def test_environment():
user = USER_2
test_environment_file = get_test_environment_file(user=user)
with ExitStack() as stack:
stack.enter_context(clear_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
clusters = show_clusters()
assert clusters == {}
cluster = add_cluster(name=TEST_CLUSTER,
user=user,
host='localhost',
port=2222)
clusters = show_clusters()
assert show_cluster(name=TEST_CLUSTER) is cluster
assert len(show_clusters()) == 1
assert clusters[TEST_CLUSTER] == cluster
assert cluster.name == TEST_CLUSTER
try:
save_environment(path=test_environment_file)
with clear_environment(user):
assert show_clusters() == {}
load_environment(path=test_environment_file)
cluster2 = show_cluster(name=TEST_CLUSTER)
assert cluster2 is not cluster
assert cluster2 == cluster
finally:
os.remove(test_environment_file)
def test_environment_create_modify_save_load():
user = USER_19
test_environment_file = get_test_environment_file(user=user)
test_environment_file2 = get_test_environment_file(user=user) + '_2'
with ExitStack() as stack:
stack.enter_context(clear_environment(user))
assert show_clusters() == {}
add_cluster(name=TEST_CLUSTER, user=user, host='localhost')
config = show_cluster(TEST_CLUSTER).config
check_config_is_default(config=config, user=user)
try:
save_environment(path=test_environment_file)
with open(test_environment_file, 'r') as test_file:
contents = test_file.read().splitlines()
pprint(contents)
assert contents == get_default_config_contents(user=user)
config = show_cluster(TEST_CLUSTER).config
config.host = 'localhost2'
config.port = 2222
config.user = '******'
config.auth = AuthMethod.PUBLIC_KEY
config.key = './fake-key'
config.install_key = False
config.disable_sshd = True
config.setup_actions.jupyter = ['abc']
config.setup_actions.dask = ['abc', 'def']
config.scratch = '$HOME2'
config.use_jupyter_lab = False
set_log_level(logging.INFO)
check_config_is_modified(config=config)
save_environment(path=test_environment_file2)
with open(test_environment_file2, 'r') as test_file:
contents = test_file.read().splitlines()
pprint(contents)
assert contents == get_modified_config_contents()
load_environment(test_environment_file)
config = show_cluster(TEST_CLUSTER).config
check_config_is_default(config=config, user=user)
load_environment(test_environment_file2)
config = show_cluster(TEST_CLUSTER).config
check_config_is_modified(config=config)
finally:
try:
os.remove(test_environment_file)
finally:
os.remove(test_environment_file2)
def test_node_tunnel_fall_back_when_local_port_taken():
"""Checks that a tunnel will fall back to a random port if local port is
taken."""
user = USER_53
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes(nodes=1,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
stack.enter_context(cancel_on_exit(nodes))
node = nodes[0]
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
there = get_free_remote_port(node=node)
here = get_free_local_port()
tunnel_1 = node.tunnel(there=there, here=here)
stack.enter_context(close_tunnel_on_exit(tunnel_1))
assert tunnel_1.here == here
tunnel_2 = node.tunnel(there=there, here=here)
stack.enter_context(close_tunnel_on_exit(tunnel_2))
assert tunnel_2.here != here
def check_remote_key_and_node_access(stack: ExitStack, user: str):
public_key_value = get_public_key_value()
cluster = show_cluster(name=TEST_CLUSTER)
node = cluster.get_access_node()
with set_password(get_test_user_password(user)):
assert node.run('whoami') == user
assert node.run('whoami') == user
node.run("grep '{public_key_value}' ~/.ssh/authorized_keys".format(
public_key_value=public_key_value))
with pytest.raises(RuntimeError):
node.run(
"grep '{public_key_value}' ~/.ssh/authorized_keys.idact".format(
public_key_value=public_key_value))
nodes = cluster.allocate_nodes(nodes=2,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
stack.enter_context(cancel_on_exit(nodes))
print(nodes)
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
node.run("grep '{public_key_value}' ~/.ssh/authorized_keys.idact".format(
public_key_value=public_key_value))
# Access to node without password works.
assert nodes[0].run('whoami') == user
check_direct_access_from_access_node_does_not_work(nodes[0])
def test_able_to_sync_dask():
user = USER_55
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
stack.enter_context(clear_deployment_sync_data(user))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes()
stack.enter_context(cancel_on_exit(nodes))
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
dask = deploy_dask(nodes)
stack.enter_context(cancel_on_exit(dask))
deployments = cluster.pull_deployments()
assert not deployments.dask_deployments
cluster.push_deployment(deployment=dask)
deployments = cluster.pull_deployments()
print(deployments)
assert len(deployments.dask_deployments) == 1
dask_2 = deployments.dask_deployments[0]
try:
assert dask.diagnostics.addresses != dask_2.diagnostics.addresses
for url in dask.diagnostics.addresses:
check_http_connection(url=url)
for url in dask_2.diagnostics.addresses:
check_http_connection(url=url)
finally:
dask_2.cancel_local()
def test_dask_deployment_with_redeploy_failure():
user = USER_42
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes(nodes=2,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
stack.enter_context(cancel_on_exit(nodes))
assert idact.detail.dask.deploy_dask_impl.validate_worker
stored_validate_worker = \
idact.detail.dask.deploy_dask_impl.validate_worker
def fake_validate_worker(worker: DaskWorkerDeployment):
print("Fake worker validation.")
raise ValueError("Fake worker validation fail.")
try:
idact.detail.dask.deploy_dask_impl.validate_worker = \
fake_validate_worker
with pytest.raises(RuntimeError):
with deploy_dask_on_testing_cluster(nodes):
pass
finally:
idact.detail.dask.deploy_dask_impl.validate_worker = \
stored_validate_worker
def test_able_to_sync_jupyter():
user = USER_47
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
stack.enter_context(clear_deployment_sync_data(user))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes()
stack.enter_context(cancel_on_exit(nodes))
node = nodes[0]
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
local_port = get_free_local_port()
jupyter = node.deploy_notebook(local_port=local_port)
stack.enter_context(cancel_on_exit(jupyter))
deployments = cluster.pull_deployments()
assert not deployments.jupyter_deployments
cluster.push_deployment(deployment=jupyter)
deployments = cluster.pull_deployments()
print(deployments)
assert len(deployments.jupyter_deployments) == 1
jupyter_2 = deployments.jupyter_deployments[0]
try:
assert jupyter.local_port != jupyter_2.local_port
check_local_http_connection(port=jupyter.local_port)
check_local_http_connection(port=jupyter_2.local_port)
finally:
jupyter_2.cancel_local()
def test_cancelled_dask_allocation_is_discarded_on_pull():
user = USER_56
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
stack.enter_context(clear_deployment_sync_data(user))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes()
stack.enter_context(cancel_on_exit(nodes))
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
dask = deploy_dask(nodes)
stack.enter_context(cancel_on_exit(dask))
try:
deployments = cluster.pull_deployments()
assert not deployments.jupyter_deployments
cluster.push_deployment(deployment=dask)
dask.cancel()
dask = None
deployments = cluster.pull_deployments()
assert not deployments.jupyter_deployments
finally:
if dask is not None:
dask.cancel()
def test_cancelled_jupyter_allocation_is_discarded_on_pull():
user = USER_48
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
stack.enter_context(clear_deployment_sync_data(user))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes()
stack.enter_context(cancel_on_exit(nodes))
node = nodes[0]
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
local_port = get_free_local_port()
jupyter = node.deploy_notebook(local_port=local_port)
try:
deployments = cluster.pull_deployments()
assert not deployments.jupyter_deployments
cluster.push_deployment(deployment=jupyter)
jupyter.cancel()
jupyter = None
deployments = cluster.pull_deployments()
assert not deployments.jupyter_deployments
finally:
if jupyter is not None:
jupyter.cancel()
def test_cancelled_node_allocation_is_discarded_on_pull():
user = USER_45
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
stack.enter_context(clear_deployment_sync_data(user))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes()
try:
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
assert nodes.running()
cluster.push_deployment(deployment=nodes)
deployments = cluster.pull_deployments()
assert len(deployments.nodes) == 1
assert deployments.nodes[0].running()
nodes.cancel()
nodes = None
deployments = cluster.pull_deployments()
assert not deployments.nodes
finally:
if nodes is not None:
nodes.cancel()
def test_nodes_sync_does_not_work_when_waiting_twice():
"""Port info was already deleted, so waiting for the second time defaults
to port 22."""
user = USER_44
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
stack.enter_context(clear_deployment_sync_data(user))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes()
with cancel_on_exit(nodes):
cluster.push_deployment(deployment=nodes)
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
assert nodes.running()
node = nodes[0]
assert node.port != 22
deployments = cluster.pull_deployments()
assert len(deployments.nodes) == 1
nodes_2 = deployments.nodes[0]
nodes_2.wait(timeout=SLURM_WAIT_TIMEOUT)
assert nodes_2.running()
node_2 = nodes_2[0]
assert node_2.port == 22
assert node_2.host == node.host
def test_remove_runtime_dir_test():
user = USER_15
with ExitStack() as stack:
stack.enter_context(set_up_key_location(user))
stack.enter_context(disable_pytest_stdin())
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes(nodes=1,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
stack.enter_context(cancel_on_exit(nodes))
node = nodes[0]
try:
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
assert nodes.running()
check_will_remove_empty(node=node)
check_will_ignore_non_existent(node=node)
check_will_remove_files(node=node)
check_will_not_remove_dotfiles(node=node)
check_will_not_remove_nested_dirs(node=node)
finally:
node.run("rm -rf *")
def test_environment_missing_and_defaults():
user = USER_25
with ExitStack() as stack:
stack.enter_context(clear_environment(user))
EnvironmentProvider()._environment = None # noqa, pylint: disable=protected-access,line-too-long
assert show_clusters() == {}
assert not os.path.isfile(os.environ['IDACT_CONFIG_PATH'])
with pytest.raises(ValueError):
load_environment()
add_cluster(name=TEST_CLUSTER, user=user, host='localhost')
config = show_cluster(TEST_CLUSTER).config
set_log_level(logging.DEBUG)
check_config_is_default(config=config, user=user)
try:
save_environment()
with open(os.environ['IDACT_CONFIG_PATH'], 'r') as test_file:
contents = test_file.read().splitlines()
pprint(contents)
assert contents == get_default_config_contents(user=user)
finally:
os.remove(os.environ['IDACT_CONFIG_PATH'])
def test_able_to_sync_nodes_before_and_after_wait():
user = USER_43
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
stack.enter_context(clear_deployment_sync_data(user))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes()
nodes_2 = None
nodes_3 = None
try:
deployments = cluster.pull_deployments()
assert not deployments.nodes
cluster.push_deployment(deployment=nodes)
nodes = None
deployments = cluster.pull_deployments()
print(deployments)
assert len(deployments.nodes) == 1
nodes_2 = deployments.nodes[0]
assert len(nodes_2) == 1
nodes_2.wait(timeout=SLURM_WAIT_TIMEOUT)
assert nodes_2.running()
node_2 = nodes_2[0]
cluster.push_deployment(deployment=nodes_2)
nodes_2 = None
deployments = cluster.pull_deployments()
print(deployments)
assert len(deployments.nodes) == 1
nodes_3 = deployments.nodes[0]
assert nodes_3.running()
with pytest.raises(RuntimeError):
nodes_3.wait()
assert len(nodes_3) == 1
node_3 = nodes_3[0]
assert node_2.host == node_3.host
assert node_2.port == node_3.port
assert node_3.resources.cpu_cores == 1
assert node_3.resources.memory_total == bitmath.GiB(1)
print(node_3)
assert node_3.run('whoami') == user
finally:
if nodes is not None:
nodes.cancel()
if nodes_2 is not None:
nodes_2.cancel()
if nodes_3 is not None:
nodes_3.cancel()
def run_invalid_squeue_output_test(user: str, output: str):
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
assert extract_squeue_line(
now=utc_now(), line=output, node=cluster.get_access_node()) is None
def test_cannot_pull_environment_when_missing_on_cluster():
user = USER_27
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
with pytest.raises(RuntimeError):
pull_environment(cluster=cluster)
def test_cannot_deploy_dask_on_zero_nodes():
user = USER_20
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
print(show_cluster(name=TEST_CLUSTER))
with pytest.raises(ValueError):
deploy_dask(nodes=[])
def test_cannot_pull_environment_when_invalid_format_on_cluster():
user = USER_33
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
node = cluster.get_access_node()
node.run("echo abc > ~/idact-bad-config")
with pytest.raises(JSONDecodeError):
pull_environment(cluster=cluster, path="~/idact-bad-config")
def test_environment_add_cluster_and_remove():
user = USER_26
with ExitStack() as stack:
stack.enter_context(clear_environment(user))
assert show_clusters() == {}
add_cluster(name=TEST_CLUSTER, user=user, host='localhost')
add_cluster(name='fake cluster', user=user, host='localhost2')
assert len(show_clusters()) == 2
cluster = show_cluster('fake cluster')
assert cluster.config.host == 'localhost2'
remove_cluster('fake cluster')
assert len(show_clusters()) == 1
add_cluster(name='fake cluster', user=user, host='localhost3')
cluster2 = show_cluster('fake cluster')
assert cluster.config.host == 'localhost2'
assert cluster2.config.host == 'localhost3'
cluster3 = show_cluster(TEST_CLUSTER)
assert cluster3.config.host == 'localhost'
remove_cluster('fake cluster')
remove_cluster(TEST_CLUSTER)
assert show_clusters() == {}
with pytest.raises(KeyError):
remove_cluster('fake cluster')
def test_node_tunnel_stress():
user = USER_40
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes(nodes=1,
cores=1,
memory_per_node=MiB(100))
stack.enter_context(cancel_on_exit(nodes))
run_tunnel_stress_test(stack=stack, user=user, nodes=nodes)
def test_node_tunnel():
"""Allocates a node and creates a tunnel."""
user = USER_5
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes(nodes=1,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
run_tunnel_test(user=user, nodes=nodes)
def test_node_tunnel_fall_back_when_local_port_free_but_fails():
"""Checks that a tunnel will fall back to a random port if local port is
is initially free, but tunnel cannot be created anyway (e.g. another
process binds to it at the last moment)."""
user = USER_54
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes(nodes=1,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
stack.enter_context(cancel_on_exit(nodes))
node = nodes[0]
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
there = get_free_remote_port(node=node)
here = get_free_local_port()
real_build_tunnel = idact.detail.nodes.node_impl.build_tunnel
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
tries = [0]
def fake_build_tunnel(*args, **kwargs) -> TunnelInternal:
tries[0] += 1
if tries[0] == 1:
raise RuntimeError("Fake failure.")
if tries[0] != 2:
assert False
return real_build_tunnel(*args, **kwargs)
try:
idact.detail.nodes.node_impl.build_tunnel = fake_build_tunnel
tunnel = node.tunnel(there=there, here=here)
stack.enter_context(close_tunnel_on_exit(tunnel))
assert tries[0] == 2
assert tunnel.here != here
finally:
idact.detail.nodes.node_impl.build_tunnel = real_build_tunnel
sock.close()
def test_jupyter_deployment():
user = USER_6
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes(nodes=1,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
stack.enter_context(cancel_on_exit(nodes))
with deploy_jupyter(nodes):
pass
def test_node_tunnel_public_key():
"""Allocates a node and creates a tunnel, uses public key authentication.
"""
user = USER_13
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(
reset_environment(user=user, auth=AuthMethod.PUBLIC_KEY))
cluster = show_cluster(name=TEST_CLUSTER)
with set_password(get_test_user_password(user)):
nodes = cluster.allocate_nodes(nodes=1,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
run_tunnel_test(user=user, nodes=nodes)
def test_able_to_reach_nodes_when_using_password_based_authentication():
"""It should be possible to connect to compute nodes even when using
password-based authentication, because local public key is authorized
for the compute nodes after initial connection.
However, direct connection from access node should fail.
Password is still used between the client and the access node."""
user = USER_10
with ExitStack() as stack:
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user=user, auth=AuthMethod.ASK))
stack.enter_context(set_password(get_test_user_password(user)))
stack.enter_context(disable_pytest_stdin())
cluster = show_cluster(TEST_CLUSTER)
node = cluster.get_access_node()
nodes = cluster.allocate_nodes(nodes=2,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
stack.enter_context(cancel_on_exit(nodes))
print(nodes)
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
compute_node = nodes[0]
assert isinstance(compute_node, NodeInternal)
public_key_value = get_public_key_value()
# Local key was installed for the deployed sshd, allowing access
# between the access node and compute nodes.
assert nodes[0].run('whoami') == user
# Local key was not installed for the access node
with pytest.raises(RuntimeError):
node.run("grep '{public_key_value}' ~/.ssh/authorized_keys".format(
public_key_value=public_key_value))
# But it was installed for compute nodes.
node.run("grep '{public_key_value}'"
" ~/.ssh/authorized_keys.idact".format(
public_key_value=public_key_value))
check_direct_access_from_access_node_does_not_work(nodes[0])
def test_dask_deployment_with_redeploy_on_validation_failure():
user = USER_41
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes(nodes=2,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
stack.enter_context(cancel_on_exit(nodes))
assert idact.detail.dask.deploy_dask_impl.validate_worker
stored_validate_worker = \
idact.detail.dask.deploy_dask_impl.validate_worker
fake_validation_counter = [0]
# pylint: disable=unused-argument
def fake_validate_worker(worker: DaskWorkerDeployment):
current_count = fake_validation_counter[0]
fake_validation_counter[0] = current_count + 1
print("Fake worker validation.")
if current_count == 0:
raise RuntimeError("Fake worker validation: First node fail.")
print("Deciding the worker is valid.")
try:
idact.detail.dask.deploy_dask_impl.validate_worker = \
fake_validate_worker
with deploy_dask_on_testing_cluster(nodes):
pass
assert fake_validation_counter[0] == 3
finally:
idact.detail.dask.deploy_dask_impl.validate_worker = \
stored_validate_worker
def test_migrate_deployments():
"""Migrating from an old version of the deployments file should work
without fatal errors."""
user = USER_57
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
stack.enter_context(clear_deployment_sync_data(user))
cluster = show_cluster(name=TEST_CLUSTER)
access_node = cluster.get_access_node()
assert isinstance(access_node, NodeInternal)
def check_deployments_file_exists():
access_node.run("cat ~/.idact/.deployments")
nodes = cluster.allocate_nodes()
stack.enter_context(cancel_on_exit(nodes))
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
with pytest.raises(RuntimeError):
check_deployments_file_exists()
remote_path = access_node.run("echo ~/.idact/.deployments")
put_file_on_node(node=access_node,
remote_path=remote_path,
contents='{{"type": "{type}"}}'.format(
type=SerializableTypes.DEPLOYMENT_DEFINITIONS))
deployments = cluster.pull_deployments()
assert not deployments.nodes
cluster.push_deployment(deployment=nodes)
deployments = cluster.pull_deployments()
assert len(deployments.nodes) == 1
cluster.clear_pushed_deployments()
with pytest.raises(RuntimeError):
check_deployments_file_exists()
def test_dask_deployment_with_setup_actions():
user = USER_18
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
nodes = cluster.allocate_nodes(nodes=2,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30))
stack.enter_context(cancel_on_exit(nodes))
cluster.config.setup_actions.dask = ['echo ABC > file.txt',
'mv file.txt file2.txt']
with deploy_dask_on_testing_cluster(nodes) as node:
assert node.run("cat file2.txt") == "ABC"
def test_dask_deployment_with_absolute_scratch_path():
user = USER_24
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
cluster.config.scratch = '/home/user-24'
nodes = cluster.allocate_nodes(nodes=1,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=10))
stack.enter_context(cancel_on_exit(nodes))
with deploy_dask_on_testing_cluster(nodes):
pass
def test_basic():
user = USER_1
with ExitStack() as stack:
stack.enter_context(disable_pytest_stdin())
stack.enter_context(set_up_key_location(user))
stack.enter_context(reset_environment(user))
stack.enter_context(set_password(get_test_user_password(user)))
clusters = show_clusters()
print(clusters)
assert len(clusters) == 1
cluster = show_cluster(name=TEST_CLUSTER)
print(cluster)
assert clusters[TEST_CLUSTER] == cluster
nodes = cluster.allocate_nodes(nodes=2,
cores=1,
memory_per_node=MiB(100),
walltime=Walltime(minutes=30),
native_args={'--partition': 'debug'})
with cancel_on_exit(nodes):
assert len(nodes) == 2
assert nodes[0] in nodes
print(nodes)
assert str(nodes) == repr(nodes)
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
assert nodes.running()
print(nodes)
print(nodes[0])
assert nodes[0].run('whoami') == user
assert nodes[1].run('whoami') == user
assert not nodes.running()
with pytest.raises(RuntimeError):
nodes.wait()
with pytest.raises(RuntimeError):
nodes[0].run('whoami')
