Completely remove GPU A2 type. It doesn't exist on the nerc-prod clus…

…ter nor do we know what it would be named, so I removing it. We can add it if it's added to the cluster. Also update the test to include A100SXM4 instead of A2
CCI-MOC · Mar 26, 2024 · 902162e · 902162e
1 parent 1451c1e
commit 902162e
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 11 deletions.
diff --git a/openshift_metrics/tests/test_utils.py b/openshift_metrics/tests/test_utils.py
@@ -643,7 +643,7 @@ def test_write_metrics_log(self, mock_gna):
             },
             "pod5": {
                 "namespace": "namespace2",
-                "gpu_type": utils.GPU_A2,
+                "gpu_type": utils.GPU_A100_SXM4,
                 "metrics": {
                     0: {
                         "cpu_request": 24,
@@ -659,7 +659,7 @@ def test_write_metrics_log(self, mock_gna):
                             "2023-01,namespace1,namespace1,PI1,,,,76,1128,OpenShift CPU,0.013,14.66\n"
                             "2023-01,namespace2,namespace2,PI2,,,,,96,OpenShift CPU,0.013,1.25\n"
                             "2023-01,namespace2,namespace2,PI2,,,,,48,OpenShift GPUA100,1.803,86.54\n"
-                            "2023-01,namespace2,namespace2,PI2,,,,,144,OpenShift GPUA2,0.466,67.1\n")
+                            "2023-01,namespace2,namespace2,PI2,,,,,48,OpenShift GPUA100SXM4,2.078,99.74\n")
 
         with tempfile.NamedTemporaryFile(mode="w+") as tmp:
             utils.write_metrics_by_namespace(test_metrics_dict, tmp.name, "2023-01")

diff --git a/openshift_metrics/utils.py b/openshift_metrics/utils.py
@@ -28,15 +28,13 @@
 GPU_A100 = "NVIDIA-A100-40GB"
 GPU_A100_SXM4 = "NVIDIA-A100-SXM4-40GB"
 GPU_V100 = "Tesla-V100-PCIE-32GB"
-GPU_A2 = "nvidia.com/gpu_A2"
 GPU_GENERIC = "nvidia.com/gpu"
 NO_GPU = "No GPU"
 
 # SU Types
 SU_CPU = "OpenShift CPU"
 SU_A100_GPU = "OpenShift GPUA100"
 SU_A100_SXM4_GPU = "OpenShift GPUA100SXM4"
-SU_A2_GPU = "OpenShift GPUA2"
 SU_V100_GPU = "OpenShift GPUV100"
 SU_UNKNOWN_GPU = "OpenShift Unknown GPU"
 SU_UNKNOWN = "Openshift Unknown"
@@ -45,7 +43,6 @@
     SU_CPU: 0.013,
     SU_A100_GPU: 1.803,
     SU_A100_SXM4_GPU: 2.078,
-    SU_A2_GPU: 0.466,
     SU_V100_GPU: 1.214,
     SU_UNKNOWN_GPU: 0,
 }
@@ -358,7 +355,6 @@ def write_metrics_by_namespace(condensed_metrics_dict, file_name, report_month):
                 "SU_CPU_HOURS": 0,
                 "SU_A100_GPU_HOURS": 0,
                 "SU_A100_SXM4_GPU_HOURS": 0,
-                "SU_A2_GPU_HOURS": 0,
                 "SU_V100_GPU_HOURS": 0,
                 "SU_UNKNOWN_GPU_HOURS": 0,
                 "total_cost": 0,
@@ -378,8 +374,6 @@ def write_metrics_by_namespace(condensed_metrics_dict, file_name, report_month):
                 metrics_by_namespace[namespace]["SU_A100_SXM4_GPU_HOURS"] += su_count * duration_in_hours
             elif gpu_type == GPU_V100:
                 metrics_by_namespace[namespace]["SU_V100_GPU_HOURS"] += su_count * duration_in_hours
-            elif gpu_type == GPU_A2:
-                metrics_by_namespace[namespace]["SU_A2_GPU_HOURS"] += su_count * duration_in_hours
             elif gpu_type == GPU_GENERIC:
                 metrics_by_namespace[namespace]["SU_UNKNOWN_GPU_HOURS"] += su_count * duration_in_hours
             else:
@@ -404,9 +398,6 @@ def write_metrics_by_namespace(condensed_metrics_dict, file_name, report_month):
         if metrics["SU_A100_SXM4_GPU_HOURS"] != 0:
             add_row(hours=metrics["SU_A100_SXM4_GPU_HOURS"], su_type=SU_A100_SXM4_GPU, **common_args)
 
-        if metrics["SU_A2_GPU_HOURS"] != 0:
-            add_row(hours=metrics["SU_A2_GPU_HOURS"], su_type=SU_A2_GPU, **common_args)
-
         if metrics["SU_V100_GPU_HOURS"] != 0:
             add_row(hours=metrics["SU_V100_GPU_HOURS"], su_type=SU_V100_GPU, **common_args)