gpustack-runtime 0.2.2.post6__tar.gz → 0.2.2.post7__tar.gz

{gpustack_runtime-0.2.2.post6 → gpustack_runtime-0.2.2.post7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpustack-runtime
-Version: 0.2.2.post6
+Version: 0.2.2.post7
 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

{gpustack_runtime-0.2.2.post6 → gpustack_runtime-0.2.2.post7}/gpustack_runtime/_version.py

@@ -27,8 +27,8 @@ version_tuple: VERSION_TUPLE
 __commit_id__: COMMIT_ID
 commit_id: COMMIT_ID
 
-__version__ = version = '0.2.2.post6'
-__version_tuple__ = version_tuple = (0, 2, 2, 'post6')
+__version__ = version = '0.2.2.post7'
+__version_tuple__ = version_tuple = (0, 2, 2, 'post7')
 try:
     from ._version_appendix import git_commit
     __commit_id__ = commit_id = git_commit

gpustack_runtime-0.2.2.post7/gpustack_runtime/_version_appendix.py

@@ -0,0 +1 @@
+git_commit = "9147850"

{gpustack_runtime-0.2.2.post6 → gpustack_runtime-0.2.2.post7}/gpustack_runtime/deployer/kuberentes.py

@@ -325,6 +325,61 @@ def _pin_pod_for_kueue(
     }
 
 
+def _is_device_plugin_resource(resource_key: str) -> bool:
+    """
+    Report whether a resource key belongs to a device plugin resource family,
+    which is either a CDI kind ("nvidia.com/gpu") or one of its suffixed
+    variants ("nvidia.com/gpu.shared", "nvidia.com/gpu.sliced.units",
+    "nvidia.com/gpu.partitioned.mig-1g.20gb").
+    """
+    return any(
+        resource_key == cdi or resource_key.startswith(f"{cdi}.")
+        for cdi in envs.GPUSTACK_RUNTIME_DEPLOY_RESOURCE_KEY_MAP_CDI.values()
+    )
+
+
+def _resolve_privileged(container: Container) -> bool:
+    """
+    Resolve whether a container runs privileged.
+
+    Privilege is dropped when the container's devices are handed out by a
+    device plugin, which is the case for every device plugin resource family
+    and, under the KDP injection policy, for every mapped device request.
+
+    A privileged container receives all device nodes of the host, so it
+    enumerates -- and can use -- every accelerator on the node, no matter
+    which one the device plugin allocated to it. That silently undoes
+    slicing: a workload holding a single MIG device or a single memory slice
+    still sees the untouched cards next to it, and a soft-slicing limit
+    lands on whichever device comes first instead of the allocated one.
+    """
+    if not container.execution or not container.execution.privileged:
+        return False
+    if not container.resources:
+        return True
+
+    kdp = get_resource_injection_policy() == "kdp"
+    for r_k in container.resources:
+        if r_k in ("cpu", "memory"):
+            continue
+        if _is_device_plugin_resource(r_k) or (
+            kdp
+            and (
+                r_k
+                in envs.GPUSTACK_RUNTIME_DEPLOY_RESOURCE_KEY_MAP_RUNTIME_VISIBLE_DEVICES
+                or r_k == envs.GPUSTACK_RUNTIME_DEPLOY_AUTOMAP_RESOURCE_KEY
+            )
+        ):
+            clogger.info(
+                "Dropping privilege of container '%s', "
+                "as its device request '%s' is allocated by a device plugin",
+                container.name,
+                r_k,
+            )
+            return False
+    return True
+
+
 class KubernetesDeployer(EndoscopicDeployer):
     """
     Deployer implementation for Kubernetes.
@@ -1048,7 +1103,7 @@ class KubernetesDeployer(EndoscopicDeployer):
                     run_as_user=c.execution.run_as_user,
                     run_as_group=c.execution.run_as_group,
                     read_only_root_filesystem=c.execution.readonly_rootfs,
-                    privileged=c.execution.privileged,
+                    privileged=_resolve_privileged(c),
                     capabilities=(
                         kubernetes.client.V1Capabilities(
                             add=c.execution.capabilities.add,

gpustack_runtime-0.2.2.post7/tests/gpustack_runtime/deployer/test_privileged.py

@@ -0,0 +1,137 @@
+import pytest
+
+from gpustack_runtime.deployer.__types__ import (
+    Container,
+    ContainerExecution,
+    ContainerResources,
+)
+from gpustack_runtime.deployer.kuberentes import _resolve_privileged
+
+
+def _container(privileged: bool | None, resources: dict | None = None) -> Container:
+    container_resources = None
+    if resources is not None:
+        container_resources = ContainerResources()
+        container_resources.update(resources)
+    return Container(
+        name="default",
+        image="gpustack/runner:latest",
+        execution=(
+            ContainerExecution(privileged=privileged)
+            if privileged is not None
+            else None
+        ),
+        resources=container_resources,
+    )
+
+
+@pytest.mark.parametrize(
+    "name, privileged, resources, policy, expected",
+    [
+        (
+            "no execution",
+            None,
+            None,
+            "env",
+            False,
+        ),
+        (
+            "not requested",
+            False,
+            {"nvidia.com/devices": "0"},
+            "env",
+            False,
+        ),
+        (
+            "no resources",
+            True,
+            None,
+            "env",
+            True,
+        ),
+        (
+            "non-device resources only",
+            True,
+            {"cpu": "2", "memory": "4Gi"},
+            "env",
+            True,
+        ),
+        (
+            "specific whole cards, env injection",
+            True,
+            {"cpu": "2", "nvidia.com/devices": "0,1"},
+            "env",
+            True,
+        ),
+        (
+            "all devices, env injection",
+            True,
+            {"nvidia.com/devices": "all"},
+            "env",
+            True,
+        ),
+        (
+            "specific whole cards, kdp injection",
+            True,
+            {"nvidia.com/devices": "0,1"},
+            "kdp",
+            False,
+        ),
+        (
+            "auto-mapped devices, kdp injection",
+            True,
+            {"gpustack.ai/devices": "0"},
+            "kdp",
+            False,
+        ),
+        (
+            "exclusive whole card",
+            True,
+            {"nvidia.com/gpu": "1"},
+            "env",
+            False,
+        ),
+        (
+            "soft slice",
+            True,
+            {
+                "nvidia.com/gpu.sliced": "1",
+                "nvidia.com/gpu.sliced.memory-percentage": "50",
+                "nvidia.com/gpu.sliced.cores-percentage": "50",
+            },
+            "env",
+            False,
+        ),
+        (
+            "hard partition",
+            True,
+            {
+                "nvidia.com/gpu.partitioned": "1",
+                "nvidia.com/gpu.partitioned.mig-1g.20gb": "1",
+            },
+            "env",
+            False,
+        ),
+        (
+            "non-NVIDIA soft slice",
+            True,
+            {"amd.com/gpu.sliced": "1"},
+            "env",
+            False,
+        ),
+        (
+            "resource key merely prefixed by a CDI kind",
+            True,
+            {"nvidia.com/gpu-alike": "1"},
+            "env",
+            True,
+        ),
+    ],
+)
+def test_resolve_privileged(name, privileged, resources, policy, expected, monkeypatch):
+    monkeypatch.setattr(
+        "gpustack_runtime.deployer.kuberentes.get_resource_injection_policy",
+        lambda: policy,
+    )
+    actual = _resolve_privileged(_container(privileged, resources))
+    assert actual == expected, f"case {name} expected {expected}, but got {actual}"

gpustack_runtime-0.2.2.post6/gpustack_runtime/_version_appendix.py

@@ -1 +0,0 @@
-git_commit = "79ee112"