gpustack-runtime 0.1.44.post4__tar.gz → 0.2.0__tar.gz

{gpustack_runtime-0.1.44.post4 → gpustack_runtime-0.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpustack-runtime
-Version: 0.1.44.post4
+Version: 0.2.0
 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
@@ -16,17 +16,13 @@ Requires-Python: >=3.10
 Requires-Dist: argcomplete>=3.6.3
 Requires-Dist: cachetools>=5.5.2
 Requires-Dist: docker>=7.1.0
-Requires-Dist: gpustack-runner>=0.1.25.post7
-Requires-Dist: grpc-interceptor>=0.15.4
-Requires-Dist: grpcio>=1.76.0
+Requires-Dist: gpustack-runner>=0.1.26
 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<=3.19.6
 Requires-Dist: pyyaml
 Requires-Dist: tqdm
-Requires-Dist: types-protobuf<=5.28.3.20241203
 Description-Content-Type: text/markdown
 
 # GPUStack Runtime

{gpustack_runtime-0.1.44.post4 → gpustack_runtime-0.2.0}/deploy/manifests/docker-compose.yaml

@@ -32,15 +32,12 @@ services:
   #    see https://github.com/gpustack/runtime/blob/016531084000df8075d53772a3c33f41d829ba37/gpustack_runtime/envs.py#L48-L55 for details.
   # - GPUSTACK_RUNTIME_DEPLOY_MIRRORED_NAME:
   #    Specify the name of this Container for gpustack-runtime identify itself.
-  # - GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY:
-  #    Specify the directory of CDI device interface specs.
   #
     container_name: gpustack-runtime
     environment:
       - GPUSTACK_RUNTIME_DEPLOY=Docker
       - GPUSTACK_RUNTIME_DEPLOY_MIRRORED_DEPLOYMENT=true
       - GPUSTACK_RUNTIME_DEPLOY_MIRRORED_NAME=gpustack-runtime
-      - GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY=/var/run/cdi
 
   # Mount necessary volume for gpustack-runtime to work properly.
   # - Mount docker socket for docker runtime.
@@ -50,7 +47,6 @@ services:
     volumes:
       - gpustack-runtime-data:/var/lib/gpustack
       - /var/run/docker.sock:/var/run/docker.sock:ro
-      - /var/run/cdi:/var/run/cdi
   #   - /opt/...:/opt/...:ro
 volumes:
   gpustack-runtime-data: { }

{gpustack_runtime-0.1.44.post4 → gpustack_runtime-0.2.0}/deploy/manifests/kubernetes.yaml

@@ -109,13 +109,9 @@ spec:
     #     Change this if you want to deploy workloads into another namespace.
     # - GPUSTACK_RUNTIME_KUBERNETES_NODE_NAME:
     #     Make sure that deploy workloads on the same node.
-    # - GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY:
-    #     Specify the directory of CDI device interface specs.
     # - GPUSTACK_RUNTIME_KUBERNETES_RESOURCE_INJECTION_POLICY:
     #     Specify the resource injection policy, "KDP" is recommended.
     #     Change to "Env" if you want to use environment variable injection policy.
-    # - GPUSTACK_RUNTIME_KUBERNETES_KDP_PER_DEVICE_MAX_ALLOCATIONS:
-    #     Specify the maximum allocations for one device in "KDP" policy.
     #
           env:
             - name: GPUSTACK_RUNTIME_DEPLOY
@@ -134,12 +130,8 @@ spec:
               valueFrom:
                 fieldRef:
                   fieldPath: spec.nodeName
-            - name: GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY
-              value: "/var/run/cdi"
             - name: GPUSTACK_RUNTIME_KUBERNETES_RESOURCE_INJECTION_POLICY
               value: "KDP"
-            - name: GPUSTACK_RUNTIME_KUBERNETES_KDP_PER_DEVICE_MAX_ALLOCATIONS
-              value: "10"
 
     # Mount necessary volume for gpustack-runtime to work properly.
     # - Mount gpustack data volume for persistent data storage.
@@ -148,23 +140,11 @@ spec:
           volumeMounts:
             - name: gpustack-runtime-data
               mountPath: /var/lib/gpustack
-            - name: cdi
-              mountPath: /var/run/cdi
-            - name: kubelet-device-plugins
-              mountPath: /var/lib/kubelet/device-plugins
     #       - name: toolkit
     #         mountPath: /opt/...
       volumes:
         - name: gpustack-runtime-data
           emptyDir: { }
-        - name: cdi
-          hostPath:
-            path: /var/run/cdi
-            type: DirectoryOrCreate
-        - name: kubelet-device-plugins
-          hostPath:
-            path: /var/lib/kubelet/device-plugins
-            type: DirectoryOrCreate
     #   - name: toolkit
     #     hostPath:
     #       path: /opt/...

{gpustack_runtime-0.1.44.post4 → gpustack_runtime-0.2.0}/gpustack_runtime/__main__.py

@@ -19,7 +19,6 @@ from .cmds import (
     DetectDevicesSubCommand,
     ExecSelfSubCommand,
     ExecWorkloadSubCommand,
-    GenerateCDIConfigSubCommand,
     GetDevicesTopologySubCommand,
     GetWorkloadSubCommand,
     InspectSelfSubCommand,
@@ -29,7 +28,6 @@ from .cmds import (
     LogsSelfSubCommand,
     LogsWorkloadSubCommand,
     SaveImagesSubCommand,
-    ServeDevicePluginSubCommand,
 )
 from .logging import setup_logging
 
@@ -77,8 +75,6 @@ def main():
     LogsSelfSubCommand.register(subcommand_parser)
     ExecSelfSubCommand.register(subcommand_parser)
     InspectSelfSubCommand.register(subcommand_parser)
-    GenerateCDIConfigSubCommand.register(subcommand_parser)
-    ServeDevicePluginSubCommand.register(subcommand_parser)
     ListImagesSubCommand.register(subcommand_parser)
     SaveImagesSubCommand.register(subcommand_parser)
     LoadImagesSubCommand.register(subcommand_parser)

{gpustack_runtime-0.1.44.post4 → gpustack_runtime-0.2.0}/gpustack_runtime/_version.py

@@ -27,8 +27,8 @@ version_tuple: VERSION_TUPLE
 __commit_id__: COMMIT_ID
 commit_id: COMMIT_ID
 
-__version__ = version = '0.1.44.post4'
-__version_tuple__ = version_tuple = (0, 1, 44, 'post4')
+__version__ = version = '0.2.0'
+__version_tuple__ = version_tuple = (0, 2, 0)
 try:
     from ._version_appendix import git_commit
     __commit_id__ = commit_id = git_commit

gpustack_runtime-0.2.0/gpustack_runtime/_version_appendix.py

@@ -0,0 +1 @@
+git_commit = "54957e1"

{gpustack_runtime-0.1.44.post4 → gpustack_runtime-0.2.0}/gpustack_runtime/cmds/__init__.py

@@ -6,14 +6,12 @@ from .deployer import (
     DeleteWorkloadSubCommand,
     ExecSelfSubCommand,
     ExecWorkloadSubCommand,
-    GenerateCDIConfigSubCommand,
     GetWorkloadSubCommand,
     InspectSelfSubCommand,
     InspectWorkloadSubCommand,
     ListWorkloadsSubCommand,
     LogsSelfSubCommand,
     LogsWorkloadSubCommand,
-    ServeDevicePluginSubCommand,
 )
 from .detector import DetectDevicesSubCommand, GetDevicesTopologySubCommand
 from .images import (
@@ -34,7 +32,6 @@ __all__ = [
     "DetectDevicesSubCommand",
     "ExecSelfSubCommand",
     "ExecWorkloadSubCommand",
-    "GenerateCDIConfigSubCommand",
     "GetDevicesTopologySubCommand",
     "GetWorkloadSubCommand",
     "InspectSelfSubCommand",
@@ -46,7 +43,6 @@ __all__ = [
     "LogsWorkloadSubCommand",
     "PlatformedImage",
     "SaveImagesSubCommand",
-    "ServeDevicePluginSubCommand",
     "append_images",
     "list_images",
 ]

{gpustack_runtime-0.1.44.post4 → gpustack_runtime-0.2.0}/gpustack_runtime/cmds/deployer.py

@@ -5,7 +5,7 @@ import json
 import os
 import sys
 import time
-from argparse import OPTIONAL, REMAINDER
+from argparse import REMAINDER
 from pathlib import Path
 from typing import TYPE_CHECKING
 
@@ -25,10 +25,6 @@ from ..deployer import (
     WorkloadStatusStateEnum,
     async_logs_self,
     async_logs_workload,
-    cdi_available_backends,
-    cdi_available_manufacturers,
-    cdi_dump_config,
-    cdi_supported_manufacturers,
     create_workload,
     delete_workload,
     exec_self,
@@ -36,7 +32,6 @@ from ..deployer import (
     get_workload,
     inspect_self,
     inspect_workload,
-    k8s_deviceplugin_serve,
     list_workloads,
 )
 from ..detector import (
@@ -992,163 +987,6 @@ class InspectSelfSubCommand(SubCommand):
         print(inspect_self())
 
 
-class GenerateCDIConfigSubCommand(SubCommand):
-    """
-    Command to generate CDI configurations.
-    """
-
-    manufacturer: ManufacturerEnum
-    format: str
-    output: Path | None
-
-    @staticmethod
-    def register(parser: _SubParsersAction):
-        generate_parser = parser.add_parser(
-            "generate-cdi",
-            help="Generate Container Device Interface(CDI) configurations according to the current environment",
-            aliases=["gen-cdi"],
-        )
-
-        generate_parser.add_argument(
-            "--manufacturer",
-            "--manu",
-            type=ManufacturerEnum,
-            help="Manufacturer to generate (default: all)",
-            choices=list(map(str, cdi_available_manufacturers())),
-        )
-
-        generate_parser.add_argument(
-            "--backend",
-            type=str,
-            help="Backend to generate (default: all)",
-            choices=cdi_available_backends(),
-        )
-
-        generate_parser.add_argument(
-            "--format",
-            type=str,
-            choices=["yaml", "json"],
-            default="yaml",
-            help="Format of the CDI configurations",
-        )
-
-        generate_parser.add_argument(
-            "output",
-            nargs=OPTIONAL,
-            help="Output directory to save CDI configurations (default: current directory)",
-        )
-
-        generate_parser.set_defaults(func=GenerateCDIConfigSubCommand)
-
-    def __init__(self, args: Namespace):
-        self.manufacturer = args.manufacturer
-        self.format = args.format
-        self.output = Path(args.output) if args.output else None
-
-        if args.backend:
-            if not self.manufacturer:
-                self.manufacturer = backend_to_manufacturer(args.backend)
-            elif args.backend != manufacturer_to_backend(self.manufacturer):
-                msg = (
-                    f"The backend '{args.backend}' is not compatible with "
-                    f"the manufacturer '{args.manufacturer}'."
-                )
-                raise ValueError(msg)
-
-        if self.output:
-            try:
-                if not self.output.exists():
-                    self.output.mkdir(parents=True, exist_ok=True)
-            except OSError as e:
-                msg = f"Failed to prepare output directory '{self.output}' for CDI configurations"
-                raise RuntimeError(msg) from e
-
-            if not self.output.is_dir():
-                msg = f"The output path '{self.output}' is not a directory"
-                raise RuntimeError(msg)
-
-    def run(self):
-        print("\033[2J\033[H", end="")
-
-        generated = False
-        for manu in cdi_supported_manufacturers():
-            if self.manufacturer and self.manufacturer != manu:
-                continue
-            content, path = cdi_dump_config(
-                manufacturer=manu,
-                output=self.output,
-            )
-            if content:
-                generated = True
-                if path:
-                    print(f"Generated CDI configuration for '{manu}' at {path}:\n")
-                else:
-                    print(f"Generated CDI configuration for '{manu}':\n")
-                print(content)
-                print()
-
-        if not generated:
-            print("No CDI configurations were generated.")
-
-
-class ServeDevicePluginSubCommand(SubCommand):
-    """
-    Command to serve Device Plugin.
-    """
-
-    manufacturer: ManufacturerEnum
-
-    @staticmethod
-    def register(parser: _SubParsersAction):
-        serve_parser = parser.add_parser(
-            "serve-device-plugin",
-            help="Serve Device Plugin according to the current environment",
-            aliases=[
-                "start-device-plugin",
-                "serve-dp",
-                "start-dp",
-            ],
-        )
-
-        serve_parser.add_argument(
-            "--manufacturer",
-            "--manu",
-            type=ManufacturerEnum,
-            help="Manufacturer to generate (default: all)",
-            choices=list(map(str, cdi_available_manufacturers())),
-        )
-
-        serve_parser.add_argument(
-            "--backend",
-            type=str,
-            help="Backend to generate (default: all)",
-            choices=cdi_available_backends(),
-        )
-
-        serve_parser.set_defaults(func=ServeDevicePluginSubCommand)
-
-    def __init__(self, args: Namespace):
-        self.manufacturer = args.manufacturer
-
-        if args.backend:
-            if not self.manufacturer:
-                self.manufacturer = backend_to_manufacturer(args.backend)
-            elif args.backend != manufacturer_to_backend(self.manufacturer):
-                msg = (
-                    f"The backend '{args.backend}' is not compatible with "
-                    f"the manufacturer '{args.manufacturer}'."
-                )
-                raise ValueError(msg)
-
-    def run(self):
-        print("\033[2J\033[H", end="")
-
-        k8s_deviceplugin_serve(
-            manufacturer=self.manufacturer,
-            cdi_generation_output=envs.GPUSTACK_RUNTIME_DEPLOY_CDI_SPECS_DIRECTORY,
-        )
-
-
 def format_workloads_json(sts: list[WorkloadStatus]) -> str:
     return json.dumps([st.to_dict() for st in sts], indent=2)
 

{gpustack_runtime-0.1.44.post4 → gpustack_runtime-0.2.0}/gpustack_runtime/deployer/__init__.py

@@ -31,32 +31,11 @@ from .__types__ import (
     WorkloadStatus,
     WorkloadStatusStateEnum,
 )
-from .cdi import (
-    available_backends as cdi_available_backends,
-)
-from .cdi import (
-    available_manufacturers as cdi_available_manufacturers,
-)
-from .cdi import (
-    dump_config as cdi_dump_config,
-)
-from .cdi import (
-    generate_config as cdi_generate_config,
-)
-from .cdi import (
-    supported_manufacturers as cdi_supported_manufacturers,
-)
 from .docker import (
     DockerDeployer,
     DockerWorkloadPlan,
     DockerWorkloadStatus,
 )
-from .k8s.deviceplugin import (
-    serve as k8s_deviceplugin_serve,
-)
-from .k8s.deviceplugin import (
-    serve_async as k8s_deviceplugin_serve_async,
-)
 from .kuberentes import (
     KubernetesDeployer,
     KubernetesWorkloadPlan,
@@ -623,11 +602,6 @@ __all__ = [
     "WorkloadStatusStateEnum",
     "async_logs_self",
     "async_logs_workload",
-    "cdi_available_backends",
-    "cdi_available_manufacturers",
-    "cdi_dump_config",
-    "cdi_generate_config",
-    "cdi_supported_manufacturers",
     "create_workload",
     "delete_workload",
     "exec_self",
@@ -635,8 +609,6 @@ __all__ = [
     "get_workload",
     "inspect_self",
     "inspect_workload",
-    "k8s_deviceplugin_serve",
-    "k8s_deviceplugin_serve_async",
     "list_workloads",
     "logs_self",
     "logs_workload",

{gpustack_runtime-0.1.44.post4 → gpustack_runtime-0.2.0}/gpustack_runtime/deployer/__types__.py

@@ -30,7 +30,7 @@ from .__utils__ import (
     safe_yaml,
     validate_rfc1123_subdomain_name,
 )
-from .k8s.deviceplugin import cdi_kind_to_kdp_resource
+from .k8s.devicemanager import cdi_kind_to_kdp_resource
 
 if TYPE_CHECKING:
     from collections.abc import AsyncGenerator, Generator
@@ -1491,7 +1491,7 @@ class Deployer(ABC):
         self,
         runtime_env: str,
         fmt: str = "plain",
-    ) -> list[str]:
+    ) -> list[str] | dict[str, str]:
         """
         Return the runtime visible devices values for the given runtime visible devices env name.
 
@@ -1500,10 +1500,11 @@ class Deployer(ABC):
                 The runtime visible devices environment variable name.
             fmt:
                 The format of the returned values,
-                can be "cdi", "kdp", or "plain".
+                can be "cdi", "plain" or "kdp".
 
         Returns:
-            A list of runtime visible devices values.
+            A list of runtime visible devices values if fmt is "cdi" or "plain",
+            or a dictionary mapping CDI resource to the count of runtime visible devices if fmt is "kdp".
 
         """
         m = self._get_materials()
@@ -1512,22 +1513,22 @@ class Deployer(ABC):
             return []
 
         rm = m[runtime_env]
-        match fmt:
-            case "cdi":
-                return [f"{rm.cdi}={v}" for v in rm.runtime_values.values()]
-            case "kdp":
-                return [
-                    cdi_kind_to_kdp_resource(rm.cdi, v)
-                    for v in rm.runtime_values.values()
-                ]
-        return list(rm.runtime_values.values())
+
+        if fmt == "cdi":
+            return [f"{rm.cdi}={v}" for v in rm.runtime_values.values()]
+        if fmt == "plain":
+            return list(rm.runtime_values.values())
+
+        return {
+            cdi_kind_to_kdp_resource(rm.cdi): str(len(rm.runtime_values)),
+        }
 
     def map_runtime_visible_devices(
         self,
         runtime_env: str,
         resource_values: list[str],
         fmt: str = "plain",
-    ) -> list[str]:
+    ) -> list[str] | dict[str, str]:
         """
         Map the given resource values to runtime visible devices values
         for the given runtime visible devices env name.
@@ -1539,10 +1540,11 @@ class Deployer(ABC):
                 The resource values to map.
             fmt:
                 The format of the returned values,
-                can be "cdi", "kdp", or "plain".
+                can be "cdi", "plain" or "kdp".
 
         Returns:
-            A list of mapped runtime visible devices values.
+            A list of mapped runtime visible devices values if fmt is "cdi" or "plain",
+            or a dictionary mapping CDI resource to the count of runtime visible devices if fmt is "kdp".
 
         """
         m = self._get_materials()
@@ -1551,17 +1553,13 @@ class Deployer(ABC):
             return []
 
         rm = m[runtime_env]
-        match fmt:
-            case "cdi":
-                return [
-                    f"{rm.cdi}={rm.runtime_values.get(v, v)}" for v in resource_values
-                ]
-            case "kdp":
-                return [
-                    cdi_kind_to_kdp_resource(rm.cdi, rm.runtime_values.get(v, v))
-                    for v in resource_values
-                ]
-        return [rm.runtime_values.get(v, v) for v in resource_values]
+        if fmt == "cdi":
+            return [f"{rm.cdi}={rm.runtime_values.get(v, v)}" for v in resource_values]
+        if fmt == "plain":
+            return [rm.runtime_values.get(v, v) for v in resource_values]
+        return {
+            cdi_kind_to_kdp_resource(rm.cdi): str(len(resource_values)),
+        }
 
     def map_backend_visible_devices(
         self,

gpustack_runtime-0.2.0/gpustack_runtime/deployer/k8s/devicemanager/__init__.py

@@ -0,0 +1,78 @@
+from __future__ import annotations as __future_annotations__
+
+import stat
+from functools import lru_cache
+from pathlib import Path
+from typing import Literal
+
+from gpustack_runtime import envs
+
+
+def is_kubelet_socket_accessible(
+    kubelet_endpoint: Path | None = None,
+) -> bool:
+    """
+    Check if the kubelet socket is accessible.
+
+    Args:
+        kubelet_endpoint:
+            The path to the kubelet endpoint.
+
+    Returns:
+        True if the socket is accessible, False otherwise.
+
+    """
+    if not kubelet_endpoint:
+        kubelet_endpoint = Path("/var/lib/kubelet/device-plugins/kubelet.sock")
+
+    if kubelet_endpoint.exists():
+        path_stat = kubelet_endpoint.lstat()
+        if path_stat and stat.S_ISSOCK(path_stat.st_mode):
+            return True
+    return False
+
+
+@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"
+
+
+@lru_cache
+def cdi_kind_to_kdp_resource(
+    cdi_kind: str,
+    mode: Literal["shared", "sliced"] = "shared",
+):
+    """
+    Convert a CDI kind to a KDP resource name.
+
+    Args:
+        cdi_kind:
+            The CDI kind to convert.
+        mode:
+            The mode of the resource, either "shared" or "sliced".
+
+    Returns:
+        The corresponding KDP resource name.
+
+    """
+    if mode == "shared":
+        return f"{cdi_kind}.shared"
+    return f"{cdi_kind}.sliced.units"
+
+
+__all__ = [
+    "cdi_kind_to_kdp_resource",
+    "get_resource_injection_policy",
+    "is_kubelet_socket_accessible",
+]