[core][refactor] Move accelerator-specific environment variables to ray_constants.py to avoid redefining them

python/ray/_private/accelerators/amd_gpu.py

@@ -3,12 +3,13 @@
 from typing import Optional, List, Tuple
 
 from ray._private.accelerators.accelerator import AcceleratorManager
+from ray._private.ray_constants import (
+    ROCR_VISIBLE_DEVICES_ENV_VAR,
+    NOSET_ROCR_VISIBLE_DEVICES_ENV_VAR,
+)
 
 logger = logging.getLogger(__name__)
 
-ROCR_VISIBLE_DEVICES_ENV_VAR = "ROCR_VISIBLE_DEVICES"
-NOSET_ROCR_VISIBLE_DEVICES_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES"
-
 amd_product_dict = {
     "0x738c": "AMD-Instinct-MI100",
     "0x7408": "AMD-Instinct-MI250X",

python/ray/_private/accelerators/hpu.py

@@ -5,12 +5,13 @@
 from importlib.util import find_spec
 
 from ray._private.accelerators.accelerator import AcceleratorManager
+from ray._private.ray_constants import (
+    HABANA_VISIBLE_DEVICES_ENV_VAR,
+    NOSET_HABANA_VISIBLE_MODULES_ENV_VAR,
+)
 
 logger = logging.getLogger(__name__)
 
-HABANA_VISIBLE_DEVICES_ENV_VAR = "HABANA_VISIBLE_MODULES"
-NOSET_HABANA_VISIBLE_MODULES_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_HABANA_VISIBLE_MODULES"
-
 
 @lru_cache()
 def is_package_present(package_name: str) -> bool:

python/ray/_private/accelerators/intel_gpu.py

@@ -3,11 +3,13 @@
 from typing import Optional, List, Tuple
 
 from ray._private.accelerators.accelerator import AcceleratorManager
+from ray._private.ray_constants import (
+    ONEAPI_DEVICE_SELECTOR_ENV_VAR,
+    NOSET_ONEAPI_DEVICE_SELECTOR_ENV_VAR,
+)
 
 logger = logging.getLogger(__name__)
 
-ONEAPI_DEVICE_SELECTOR_ENV_VAR = "ONEAPI_DEVICE_SELECTOR"
-NOSET_ONEAPI_DEVICE_SELECTOR_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_ONEAPI_DEVICE_SELECTOR"
 ONEAPI_DEVICE_BACKEND_TYPE = "level_zero"
 ONEAPI_DEVICE_TYPE = "gpu"
 

python/ray/_private/accelerators/neuron.py

@@ -6,14 +6,13 @@
 from typing import Optional, List, Tuple
 
 from ray._private.accelerators.accelerator import AcceleratorManager
+from ray._private.ray_constants import (
+    NEURON_RT_VISIBLE_CORES_ENV_VAR,
+    NOSET_AWS_NEURON_RT_VISIBLE_CORES_ENV_VAR,
+)
 
 logger = logging.getLogger(__name__)
 
-NEURON_RT_VISIBLE_CORES_ENV_VAR = "NEURON_RT_VISIBLE_CORES"
-NOSET_AWS_NEURON_RT_VISIBLE_CORES_ENV_VAR = (
-    "RAY_EXPERIMENTAL_NOSET_NEURON_RT_VISIBLE_CORES"
-)
-
 # https://awsdocs-neuron.readthedocs-hosted.com/en/latest/general/arch/neuron-hardware/inf2-arch.html#aws-inf2-arch
 # https://awsdocs-neuron.readthedocs-hosted.com/en/latest/general/arch/neuron-hardware/trn1-arch.html#aws-trn1-arch
 # Subject to removal after the information is available via public API

python/ray/_private/accelerators/npu.py

@@ -4,14 +4,13 @@
 from typing import Optional, List, Tuple
 
 from ray._private.accelerators.accelerator import AcceleratorManager
+from ray._private.ray_constants import (
+    ASCEND_RT_VISIBLE_DEVICES_ENV_VAR,
+    NOSET_ASCEND_RT_VISIBLE_DEVICES_ENV_VAR,
+)
 
 logger = logging.getLogger(__name__)
 
-ASCEND_RT_VISIBLE_DEVICES_ENV_VAR = "ASCEND_RT_VISIBLE_DEVICES"
-NOSET_ASCEND_RT_VISIBLE_DEVICES_ENV_VAR = (
-    "RAY_EXPERIMENTAL_NOSET_ASCEND_RT_VISIBLE_DEVICES"
-)
-
 
 class NPUAcceleratorManager(AcceleratorManager):
     """Ascend NPU accelerators."""

python/ray/_private/accelerators/nvidia_gpu.py

@@ -4,12 +4,13 @@
 from typing import Optional, List, Tuple
 
 from ray._private.accelerators.accelerator import AcceleratorManager
+from ray._private.ray_constants import (
+    CUDA_VISIBLE_DEVICES_ENV_VAR,
+    NOSET_CUDA_VISIBLE_DEVICES_ENV_VAR,
+)
 
 logger = logging.getLogger(__name__)
 
-CUDA_VISIBLE_DEVICES_ENV_VAR = "CUDA_VISIBLE_DEVICES"
-NOSET_CUDA_VISIBLE_DEVICES_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_CUDA_VISIBLE_DEVICES"
-
 # TODO(Alex): This pattern may not work for non NVIDIA Tesla GPUs (which have
 # the form "Tesla V100-SXM2-16GB" or "Tesla K80").
 NVIDIA_GPU_NAME_PATTERN = re.compile(r"\w+\s+([A-Z0-9]+)")

python/ray/_private/accelerators/tpu.py

@@ -7,6 +7,10 @@
 from typing import Dict, Optional, List, Tuple
 
 from ray._private.accelerators.accelerator import AcceleratorManager
+from ray._private.ray_constants import (
+    TPU_VISIBLE_CHIPS_ENV_VAR,
+    NOSET_TPU_VISIBLE_CHIPS_ENV_VAR,
+)
 
 logger = logging.getLogger(__name__)
 
@@ -28,10 +32,6 @@
 GCE_TPU_INSTANCE_ID_KEY = "instance-id"
 GCE_TPU_WORKER_ID_KEY = "agent-worker-number"
 
-TPU_VISIBLE_CHIPS_ENV_VAR = "TPU_VISIBLE_CHIPS"
-
-NOSET_TPU_VISIBLE_CHIPS_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_TPU_VISIBLE_CHIPS"
-
 # The following defines environment variables that allow
 # us to access a subset of TPU visible chips.
 #

python/ray/_private/ray_constants.py

@@ -428,13 +428,26 @@ def env_set_by_user(key):
 LANGUAGE_WORKER_TYPES = ["python", "java", "cpp"]
 
 # Accelerator constants
+NOSET_ASCEND_RT_VISIBLE_DEVICES_ENV_VAR = (
+    "RAY_EXPERIMENTAL_NOSET_ASCEND_RT_VISIBLE_DEVICES"
+)
+NOSET_AWS_NEURON_RT_VISIBLE_CORES_ENV_VAR = (
+    "RAY_EXPERIMENTAL_NOSET_NEURON_RT_VISIBLE_CORES"
+)
 NOSET_CUDA_VISIBLE_DEVICES_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_CUDA_VISIBLE_DEVICES"
+NOSET_HABANA_VISIBLE_MODULES_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_HABANA_VISIBLE_MODULES"
+NOSET_ONEAPI_DEVICE_SELECTOR_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_ONEAPI_DEVICE_SELECTOR"
+NOSET_ROCR_VISIBLE_DEVICES_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES"
+NOSET_TPU_VISIBLE_CHIPS_ENV_VAR = "RAY_EXPERIMENTAL_NOSET_TPU_VISIBLE_CHIPS"
 
+ASCEND_RT_VISIBLE_DEVICES_ENV_VAR = "ASCEND_RT_VISIBLE_DEVICES"
 CUDA_VISIBLE_DEVICES_ENV_VAR = "CUDA_VISIBLE_DEVICES"
-ROCR_VISIBLE_DEVICES_ENV_VAR = "ROCR_VISIBLE_DEVICES"
+HABANA_VISIBLE_DEVICES_ENV_VAR = "HABANA_VISIBLE_MODULES"
 NEURON_RT_VISIBLE_CORES_ENV_VAR = "NEURON_RT_VISIBLE_CORES"
+ONEAPI_DEVICE_SELECTOR_ENV_VAR = "ONEAPI_DEVICE_SELECTOR"
+ROCR_VISIBLE_DEVICES_ENV_VAR = "ROCR_VISIBLE_DEVICES"
 TPU_VISIBLE_CHIPS_ENV_VAR = "TPU_VISIBLE_CHIPS"
-NPU_RT_VISIBLE_DEVICES_ENV_VAR = "ASCEND_RT_VISIBLE_DEVICES"
+
 
 NEURON_CORES = "neuron_cores"
 GPU = "GPU"

python/ray/air/_internal/device_manager/npu.py

@@ -55,7 +55,7 @@ def get_devices(self) -> List[torch.device]:
 
             if len(npu_ids) > 0:
                 npu_visible_str = os.environ.get(
-                    ray_constants.NPU_RT_VISIBLE_DEVICES_ENV_VAR, ""
+                    ray_constants.ASCEND_RT_VISIBLE_DEVICES_ENV_VAR, ""
                 )
                 if npu_visible_str and npu_visible_str != "NoDevFiles":
                     npu_visible_list = npu_visible_str.split(",")

python/ray/train/_internal/backend_executor.py

@@ -123,7 +123,7 @@ def __init__(
             ResourceConfig(
                 ray_constants.NPU,
                 ENABLE_SHARE_NPU_RT_VISIBLE_DEVICES_ENV,
-                ray_constants.NPU_RT_VISIBLE_DEVICES_ENV_VAR,
+                ray_constants.ASCEND_RT_VISIBLE_DEVICES_ENV_VAR,
             ),
             # For AMD GPUs, they are using ROCR_VISIBLE_DEVICES env var.
             ResourceConfig(