gpustack-runtime 0.1.41__py3-none-any.whl → 0.1.41.post2__py3-none-any.whl

gpustack_runtime/deployer/kuberentes.py

@@ -7,9 +7,9 @@ import operator
 import os
 from dataclasses import dataclass, field
 from enum import Enum
-from functools import reduce
+from functools import lru_cache, reduce
 from pathlib import Path
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Literal
 
 import kubernetes
 import kubernetes.stream.ws_client
@@ -43,7 +43,7 @@ from .__utils__ import (
     sensitive_env_var,
     validate_rfc1123_domain_name,
 )
-from .k8s.deviceplugin import cdi_kind_to_kdp_resource
+from .k8s.deviceplugin import cdi_kind_to_kdp_resource, is_kubelet_socket_accessible
 
 if TYPE_CHECKING:
     from collections.abc import Callable, Generator
@@ -381,6 +381,22 @@ class KubernetesDeployer(EndoscopicDeployer):
 
         return wrapper
 
+    @staticmethod
+    @lru_cache
+    def _get_resource_injection_policy() -> Literal["env", "kdp"]:
+        """
+        Get the resource injection policy (in lowercase) for the deployer.
+
+        Returns:
+            The resource injection policy.
+
+        """
+        policy = envs.GPUSTACK_RUNTIME_KUBERNETES_RESOURCE_INJECTION_POLICY.lower()
+        if policy != "auto":
+            return policy
+
+        return "kdp" if is_kubelet_socket_accessible() else "env"
+
     def _create_ephemeral_configmaps(
         self,
         workload: KubernetesWorkloadPlan,
@@ -992,10 +1008,7 @@ class KubernetesDeployer(EndoscopicDeployer):
 
             # Parameterize resources
             if c.resources:
-                kdp = (
-                    envs.GPUSTACK_RUNTIME_KUBERNETES_RESOURCE_INJECTION_POLICY.lower()
-                    == "kdp"
-                )
+                kdp = self._get_resource_injection_policy() == "kdp"
 
                 resources: dict[str, str] = {}
                 r_k_runtime_env = workload.resource_key_runtime_env_mapping or {}

gpustack_runtime/envs.py

@@ -280,7 +280,8 @@ if TYPE_CHECKING:
     """
     GPUSTACK_RUNTIME_KUBERNETES_RESOURCE_INJECTION_POLICY: str | None = None
     """
-    Resource injection policy for the Kubernetes deployer (e.g., Env, KDP).
+    Resource injection policy for the Kubernetes deployer (e.g., Auto, Env, KDP).
+    `Auto`: Automatically choose the resource injection policy based on the environment.
     `Env`: Injects resources using standard environment variable, depends on underlying Container Toolkit, based on `GPUSTACK_RUNTIME_DEPLOY_RESOURCE_KEY_MAP_RUNTIME_VISIBLE_DEVICES`.
     `KDP`: Injects resources using Kubernetes Device Plugin, based on `GPUSTACK_RUNTIME_DEPLOY_RESOURCE_KEY_MAP_CDI`.
     """
@@ -292,8 +293,9 @@ if TYPE_CHECKING:
     GPUSTACK_RUNTIME_KUBERNETES_KDP_DEVICE_ALLOCATION_POLICY: str | None = None
     """
     Device allocation policy for the Kubernetes Device Plugin (e.g., CDI, Env, Opaque).
-    `CDI`: Allocates devices using generated CDI specifications, making it easy to debug and troubleshoot; requires `GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY` to exist.
+    `Auto`: Automatically choose the device allocation policy based on the environment.
     `Env`: Allocates devices using runtime-visible environment variables; requires Container Toolkit support.
+    `CDI`: Allocates devices using generated CDI specifications, making it easy to debug and troubleshoot; requires `GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY` to exist.
     `Opaque`: Uses internal logic for allocation, which is convenient for deployment but difficult to troubleshoot.
     """
     GPUSTACK_RUNTIME_KUBERNETES_KDP_CDI_SPECS_GENERATE: bool = True
@@ -685,8 +687,8 @@ variables: dict[str, Callable[[], Any]] = {
         getenv(
             "GPUSTACK_RUNTIME_KUBERNETES_RESOURCE_INJECTION_POLICY",
         ),
-        options=["Env", "KDP"],
-        default="Env",
+        options=["Auto", "Env", "KDP"],
+        default="Auto",
     ),
     "GPUSTACK_RUNTIME_KUBERNETES_KDP_PER_DEVICE_MAX_ALLOCATIONS": lambda: to_int(
         getenv(
@@ -698,12 +700,12 @@ variables: dict[str, Callable[[], Any]] = {
         getenv(
             "GPUSTACK_RUNTIME_KUBERNETES_KDP_DEVICE_ALLOCATION_POLICY",
         ),
-        options=["CDI", "Env", "Opaque"],
-        default="CDI",
+        options=["Auto", "Env", "CDI", "Opaque"],
+        default="Auto",
     ),
     "GPUSTACK_RUNTIME_KUBERNETES_KDP_CDI_SPECS_GENERATE": lambda: ternary(
         lambda: (
-            getenv("GPUSTACK_RUNTIME_KUBERNETES_RESOURCE_INJECTION_POLICY", "Env")
+            getenv("GPUSTACK_RUNTIME_KUBERNETES_RESOURCE_INJECTION_POLICY", "Auto")
             == "Env"
         ),
         lambda: False,
@@ -711,9 +713,9 @@ variables: dict[str, Callable[[], Any]] = {
             lambda: (
                 getenv(
                     "GPUSTACK_RUNTIME_KUBERNETES_KDP_DEVICE_ALLOCATION_POLICY",
-                    "Opaque",
+                    "Auto",
                 )
-                == "Opaque"
+                not in ["Auto", "CDI"]
             ),
             lambda: False,
             lambda: to_bool(

{gpustack_runtime-0.1.41.dist-info → gpustack_runtime-0.1.41.post2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpustack-runtime
-Version: 0.1.41
+Version: 0.1.41.post2
 Summary: GPUStack Runtime is library for detecting GPU resources and launching GPU workloads.
 Project-URL: Homepage, https://github.com/gpustack/runtime
 Project-URL: Bug Tracker, https://github.com/gpustack/gpustack/issues
@@ -23,10 +23,10 @@ Requires-Dist: kubernetes>=33.1.0
 Requires-Dist: mthreads-ml-py>=2.2.10
 Requires-Dist: nvidia-ml-py>=13.580.65
 Requires-Dist: podman==5.6.0
-Requires-Dist: protobuf>=5.28.3
+Requires-Dist: protobuf<=3.19.6
 Requires-Dist: pyyaml
 Requires-Dist: tqdm
-Requires-Dist: types-protobuf>=6.32.1.20251210
+Requires-Dist: types-protobuf<=5.28.3.20241203
 Description-Content-Type: text/markdown
 
 # GPUStack Runtime

{gpustack_runtime-0.1.41.dist-info → gpustack_runtime-0.1.41.post2.dist-info}/RECORD

@@ -1,8 +1,8 @@
 gpustack_runtime/__init__.py,sha256=Xw_PVWneitx-8QmW6sJQeymj6zVbEgEndGhIB_km6TI,186
 gpustack_runtime/__main__.py,sha256=O9yJKcN7vg0Ppgc13qesxHwST2wkH3ccOkTQXPWHnNA,3939
-gpustack_runtime/_version.py,sha256=3VyVzUi8Smv4sZH0Q9rdGQXWEqqoTm7L84ArFLnPqZM,777
+gpustack_runtime/_version.py,sha256=fURzjTdczs5t1qsDdLl8wAzcF1VVhLtnm_S0ghrUvIE,792
 gpustack_runtime/_version.pyi,sha256=A42NoSgcqEXVy2OeNm4LXC9CbyonbooYrSUBlPm2lGY,156
-gpustack_runtime/envs.py,sha256=cprrZ-c4owsRLq7M__DlfpK6Ob9zbqMzRGVOsPTMJX0,38583
+gpustack_runtime/envs.py,sha256=Q8vK42OpkY4T72zN6pOz_eCS_hnQElhAmxZ1wdks0xQ,38794
 gpustack_runtime/logging.py,sha256=wMPriPpOuVsuClsjMh0qwEPQKyJiJa89ggdDjqkk7i0,6934
 gpustack_runtime/cmds/__init__.py,sha256=-_X2O2lBn6KcdLGUzhL3lEjQC4_cwA36fvWDnFAgtVM,1382
 gpustack_runtime/cmds/__types__.py,sha256=TBnUWUqzTkDtJnsMv363kdw-H8fOf-XQYbOvrmQif-M,815
@@ -14,23 +14,23 @@ gpustack_runtime/deployer/__patches__.py,sha256=cTBge8BT6IsY5MzETKY3kN28k3igYfNj
 gpustack_runtime/deployer/__types__.py,sha256=PgIWogHOvHKsHoeBjmKFEEM3JrKck89Mmnwlfx01BbE,72248
 gpustack_runtime/deployer/__utils__.py,sha256=paQu2M1UeoSfQPsiskmAqJSiln-8qwibTssEoWFMLec,21109
 gpustack_runtime/deployer/docker.py,sha256=bOaXbTnaalbO42FlyWR1Ha26Y30LGWPzWKPV5Q-Nk7g,85039
-gpustack_runtime/deployer/kuberentes.py,sha256=2M0RgsR1TlzkZ2QsElKZulftUnUMipYQPeQKKSm_Alo,89387
+gpustack_runtime/deployer/kuberentes.py,sha256=VkaAvuQJ5rRiNVD6OfM2pE3rmyT_a6oEvp-G8gW8Ojo,89816
 gpustack_runtime/deployer/podman.py,sha256=_qdbsTezacRmiXa3n04OUPUsgVy1pSFgJSKxous4s14,82156
-gpustack_runtime/deployer/cdi/__init__.py,sha256=y5fFNLhlrVhWzW-GTazT7xXBukHkBQ0DGfncKqKpBzk,3986
+gpustack_runtime/deployer/cdi/__init__.py,sha256=2wHrxkud3GJokE3ytNc3jvjddemXkNuuz_oIKzxD3-I,4000
 gpustack_runtime/deployer/cdi/__types__.py,sha256=04DKvcogk7OoHS7TU2Bmht3VVMu7iOEBWTEOvxpHt4w,18399
-gpustack_runtime/deployer/cdi/__utils__.py,sha256=cgqub-TQkXEjujlcQC4hJpignmmgxN5Yq9Rn6Z-TAh8,3866
+gpustack_runtime/deployer/cdi/__utils__.py,sha256=mvdOqkbhaSkphl0K-VpNwtFviAkttS9UrmKEA285kRw,3908
 gpustack_runtime/deployer/cdi/amd.py,sha256=-eq_SOlC56VX2QscZXvnoeffWSRindhr8zFZmaIcKrE,4082
 gpustack_runtime/deployer/cdi/ascend.py,sha256=lDs75a9--c0lM34xfJqu-_QbfWNFrf4zE-GXPKReBe4,4538
 gpustack_runtime/deployer/cdi/hygon.py,sha256=h6-vQfv03sgxYjMJAf_JOMq9cHFPaNjK1YbUYIiSXck,4117
 gpustack_runtime/deployer/cdi/iluvatar.py,sha256=6nNECZpU5IPP6-5l-O1rzU-ib-WcuwKvDg7ZV__1NE4,3650
 gpustack_runtime/deployer/cdi/metax.py,sha256=tmJBvr-n9pERAp-dXsa54qv6xmxt0rJoJwY36TFdoWk,4143
 gpustack_runtime/deployer/cdi/thead.py,sha256=SvIDKNYZx7FwMPTTxyJ2RRjlr9LXLN8BUYCUhidmiQk,3671
-gpustack_runtime/deployer/k8s/deviceplugin/__init__.py,sha256=RTQfArfX3FmS9lXKTwoWnb2gyqTIiDhRHEGe7MPxLzE,8040
-gpustack_runtime/deployer/k8s/deviceplugin/__types__.py,sha256=QaswnGns_hqMdVPeryDqmyhJrAHu_pfAlIC_25NHp4k,3057
-gpustack_runtime/deployer/k8s/deviceplugin/plugin.py,sha256=ktQ44wdNJNnv0Fn-TAnTQN0Jq5ZSqRA09WV407CrPvc,17931
+gpustack_runtime/deployer/k8s/deviceplugin/__init__.py,sha256=KoFztB0MmwO-lHFgEa0wLRsQyG2BN8i5w3mzo43GkbE,10498
+gpustack_runtime/deployer/k8s/deviceplugin/__types__.py,sha256=LCkgPDZ64Mra7bo5jmtsAO2Ypbc4qK99lMl6R_nQhnY,3043
+gpustack_runtime/deployer/k8s/deviceplugin/plugin.py,sha256=ipZ_V6pgJ2pzyEYUgAizZ7_W3a4noKEdTiZ9GAeuiRY,17728
 gpustack_runtime/deployer/k8s/types/kubelet/deviceplugin/v1beta1/__init__.py,sha256=3rOYmgDIIJ4idEtwgnumGStH7PaK-J7EYrOnLa9A-8o,118
 gpustack_runtime/deployer/k8s/types/kubelet/deviceplugin/v1beta1/api.proto,sha256=rmB8RDe4LN5FCVkQ608uS-pl32mk5tt6iGe-g2lKtPs,7919
-gpustack_runtime/deployer/k8s/types/kubelet/deviceplugin/v1beta1/api_pb2.py,sha256=DzmeyfwekwE0U0dK6WfwX1NQYLEr2mFa1hdaUTxt2f8,8404
+gpustack_runtime/deployer/k8s/types/kubelet/deviceplugin/v1beta1/api_pb2.py,sha256=dNkzjTE-2y25q8NF0QRznNJ5r1-5ZxxJS598MHbjx98,45998
 gpustack_runtime/deployer/k8s/types/kubelet/deviceplugin/v1beta1/api_pb2.pyi,sha256=lq1dbSgBYqJ7zyGfoKKHCyfr6R5vcCGzJxteeyQpbuI,8232
 gpustack_runtime/deployer/k8s/types/kubelet/deviceplugin/v1beta1/api_pb2_grpc.py,sha256=GM6EyCEFeyOjL0XOCisbcHurRoLKqKDUI5obsUyTxpE,17446
 gpustack_runtime/deployer/k8s/types/kubelet/deviceplugin/v1beta1/constants.py,sha256=tpNk3e_cvY67C9RwVsdTNl75YuNXBgsn53fSJIzeTR4,828
@@ -59,9 +59,9 @@ gpustack_runtime/detector/pyixml/__init__.py,sha256=6ss_Dyl8lIT4WrKpfwmQqzBmg4Bx
 gpustack_runtime/detector/pymxsml/__init__.py,sha256=YxfNHq7TWd7CpNroP45BGXhcWNpY_sXgVzNGtx68DII,45409
 gpustack_runtime/detector/pyrocmcore/__init__.py,sha256=rgwIdPS-7GG7_5luRMR1XG9QyNM3lJh5ryD7kfZqpWg,2523
 gpustack_runtime/detector/pyrocmsmi/__init__.py,sha256=ACwRtJWVIuJ4NTcBJxk0zrVb_qtDOMkApMdbJoag5g0,11906
-gpustack_runtime/_version_appendix.py,sha256=C8KJWDWnWAHNEMYzKYXyeA2PTIO31o7aTAhQlRkmhUA,23
-gpustack_runtime-0.1.41.dist-info/METADATA,sha256=RDItHofAkys1Iqn6uwI03D6nqq-QrrMOcaJ43Iu2XH4,2358
-gpustack_runtime-0.1.41.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-gpustack_runtime-0.1.41.dist-info/entry_points.txt,sha256=bBO_61GxP6dIT74uZwbSDgW5Vt2pTePUS3CgjUJkUgg,68
-gpustack_runtime-0.1.41.dist-info/licenses/LICENSE,sha256=OiPibowBvB-NHV3TP_NOj18XNBlXcshXZFMpa3uvKVE,10362
-gpustack_runtime-0.1.41.dist-info/RECORD,,
+gpustack_runtime/_version_appendix.py,sha256=VLCosBR_TL4DtCZkpcNTzR57fiBiNckG8dZUua523Ok,23
+gpustack_runtime-0.1.41.post2.dist-info/METADATA,sha256=KgdFo9ZEq09i9TQWXLYgfmd7mbvAUxY5SYVocoAOJSM,2364
+gpustack_runtime-0.1.41.post2.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+gpustack_runtime-0.1.41.post2.dist-info/entry_points.txt,sha256=bBO_61GxP6dIT74uZwbSDgW5Vt2pTePUS3CgjUJkUgg,68
+gpustack_runtime-0.1.41.post2.dist-info/licenses/LICENSE,sha256=OiPibowBvB-NHV3TP_NOj18XNBlXcshXZFMpa3uvKVE,10362
+gpustack_runtime-0.1.41.post2.dist-info/RECORD,,