', '35'],
env_vars={'LD_CONFIG_PATH': '/etc/ld'},
master_hostname='algo-1')
worker.run()
ssh_client().load_system_host_keys.assert_called()
ssh_client().set_missing_host_key_policy.assert_called_with(policy())
ssh_client().connect.assert_called_with('algo-1', port=22)
ssh_client().close.assert_called()
wait_procs.assert_called_with([process])
popen.assert_called_with(['/usr/sbin/sshd', '-D'])
def _get_by_runner_type(identifier,
user_entry_point=None,
args=None,
env_vars=None,
extra_opts=None):
"""Placeholder docstring"""
env = sagemaker_containers.training_env()
user_entry_point = user_entry_point or env.user_entry_point
args = args or env.to_cmd_args()
env_vars = env_vars or env.to_env_vars()
if identifier is RunnerType.MPI and env.is_master:
mpi_args = extra_opts or {}
# Default to single process for CPU
default_processes_per_host = env.num_gpus if env.num_gpus > 0 else 1
processes_per_host = _mpi_param_value(mpi_args, env,
_params.MPI_PROCESSES_PER_HOST,
default_processes_per_host)
num_processes = _mpi_param_value(mpi_args, env,
_params.MPI_NUM_PROCESSES)
custom_mpi_options = _mpi_param_value(mpi_args, env,
_params.MPI_CUSTOM_OPTIONS, "")
return _mpi.MasterRunner(
user_entry_point,
args,
env_vars,
env.master_hostname,
env.hosts,
processes_per_host,
custom_mpi_options,
env.network_interface_name,
num_processes=num_processes,
)
elif identifier is RunnerType.MPI:
return _mpi.WorkerRunner(user_entry_point, args, env_vars,
env.master_hostname)
elif identifier is RunnerType.Process:
return _process.ProcessRunner(user_entry_point, args, env_vars)
else:
raise ValueError("Invalid identifier %s" % identifier)
def _get_by_runner_type(identifier):
env = sagemaker_containers.training_env()
if identifier is RunnerType.MPI and env.is_master:
processes_per_host = env.additional_framework_parameters.get(
_params.MPI_PROCESSES_PER_HOST, 1)
custom_mpi_options = env.additional_framework_parameters.get(
_params.MPI_CUSTOM_OPTIONS, '')
return _mpi.MasterRunner(env.user_entry_point, env.to_cmd_args(),
env.to_env_vars(), env.master_hostname,
env.hosts, processes_per_host,
custom_mpi_options,
env.network_interface_name)
elif identifier is RunnerType.MPI:
return _mpi.WorkerRunner(env.user_entry_point, env.to_cmd_args(),
env.to_env_vars(), env.master_hostname)
elif identifier is RunnerType.Process:
return _process.ProcessRunner(env.user_entry_point, env.to_cmd_args(),
env.to_env_vars())
else:
raise ValueError('Invalid identifier %s' % identifier)
def test_mpi_worker_run(popen, policy, process_iter, wait_procs, ssh_client, sleep):
process = MagicMock(info={"name": "orted"})
process_iter.side_effect = lambda attrs: [process]
worker = _mpi.WorkerRunner(
user_entry_point="train.sh",
args=["-v", "--lr", "35"],
env_vars={"LD_CONFIG_PATH": "/etc/ld"},
master_hostname="algo-1",
)
worker.run()
ssh_client().load_system_host_keys.assert_called()
ssh_client().set_missing_host_key_policy.assert_called_with(policy())
ssh_client().connect.assert_called_with("algo-1", port=22)
ssh_client().close.assert_called()
wait_procs.assert_called_with([process])
popen.assert_called_with(["/usr/sbin/sshd", "-D"])
def _get_by_runner_type(identifier,
user_entry_point=None,
args=None,
env_vars=None,
extra_opts=None):
env = sagemaker_containers.training_env()
user_entry_point = user_entry_point or env.user_entry_point
args = args or env.to_cmd_args()
env_vars = env_vars or env.to_env_vars()
if identifier is RunnerType.MPI and env.is_master:
mpi_args = extra_opts or {}
processes_per_host = _mpi_param_value(mpi_args, env,
_params.MPI_PROCESSES_PER_HOST,
1)
num_processes = _mpi_param_value(mpi_args, env,
_params.MPI_NUM_PROCESSES)
custom_mpi_options = _mpi_param_value(mpi_args, env,
_params.MPI_CUSTOM_OPTIONS, '')
return _mpi.MasterRunner(user_entry_point,
args,
env_vars,
env.master_hostname,
env.hosts,
processes_per_host,
custom_mpi_options,
env.network_interface_name,
num_processes=num_processes)
elif identifier is RunnerType.MPI:
return _mpi.WorkerRunner(user_entry_point, args, env_vars,
env.master_hostname)
elif identifier is RunnerType.Process:
return _process.ProcessRunner(user_entry_point, args, env_vars)
else:
raise ValueError('Invalid identifier %s' % identifier)
