def test_create_docker_task():
dp = DockerProvider()
docker = dp.docker
env_vars = {'hello': 'team'}
taskdef = TaskDefinition(
name=TEST_TASK,
image=TEST_IMAGE,
parent='parent',
env=env_vars,
inputs={
'hello': '123',
'child': False,
},
# disables any output.
# this is hacky and should be refactored
# we need a proper way to disable all logging
upstream='disabled',
)
# run task
task = dp.spawn(taskdef)
assert task.id == taskdef.id
assert hasattr(task, 'container')
assert hasattr(task.container, 'id')
# try to grab the container from docker api
container = docker.containers.get(task.container.id)
assert task.container == container
# make sure container is properly labeled
assert container.labels == {
LABEL_TASK_ID: task.id,
LABEL_PARENT_ID: 'parent',
}
# wait for container to execute
result = container.wait()
assert result['StatusCode'] == 0
# test task will dump info as json, so we can pick it up
# make sure it matches what we put in.
logs = container.logs()
task_dump = json.loads(logs)
# taskdef
assert taskdef.serialize() == task_dump['taskdef']
# actual environment variables
for key, val in env_vars.items():
assert task_dump['env'][key] == val
def test_create_docker_task():
dp = DockerProvider()
docker = dp.docker
env_vars = {'hello': 'team'}
taskdef = TaskDefinition(
name=TEST_TASK,
image=TEST_IMAGE,
parent='parent',
env=env_vars,
inputs={
'hello': '123',
'child': False,
},
)
# run task
task = dp.spawn(taskdef)
assert task.id == taskdef.id
assert hasattr(task, 'container')
assert hasattr(task.container, 'id')
# try to grab the container from docker api
container = docker.containers.get(task.container.id)
assert task.container == container
# make sure container is properly labeled
assert container.labels[LABEL_TASK_ID] == task.id
assert container.labels[LABEL_PARENT_ID] == 'parent'
# test task will dump info as json, so we can pick it up
# make sure it matches what we put in.
task_dump = None
for msg in task.logs():
print(msg)
if msg['type'] == 'task/log':
task_dump = json.loads(msg['data'])
# wait for container to execute
result = container.wait()
assert result['StatusCode'] == 0
# taskdef
assert task_dump is not None
assert taskdef.serialize() == task_dump['taskdef']
# actual environment variables
for key, val in env_vars.items():
assert task_dump['env'][key] == val
async def run(self,
name: str,
image: str,
env: dict = {},
routes: dict = {},
ports: dict = {},
cpu: any = 0,
memory: any = 0,
**inputs):
taskdef = TaskDefinition(
name=name,
image=image,
parent=self.id,
inputs=inputs,
ports=ports,
routes=routes,
cpu=cpu,
memory=memory,
env={
**self.env,
**env,
},
)
# run it
task = self.cluster.spawn(taskdef)
# wait for container to exit
await self.watch(task)
# clean up
self.cluster.destroy(task)
def extract_pod_taskdef(pod) -> TaskDefinition:
for container in pod.spec.containers:
for env in container.env:
if env.name == ENV_TASK_DEFINITION:
taskdef = env_unpack(env.value)
return TaskDefinition(**taskdef)
raise Exception('Failed to extract pod task definition')
def test_docker_child_task():
dp = DockerProvider()
task = dp.spawn(TaskDefinition(
name=TEST_TASK,
image=TEST_IMAGE,
inputs={'child': True},
))
# wait for the child to spawn
child = None
for i in range(0, 10):
children = dp.find_child_containers(task.id)
if len(children) > 0:
child = children[0]
break
time.sleep(0.5)
# make sure we got a child
assert child is not None
# test list tasks
tasks = dp.list_all()
task_ids = list(map(lambda t: t.id, tasks))
assert task.id in task_ids
assert child.labels[LABEL_TASK_ID] in task_ids
# kill the whole family
dp.destroy(task.id)
children = dp.find_child_containers(task.id)
assert len(children) == 0
async def init_node(self):
cluster = env_get_cluster_provider()
parent = env_get_task_definition()
token = os.getenv(ENV_KERNEL_TOKEN)
taskdef = TaskDefinition(
name='kernel',
image=parent.image,
parent=parent.id,
volumes=parent.volumes,
env=parent.env,
upstream=get_local_url(),
meta={
**parent.meta,
'virtual': True,
},
)
# set up notebook node
self.node = NotebookNode(taskdef)
await self.node.start(token)
# instantiate kernel task
self.task = KernelTask(node=self.node,
cluster=cluster,
taskdef=taskdef)
# write globals
self.shell.push({
'cowait': cowait,
'kernel': self.task,
'tasks': self.task.subtasks,
'NotebookRunner': NotebookRunner,
})
def create_env(self, taskdef: TaskDefinition) -> dict:
"""
Create a container environment dict from a task definition.
Arguments:
taskdef (TaskDefinition): Task definition
Returns:
env (dict): Environment variable dict
"""
env = {
**taskdef.env,
ENV_GZIP_ENABLED: '1',
ENV_TASK_CLUSTER: env_pack(self.serialize()),
ENV_TASK_DEFINITION: env_pack(taskdef.serialize()),
}
# check total length of environment data
length = 0
for key, value in env.items():
length += len(str(key)) + len(str(value))
if length > MAX_ENV_LENGTH:
raise ProviderError(f'Task environment too long. Was {length}, max: {MAX_ENV_LENGTH}')
return env
def env_get_task_definition():
if ENV_TASK_DEFINITION not in os.environ:
raise ValueError(
f'Task definition must be passed in the '
f'{ENV_TASK_DEFINITION} environment variable.')
taskdef_json = json.loads(os.environ[ENV_TASK_DEFINITION])
return TaskDefinition.deserialize(taskdef_json)
async def send_init(self, taskdef: TaskDefinition) -> None:
"""
Send a task initialization message.
Arguments:
taskdef (TaskDefinition): New task definition
"""
await self.msg(TASK_INIT, task=taskdef.serialize())
def agent(
config: CowaitConfig,
detach: bool = False,
upstream: str = None,
) -> None:
logger = RunLogger(quiet=False, raw=False)
try:
cluster = config.get_cluster()
if cluster.type == 'api':
raise CliError('Error: Cant deploy agent using an API cluster')
token = uuid()
if cluster.type == 'docker':
token = ''
cluster.destroy('agent')
# create task definition
taskdef = TaskDefinition(
id='agent',
name='cowait.tasks.agent',
image=DEFAULT_BASE_IMAGE,
upstream=upstream,
routes={
'/': 80,
},
meta={
'http_token': token,
},
)
# submit task to cluster
task = cluster.spawn(taskdef)
if detach:
logger.header('detached')
return
def destroy(*args):
logger.header('interrupt')
cluster.destroy(task.id)
sys.exit(0)
with ExitTrap(destroy):
# capture & print logs
logs = cluster.logs(task)
logger.header('task output')
for log in logs:
logger.handle(log)
logger.header()
except ProviderError as e:
raise CliError(f'Provider error: {e}')
except TaskCreationError as e:
raise CliError(f'Task creation error: {e}')
def spawn_test_task(name, **taskdef: dict) -> Task:
""" Spawns a task using the test cluster provider """
provider = get_test_provider()
return provider.spawn(
TaskDefinition(**{
'name': name,
'image': DEFAULT_BASE_IMAGE,
**taskdef
}))
def agent(
config: CowaitConfig,
detach: bool = False,
upstream: str = None,
) -> None:
try:
context = CowaitContext.open()
cluster_name = context.get('cluster', config.default_cluster)
cluster = config.get_cluster(cluster_name)
if cluster.type == 'api':
raise CliError('Error: Cant deploy agent using an API cluster')
cluster.destroy('agent')
# create task definition
taskdef = TaskDefinition(
id='agent',
name='cowait.tasks.agent',
image=DEFAULT_BASE_IMAGE,
upstream=upstream,
routes={
'/': 80,
},
meta={
'http_token': uuid(),
},
)
# submit task to cluster
task = cluster.spawn(taskdef)
if detach:
printheader('detached')
return
def destroy(*args):
print()
printheader('interrupt')
cluster.destroy(task.id)
sys.exit(0)
with ExitTrap(destroy):
# capture & print logs
logs = cluster.logs(task)
printheader('task output')
for log in logs:
print(log, flush=True)
printheader()
except ProviderError as e:
raise CliError(f'Provider error: {e}')
except TaskCreationError as e:
raise CliError(f'Task creation error: {e}')
def test_rpc():
dp = DockerProvider()
task = dp.spawn(TaskDefinition(
name='cowait.test.tasks.rpc_parent',
image=DEFAULT_BASE_IMAGE,
))
# wait for execution
result = task.container.wait()
assert result['StatusCode'] == 0
def test_max_env_length():
""" Passing too large inputs should raise a ProviderError """
random_data = uuid(2 * MAX_ENV_LENGTH, lower=False)
with pytest.raises(ProviderError):
cp = ClusterProvider('test')
cp.create_env(TaskDefinition(
'test-task',
image='imaginary-image',
inputs={
'ohshit': random_data,
},
))
def test_docker_task_error():
dp = DockerProvider()
task = dp.spawn(TaskDefinition(
name=TEST_TASK,
image=TEST_IMAGE,
inputs={'error': True},
))
container = dp.docker.containers.get(task.container.id)
assert task.container == container
result = container.wait()
assert result['StatusCode'] != 0
def create_env(self, taskdef: TaskDefinition) -> dict:
"""
Create a container environment dict from a task definition.
Arguments:
taskdef (TaskDefinition): Task definition
Returns:
env (dict): Environment variable dict
"""
return {
**taskdef.env,
ENV_TASK_CLUSTER: json.dumps(self.serialize()),
ENV_TASK_DEFINITION: json.dumps(taskdef.serialize()),
}
def test_docker_child_error():
dp = DockerProvider()
task = dp.spawn(TaskDefinition(
name=TEST_TASK,
image=TEST_IMAGE,
inputs={'child_error': True},
))
container = dp.docker.containers.get(task.container.id)
assert task.container == container
# child error should cause the parent to fail
result = container.wait()
assert result['StatusCode'] != 0
def test(
config: CowaitConfig,
push: bool,
):
logger = TestLogger()
try:
context = CowaitContext.open()
cluster = config.get_cluster()
if push:
run_push()
else:
run_build()
# execute the test task within the current image
task = cluster.spawn(
TaskDefinition(
name='cowait.test',
image=context.image,
))
def destroy(*args):
logger.header('interrupt')
cluster.destroy(task.id)
sys.exit(1)
with ExitTrap(destroy):
# capture & print logs
logs = cluster.logs(task)
logger.header('task output')
for msg in logs:
logger.handle(msg)
logger.header()
# grab task result
passing = cluster.wait(task)
sys.exit(0 if passing else 1)
except ProviderError as e:
logger.print_exception(f'Provider Error: {e}')
sys.exit(1)
except TaskCreationError as e:
logger.print_exception(f'Error creating task: {e}')
sys.exit(1)
async def spawn(
self,
name: str,
image: str,
id: str = None,
ports: dict = {},
routes: dict = {},
inputs: dict = {},
meta: dict = {},
env: dict = {},
cpu: str = '0',
memory: str = '0',
owner: str = '',
**kwargs: dict,
) -> dict:
if not isinstance(name, str) and issubclass(name, Task):
name = name.__module__
# todo: throw error if any input is a coroutine
task = self.cluster.spawn(
TaskDefinition(
id=id,
name=name,
image=image,
upstream=get_local_url(),
meta=meta,
ports=ports,
routes=routes,
env=env,
cpu=cpu,
memory=memory,
owner=owner,
inputs={
**inputs,
**kwargs,
},
))
# authorize id
self.node.server.auth.add_token(id)
# register with subtask manager
self.subtasks.watch(task)
return task.serialize()
def base_environment(cluster, taskdef: TaskDefinition) -> dict:
"""
Create a container environment dict from a task definition.
Arguments:
taskdef (TaskDefinition): Task definition
Returns:
env (dict): Environment variable dict
"""
return {
**taskdef.env,
ENV_GZIP_ENABLED: '1',
ENV_TASK_CLUSTER: env_pack(cluster.serialize()),
ENV_TASK_DEFINITION: env_pack(taskdef.serialize()),
}
def test(
config: CowaitConfig,
push: bool,
):
try:
context = CowaitContext.open()
cluster_name = context.get('cluster', config.default_cluster)
cluster = config.get_cluster(cluster_name)
if push:
run_push()
else:
run_build()
# execute the test task within the current image
task = cluster.spawn(
TaskDefinition(
name='cowait.test',
image=context.get_image_name(),
))
def destroy(*args):
print()
printheader('interrupt')
cluster.destroy(task.id)
os._exit(1)
with ExitTrap(destroy):
# capture & print logs
logs = cluster.logs(task)
printheader('task output')
for log in logs:
print(log, flush=True)
except TaskCreationError as e:
printheader('error')
print('Error creating task:', str(e))
except ProviderError as e:
printheader('error')
print('Provider error:', str(e))
finally:
printheader()
def test_kill_docker_task():
dp = DockerProvider()
task = dp.spawn(
TaskDefinition(
name=TEST_TASK,
image=TEST_IMAGE,
inputs={'forever': True},
))
# ensure container exists
dp.docker.containers.get(task.container.id)
# destroy it
dp.destroy(task.id)
# ensure it no longer exists
with pytest.raises(docker.errors.NotFound):
dp.docker.containers.get(task.container.id)
def test_kill_docker_task():
dp = DockerProvider()
task = dp.spawn(TaskDefinition(
name=TEST_TASK,
image=TEST_IMAGE,
inputs={'forever': True},
))
# ensure container exists
dp.docker.containers.get(task.container.id)
# destroy it
dp.destroy(task.id)
# ensure it no longer exists
with pytest.raises(docker.errors.NotFound):
try:
dp.docker.containers.get(task.container.id)
except requests.exceptions.ChunkedEncodingError:
# workaround for docker for mac bug:
# https://github.com/docker/docker-py/issues/2696
raise docker.errors.NotFound('Not found')
def run(
config: CowaitConfig,
task: str,
name: str = None,
cluster_name: str = None,
inputs: dict = {},
env: dict = {},
ports: dict = {},
routes: dict = {},
build: bool = False,
upstream: str = None,
detach: bool = False,
cpu: str = '0',
memory: str = '0',
):
try:
context = CowaitContext.open()
cluster_name = context.get('cluster', config.default_cluster)
cluster = config.get_cluster(cluster_name)
# figure out image name
image, task = parse_task_image_name(task, None)
if image is None:
if build:
build_cmd()
image = context.get_image_name()
# default to agent as upstream
agent = cluster.find_agent()
# create task definition
taskdef = TaskDefinition(
id=name,
name=task,
image=image,
inputs=inputs,
env={
**context.get('environment', {}),
**env,
},
ports=ports,
routes=routes,
upstream=context.coalesce('upstream', upstream, agent),
parent=None, # root task
owner=os.getlogin(),
cpu=cpu,
memory=memory,
)
# print execution info
printheader('task')
print(' task: ', taskdef.id)
print(' cluster: ', cluster_name)
if taskdef.upstream:
print(' upstream: ', taskdef.upstream)
print(' image: ', image)
print(' inputs: ', inputs)
print(' env: ', env)
# submit task to cluster
task = cluster.spawn(taskdef)
if detach:
printheader('detached')
return
def destroy(*args):
print()
printheader('interrupt')
cluster.destroy(task.id)
sys.exit(0)
with ExitTrap(destroy):
# capture & print logs
logs = cluster.logs(task)
printheader('task output')
for log in logs:
print(log, flush=True)
except ProviderError as e:
printheader('error')
print('Provider error:', str(e))
except TaskCreationError as e:
printheader('error')
print('Error creating task:', str(e))
finally:
printheader()
def extract_container_taskdef(container) -> TaskDefinition:
for env in container.attrs['Config']['Env']:
if ENV_TASK_DEFINITION == env[0:len(ENV_TASK_DEFINITION)]:
data = env[len(ENV_TASK_DEFINITION) + 1:]
return TaskDefinition(**env_unpack(data))
raise Exception('Unable to unpack container task definition')
def spawn(self, taskdef: TaskDefinition) -> RemoteTask:
try:
task = self.rpc('spawn', **taskdef.serialize())
return RemoteTask(TaskDefinition.deserialize(task), self)
except RpcError as e:
raise TaskCreationError(str(e))
def test(
config: Config,
cluster_name: str = None,
mount: bool = True,
cpu: str = None,
cpu_limit: str = None,
memory: str = None,
memory_limit: str = None,
marks: str = None,
verbose: bool = None,
capture: bool = None,
):
logger = TestLogger()
try:
context = Context.open(config)
cluster = context.get_cluster(cluster_name)
volumes = {}
if mount and cluster.type == 'docker':
# when testing in docker, mount the local directory
# this avoids the problem of having to constantly rebuild in order to test
print('** Mounting', context.root_path)
volumes['/var/task'] = {
'bind': {
'src': context.root_path,
'mode': 'rw',
'inherit': 'same-image',
},
}
# execute the test task within the current image
task = cluster.spawn(
TaskDefinition(
name='cowait.test',
image=context.image,
owner=getpass.getuser(),
env={
**context.environment,
**context.dotenv,
},
volumes={
**context.get('volumes', {}),
**volumes,
},
inputs={
'marks': marks,
'verbose': verbose,
'capture': capture,
},
cpu=context.override('cpu', cpu),
cpu_limit=context.override('cpu_limit', cpu_limit),
memory=context.override('memory', memory),
memory_limit=context.override('memory_limit', memory_limit),
))
def destroy(*args):
logger.header('interrupt')
cluster.destroy(task.id)
sys.exit(1)
with ExitTrap(destroy):
# capture & print logs
logs = cluster.logs(task.id)
logger.header('task output')
for msg in logs:
logger.handle(msg)
logger.header()
# grab task result
passing = cluster.wait(task)
sys.exit(0 if passing else 1)
except ProviderError as e:
logger.print_exception(f'Provider Error: {e}')
sys.exit(1)
except TaskCreationError as e:
logger.print_exception(f'Error creating task: {e}')
sys.exit(1)
def run(
config: CowaitConfig,
task: str,
name: str = None,
inputs: dict = {},
env: dict = {},
ports: dict = {},
routes: dict = {},
build: bool = False,
upstream: str = None,
detach: bool = False,
cpu: str = None,
cpu_limit: str = None,
memory: str = None,
memory_limit: str = None,
raw: bool = False,
quiet: bool = False,
affinity: str = None,
):
logger = RunLogger(raw, quiet)
try:
context = CowaitContext.open()
cluster = config.get_cluster()
# figure out image name
remote_image = True
image, task = parse_task_image_name(task, None)
if image is None:
if build:
build_cmd(quiet=quiet or raw)
image = context.image
remote_image = False
volumes = context.get('volumes', {})
if not isinstance(volumes, dict):
raise TaskCreationError('Invalid volume configuration')
if not remote_image:
volumes['/var/task'] = {
'bind': {
'src': context.root_path,
'mode': 'rw',
},
}
if not affinity:
affinity = {}
elif affinity.lower() == 'spread':
affinity = {
"type": "spread",
"label": {
"key": task + "-key",
"value": task + "-value"
}
}
elif affinity.lower() == 'group':
affinity = {
"type": "group",
"label": {
"key": task + "-key",
"value": task + "-value"
}
}
# default to agent as upstream
agent = cluster.find_agent()
# create task definition
taskdef = TaskDefinition(
id=name,
name=task,
image=image,
inputs=inputs,
env={
**context.environment,
**env,
},
ports=ports,
routes=routes,
parent=None, # root task
upstream=context.coalesce('upstream', upstream, agent),
owner=getpass.getuser(),
volumes=volumes,
cpu=context.override('cpu', cpu),
cpu_limit=context.override('cpu_limit', cpu_limit),
memory=context.override('memory', memory),
memory_limit=context.override('memory_limit', memory_limit),
storage=context.get('storage', {}),
affinity=affinity)
# print execution info
logger.print_info(taskdef, config.default_cluster)
# submit task to cluster
task = cluster.spawn(taskdef)
if detach:
logger.header('detached')
return
def destroy(*args):
logger.header('interrupt')
cluster.destroy(task.id)
sys.exit(1)
with ExitTrap(destroy):
# capture & print logs
logs = cluster.logs(task)
logger.header('task output')
for msg in logs:
logger.handle(msg)
logger.header()
except ProviderError as e:
print('Provider error:', str(e))
logger.print_exception(f'Provider Error: {e}')
except TaskCreationError as e:
logger.print_exception(f'Error creating task: {e}')
def env_get_task_definition():
taskdef = env_get(ENV_TASK_DEFINITION)
return TaskDefinition.deserialize(taskdef)
def notebook(config, image: str = None, cluster_name: str = None) -> None:
context = Context.open(config)
if not context.notebook:
print('Notebook funcitonaility is not enabled.')
print('To enable, set features.notebook to True in cowait.yml and rebuild.')
sys.exit(1)
if image is not None:
print('Remote images are currently not supported')
sys.exit(1)
volumes = {
'/var/task': {
'bind': {
'src': os.getcwd(),
'mode': 'rw',
'inherit': 'same-image',
},
}
}
cluster = context.get_cluster(cluster_name)
# Docker
if cluster.type == 'docker':
return run_cmd(
config=config,
task='cowait.notebook',
build=False,
image=image,
routes={
'/': '8888',
},
cluster_name=cluster_name,
volumes=volumes,
)
# check for clientfs
clientfs_executable = './clientfs-' + platform.system().lower()
if not os.path.exists(clientfs_executable):
print('Kubernetes notebooks are not supported in this build of Cowait')
sys.exit(1)
# Kubernetes
core = client.CoreV1Api()
notebook_id = 'notebook-' + uuid(4)
core.create_namespaced_persistent_volume_claim(
namespace=cluster.namespace,
body=client.V1PersistentVolumeClaim(
metadata=client.V1ObjectMeta(
name=notebook_id,
namespace=cluster.namespace,
),
spec=client.V1PersistentVolumeClaimSpec(
storage_class_name='clientfs',
access_modes=['ReadWriteMany'],
resources=client.V1ResourceRequirements(
requests={
'storage': '1G',
},
),
),
),
)
def delete_pvc(task_id):
print('destroy', task_id)
if task_id != notebook_id:
return
print('* stopping clientfs')
clientfs.terminate()
print('* deleting volume')
core.delete_namespaced_persistent_volume_claim(notebook_id, cluster.namespace)
cluster.on('kill', delete_pvc)
pvc_id = None
while True:
time.sleep(1)
volume = core.read_namespaced_persistent_volume_claim(notebook_id, cluster.namespace)
if volume.status.phase == 'Bound':
pvc_id = 'pvc-' + volume.metadata.uid
print('* created volume', notebook_id, '/', pvc_id)
break
volumes['/var/task'] = {
'persistent_volume_claim': {
'claim_name': notebook_id,
},
}
# start clientfs
clientfs_host = cluster.args.get('clientfs', {}).get('host')
print(f'* connecting clientfs volume to {clientfs_host}...')
clientfs = subprocess.Popen([
clientfs_executable,
f"--proxy={clientfs_host}",
f"--volume={pvc_id}"
])
logger = RunLogger()
try:
# default to agent as upstream
agent = cluster.find_agent()
# create task definition
taskdef = TaskDefinition(
id=notebook_id,
name='cowait.notebook',
image=context.image,
env={
**context.extend('environment', {}),
**context.dotenv,
},
routes={
'/': '8888',
},
parent=None, # root task
upstream=agent,
owner=getpass.getuser(),
volumes=context.extend('volumes', volumes),
)
# print execution info
logger.print_info(taskdef, cluster)
# submit task to cluster
task = cluster.spawn(taskdef)
detach = False
if detach:
logger.header('detached')
return
def destroy(*args):
logger.header('interrupt')
cluster.destroy(task.id)
sys.exit(1)
with ExitTrap(destroy):
# capture & print logs
logs = cluster.logs(task.id)
logger.header('task output')
for msg in logs:
logger.handle(msg)
except Exception:
traceback.print_exc()
sys.exit(1)
