create minicube integration test (#903)

* create minicube integration test

* mini_v2

* mini_v3

* mini_v4

* mini_v5

* mini_v6

* mini_v7

* mini_v8

* mini_v9

* mini_v10

* mini_v11

* mini_v12

* mini_v12

* mini_v14

* mini_v15

* mini_v16

* mini_v17

* mini_v18

* mini_v19

* mini_v20

* mini_v21

* mini_v22

Author: yikaiMeta

.github/workflows/kubernetes-minicube-integration-test.yaml

@@ -0,0 +1,35 @@
+name: Kubernetes Minikube Integration Test
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+
+jobs:
+  kubernetes-launch:
+    runs-on: "linux.20_04.16x"
+    permissions:
+      id-token: write
+      contents: read
+    steps:
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.10"
+          architecture: x64
+      - name: Checkout TorchX
+        uses: actions/checkout@v4
+      - name: Install dependencies
+        run: |
+          set -eux
+          pip install -e .[kubernetes]
+      - name: Start Kubernetes
+        run: |
+          scripts/setup_minikube.sh
+      - name: Run Kubernetes Integration Tests
+        env:
+          CONTAINER_REPO: localhost:5000/torchx
+        run: |
+          chmod +x scripts/minikube_trainer.py
+          scripts/minikube_trainer.py

scripts/minikube_trainer.py

@@ -0,0 +1,72 @@
+#!/usr/bin/env python3
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+# pyre-strict
+
+"""
+Kubernetes integration tests.
+"""
+
+from component_integration_tests import build_and_push_image
+
+from integ_test_utils import getenv_asserts, MissingEnvError
+from torchx.components.dist import ddp as dist_ddp
+from torchx.runner import get_runner
+from torchx.specs import _named_resource_factories, AppState, Resource
+from torchx.util.types import none_throws
+
+GiB: int = 1024
+
+
+def register_gpu_resource() -> None:
+    res = Resource(
+        cpu=2,
+        gpu=1,
+        memMB=8 * GiB,
+    )
+    print(f"Registering resource: {res}")
+    _named_resource_factories["GPU_X1"] = lambda: res
+
+
+def run_job() -> None:
+    register_gpu_resource()
+    build = build_and_push_image(container_repo=getenv_asserts("CONTAINER_REPO"))
+    image = build.torchx_image
+    runner = get_runner()
+    train_app = dist_ddp(
+        m="torchx.examples.apps.compute_world_size.main",
+        name="ddp-trainer",
+        image=image,
+        cpu=1,
+        j="2x2",
+        max_retries=3,
+        env={
+            "LOGLEVEL": "INFO",
+        },
+    )
+    cfg = {
+        "namespace": "torchx-dev",
+        "queue": "default",
+    }
+    app_handle = runner.run(train_app, "kubernetes", cfg)
+    print("Start waiting for app to finish")
+    runner.wait(app_handle)
+    final_status = runner.status(app_handle)
+    print(f"Final status: {final_status}")
+    if none_throws(final_status).state != AppState.SUCCEEDED:
+        raise Exception(f"Dist app failed with status: {final_status}")
+
+
+def main() -> None:
+    try:
+        run_job()
+    except MissingEnvError:
+        print("Skip running tests, executed only docker buid step")
+
+
+if __name__ == "__main__":
+    main()