[Ray] Optimize Ray CI execution time and stability (#3102)

Author: chaokunyang

.github/workflows/platform-ci.yml

@@ -18,7 +18,7 @@ jobs:
       fail-fast: false
       matrix:
         os: [ubuntu-latest]
-        python-version: [3.8-kubernetes, 3.8-hadoop, 3.8-ray, 3.8-ray-dag, 3.8-vineyard, 3.8-dask]
+        python-version: [3.8-kubernetes, 3.8-hadoop, 3.8-ray, 3.8-ray-deploy, 3.8-ray-dag, 3.8-vineyard, 3.8-dask]
         include:
           - { os: ubuntu-latest, python-version: 3.8-kubernetes, no-common-tests: 1,
               no-deploy: 1, with-kubernetes: "with Kubernetes" }
@@ -28,6 +28,8 @@ jobs:
               no-deploy: 1, with-vineyard: "with vineyard" }
           - { os: ubuntu-latest, python-version: 3.8-ray, no-common-tests: 1,
               no-deploy: 1, with-ray: "with ray" }
+          - { os: ubuntu-latest, python-version: 3.8-ray-deploy, no-common-tests: 1,
+              no-deploy: 1, with-ray-deploy: "with ray deploy" }
           - { os: ubuntu-latest, python-version: 3.8-ray-dag, no-common-tests: 1,
               no-deploy: 1, with-ray-dag: "with ray dag" }
           - { os: ubuntu-latest, python-version: 3.8-dask, no-common-tests: 1,
@@ -53,6 +55,7 @@ jobs:
           WITH_KUBERNETES: ${{ matrix.with-kubernetes }}
           WITH_VINEYARD: ${{ matrix.with-vineyard }}
           WITH_RAY: ${{ matrix.with-ray }}
+          WITH_RAY_DEPLOY: ${{ matrix.with-ray-deploy }}
           WITH_RAY_DAG: ${{ matrix.with-ray-dag }}
           RUN_DASK: ${{ matrix.run-dask }}
           NO_COMMON_TESTS: ${{ matrix.no-common-tests }}
@@ -93,7 +96,7 @@ jobs:
               sudo mv /tmp/etcd-download-test/etcdctl /usr/local/bin/
               rm -fr /tmp/etcd-$ETCD_VER-linux-amd64.tar.gz /tmp/etcd-download-test
             fi
-            if [ -n "$WITH_RAY" ] || [ -n "$WITH_RAY_DAG" ]; then
+            if [ -n "$WITH_RAY" ] || [ -n "$WITH_RAY_DAG" ] || [ -n "$WITH_RAY_DEPLOY" ]; then
               pip install ray[default]==1.9.2 "protobuf<4"
               pip install "xgboost_ray==0.1.5" "xgboost<1.6.0"
             fi
@@ -110,6 +113,7 @@ jobs:
           WITH_CYTHON: ${{ matrix.with-cython }}
           WITH_VINEYARD: ${{ matrix.with-vineyard }}
           WITH_RAY: ${{ matrix.with-ray }}
+          WITH_RAY_DEPLOY: ${{ matrix.with-ray-deploy }}
           WITH_RAY_DAG: ${{ matrix.with-ray-dag }}
           RUN_DASK: ${{ matrix.run-dask }}
           NO_COMMON_TESTS: ${{ matrix.no-common-tests }}
@@ -144,13 +148,31 @@ jobs:
             coverage combine build/ && coverage report
           fi
           if [ -n "$WITH_RAY" ]; then
-            pytest $PYTEST_CONFIG --durations=0 --timeout=200 -v -s -m ray
+            pytest $PYTEST_CONFIG --durations=0 --timeout=200 -v -s --ignore=mars/deploy/oscar/ -m ray 
             coverage report
           fi
+          if [ -n "$WITH_RAY_DEPLOY" ]; then
+            pytest $PYTEST_CONFIG --durations=0 --timeout=200 -v -s mars/deploy/oscar/tests/test_ray.py -m ray
+            mv .coverage build/.coverage.test_ray.file
+            pytest $PYTEST_CONFIG --durations=0 --timeout=200 -v -s mars/deploy/oscar/tests/test_ray_client.py -m ray
+            mv .coverage build/.coverage.test_ray_client.file
+            pytest $PYTEST_CONFIG --durations=0 --timeout=200 -v -s mars/deploy/oscar/tests/test_ray_fault_injection.py -m ray
+            mv .coverage build/.coverage.test_ray_fault_injection.file
+            pytest $PYTEST_CONFIG --durations=0 --timeout=200 -v -s mars/deploy/oscar/tests/test_ray_scheduling.py -m ray
+            mv .coverage build/.coverage.test_ray_scheduling.file
+          
+            coverage combine build/ && coverage report
+          fi
           if [ -n "$WITH_RAY_DAG" ]; then
             export MARS_CI_BACKEND=ray
-            pytest $PYTEST_CONFIG --durations=0 --timeout=600 -v -s -m ray_dag
-            coverage report
+            pytest $PYTEST_CONFIG --durations=0 --timeout=200 -v -s -m ray_dag
+            mv .coverage build/.coverage.ray_dag.file
+            pytest $PYTEST_CONFIG --durations=0 --timeout=200 -v -s mars/deploy/oscar/tests/test_ray_dag.py
+            mv .coverage build/.coverage.test_ray_dag.file
+            pytest $PYTEST_CONFIG --durations=0 --timeout=200 -v -s mars/deploy/oscar/tests/test_ray_dag_failover.py
+            mv .coverage build/.coverage.test_ray_dag_failover.file
+            
+            coverage combine build/ && coverage report
           fi
           if [ -n "$RUN_DASK" ]; then
             pytest $PYTEST_CONFIG mars/contrib/dask/tests/test_dask.py

mars/conftest.py

@@ -165,7 +165,6 @@ async def ray_create_mars_cluster(request, check_router_cleaned):
     worker_mem = param.get("worker_mem", 256 * 1024**2)
     ray_config.update(param.get("config", {}))
     client = await new_cluster(
-        "test_cluster",
         supervisor_mem=supervisor_mem,
         worker_num=worker_num,
         worker_cpu=worker_cpu,