def sanitize_resources(
resources: Union[k8s_schemas.V1ResourceRequirements, Dict]
) -> Optional[k8s_schemas.V1ResourceRequirements]:
def validate_resources(r_field: Dict) -> Dict:
if not r_field:
return r_field
for k in r_field:
r_field[k] = str(r_field[k])
return r_field
if not resources:
return None
if isinstance(resources, Dict):
return k8s_schemas.V1ResourceRequirements(
limits=validate_resources(resources.get("limits", None)),
requests=validate_resources(resources.get("requests", None)),
)
else:
return k8s_schemas.V1ResourceRequirements(
limits=validate_resources(resources.limits),
requests=validate_resources(resources.requests),
)
def test_requests_gpu(self):
assert (requests_gpu(
k8s_schemas.V1ResourceRequirements(limits={"cpu": 1})) is False)
assert (requests_gpu(
k8s_schemas.V1ResourceRequirements(limits={"amd.com/gpu": 1})) is
True)
assert (requests_gpu(
k8s_schemas.V1ResourceRequirements(requests={"nvidia.com/gpu": 1}))
is True)
def test_requests_tpu(self):
assert (requests_tpu(
k8s_schemas.V1ResourceRequirements(limits={"cpu": 1})) is False)
assert (requests_tpu(
k8s_schemas.V1ResourceRequirements(
limits={"cloud-tpus.google.com/v2": 1})) is True)
assert (requests_tpu(
k8s_schemas.V1ResourceRequirements(
requests={"cloud-tpus.google.com/v2:": 32})) is True)
def get_default_notification_container():
return V1Container(
name=MAIN_JOB_CONTAINER,
image="polyaxon/polyaxon-events-handlers:{}".format("dev"),
image_pull_policy=PullPolicy.ALWAYS.value,
command=["polyaxon", "notify"],
args=[
"--kind={{kind}}",
"--owner={{owner}}",
"--project={{project}}",
"--run_uuid={{run_uuid}}",
"--run_name={{run_name}}",
"--condition={{condition}}",
],
resources=k8s_schemas.V1ResourceRequirements(
limits={
"cpu": "0.5",
"memory": "100Mi"
},
requests={
"cpu": "0.1",
"memory": "20Mi"
},
),
)
def get_default_notification_container():
return V1Container(
name=MAIN_JOB_CONTAINER,
image="polyaxon/polyaxon-events-handlers:{}".format(pkg.VERSION),
image_pull_policy=PullPolicy.IF_NOT_PRESENT.value,
command=["polyaxon", "notify"],
args=[
"--kind={{kind}}",
"--owner={{owner}}",
"--project={{project}}",
"--run-uuid={{run_uuid}}",
"{{params.condition.as_arg}}",
"{{params.run_name.as_arg}}",
],
resources=k8s_schemas.V1ResourceRequirements(
limits={
"cpu": "0.5",
"memory": "100Mi"
},
requests={
"cpu": "0.1",
"memory": "20Mi"
},
),
)
def get_batch_cleaner_container(
store: V1ConnectionType,
paths: List[str],
):
subpaths = [os.path.join(store.store_path, subpath) for subpath in paths]
subpaths = " ".join(["-sp={}".format(sp) for sp in subpaths])
clean_args = "polyaxon clean-artifacts {} {}".format(
store.kind.replace("_", "-"), subpaths)
return V1Container(
name=MAIN_JOB_CONTAINER,
image="polyaxon/polyaxon-init:{}".format(pkg.VERSION),
image_pull_policy=PullPolicy.IF_NOT_PRESENT.value,
command=["/bin/bash", "-c"],
args=[clean_args],
resources=k8s_schemas.V1ResourceRequirements(
limits={
"cpu": "0.5",
"memory": "160Mi"
},
requests={
"cpu": "0.1",
"memory": "80Mi"
},
),
)
def test_get_main_container_simple_params(self):
initial_mounts = [
k8s_schemas.V1VolumeMount(name="test",
mount_path="/mount_test",
read_only=True)
]
resources = k8s_schemas.V1ResourceRequirements(
requests={
"cpu": "1",
"memory": "256Mi"
},
limits={
"cpu": "1",
"memory": "256Mi"
},
)
container = get_main_container(
container_id="new-name",
main_container=k8s_schemas.V1Container(
name="main",
image="job_docker_image",
image_pull_policy="IfNotPresent",
command=["cmd", "-p", "-c"],
args=["arg1", "arg2"],
resources=resources,
),
contexts=None,
volume_mounts=initial_mounts,
log_level="info",
artifacts_store=None,
init=None,
connection_by_names=None,
connections=None,
secrets=None,
config_maps=None,
kv_env_vars=None,
env=None,
ports=23,
run_path=None,
)
assert container.name == "new-name"
assert container.image == "job_docker_image"
assert container.image_pull_policy == "IfNotPresent"
assert container.command == ["cmd", "-p", "-c"]
assert container.args == ["arg1", "arg2"]
assert container.ports == [
k8s_schemas.V1ContainerPort(container_port=23)
]
assert container.env == [
get_env_var(name=POLYAXON_KEYS_LOG_LEVEL, value="info")
]
assert container.env_from == []
assert container.resources == resources
assert container.volume_mounts == initial_mounts
def get_init_resources() -> k8s_schemas.V1ResourceRequirements:
return k8s_schemas.V1ResourceRequirements(
limits={
"cpu": "1",
"memory": "200Mi"
},
requests={
"cpu": "0.1",
"memory": "20Mi"
},
)
def test_get_init_resources(self):
assert get_init_resources() == k8s_schemas.V1ResourceRequirements(
limits={
"cpu": "1",
"memory": "200Mi"
},
requests={
"cpu": "0.1",
"memory": "20Mi"
},
)
def test_get_resources_env_vars(self):
env_vars = get_resources_env_vars(None)
assert len(env_vars) == 1
assert env_vars[0].name == "NVIDIA_VISIBLE_DEVICES"
assert env_vars[0].value == "none"
resources = k8s_schemas.V1ResourceRequirements(limits={"cpu": 1})
env_vars = get_resources_env_vars(resources)
assert len(env_vars) == 1
assert env_vars[0].name == "NVIDIA_VISIBLE_DEVICES"
assert env_vars[0].value == "none"
resources = k8s_schemas.V1ResourceRequirements(limits={"memory": 1})
env_vars = get_resources_env_vars(resources)
assert len(env_vars) == 1
assert env_vars[0].name == "NVIDIA_VISIBLE_DEVICES"
assert env_vars[0].value == "none"
resources = k8s_schemas.V1ResourceRequirements(
requests={"nvidia.com/gpu": 1})
env_vars = get_resources_env_vars(resources)
assert len(env_vars) == 0
assert env_vars == []
def get_default_tuner_container(command):
return V1Container(
name=MAIN_JOB_CONTAINER,
image="polyaxon/polyaxon-hpsearch:{}".format("dev"),
image_pull_policy=PullPolicy.ALWAYS.value,
command=command,
args=[
"--parallel={{parallel}}",
"--configs={{configs}}",
"--metrics={{metrics}}",
],
resources=k8s_schemas.V1ResourceRequirements(
requests={"cpu": "0.1", "memory": "180Mi"},
),
)
def requests_tpu(
resources: Union[k8s_schemas.V1ResourceRequirements, Dict]) -> bool:
if not resources:
return False
if not isinstance(resources, k8s_schemas.V1ResourceRequirements):
resources = k8s_schemas.V1ResourceRequirements(**resources)
if resources.requests:
for key in resources.requests.keys():
if "tpu" in key:
return True
if resources.limits:
for key in resources.limits.keys():
if "tpu" in key:
return True
return False
def get_default_tuner_container(command, bracket_iteration: int = None):
args = [
"{{params.matrix.as_arg}}",
"{{params.search.as_arg}}",
"{{params.iteration.as_arg}}",
]
if bracket_iteration is not None:
args.append("{{params.bracket_iteration.as_arg}}")
return V1Container(
name=MAIN_JOB_CONTAINER,
image="polyaxon/polyaxon-hpsearch:{}".format(pkg.VERSION),
image_pull_policy=PullPolicy.IF_NOT_PRESENT.value,
command=command,
args=args,
resources=k8s_schemas.V1ResourceRequirements(requests={
"cpu": "0.1",
"memory": "180Mi"
}, ),
)
def get_default_cleaner_container(store: V1ConnectionType, run_path: str):
subpath = os.path.join(store.store_path, run_path)
return V1Container(
name=MAIN_JOB_CONTAINER,
image="polyaxon/polyaxon-init:{}".format(pkg.VERSION),
image_pull_policy=PullPolicy.ALWAYS.value,
command=["polyaxon", "clean-artifacts",
store.kind.replace('_', '-')],
args=["--subpath={}".format(subpath)],
resources=k8s_schemas.V1ResourceRequirements(
limits={
"cpu": "0.5",
"memory": "100Mi"
},
requests={
"cpu": "0.1",
"memory": "20Mi"
},
),
)
def get_default_cleaner_container(store: V1ConnectionType, run_uuid: str,
run_kind: str):
subpath = os.path.join(store.store_path, run_uuid)
clean_args = "polyaxon clean-artifacts {} --subpath={}".format(
store.kind.replace("_", "-"), subpath)
wait_args = "polyaxon wait --uuid={} --kind={}".format(run_uuid, run_kind)
return V1Container(
name=MAIN_JOB_CONTAINER,
image="polyaxon/polyaxon-init:{}".format(pkg.VERSION),
image_pull_policy=PullPolicy.IF_NOT_PRESENT.value,
command=["/bin/bash", "-c"],
args=["{} && {}".format(wait_args, clean_args)],
resources=k8s_schemas.V1ResourceRequirements(
limits={
"cpu": "0.5",
"memory": "160Mi"
},
requests={
"cpu": "0.1",
"memory": "80Mi"
},
),
)
