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_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_generate_key_when_location_is_taken():
"""Key location is taken, must fall back."""
user = LOCAL_USER_7
random.seed(571303)
with set_up_key_location(user):
def get_expected_path(file_name: str) -> str:
return os.path.join(os.environ['IDACT_KEY_LOCATION'], file_name)
files_to_create = ['id_rsa_in',
'id_rsa_sy.pub',
'id_rsa_ov',
'id_rsa_ov.pub',
'id_rsa_d9']
for file in files_to_create:
with open(get_expected_path(file), 'w'):
pass
assert generate_key(host='host0', key_type=KeyType.RSA) == (
get_expected_path('id_rsa_wg9o'))
assert generate_key(host='host1', key_type=KeyType.RSA) == (
get_expected_path('id_rsa_t1'))
check_key_pair(host='host0',
path=get_expected_path('id_rsa_wg9o'))
check_key_pair(host='host1',
path=get_expected_path('id_rsa_t1'))
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_get_free_private_key_location_when_location_is_taken():
"""Some paths are taken, use fallbacks."""
user = LOCAL_USER_4
random.seed(571303)
assert KEY_NAME_SUFFIX_RETRIES == 4
with set_up_key_location(user):
def get_expected_path(file_name: str) -> str:
return os.path.join(os.environ['IDACT_KEY_LOCATION'], file_name)
files_to_create = [
'id_rsa_in', 'id_rsa_sy.pub', 'id_rsa_ov', 'id_rsa_ov.pub',
'id_rsa_d9'
]
for file in files_to_create:
with open(get_expected_path(file), 'w'):
pass
def get_path():
return get_free_private_key_location(key_type=KeyType.RSA)
assert get_path() == get_expected_path('id_rsa_wg9o')
assert get_path() == get_expected_path('id_rsa_t1')
assert get_path() == get_expected_path('id_rsa_4j')
def test_get_free_private_key_location_when_all_locations_are_taken():
"""All generated file paths are taken, even with fallbacks."""
user = LOCAL_USER_5
random.seed(571303)
with set_up_key_location(user):
def get_expected_path(file_name: str) -> str:
return os.path.join(os.environ['IDACT_KEY_LOCATION'], file_name)
files_to_create = [
'id_rsa_in', 'id_rsa_sy', 'id_rsa_ov', 'id_rsa_d9', 'id_rsa_wg9o',
'id_rsa_t14j', 'id_rsa_dwud', 'id_rsa_oa2v', 'id_rsa_9tybnbx2',
'id_rsa_aspfb1np', 'id_rsa_vy5345nr', 'id_rsa_c7gl4y4p',
'id_rsa_3w95f8974vbce8um', 'id_rsa_a56h9p6ez6e5p3pk',
'id_rsa_vx3bq5hwplvbwzgo', 'id_rsa_1ypfko2yz2vypimt',
'id_rsa_h8sqyu1avr7f5hwo3geil5nt8rkb9rx4',
'id_rsa_4xq0hw5qvhb9edcp9o2g5id1wmjbq1ro',
'id_rsa_f7c2epry1u1qzpac4hb7gymb3r3s3iex',
'id_rsa_im4sg2s7euj8j4h2no03qvhel4lcg9l1'
]
for file in files_to_create:
with open(get_expected_path(file), 'w'):
pass
with pytest.raises(RuntimeError):
print(get_free_private_key_location(key_type=KeyType.RSA))
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_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_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_generate_and_install_key_on_access_node():
with ExitStack() as stack:
user = USER_8
stack.enter_context(clear_environment(user))
stack.enter_context(set_up_key_location(user))
stack.enter_context(disable_pytest_stdin())
with pytest.raises(ValueError): # No key provided.
add_cluster(name=TEST_CLUSTER,
user=user,
host=get_testing_host(),
port=get_testing_port(),
auth=AuthMethod.PUBLIC_KEY,
key=None,
install_key=True)
# Generate RSA key.
add_cluster(name=TEST_CLUSTER,
user=user,
host=get_testing_host(),
port=get_testing_port(),
auth=AuthMethod.PUBLIC_KEY,
key=KeyType.RSA,
install_key=True,
retries={Retry.PORT_INFO: set_retry(count=0)})
check_remote_key_and_node_access(stack=stack, user=user)
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 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_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_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 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_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_generate_and_install_key_no_sshd():
with ExitStack() as stack:
user = USER_14
stack.enter_context(clear_environment(user))
stack.enter_context(set_up_key_location(user))
stack.enter_context(disable_pytest_stdin())
add_cluster(name=TEST_CLUSTER,
user=user,
host=get_testing_host(),
port=get_testing_port(),
auth=AuthMethod.PUBLIC_KEY,
key=KeyType.RSA,
install_key=True,
disable_sshd=True,
retries={Retry.PORT_INFO: set_retry(count=0)})
check_remote_key_and_node_access(stack=stack, user=user)
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_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')
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_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_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_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_get_free_private_key_location_when_location_is_free():
"""Free paths."""
user = LOCAL_USER_3
random.seed(571303)
result_paths = []
with set_up_key_location(user):
for _ in range(8):
result_paths += [
get_free_private_key_location(key_type=KeyType.RSA)
]
expected_file_names = [
'id_rsa_in', 'id_rsa_sy', 'id_rsa_ov', 'id_rsa_d9', 'id_rsa_wg',
'id_rsa_9o', 'id_rsa_t1', 'id_rsa_4j'
]
expected_paths = [
os.path.join(os.environ['IDACT_KEY_LOCATION'], i)
for i in expected_file_names
]
assert result_paths == expected_paths
def test_can_read_node_resources():
user = USER_39
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)
access_node = cluster.get_access_node()
print(str(access_node))
assert str(access_node) == repr(access_node)
assert access_node.resources.cpu_cores is None
assert access_node.resources.memory_total is None
start_stress_cpu(user=user, timeout=10)
try:
check_resources_in_believable_range(access_node.resources)
finally:
stop_stress_cpu(user=user)
nodes = cluster.allocate_nodes(cores=1,
memory_per_node=bitmath.GiB(0.8))
assert len(nodes) == 1
node = nodes[0]
stack.enter_context(cancel_on_exit(nodes))
nodes.wait(timeout=SLURM_WAIT_TIMEOUT)
assert nodes.running()
assert node.resources.cpu_cores == 1
assert node.resources.memory_total == bitmath.GiB(0.8)
start_stress_cpu(user=user, timeout=10)
try:
check_resources_in_believable_range(access_node.resources)
finally:
stop_stress_cpu(user=user)
assert node.run('whoami') == user
def run_notebook_app(user: str,
environment_file: str,
args: List[str],
notebook_defaults: dict = None) -> Result:
"""Runs the notebook app. Returns the result from CliRunner."""
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))
stack.enter_context(set_password(get_test_user_password(user)))
cluster = show_cluster(name=TEST_CLUSTER)
cluster.get_access_node().connect()
set_log_level(logging.INFO)
if notebook_defaults:
config = cluster.config
assert isinstance(config, ClusterConfigImpl)
config.notebook_defaults = notebook_defaults
save_environment(path=environment_file)
try:
runner = CliRunner()
opened_in_browser = []
def fake_open_in_browser(_):
opened_in_browser.append(True)
saved_open_in_browser = JupyterDeploymentImpl.open_in_browser
JupyterDeploymentImpl.open_in_browser = fake_open_in_browser
try:
os.environ[NOTEBOOK_TEST_RUN_ENVIRONMENT_VARIABLE] = ''
result = runner.invoke(main, args=args)
finally:
JupyterDeploymentImpl.open_in_browser = saved_open_in_browser
del os.environ[NOTEBOOK_TEST_RUN_ENVIRONMENT_VARIABLE]
print("\n\n\nClick output of the notebook app run:")
print(result.output)
yield result
finally:
os.remove(environment_file)
