attempting to bump coverage

Signed-off-by: Pat O'Connor <[email protected]>

Author: chipspeak

src/codeflare_sdk/ray/job/test_job.py

@@ -0,0 +1,230 @@
+import pytest
+import yaml
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+from codeflare_sdk.ray.job.job import RayJob, RayJobSpec, RayJobStatus
+
+# Path to expected YAML files
+parent = Path(__file__).resolve().parents[4]  # project directory
+expected_yamls_dir = f"{parent}/tests/test_cluster_yamls/ray"
+
+
+def test_rayjob_to_dict_minimal():
+    """Test RayJob.to_dict() with minimal configuration using YAML comparison"""
+    spec = RayJobSpec(entrypoint="python test.py")
+    job = RayJob(
+        metadata={"name": "test-job", "namespace": "default"},
+        spec=spec
+    )
+    
+    result = job.to_dict()
+    
+    # Load expected YAML
+    with open(f"{expected_yamls_dir}/rayjob-minimal.yaml") as f:
+        expected = yaml.safe_load(f)
+    
+    assert result == expected
+
+
+def test_rayjob_to_dict_full_spec():
+    """Test RayJob.to_dict() with complete configuration using YAML comparison"""
+    spec = RayJobSpec(
+        entrypoint="python main.py --config config.yaml",
+        submission_id="job-12345",
+        runtime_env={"working_dir": "/app", "pip": ["numpy", "pandas"]},
+        metadata={"experiment": "test-run", "version": "1.0"},
+        entrypoint_num_cpus=2,
+        entrypoint_num_gpus=1,
+        entrypoint_memory=4096,
+        entrypoint_resources={"custom_resource": 1.0},
+        cluster_name="test-cluster",
+        cluster_namespace="ml-jobs",
+        status=RayJobStatus.RUNNING,
+        message="Job is running successfully",
+        start_time="2024-01-01T10:00:00Z",
+        end_time=None,
+        driver_info={"id": "driver-123", "node_ip_address": "10.0.0.1", "pid": "12345"}
+    )
+    
+    job = RayJob(
+        metadata={
+            "name": "ml-training-job",
+            "namespace": "ml-jobs",
+            "labels": {"app": "ml-training", "version": "v1"},
+            "annotations": {"description": "Machine learning training job"}
+        },
+        spec=spec
+    )
+    
+    result = job.to_dict()
+    
+    # Load expected YAML
+    with open(f"{expected_yamls_dir}/rayjob-full-spec.yaml") as f:
+        expected = yaml.safe_load(f)
+    
+    assert result == expected
+
+
+def test_rayjob_to_dict_with_existing_status():
+    """Test RayJob.to_dict() when status is already set using YAML comparison"""
+    spec = RayJobSpec(entrypoint="python test.py")
+    
+    # Pre-existing status from Kubernetes
+    existing_status = {
+        "jobStatus": "SUCCEEDED",
+        "jobId": "raysubmit_12345",
+        "message": "Job completed successfully",
+        "startTime": "2024-01-01T10:00:00Z",
+        "endTime": "2024-01-01T10:05:00Z"
+    }
+    
+    job = RayJob(
+        metadata={"name": "completed-job", "namespace": "default"},
+        spec=spec,
+        status=existing_status
+    )
+    
+    result = job.to_dict()
+    
+    # Load expected YAML
+    with open(f"{expected_yamls_dir}/rayjob-existing-status.yaml") as f:
+        expected = yaml.safe_load(f)
+    
+    assert result == expected
+
+
+def test_rayjob_status_enum_values():
+    """Test all RayJobStatus enum values"""
+    assert RayJobStatus.PENDING == "PENDING"
+    assert RayJobStatus.RUNNING == "RUNNING"
+    assert RayJobStatus.STOPPED == "STOPPED"
+    assert RayJobStatus.SUCCEEDED == "SUCCEEDED"
+    assert RayJobStatus.FAILED == "FAILED"
+
+
+@patch('codeflare_sdk.common.kubernetes_cluster.auth.config_check')
+@patch('codeflare_sdk.common.kubernetes_cluster.auth.get_api_client')
+@patch('kubernetes.dynamic.DynamicClient')
+def test_rayjob_apply_success(mock_dynamic_client, mock_get_api_client, mock_config_check):
+    """Test RayJob.apply() method successful execution"""
+    # Mock the Kubernetes API components
+    mock_api_instance = MagicMock()
+    mock_dynamic_client.return_value.resources.get.return_value = mock_api_instance
+    
+    spec = RayJobSpec(entrypoint="python test.py")
+    job = RayJob(
+        metadata={"name": "test-job", "namespace": "test-ns"},
+        spec=spec
+    )
+    
+    # Test successful apply
+    job.apply()
+    
+    # Verify the API was called correctly
+    mock_config_check.assert_called_once()
+    mock_get_api_client.assert_called_once()
+    mock_dynamic_client.assert_called_once()
+    mock_api_instance.server_side_apply.assert_called_once()
+    
+    # Check the server_side_apply call arguments
+    call_args = mock_api_instance.server_side_apply.call_args
+    assert call_args[1]['field_manager'] == 'codeflare-sdk'
+    assert call_args[1]['group'] == 'ray.io'
+    assert call_args[1]['version'] == 'v1'
+    assert call_args[1]['namespace'] == 'test-ns'
+    assert call_args[1]['plural'] == 'rayjobs'
+    assert call_args[1]['force_conflicts'] == False
+    
+    # Verify the body contains the expected job structure
+    body = call_args[1]['body']
+    assert body['apiVersion'] == 'ray.io/v1'
+    assert body['kind'] == 'RayJob'
+    assert body['metadata']['name'] == 'test-job'
+    assert body['spec']['entrypoint'] == 'python test.py'
+
+
+@patch('codeflare_sdk.common.kubernetes_cluster.auth.config_check')
+@patch('codeflare_sdk.common.kubernetes_cluster.auth.get_api_client')
+@patch('kubernetes.dynamic.DynamicClient')
+def test_rayjob_apply_with_force(mock_dynamic_client, mock_get_api_client, mock_config_check):
+    """Test RayJob.apply() method with force=True"""
+    mock_api_instance = MagicMock()
+    mock_dynamic_client.return_value.resources.get.return_value = mock_api_instance
+    
+    spec = RayJobSpec(entrypoint="python test.py")
+    job = RayJob(
+        metadata={"name": "test-job", "namespace": "default"},
+        spec=spec
+    )
+    
+    # Test apply with force=True
+    job.apply(force=True)
+    
+    # Verify force_conflicts was set to True
+    call_args = mock_api_instance.server_side_apply.call_args
+    assert call_args[1]['force_conflicts'] == True
+
+
+@patch('codeflare_sdk.common.kubernetes_cluster.auth.config_check')
+@patch('codeflare_sdk.common.kubernetes_cluster.auth.get_api_client')
+@patch('kubernetes.dynamic.DynamicClient')
+def test_rayjob_apply_dynamic_client_error(mock_dynamic_client, mock_get_api_client, mock_config_check):
+    """Test RayJob.apply() method with DynamicClient initialization error"""
+    # Mock DynamicClient to raise AttributeError
+    mock_dynamic_client.side_effect = AttributeError("Failed to connect")
+    
+    spec = RayJobSpec(entrypoint="python test.py")
+    job = RayJob(
+        metadata={"name": "test-job", "namespace": "default"},
+        spec=spec
+    )
+    
+    # Test that RuntimeError is raised
+    with pytest.raises(RuntimeError, match="Failed to initialize DynamicClient"):
+        job.apply()
+
+
+@patch('codeflare_sdk.common.kubernetes_cluster.auth.config_check')
+@patch('codeflare_sdk.common.kubernetes_cluster.auth.get_api_client')
+@patch('kubernetes.dynamic.DynamicClient')
+@patch('codeflare_sdk.common._kube_api_error_handling')
+def test_rayjob_apply_api_error(mock_error_handling, mock_dynamic_client, mock_get_api_client, mock_config_check):
+    """Test RayJob.apply() method with Kubernetes API error"""
+    # Mock the API to raise an exception
+    mock_api_instance = MagicMock()
+    mock_api_instance.server_side_apply.side_effect = Exception("API Error")
+    mock_dynamic_client.return_value.resources.get.return_value = mock_api_instance
+    
+    spec = RayJobSpec(entrypoint="python test.py")
+    job = RayJob(
+        metadata={"name": "test-job", "namespace": "default"},
+        spec=spec
+    )
+    
+    # Test that error handling is called
+    job.apply()
+    
+    # Verify error handling was called
+    mock_error_handling.assert_called_once()
+
+
+def test_rayjob_default_namespace_in_apply():
+    """Test that RayJob.apply() uses 'default' namespace when not specified in metadata"""
+    with patch('codeflare_sdk.common.kubernetes_cluster.auth.config_check'), \
+         patch('codeflare_sdk.common.kubernetes_cluster.auth.get_api_client'), \
+         patch('kubernetes.dynamic.DynamicClient') as mock_dynamic_client:
+        
+        mock_api_instance = MagicMock()
+        mock_dynamic_client.return_value.resources.get.return_value = mock_api_instance
+        
+        spec = RayJobSpec(entrypoint="python test.py")
+        job = RayJob(
+            metadata={"name": "test-job"},  # No namespace specified
+            spec=spec
+        )
+        
+        job.apply()
+        
+        # Verify default namespace was used
+        call_args = mock_api_instance.server_side_apply.call_args
+        assert call_args[1]['namespace'] == 'default' 

src/codeflare_sdk/ray/rayjobs/test_rayjob.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import pytest
 from unittest.mock import MagicMock
 from codeflare_sdk.ray.rayjobs.rayjob import RayJob
 
@@ -58,3 +59,30 @@ def test_rayjob_submit_success(mocker):
     assert job_cr["spec"]["entrypoint"] == "python -c 'print(\"hello world\")'"
     assert job_cr["spec"]["clusterSelector"]["ray.io/cluster"] == "test-ray-cluster"
     assert job_cr["spec"]["runtimeEnvYAML"] == "{'pip': ['requests']}"
+
+
+def test_rayjob_submit_failure(mocker):
+    """Test RayJob submission failure."""
+    # Mock kubernetes config loading
+    mocker.patch("kubernetes.config.load_kube_config")
+
+    # Mock the RayjobApi class entirely
+    mock_api_class = mocker.patch("codeflare_sdk.ray.rayjobs.rayjob.RayjobApi")
+    mock_api_instance = MagicMock()
+    mock_api_class.return_value = mock_api_instance
+
+    # Configure the mock to return failure (False/None) when submit_job is called
+    mock_api_instance.submit_job.return_value = None
+
+    # Create a RayJob instance
+    rayjob = RayJob(
+        job_name="test-rayjob",
+        cluster_name="test-ray-cluster",
+        namespace="default",
+        entrypoint="python script.py",
+        runtime_env={"pip": ["numpy"]},
+    )
+
+    # Test that RuntimeError is raised on failure
+    with pytest.raises(RuntimeError, match="Failed to submit RayJob test-rayjob"):
+        rayjob.submit()

tests/test_cluster_yamls/ray/rayjob-existing-status.yaml

@@ -0,0 +1,22 @@
+apiVersion: ray.io/v1
+kind: RayJob
+metadata:
+  name: completed-job
+  namespace: default
+spec:
+  entrypoint: python test.py
+  submission_id: null
+  runtime_env: null
+  metadata: null
+  entrypoint_num_cpus: null
+  entrypoint_num_gpus: null
+  entrypoint_memory: null
+  entrypoint_resources: null
+  cluster_name: null
+  cluster_namespace: null
+status:
+  jobStatus: SUCCEEDED
+  jobId: raysubmit_12345
+  message: Job completed successfully
+  startTime: '2024-01-01T10:00:00Z'
+  endTime: '2024-01-01T10:05:00Z' 

tests/test_cluster_yamls/ray/rayjob-full-spec.yaml

@@ -0,0 +1,37 @@
+apiVersion: ray.io/v1
+kind: RayJob
+metadata:
+  name: ml-training-job
+  namespace: ml-jobs
+  labels:
+    app: ml-training
+    version: v1
+  annotations:
+    description: Machine learning training job
+spec:
+  entrypoint: python main.py --config config.yaml
+  submission_id: job-12345
+  runtime_env:
+    working_dir: /app
+    pip:
+      - numpy
+      - pandas
+  metadata:
+    experiment: test-run
+    version: '1.0'
+  entrypoint_num_cpus: 2
+  entrypoint_num_gpus: 1
+  entrypoint_memory: 4096
+  entrypoint_resources:
+    custom_resource: 1.0
+  cluster_name: test-cluster
+  cluster_namespace: ml-jobs
+status:
+  status: RUNNING
+  message: Job is running successfully
+  start_time: '2024-01-01T10:00:00Z'
+  end_time: null
+  driver_info:
+    id: driver-123
+    node_ip_address: 10.0.0.1
+    pid: '12345' 

tests/test_cluster_yamls/ray/rayjob-minimal.yaml

@@ -0,0 +1,22 @@
+apiVersion: ray.io/v1
+kind: RayJob
+metadata:
+  name: test-job
+  namespace: default
+spec:
+  entrypoint: python test.py
+  submission_id: null
+  runtime_env: null
+  metadata: null
+  entrypoint_num_cpus: null
+  entrypoint_num_gpus: null
+  entrypoint_memory: null
+  entrypoint_resources: null
+  cluster_name: null
+  cluster_namespace: null
+status:
+  status: PENDING
+  message: null
+  start_time: null
+  end_time: null
+  driver_info: null