def _apply_ray_cr(
self,
min_replicas=0,
max_replicas=300,
replicas=0,
validate_replicas: bool = False,
) -> None:
"""Apply Ray CR config yaml, with configurable replica fields for the cpu
workerGroup.
If the CR does not yet exist, `replicas` can be set as desired.
If the CR does already exist, the recommended usage is this:
(1) Set `replicas` to what we currently expect it to be.
(2) Set `validate_replicas` to True. We will then check that the replicas
set on the CR coincides with `replicas`.
"""
with tempfile.NamedTemporaryFile("w") as config_file:
if validate_replicas:
raycluster = get_raycluster("raycluster-complete", namespace="default")
assert raycluster["spec"]["workerGroupSpecs"][0]["replicas"] == replicas
logger.info(
f"Validated that worker replicas for raycluster-complete"
f" is currently {replicas}."
)
cr_config = self._get_ray_cr_config(
min_replicas=min_replicas, max_replicas=max_replicas, replicas=replicas
)
yaml.dump(cr_config, config_file)
config_file.flush()
subprocess.check_call(["kubectl", "apply", "-f", config_file.name])
def _apply_ray_cr(
self,
min_replicas=0,
cpu_replicas=0,
gpu_replicas=0,
validate_replicas: bool = False,
) -> None:
"""Apply Ray CR config yaml, with configurable replica fields for the cpu
workerGroup.
If the CR does not yet exist, `replicas` can be set as desired.
If the CR does already exist, the recommended usage is this:
(1) Set `cpu_replicas` and `gpu_replicas` to what we currently expect them
to be.
(2) Set `validate_replicas` to True. We will then check that the replicas
set on the CR coincides with `replicas`.
"""
with tempfile.NamedTemporaryFile("w") as config_file:
if validate_replicas:
raycluster = get_raycluster(RAY_CLUSTER_NAME,
namespace=RAY_CLUSTER_NAMESPACE)
assert (raycluster["spec"]["workerGroupSpecs"][0]["replicas"]
== cpu_replicas)
assert (raycluster["spec"]["workerGroupSpecs"][1]["replicas"]
== gpu_replicas)
logger.info(
f"Validated that cpu and gpu worker replicas for "
f"{RAY_CLUSTER_NAME} are currently {cpu_replicas} and"
f" {gpu_replicas}, respectively.")
cr_config = self._get_ray_cr_config(
min_replicas=min_replicas,
cpu_replicas=cpu_replicas,
gpu_replicas=gpu_replicas,
)
yaml.dump(cr_config, config_file)
config_file.flush()
subprocess.check_call(["kubectl", "apply", "-f", config_file.name])