Merge branch 'main' into k8s-autoscaling

Author: maggiezha

src/lib/onboard/sandbox-gpu-preflight-routing.test.ts

@@ -92,6 +92,61 @@ describe("sandbox GPU preflight routing", () => {
     expect(dockerInfo).not.toHaveBeenCalled();
   });
 
+  it("falls back to Docker's default CDI spec dirs when docker info reports none (#7330)", () => {
+    const getDockerCdiSpecDirs = vi.fn(() => []);
+    const findReadableNvidiaCdiSpecFiles = (dirs: string[]) =>
+      dirs.length === 2 && dirs[0] === "/etc/cdi" && dirs[1] === "/var/run/cdi"
+        ? ["/etc/cdi/nvidia.yaml"]
+        : [];
+    const exitProcess = (code: number): never => {
+      throw new Error(`exit:${code}`);
+    };
+
+    expect(() =>
+      validateSandboxGpuPreflight(
+        sandboxGpuConfig(),
+        {
+          platform: "linux",
+          env: {},
+          release: "6.8.0-generic",
+          procVersion: "Linux version 6.8.0-generic",
+          dockerInfoFormat: vi.fn(),
+          getDockerCdiSpecDirs,
+          findReadableNvidiaCdiSpecFiles,
+        },
+        exitProcess,
+      ),
+    ).not.toThrow();
+  });
+
+  it("still fails when the fallback CDI spec dirs hold no NVIDIA spec (#7330)", () => {
+    const errorSpy = vi.spyOn(console, "error").mockImplementation(() => undefined);
+    const exitSpy = vi.spyOn(process, "exit").mockImplementation(((
+      code?: number | string | null,
+    ) => {
+      throw new Error(`exit:${code}`);
+    }) as never);
+
+    try {
+      expect(() =>
+        validateSandboxGpuPreflight(sandboxGpuConfig(), {
+          platform: "linux",
+          env: {},
+          release: "6.8.0-generic",
+          procVersion: "Linux version 6.8.0-generic",
+          dockerInfoFormat: vi.fn(),
+          getDockerCdiSpecDirs: vi.fn(() => []),
+          findReadableNvidiaCdiSpecFiles: vi.fn(() => []),
+        }),
+      ).toThrow("exit:1");
+      const message = errorSpy.mock.calls.map((call) => call[0]).join("\n");
+      expect(message).toContain("Docker CDI GPU support was not detected");
+    } finally {
+      errorSpy.mockRestore();
+      exitSpy.mockRestore();
+    }
+  });
+
   it("skips CDI spec validation on Docker Desktop WSL so Docker --gpus can be used", () => {
     const logSpy = vi.spyOn(console, "log").mockImplementation(() => undefined);
     const getDockerCdiSpecDirs = vi.fn(() => ["/etc/cdi"]);

src/lib/onboard/sandbox-gpu-preflight.ts

@@ -18,6 +18,12 @@ export { formatSandboxGpuPassthroughNote } from "./sandbox-gpu-notes";
 
 const SANDBOX_GPU_PREFLIGHT_TIMEOUT_MS = 30_000;
 
+// Docker Engine's built-in CDI spec directories. `docker info` can report no
+// CDISpecDirs (daemon unreachable from this process, or an engine that omits
+// the field) even though CDI works, so the preflight falls back to these
+// before declaring CDI unsupported (#7330).
+const FALLBACK_DOCKER_CDI_SPEC_DIRS = ["/etc/cdi", "/var/run/cdi"];
+
 export type SandboxGpuPreflightDeps = WslDockerDesktopDetectionDeps & {
   getDockerCdiSpecDirs?: () => string[];
   findReadableNvidiaCdiSpecFiles?: (dirs: string[]) => string[];
@@ -368,7 +374,9 @@ export function validateSandboxGpuPreflight(
     return;
   }
 
-  const cdiSpecDirs = (deps.getDockerCdiSpecDirs ?? getDockerCdiSpecDirs)();
+  const reportedCdiSpecDirs = (deps.getDockerCdiSpecDirs ?? getDockerCdiSpecDirs)();
+  const cdiSpecDirs =
+    reportedCdiSpecDirs.length > 0 ? reportedCdiSpecDirs : FALLBACK_DOCKER_CDI_SPEC_DIRS;
   const cdiSpecFiles = (deps.findReadableNvidiaCdiSpecFiles ?? findReadableNvidiaCdiSpecFiles)(
     cdiSpecDirs,
   );