fix: address second round of review comments

- Remove unused gpu_count from llm-d template context
- Use distinct gpu_count/tensor_parallel values in test fixture to catch regressions
- Mock ModelCatalog and use tmp_path in client fixture to avoid disk side-effects

Assisted-by: Claude <noreply@anthropic.com>
Signed-off-by: Jing Chen <jing.chen2@ibm.com>

Author: jgchn

src/planner/configuration/llmd_generator.py

@@ -63,7 +63,6 @@ def _prepare_context(
             "deployment_id": deployment_id,
             "namespace": namespace,
             "model_id": model_id,
-            "gpu_count": gpu_config.gpu_count if gpu_config else 1,
             "tensor_parallel": tensor_parallel,
             "gpus_per_replica": tensor_parallel,
             "replicas": gpu_config.replicas if gpu_config else 1,

tests/unit/test_llmd_generator.py

@@ -2,7 +2,7 @@
 
 from __future__ import annotations
 
-from unittest.mock import MagicMock
+from unittest.mock import MagicMock, patch
 
 import pytest
 import yaml
@@ -21,11 +21,15 @@
 
 
 @pytest.fixture
-def client() -> TestClient:
-    """Create a test client with mocked app state (no DB required)."""
+def client(tmp_path) -> TestClient:
+    """Create a test client with mocked app state (no DB or disk side-effects)."""
     app = FastAPI()
-    app.state.deployment_generator = DeploymentGenerator(simulator_mode=False)
-    app.state.llmd_deployment_generator = LlmdDeploymentGenerator()
+
+    with patch("planner.configuration.generator.ModelCatalog"):
+        app.state.deployment_generator = DeploymentGenerator(
+            output_dir=str(tmp_path / "vllm"), simulator_mode=False
+        )
+    app.state.llmd_deployment_generator = LlmdDeploymentGenerator(output_dir=str(tmp_path / "llmd"))
     app.state.yaml_validator = YAMLValidator()
     app.state.cluster_managers = {}
     app.state.cluster_manager_lock = MagicMock()
@@ -55,7 +59,7 @@ def sample_recommendation() -> DeploymentRecommendation:
         meets_slo=True,
         gpu_config=GPUConfig(
             gpu_type="NVIDIA-A100-80GB",
-            gpu_count=2,
+            gpu_count=6,
             tensor_parallel=2,
             replicas=3,
         ),
@@ -251,7 +255,7 @@ def test_sets_gpu_resources_per_replica(
         parsed = yaml.safe_load(result["contents"]["patch_vllm"])
 
         container = parsed["spec"]["template"]["spec"]["containers"][0]
-        # tensor_parallel=2, so each replica gets 2 GPUs (not gpu_count which is total)
+        # gpu_count=6, tensor_parallel=2, replicas=3 — each pod gets tensor_parallel GPUs
         assert container["resources"]["requests"]["nvidia.com/gpu"] == "2"
         assert container["resources"]["limits"]["nvidia.com/gpu"] == "2"