@nexusgpu/repterm-plugin-kubectl 0.1.0

package/dist/plugin-kubectl/examples/tensor-fusion/07-workload-worker-pod-generation.js

@@ -0,0 +1,146 @@
+/**
+ * 测试场景 7: Workload 创建和 Worker Pod 生成
+ *
+ * 基于 `Workload创建和WorkerPod生成测试.md`：
+ * - 创建 `replicas=2` 的 TensorFusionWorkload
+ * - 等待 `status.workerCount` 变为 2
+ * - 等待 worker Pod Ready
+ * - 验证 worker Pod 数量与 `spec.replicas` 一致
+ * - 验证 `status.workerCount` 正确更新
+ *
+ * 运行方式:
+ *   bun run repterm packages/plugin-kubectl/examples/tensor-fusion/07-workload-worker-pod-generation.ts
+ */
+import { sleep } from 'bun';
+import { test, describe, expect, step, tensorfusionworkload, TEST_NAMESPACE, TEST_GPU_POOL, } from './_config.js';
+const WORKLOAD_NAME = 'tf-workload-worker-count-test';
+const EXPECTED_REPLICAS = 2;
+const WAIT_TIMEOUT = 180000;
+const POLL_INTERVAL = 1500;
+function workloadWorkerCountYaml(name, poolName) {
+    return `
+apiVersion: tensor-fusion.ai/v1
+kind: TensorFusionWorkload
+metadata:
+  name: ${name}
+  namespace: ${TEST_NAMESPACE}
+  labels:
+    app: ${name}
+    test-type: worker-count
+spec:
+  replicas: ${EXPECTED_REPLICAS}
+  gpuCount: 1
+  poolName: ${poolName}
+  qos: medium
+  isolation: soft
+  isLocalGPU: false
+  resources:
+    requests:
+      tflops: "100m"
+      vram: "1Gi"
+    limits:
+      tflops: "100m"
+      vram: "1Gi"
+`;
+}
+async function listWorkerPods(kubectl, workloadName) {
+    const strictSelector = `tensor-fusion.ai/workload=${workloadName},tensor-fusion.ai/component=worker`;
+    const fallbackSelector = `tensor-fusion.ai/workload=${workloadName}`;
+    const workerPods = await kubectl.get('pod', undefined, {
+        selector: strictSelector,
+        jqFilter: '[.items[] | {name: .metadata.name, phase: .status.phase}]',
+    });
+    if (Array.isArray(workerPods) && workerPods.length > 0) {
+        return workerPods;
+    }
+    const fallbackPods = await kubectl.get('pod', undefined, {
+        selector: fallbackSelector,
+        jqFilter: '[.items[] | {name: .metadata.name, phase: .status.phase}]',
+    });
+    return fallbackPods ?? [];
+}
+async function waitForWorkerPods(kubectl, workloadName, expectedCount, timeoutMs) {
+    const deadline = Date.now() + timeoutMs;
+    while (Date.now() < deadline) {
+        const workerPods = await listWorkerPods(kubectl, workloadName);
+        if (workerPods.length >= expectedCount) {
+            return workerPods;
+        }
+        await sleep(POLL_INTERVAL);
+    }
+    throw new Error(`Timeout waiting worker pods for ${workloadName} reach ${expectedCount}`);
+}
+describe('测试场景 7: Workload 创建和 Worker Pod 生成', { record: true }, () => {
+    test('创建 replicas=2 的 Workload 后生成对应 worker Pods', async (ctx) => {
+        const { kubectl } = ctx.plugins;
+        try {
+            await step('创建 TensorFusionWorkload（replicas=2）', {
+                showStepTitle: false,
+                typingSpeed: 90,
+                pauseAfter: 1800,
+            }, async () => {
+                const yaml = workloadWorkerCountYaml(WORKLOAD_NAME, TEST_GPU_POOL);
+                const result = await kubectl.apply(yaml);
+                await expect(result).toBeSuccessful();
+                const specReplicas = await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.spec.replicas');
+                expect(specReplicas).toBe(EXPECTED_REPLICAS);
+            });
+            await step('等待 Workload status.workerCount=2', {
+                showStepTitle: false,
+                pauseAfter: 1800,
+            }, async () => {
+                await kubectl.waitForJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.status.workerCount', String(EXPECTED_REPLICAS), WAIT_TIMEOUT);
+                const workload = tensorfusionworkload(kubectl, WORKLOAD_NAME);
+                await expect(workload).toHaveStatusField('workerCount', EXPECTED_REPLICAS);
+            });
+            await step('等待 worker Pods 创建并 Ready', {
+                showStepTitle: false,
+                typingSpeed: 80,
+                pauseAfter: 2000,
+            }, async () => {
+                const workerPods = await waitForWorkerPods(kubectl, WORKLOAD_NAME, EXPECTED_REPLICAS, WAIT_TIMEOUT);
+                for (const podInfo of workerPods.slice(0, EXPECTED_REPLICAS)) {
+                    const waitResult = await kubectl.wait('pod', podInfo.name, 'Ready', {
+                        timeout: WAIT_TIMEOUT,
+                    });
+                    await expect(waitResult).toBeSuccessful();
+                }
+            });
+            await step('验证 worker Pod 数量与 replicas 一致', {
+                typingSpeed: 80,
+                pauseAfter: 1800,
+            }, async () => {
+                const expectedReplicas = (await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.spec.replicas')) ?? 0;
+                const workerPods = await listWorkerPods(kubectl, WORKLOAD_NAME);
+                const actualWorkerCount = workerPods.length;
+                expect(actualWorkerCount).toBe(expectedReplicas);
+            });
+            await step('验证 status.workerCount 与 replicas 一致', {
+                showStepTitle: false,
+                pauseAfter: 1500,
+            }, async () => {
+                const statusWorkerCount = (await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.status.workerCount')) ?? 0;
+                expect(statusWorkerCount).toBe(EXPECTED_REPLICAS);
+            });
+        }
+        finally {
+            await step('清理 TensorFusionWorkload', {
+                showStepTitle: false,
+                typingSpeed: 70,
+                pauseAfter: 1200,
+            }, async () => {
+                try {
+                    const exists = await kubectl.exists('tensorfusionworkload', WORKLOAD_NAME);
+                    if (!exists) {
+                        return;
+                    }
+                    const deleteResult = await kubectl.delete('tensorfusionworkload', WORKLOAD_NAME);
+                    await expect(deleteResult).toBeSuccessful();
+                }
+                catch {
+                    // 清理失败不覆盖主断言错误
+                }
+            });
+        }
+    });
+});

package/dist/plugin-kubectl/examples/tensor-fusion/08-workload-replicas-scale.d.ts

@@ -0,0 +1,13 @@
+/**
+ * 测试场景 8: Workload Replicas 扩缩容
+ *
+ * 基于 `WorkloadReplicas扩缩容测.md`：
+ * - 先创建一个 replicas=1 的 TensorFusionWorkload 并等待就绪
+ * - 将 replicas 从 1 扩容到 2，验证 worker pods 增加、status.workerCount=2
+ * - 将 replicas 从 2 缩容回 1，验证 worker pods 减少、status.workerCount=1
+ *
+ * 运行方式:
+ *   bun run repterm packages/plugin-kubectl/examples/tensor-fusion/08-workload-replicas-scale.ts
+ */
+export {};
+//# sourceMappingURL=08-workload-replicas-scale.d.ts.map

package/dist/plugin-kubectl/examples/tensor-fusion/08-workload-replicas-scale.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"08-workload-replicas-scale.d.ts","sourceRoot":"","sources":["../../../../examples/tensor-fusion/08-workload-replicas-scale.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG"}

package/dist/plugin-kubectl/examples/tensor-fusion/08-workload-replicas-scale.js

@@ -0,0 +1,141 @@
+/**
+ * 测试场景 8: Workload Replicas 扩缩容
+ *
+ * 基于 `WorkloadReplicas扩缩容测.md`：
+ * - 先创建一个 replicas=1 的 TensorFusionWorkload 并等待就绪
+ * - 将 replicas 从 1 扩容到 2，验证 worker pods 增加、status.workerCount=2
+ * - 将 replicas 从 2 缩容回 1，验证 worker pods 减少、status.workerCount=1
+ *
+ * 运行方式:
+ *   bun run repterm packages/plugin-kubectl/examples/tensor-fusion/08-workload-replicas-scale.ts
+ */
+import { sleep } from 'bun';
+import { test, describe, expect, step, tensorfusionworkload, workloadYaml, DEFAULT_TIMEOUT, } from './_config.js';
+const WORKLOAD_NAME = 'test-workload-replicas';
+/**
+ * 获取指定 workload 的 worker pod 数量
+ */
+async function getWorkerPodCount(kubectl, workloadName) {
+    const pods = await kubectl.get('pod', undefined, {
+        selector: `tensor-fusion.ai/workload=${workloadName},tensor-fusion.ai/component=worker`,
+        jqFilter: '[.items[] | {name: .metadata.name, phase: .status.phase}]',
+    });
+    return pods?.length ?? 0;
+}
+describe('测试场景 8: Workload Replicas 扩缩容', { record: true }, () => {
+    test('TensorFusionWorkload 扩缩容后 worker pods 与 workerCount 同步更新', async (ctx) => {
+        const { kubectl } = ctx.plugins;
+        // ===== Step 1: 创建 replicas=1 的 Workload 并等待就绪 =====
+        await step('创建 Workload（replicas=1）', {
+            showStepTitle: false,
+            typingSpeed: 100,
+            pauseAfter: 2000,
+        }, async () => {
+            const yaml = workloadYaml(WORKLOAD_NAME, {
+                tflopsRequest: '1000m',
+                tflopsLimit: '2000m',
+                vramRequest: '1Gi',
+                vramLimit: '2Gi',
+                replicas: 1,
+            });
+            const result = await kubectl.apply(yaml);
+            await expect(result).toBeSuccessful();
+        });
+        await step('等待 Workload Ready', {
+            showStepTitle: false,
+            pauseAfter: 1800,
+        }, async () => {
+            const waitResult = await kubectl.wait('tensorfusionworkload', WORKLOAD_NAME, 'Ready', { timeout: DEFAULT_TIMEOUT });
+            await expect(waitResult).toBeSuccessful();
+            const workload = tensorfusionworkload(kubectl, WORKLOAD_NAME);
+            await expect(workload).toHaveStatusField('phase', 'Running');
+        });
+        // ===== Step 2: 记录当前 worker pod 数量 =====
+        await step('记录当前 worker pod 数量', {
+            typingSpeed: 80,
+            pauseAfter: 1500,
+        }, async () => {
+            const beforeCount = await getWorkerPodCount(kubectl, WORKLOAD_NAME);
+            expect(beforeCount).toBe(1);
+            // 验证 status.workerCount 也为 1
+            const status = await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.status');
+            expect(status?.workerCount).toBe(1);
+        });
+        // ===== Step 3: 扩容 replicas 到 2 =====
+        await step('patch Workload replicas=2', {
+            showStepTitle: false,
+            typingSpeed: 100,
+            pauseAfter: 1800,
+        }, async () => {
+            const patchResult = await kubectl.patch('tensorfusionworkload', WORKLOAD_NAME, { spec: { replicas: 2 } }, 'merge');
+            await expect(patchResult).toBeSuccessful();
+            // 确认 spec.replicas 已更新
+            const replicas = await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.spec.replicas');
+            expect(replicas).toBe(2);
+        });
+        // ===== Step 4: 验证扩容结果 =====
+        await step('等待 workerCount=2', {
+            showStepTitle: false,
+            pauseAfter: 2000,
+        }, async () => {
+            const deadline = Date.now() + DEFAULT_TIMEOUT;
+            while (Date.now() < deadline) {
+                const status = await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.status');
+                if (status?.workerCount === 2) {
+                    break;
+                }
+                await sleep(3000);
+            }
+            const status = await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.status');
+            expect(status?.workerCount).toBe(2);
+        });
+        await step('验证扩容后 worker pod 数量为 2', {
+            typingSpeed: 80,
+            pauseAfter: 2500,
+        }, async () => {
+            const afterScaleUpCount = await getWorkerPodCount(kubectl, WORKLOAD_NAME);
+            expect(afterScaleUpCount).toBe(2);
+        });
+        // ===== Step 5: 缩容 replicas 回 1 =====
+        await step('patch Workload replicas=1', {
+            showStepTitle: false,
+            typingSpeed: 100,
+            pauseAfter: 1800,
+        }, async () => {
+            const patchResult = await kubectl.patch('tensorfusionworkload', WORKLOAD_NAME, { spec: { replicas: 1 } }, 'merge');
+            await expect(patchResult).toBeSuccessful();
+        });
+        await step('等待 workerCount=1', {
+            showStepTitle: false,
+            pauseAfter: 2000,
+        }, async () => {
+            const deadline = Date.now() + DEFAULT_TIMEOUT;
+            while (Date.now() < deadline) {
+                const status = await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.status');
+                if (status?.workerCount === 1) {
+                    break;
+                }
+                await sleep(3000);
+            }
+            const status = await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.status');
+            expect(status?.workerCount).toBe(1);
+        });
+        await step('验证缩容后 worker pod 数量为 1', {
+            typingSpeed: 80,
+            pauseAfter: 2500,
+        }, async () => {
+            const afterScaleDownCount = await getWorkerPodCount(kubectl, WORKLOAD_NAME);
+            expect(afterScaleDownCount).toBe(1);
+        });
+        // ===== Step 6: 清理 =====
+        await step('删除 TensorFusionWorkload', {
+            showStepTitle: false,
+            typingSpeed: 80,
+            pauseAfter: 2000,
+        }, async () => {
+            const deleteResult = await kubectl.delete('tensorfusionworkload', WORKLOAD_NAME);
+            await expect(deleteResult).toBeSuccessful();
+            await sleep(5000);
+        });
+    });
+});

package/dist/plugin-kubectl/examples/tensor-fusion/09-gpu-remote-invocation.d.ts

@@ -0,0 +1,15 @@
+/**
+ * 测试场景 9: GPU 远程调用
+ *
+ * 基于 `GPU远程调用测试.md`：
+ * - 创建一个远程模式（isLocalGPU=false）的 TensorFusionWorkload
+ * - 创建一个带 remote annotation 的 client pod
+ * - 验证 TensorFusionConnection 自动创建
+ * - 验证 connection 的 metadata（namespace、labels、ownerReferences）与 spec（workloadName、clientPod）
+ * - 在 client pod 内执行 nvidia-smi 和 PyTorch 验证，确认可通过远程方式使用 GPU
+ *
+ * 运行方式:
+ *   bun run repterm packages/plugin-kubectl/examples/tensor-fusion/09-gpu-remote-invocation.ts
+ */
+export {};
+//# sourceMappingURL=09-gpu-remote-invocation.d.ts.map

package/dist/plugin-kubectl/examples/tensor-fusion/09-gpu-remote-invocation.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"09-gpu-remote-invocation.d.ts","sourceRoot":"","sources":["../../../../examples/tensor-fusion/09-gpu-remote-invocation.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG"}

package/dist/plugin-kubectl/examples/tensor-fusion/09-gpu-remote-invocation.js

@@ -0,0 +1,256 @@
+/**
+ * 测试场景 9: GPU 远程调用
+ *
+ * 基于 `GPU远程调用测试.md`：
+ * - 创建一个远程模式（isLocalGPU=false）的 TensorFusionWorkload
+ * - 创建一个带 remote annotation 的 client pod
+ * - 验证 TensorFusionConnection 自动创建
+ * - 验证 connection 的 metadata（namespace、labels、ownerReferences）与 spec（workloadName、clientPod）
+ * - 在 client pod 内执行 nvidia-smi 和 PyTorch 验证，确认可通过远程方式使用 GPU
+ *
+ * 运行方式:
+ *   bun run repterm packages/plugin-kubectl/examples/tensor-fusion/09-gpu-remote-invocation.ts
+ */
+import { sleep } from 'bun';
+import { test, describe, expect, step, tensorfusionworkload, DEFAULT_TIMEOUT, TEST_GPU_POOL, TEST_NAMESPACE, } from './_config.js';
+const WORKLOAD_NAME = 'test-remote-workload';
+const CLIENT_POD_NAME = 'test-remote-client';
+/**
+ * 远程模式 TensorFusionWorkload YAML 模板
+ */
+function remoteWorkloadYaml(name) {
+    return `
+apiVersion: tensor-fusion.ai/v1
+kind: TensorFusionWorkload
+metadata:
+  name: ${name}
+  namespace: ${TEST_NAMESPACE}
+  labels:
+    app: ${name}
+    test-type: gpu-remote
+spec:
+  replicas: 1
+  gpuCount: 1
+  poolName: ${TEST_GPU_POOL}
+  qos: medium
+  isolation: soft
+  resources:
+    requests:
+      tflops: 100m
+      vram: "1Gi"
+    limits:
+      tflops: 100m
+      vram: "1Gi"
+  isLocalGPU: false
+  autoScalingConfig:
+    autoSetResources:
+      enable: false
+`;
+}
+/**
+ * 远程模式 client pod YAML 模板
+ */
+function clientPodYaml(podName, poolName) {
+    return `
+apiVersion: v1
+kind: Pod
+metadata:
+  name: ${podName}
+  namespace: ${TEST_NAMESPACE}
+  labels:
+    tensor-fusion.ai/enabled: "true"
+  annotations:
+    tensor-fusion.ai/is-local-gpu: "false"
+    tensor-fusion.ai/gpupool: "${poolName}"
+    tensor-fusion.ai/tflops-request: "100m"
+    tensor-fusion.ai/vram-request: "1Gi"
+    tensor-fusion.ai/tflops-limit: "100m"
+    tensor-fusion.ai/vram-limit: "1Gi"
+    tensor-fusion.ai/inject-container: "app"
+spec:
+  restartPolicy: Never
+  containers:
+    - name: app
+      image: pytorch/pytorch:2.4.1-cuda12.1-cudnn9-runtime
+      command: ["sh", "-c", "sleep 3600"]
+`;
+}
+/**
+ * 从 client pod 的 env 中读取 connection 信息
+ */
+async function getConnectionInfoFromPod(kubectl, podName) {
+    const envData = await kubectl.get('pod', podName, {
+        jqFilter: '[.spec.containers[0].env[] | select(.name == "TENSOR_FUSION_CONNECTION_NAME" or .name == "TENSOR_FUSION_CONNECTION_NAMESPACE") | {name: .name, value: .value}]',
+    });
+    const envMap = new Map((envData ?? []).map((e) => [e.name, e.value]));
+    return {
+        connName: envMap.get('TENSOR_FUSION_CONNECTION_NAME') ?? '',
+        connNamespace: envMap.get('TENSOR_FUSION_CONNECTION_NAMESPACE') ?? '',
+    };
+}
+describe('测试场景 9: GPU 远程调用', { record: true }, () => {
+    test('远程模式下 client pod 自动创建 TensorFusionConnection 并验证关联', async (ctx) => {
+        const { kubectl } = ctx.plugins;
+        let connName;
+        let connNamespace;
+        // ===== Step 1: 创建远程模式 Workload =====
+        await step('创建远程模式 Workload（isLocalGPU=false）', {
+            showStepTitle: false,
+            typingSpeed: 100,
+            pauseAfter: 2000,
+        }, async () => {
+            const yaml = remoteWorkloadYaml(WORKLOAD_NAME);
+            const result = await kubectl.apply(yaml);
+            await expect(result).toBeSuccessful();
+        });
+        await step('等待 Workload Ready', {
+            showStepTitle: false,
+            pauseAfter: 1800,
+        }, async () => {
+            const waitResult = await kubectl.wait('tensorfusionworkload', WORKLOAD_NAME, 'Ready', { timeout: DEFAULT_TIMEOUT });
+            await expect(waitResult).toBeSuccessful();
+            const workload = tensorfusionworkload(kubectl, WORKLOAD_NAME);
+            await expect(workload).toHaveStatusField('phase', 'Running');
+        });
+        await step('确认 Workload 为远程模式', {
+            typingSpeed: 80,
+            pauseAfter: 1500,
+        }, async () => {
+            const spec = await kubectl.getJsonPath('tensorfusionworkload', WORKLOAD_NAME, '.spec');
+            expect(spec?.isLocalGPU).toBe(false);
+        });
+        // ===== Step 2: 创建 client pod =====
+        await step('创建远程模式 client pod', {
+            showStepTitle: false,
+            typingSpeed: 100,
+            pauseAfter: 2500,
+        }, async () => {
+            const yaml = clientPodYaml(CLIENT_POD_NAME, TEST_GPU_POOL);
+            const result = await kubectl.apply(yaml);
+            await expect(result).toBeSuccessful();
+        });
+        await step('等待 client pod Ready', {
+            showStepTitle: false,
+            pauseAfter: 2000,
+        }, async () => {
+            await kubectl.waitForPod(CLIENT_POD_NAME, 'Running', DEFAULT_TIMEOUT * 3);
+        });
+        // ===== Step 3: 检查 TensorFusionConnection 自动创建 =====
+        await step('从 client pod env 读取 connection 信息', {
+            showStepTitle: false,
+            typingSpeed: 80,
+            pauseAfter: 2000,
+        }, async () => {
+            const info = await getConnectionInfoFromPod(kubectl, CLIENT_POD_NAME);
+            connName = info.connName;
+            connNamespace = info.connNamespace;
+            expect(connName.length).toBeGreaterThan(0);
+            expect(connNamespace.length).toBeGreaterThan(0);
+        });
+        await step('验证 TensorFusionConnection 资源存在', {
+            typingSpeed: 80,
+            pauseAfter: 2000,
+        }, async () => {
+            const exists = await kubectl.exists('tensorfusionconnection', connName);
+            expect(exists).toBe(true);
+        });
+        // ===== Step 4: 验证 connection metadata =====
+        await step('验证 connection metadata 字段', {
+            showStepTitle: false,
+            typingSpeed: 80,
+            pauseAfter: 2500,
+        }, async () => {
+            const metadata = await kubectl.get('tensorfusionconnection', connName, {
+                jqFilter: `{namespace: .metadata.namespace, workloadLabel: .metadata.labels["tensor-fusion.ai/workload"], ownerKind: .metadata.ownerReferences[0].kind, ownerName: .metadata.ownerReferences[0].name}`,
+            });
+            // metadata.namespace 应与测试命名空间一致
+            expect(metadata?.namespace).toBe(TEST_NAMESPACE);
+            // ownerReferences 指向 client pod
+            expect(metadata?.ownerKind).toBe('Pod');
+            expect(metadata?.ownerName).toBe(CLIENT_POD_NAME);
+        });
+        // ===== Step 5: 验证 connection spec =====
+        await step('验证 connection spec 字段', {
+            typingSpeed: 80,
+            pauseAfter: 2500,
+        }, async () => {
+            const spec = await kubectl.getJsonPath('tensorfusionconnection', connName, '.spec');
+            expect(spec?.clientPod).toBe(CLIENT_POD_NAME);
+        });
+        // ===== Step 6: （可选）查看 connection 状态 =====
+        await step('查看 connection 状态', {
+            pauseAfter: 2000,
+        }, async () => {
+            const status = await kubectl.getJsonPath('tensorfusionconnection', connName, '.status');
+            // phase 应存在
+            expect(status?.phase).toBeDefined();
+        });
+        // ===== Step 7: 在 client pod 内验证 GPU 可用 =====
+        await step('执行 nvidia-smi 验证 GPU 可见', {
+            showStepTitle: false,
+            typingSpeed: 80,
+            pauseAfter: 2500,
+        }, async () => {
+            const nvidiaSmiOutput = await kubectl.exec(CLIENT_POD_NAME, ['nvidia-smi'], { container: 'app' });
+            // nvidia-smi 输出应包含 NVIDIA 驱动信息和 GPU 信息
+            expect(nvidiaSmiOutput).toContain('NVIDIA');
+            expect(nvidiaSmiOutput).toContain('GPU');
+        });
+        await step('执行 nvidia-smi -L 列出 GPU 设备', {
+            typingSpeed: 80,
+            pauseAfter: 2000,
+        }, async () => {
+            const gpuListOutput = await kubectl.exec(CLIENT_POD_NAME, ['nvidia-smi', '-L'], { container: 'app' });
+            // 至少存在一块 GPU
+            expect(gpuListOutput).toContain('GPU 0');
+        });
+        await step('PyTorch 检测 CUDA 可用', {
+            showStepTitle: false,
+            typingSpeed: 80,
+            pauseAfter: 2500,
+        }, async () => {
+            const cudaAvailable = await kubectl.exec(CLIENT_POD_NAME, ['python3', '-c', 'import torch; print(torch.cuda.is_available())'], { container: 'app' });
+            expect(cudaAvailable.trim()).toBe('True');
+        });
+        await step('PyTorch 获取 GPU 设备信息', {
+            typingSpeed: 80,
+            pauseAfter: 2500,
+        }, async () => {
+            const gpuInfo = await kubectl.exec(CLIENT_POD_NAME, ['python3', '-c', 'import torch; print(f"device_count={torch.cuda.device_count()}, name={torch.cuda.get_device_name(0)}")'], { container: 'app' });
+            // 至少有 1 块 GPU
+            expect(gpuInfo).toContain('device_count=');
+            expect(gpuInfo).not.toContain('device_count=0');
+        });
+        await step('PyTorch GPU 张量运算验证', {
+            showStepTitle: false,
+            typingSpeed: 80,
+            pauseAfter: 3000,
+        }, async () => {
+            const tensorTest = await kubectl.exec(CLIENT_POD_NAME, [
+                'python3', '-c',
+                'import torch; a = torch.randn(2, 3, device="cuda"); b = torch.randn(3, 2, device="cuda"); c = torch.mm(a, b); print(f"shape={list(c.shape)}, device={c.device}")',
+            ], { container: 'app' });
+            // 验证运算在 GPU 上完成
+            expect(tensorTest).toContain('shape=[2, 2]');
+            expect(tensorTest).toContain('device=cuda');
+        });
+        // ===== Step 8: 清理 =====
+        await step('删除 client pod', {
+            showStepTitle: false,
+            typingSpeed: 80,
+            pauseAfter: 1500,
+        }, async () => {
+            const result = await kubectl.delete('pod', CLIENT_POD_NAME);
+            await expect(result).toBeSuccessful();
+        });
+        await step('删除 TensorFusionWorkload', {
+            showStepTitle: false,
+            typingSpeed: 80,
+            pauseAfter: 2000,
+        }, async () => {
+            const result = await kubectl.delete('tensorfusionworkload', WORKLOAD_NAME);
+            await expect(result).toBeSuccessful();
+            await sleep(5000);
+        });
+    });
+});

package/dist/plugin-kubectl/examples/tensor-fusion/_config.d.ts

@@ -0,0 +1,71 @@
+/**
+ * Tensor Fusion GPU 资源分配测试 - 共享配置
+ *
+ * 运行方式:
+ *   bun run repterm packages/plugin-kubectl/examples/tensor-fusion/
+ *
+ * 前置条件:
+ *   - 已配置 kubectl 并连接到 Kubernetes 集群
+ *   - Tensor Fusion Controller 已部署并运行
+ *   - 至少存在一个 GPUPool 和可用 GPU
+ */
+import { describe, expect, step } from 'repterm';
+import { gpupool, gpu, tensorfusionworkload, tensorfusionconnection, pod, deployment, resource, type KubectlMethods } from '../../src/index.js';
+/** 测试使用的 GPUPool 名称 */
+export declare const TEST_GPU_POOL = "tensor-fusion-shared";
+/** 测试命名空间 */
+export declare const TEST_NAMESPACE = "default";
+/** Tensor Fusion 系统命名空间 */
+export declare const TF_SYSTEM_NAMESPACE = "tensor-fusion-sys";
+/** Controller Deployment 名称 */
+export declare const TF_CONTROLLER_DEPLOYMENT = "tensor-fusion-sys-controller";
+/** 默认等待超时时间 (ms) */
+export declare const DEFAULT_TIMEOUT = 60000;
+export declare const config: import("node_modules/repterm/dist/plugin/index.js").PluginRuntime<readonly [import("repterm").PluginDefinition<"kubectl", import("@repterm/plugin-api").BasePluginContext, import("../../src/index.js").KubectlContext, KubectlMethods>]>;
+export declare const test: (name: string, fn: import("repterm").PluginTestFunction<readonly [import("repterm").PluginDefinition<"kubectl", import("@repterm/plugin-api").BasePluginContext, import("../../src/index.js").KubectlContext, KubectlMethods>]>) => void;
+export { describe, expect, step, gpupool, gpu, tensorfusionworkload, tensorfusionconnection, pod, deployment, resource, };
+export type { KubectlMethods };
+/**
+ * TensorFusionWorkload 测试模板
+ */
+export declare const workloadYaml: (name: string, options?: {
+    tflopsRequest?: string;
+    tflopsLimit?: string;
+    vramRequest?: string;
+    vramLimit?: string;
+    replicas?: number;
+    poolName?: string;
+}) => string;
+/**
+ * 带 Tensor Fusion Annotation 的 Deployment 模板
+ */
+export declare const annotatedDeploymentYaml: (name: string, options?: {
+    tflopsRequest?: string;
+    tflopsLimit?: string;
+    vramRequest?: string;
+    vramLimit?: string;
+    poolName?: string;
+}) => string;
+/**
+ * 获取 GPUPool 中第一个 GPU 名称
+ */
+export declare function getFirstGpuName(kubectl: KubectlMethods): Promise<string>;
+/**
+ * 获取 GPU 的可用资源
+ */
+export declare function getGpuAvailable(kubectl: KubectlMethods, gpuName: string): Promise<{
+    tflops: string;
+    vram: string;
+}>;
+/**
+ * 解析 TFlops 值为数字（支持带 m 后缀和不带后缀的格式）
+ */
+export declare function parseTflops(value: string | number): number;
+/**
+ * 清理测试资源
+ */
+export declare function cleanup(kubectl: KubectlMethods, resources: Array<{
+    kind: string;
+    name: string;
+}>): Promise<void>;
+//# sourceMappingURL=_config.d.ts.map

package/dist/plugin-kubectl/examples/tensor-fusion/_config.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"_config.d.ts","sourceRoot":"","sources":["../../../../examples/tensor-fusion/_config.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EACL,QAAQ,EAGR,MAAM,EAEN,IAAI,EACL,MAAM,SAAS,CAAC;AACjB,OAAO,EAEL,OAAO,EACP,GAAG,EACH,oBAAoB,EACpB,sBAAsB,EACtB,GAAG,EACH,UAAU,EACV,QAAQ,EACR,KAAK,cAAc,EACpB,MAAM,oBAAoB,CAAC;AAI5B,uBAAuB;AACvB,eAAO,MAAM,aAAa,yBAAyB,CAAC;AAEpD,aAAa;AACb,eAAO,MAAM,cAAc,YAAY,CAAC;AAExC,2BAA2B;AAC3B,eAAO,MAAM,mBAAmB,sBAAsB,CAAC;AAEvD,+BAA+B;AAC/B,eAAO,MAAM,wBAAwB,iCAAiC,CAAC;AAEvE,oBAAoB;AACpB,eAAO,MAAM,eAAe,QAAQ,CAAC;AAIrC,eAAO,MAAM,MAAM,2OAEjB,CAAC;AAEH,eAAO,MAAM,IAAI,0OAAgC,CAAC;AAIlD,OAAO,EACL,QAAQ,EACR,MAAM,EACN,IAAI,EACJ,OAAO,EACP,GAAG,EACH,oBAAoB,EACpB,sBAAsB,EACtB,GAAG,EACH,UAAU,EACV,QAAQ,GACT,CAAC;AAEF,YAAY,EAAE,cAAc,EAAE,CAAC;AAI/B;;GAEG;AACH,eAAO,MAAM,YAAY,GAAI,MAAM,MAAM,EAAE,UAAS;IAClD,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACd,WA4BL,CAAC;AAEF;;GAEG;AACH,eAAO,MAAM,uBAAuB,GAAI,MAAM,MAAM,EAAE,UAAS;IAC7D,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACd,WAwCL,CAAC;AAIF;;GAEG;AACH,wBAAsB,eAAe,CAAC,OAAO,EAAE,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,CAc9E;AAED;;GAEG;AACH,wBAAsB,eAAe,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC;IACvF,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;CACd,CAAC,CASD;AAED;;GAEG;AACH,wBAAgB,WAAW,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAQ1D;AAED;;GAEG;AACH,wBAAsB,OAAO,CAAC,OAAO,EAAE,cAAc,EAAE,SAAS,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC,iBAQtG"}