[TRTLLM-8930][infra] Force Blossom perf test stages to use 'tensorrt/test_type: perf' in the K8S template (#8752)

Signed-off-by: ZhanruiSunCh <[email protected]>
Signed-off-by: Zhanrui Sun <[email protected]>

Author: ZhanruiSunCh

jenkins/L0_Test.groovy

@@ -1296,14 +1296,14 @@ def createKubernetesPodConfig(image, type, arch = "amd64", gpuCount = 1, perfMod
                     kubernetes.io/os: linux
                     nvidia.com/gpu_type: ${gpuType}"""
         } else if (perfMode && !hasMultipleGPUs) {
-        // Not using the "perf" node currently due to hardware resource constraint.
         // Use single GPU machine with "tensorrt/test_type: perf" for stable perf testing.
         // H100 / A100 single GPU machine has this unique label in TensorRT Blossom pool.
             selectors = """
                     kubernetes.io/arch: ${arch}
                     kubernetes.io/os: linux
                     nvidia.com/gpu_type: ${gpuType}
-                    nvidia.com/driver_version: '${driverVersion}'"""
+                    nvidia.com/driver_version: '${driverVersion}'
+                    tensorrt/test_type: perf"""
         }
         else
         {