Skip to content

Update sagemaker containers #119

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Nov 19, 2018
Merged

Update sagemaker containers #119

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
960ad52
Update sagemaker containers
mvsusp Nov 19, 2018
a99251d
Add tests
mvsusp Nov 19, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@ def read(fname):
'Programming Language :: Python :: 3.6',
],

install_requires=['sagemaker-containers>=2.2.6', 'numpy', 'scipy', 'sklearn',
install_requires=['sagemaker-containers>==2.3.1', 'numpy', 'scipy', 'sklearn',
'pandas', 'Pillow', 'h5py'],
extras_require={
'test': ['tox', 'flake8', 'pytest', 'pytest-cov', 'pytest-xdist', 'mock',
Expand Down
18 changes: 6 additions & 12 deletions src/sagemaker_tensorflow_container/training.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,18 +99,12 @@ def _env_vars_with_tf_config(env, ps_task):

def _run_ps(env):
env_vars = _env_vars_with_tf_config(env, ps_task=True)
return framework.modules.run_module(
env.module_dir, env.to_cmd_args(), env_vars, env.module_name, wait=False)
framework.entry_point.run(env.module_dir, env.user_entry_point, env.to_cmd_args(), env_vars)


def _run_worker(env, install_module=False):
def _run_worker(env):
env_vars = _env_vars_with_tf_config(env, ps_task=False)
if install_module:
return framework.modules.run_module(
env.module_dir, env.to_cmd_args(), env_vars, env.module_name)
else:
framework.modules.write_env_vars(env_vars)
framework.modules.run(env.module_name, env.to_cmd_args(), env_vars)
framework.entry_point.run(env.module_dir, env.user_entry_point, env.to_cmd_args(), env_vars)


def _wait_until_master_is_down(master):
Expand Down Expand Up @@ -139,14 +133,14 @@ def train(env):
logger.info('Launching parameter server process')
_run_ps(env)
logger.info('Launching worker process')
_run_worker(env, install_module=False)
_run_worker(env)

if not _is_host_master(env.hosts, env.current_host):
_wait_until_master_is_down(env.hosts[0])

else:
framework.modules.run_module(env.module_dir, env.to_cmd_args(),
env.to_env_vars(), env.module_name)
framework.entry_point.run(env.module_dir, env.user_entry_point,
env.to_cmd_args(), env.to_env_vars())


def main():
Expand Down
176 changes: 89 additions & 87 deletions test/unit/test_training.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@

import json
import os
import subprocess

from mock import MagicMock, patch
import pytest
Expand Down Expand Up @@ -43,27 +44,24 @@

@pytest.fixture
def distributed_training_env():
env = MagicMock()

env.module_dir = MODULE_DIR
env.module_name = MODULE_NAME
env.hyperparameters = {}
env.log_level = LOG_LEVEL
env.hosts = HOST_LIST
env.current_host = CURRENT_HOST
env.additional_framework_parameters = {
training.SAGEMAKER_PARAMETER_SERVER_ENABLED: True
}

return env
return MagicMock(module_dir=MODULE_DIR,
user_entry_point=MODULE_NAME,
hyperparameters={},
log_level=LOG_LEVEL,
hosts=HOST_LIST,
current_host=CURRENT_HOST,
to_env_vars=lambda: {},
additional_framework_parameters={
training.SAGEMAKER_PARAMETER_SERVER_ENABLED: True
})


@pytest.fixture
def single_machine_training_env():
env = MagicMock()

env.module_dir = MODULE_DIR
env.module_name = MODULE_NAME
env.user_entry_point = MODULE_NAME
env.hyperparameters = {'model_dir': MODEL_DIR}
env.log_level = LOG_LEVEL

Expand All @@ -76,48 +74,87 @@ def test_is_host_master():
assert training._is_host_master(HOST_LIST, 'somehost') is False


@patch('sagemaker_containers.beta.framework.modules.run_module')
@patch('sagemaker_containers.beta.framework.entry_point.run')
def test_single_machine(run_module, single_machine_training_env):
training.train(single_machine_training_env)
run_module.assert_called_with(MODULE_DIR, single_machine_training_env.to_cmd_args(),
single_machine_training_env.to_env_vars(), MODULE_NAME)
run_module.assert_called_with(MODULE_DIR, MODULE_NAME,
single_machine_training_env.to_cmd_args(),
single_machine_training_env.to_env_vars())


@patch('sagemaker_tensorflow_container.training._wait_until_master_is_down')
@patch('sagemaker_tensorflow_container.training._run_worker')
@patch('sagemaker_tensorflow_container.training._run_ps')
def test_train_distributed_master(run_ps,
run_worker,
wait_until_master_is_down,
distributed_training_env):
@patch('sagemaker_containers.beta.framework.entry_point.run')
@patch('time.sleep', MagicMock())
def test_train_distributed_master(run, distributed_training_env):
training.train(distributed_training_env)
run_ps.assert_called_with(distributed_training_env)
run_worker.assert_called_with(distributed_training_env, install_module=False)
wait_until_master_is_down.assert_not_called


@patch('sagemaker_tensorflow_container.training._wait_until_master_is_down')
@patch('sagemaker_tensorflow_container.training._run_worker')
@patch('sagemaker_tensorflow_container.training._run_ps')
def test_train_distributed_worker(run_ps,
run_worker,
wait_until_master_is_down,
ps_tf_config = '{"cluster": {' \
'"master": ["host1:2222"], ' \
'"ps": ["host1:2223", "host2:2223"], ' \
'"worker": ["host2:2222"]}, ' \
'"environment": "cloud", ' \
'"task": {"index": 0, "type": "ps"}}'

run.assert_any_call('s3://my/bucket', 'script_name',
distributed_training_env.to_cmd_args(),
{'TF_CONFIG': ps_tf_config})

master_tf_config = '{"cluster": {' \
'"master": ["host1:2222"], ' \
'"ps": ["host1:2223", "host2:2223"], ' \
'"worker": ["host2:2222"]}, ' \
'"environment": "cloud", ' \
'"task": {"index": 0, "type": "master"}}'

run.assert_called_with('s3://my/bucket', 'script_name',
distributed_training_env.to_cmd_args(),
{
'TF_CONFIG': master_tf_config})


@patch('subprocess.check_call')
@patch('time.sleep', MagicMock())
@patch('sagemaker_containers.beta.framework.entry_point.run')
def test_train_distributed_worker(run,
check_call,
distributed_training_env):
distributed_training_env.current_host = HOST2
check_call.side_effect = subprocess.CalledProcessError(returncode=1, cmd=[])

training.train(distributed_training_env)
run_ps.assert_called_with(distributed_training_env)
run_worker.assert_called_with(distributed_training_env, install_module=False)
wait_until_master_is_down.assert_called_with(HOST1)

ps_tf_config = '{"cluster": {' \
'"master": ["host1:2222"], ' \
'"ps": ["host1:2223", "host2:2223"], ' \
'"worker": ["host2:2222"]}, ' \
'"environment": "cloud", ' \
'"task": {"index": 1, "type": "ps"}}'

@patch('sagemaker_containers.beta.framework.modules.run_module')
def test_train_distributed_no_ps(run_module, distributed_training_env):
run.assert_any_call('s3://my/bucket', 'script_name',
distributed_training_env.to_cmd_args(),
{'TF_CONFIG': ps_tf_config})

master_tf_config = '{"cluster": {' \
'"master": ["host1:2222"], ' \
'"ps": ["host1:2223", "host2:2223"], ' \
'"worker": ["host2:2222"]}, ' \
'"environment": "cloud", ' \
'"task": {"index": 0, "type": "worker"}}'

run.assert_called_with('s3://my/bucket', 'script_name',
distributed_training_env.to_cmd_args(),
{
'TF_CONFIG': master_tf_config})


@patch('sagemaker_containers.beta.framework.entry_point.run')
def test_train_distributed_no_ps(run, distributed_training_env):
distributed_training_env.additional_framework_parameters[
training.SAGEMAKER_PARAMETER_SERVER_ENABLED] = False
distributed_training_env.current_host = HOST2
training.train(distributed_training_env)
run_module.assert_called_with(MODULE_DIR, distributed_training_env.to_cmd_args(),
distributed_training_env.to_env_vars(), MODULE_NAME)

run.assert_called_with(MODULE_DIR, MODULE_NAME, distributed_training_env.to_cmd_args(),
distributed_training_env.to_env_vars())


@patch('sagemaker_tensorflow_container.training._build_tf_config')
Expand All @@ -131,61 +168,26 @@ def test_get_env_vars_with_tf_config(build_tf_config, distributed_training_env):
hosts=HOST_LIST, current_host=CURRENT_HOST, ps_task=True)


@patch('sagemaker_containers.beta.framework.modules.run_module')
@patch('sagemaker_containers.beta.framework.entry_point.run')
@patch('sagemaker_tensorflow_container.training._env_vars_with_tf_config')
def test_run_ps(env_vars_with_tf_config, run_module, distributed_training_env):
env_vars_with_tf_config.return_value = {}
distributed_training_env.to_cmd_args.return_value = CMD_ARGS
def test_run_ps(env_vars_with_tf_config, run, distributed_training_env):
training._run_ps(distributed_training_env)
env_vars_with_tf_config.assert_called_once_with(distributed_training_env, ps_task=True)
run_module.assert_called_once_with(distributed_training_env.module_dir,
CMD_ARGS,
{},
distributed_training_env.module_name,
wait=False)


@patch('sagemaker_containers.beta.framework.modules.write_env_vars')
@patch('sagemaker_containers.beta.framework.modules.run')
@patch('sagemaker_tensorflow_container.training._env_vars_with_tf_config')
def test_run_worker_no_install(get_env_vars_with_tf_config,
run,
write_env_vars,
distributed_training_env):
get_env_vars_with_tf_config.return_value = {}
distributed_training_env.to_cmd_args.return_value = CMD_ARGS
training._run_worker(distributed_training_env, install_module=False)
get_env_vars_with_tf_config.assert_called_once_with(distributed_training_env, ps_task=False)
write_env_vars.assert_called_once_with({})
run.assert_called_once_with(distributed_training_env.module_name,
CMD_ARGS,
{})


@patch('sagemaker_containers.beta.framework.modules.run_module')
@patch('sagemaker_tensorflow_container.training._env_vars_with_tf_config')
def test_run_worker_install(get_env_vars_with_tf_config,
run_module,
distributed_training_env):
get_env_vars_with_tf_config.return_value = {}
distributed_training_env.to_cmd_args.return_value = CMD_ARGS
training._run_worker(distributed_training_env, install_module=True)
get_env_vars_with_tf_config.assert_called_once_with(distributed_training_env, ps_task=False)
run_module.assert_called_once_with(distributed_training_env.module_dir,
CMD_ARGS,
{},
distributed_training_env.module_name)
run.assert_called_once_with(distributed_training_env.module_dir,
distributed_training_env.user_entry_point,
distributed_training_env.to_cmd_args(), env_vars_with_tf_config())


def test_build_tf_config():
assert training._build_tf_config(HOST_LIST, HOST1) ==\
{'cluster': CLUSTER_WITH_PS, 'environment': 'cloud', 'task': MASTER_TASK}
assert training._build_tf_config(HOST_LIST, HOST1) == \
{'cluster': CLUSTER_WITH_PS, 'environment': 'cloud', 'task': MASTER_TASK}
assert training._build_tf_config(HOST_LIST, HOST1, ps_task=True) == \
{'cluster': CLUSTER_WITH_PS, 'environment': 'cloud', 'task': PS_TASK_1}
assert training._build_tf_config(HOST_LIST, HOST2) ==\
{'cluster': CLUSTER_WITH_PS, 'environment': 'cloud', 'task': WORKER_TASK}
{'cluster': CLUSTER_WITH_PS, 'environment': 'cloud', 'task': PS_TASK_1}
assert training._build_tf_config(HOST_LIST, HOST2) == \
{'cluster': CLUSTER_WITH_PS, 'environment': 'cloud', 'task': WORKER_TASK}
assert training._build_tf_config(HOST_LIST, HOST2, ps_task=True) == \
{'cluster': CLUSTER_WITH_PS, 'environment': 'cloud', 'task': PS_TASK_2}
{'cluster': CLUSTER_WITH_PS, 'environment': 'cloud', 'task': PS_TASK_2}


def test_build_tf_config_error():
Expand Down
2 changes: 1 addition & 1 deletion tox.ini
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,7 @@ passenv =
# Can be used to specify which tests to run, e.g.: tox -- -s
commands =
coverage run --rcfile .coveragerc_{envname} --source sagemaker_tensorflow_container -m py.test {posargs}
{env:IGNORE_COVERAGE:} coverage report --fail-under=90 --include *sagemaker_tensorflow_container* --omit */tensorflow/tensorflow_serving/*
{env:IGNORE_COVERAGE:} coverage report --fail-under=90 --include *sagemaker_tensorflow_container* --show-missing
deps = .[test]

[testenv:flake8]
Expand Down