chore: use kepler_vm metrics as energy source

Signed-off-by: Huamin Chen <[email protected]>

Author: rootfs

src/kepler_model/cmd/cmd_util.py

@@ -7,6 +7,7 @@
 from kepler_model.util.prom_types import (
     SOURCE_COL,
     energy_component_to_query,
+    vm_energy_component_to_query,
     node_info_column,
     prom_responses_to_results,
 )
@@ -99,7 +100,7 @@ def summary_validation(validate_df):
         print("{} data: \t{}".format(metric, target_df[">0"].values))
 
 
-def get_validate_df(data_path, benchmark_filename, query_response):
+def get_validate_df(data_path, benchmark_filename, query_response, use_vm_metrics=False):
     items = []
     query_results = prom_responses_to_results(query_response)
     container_queries = [query for query in query_results.keys() if "container" in query]
@@ -175,10 +176,14 @@ def get_validate_df(data_path, benchmark_filename, query_response):
                     item["total"] = filtered_df[query].max()
                     items += [item]
     energy_queries = [query for query in query_results.keys() if "_joules" in query]
-    print("Energy Queries: ", container_queries)
+    print("Energy Queries: ", energy_queries)
     for energy_source, energy_components in PowerSourceMap.items():
         for component in energy_components:
             query = energy_component_to_query(component)
+            if use_vm_metrics:
+                query = vm_energy_component_to_query(component)
+            else:
+                query = energy_component_to_query(component)
             df = query_results[query]
             df = df[df[SOURCE_COL] == energy_source]
             if len(df) == 0:

src/kepler_model/cmd/main.py

@@ -156,7 +156,7 @@ def query(args):
     if args.to_csv:
         save_query_results(data_path, args.output, response)
     # try validation if applicable
-    validate_df = get_validate_df(data_path, benchmark_filename, response)
+    validate_df = get_validate_df(data_path, benchmark_filename, response, use_vm_metrics=args.vm_train)
     summary_validation(validate_df)
     save_csv(path=data_path, name=args.output + "_validate_result", data=validate_df)
 
@@ -487,7 +487,9 @@ def train(args):
         print("cannot get pipeline")
         exit()
     for energy_source in energy_sources:
+        print("energy source: ", energy_source)
         energy_components = PowerSourceMap[energy_source]
+        print("energy components: ", energy_components)
         for feature_group in valid_feature_groups:
             success, abs_data, dyn_data = pipeline.process_multiple_query(
                 input_query_results_list, energy_components, energy_source, feature_group=feature_group.name, replace_node_type=node_type

src/kepler_model/train/extractor/extractor.py

@@ -168,7 +168,6 @@ def get_workload_feature_data(self, query_results, features, use_vm_metrics=Fals
                 print("no data in ", query)
                 return None
             aggr_query_data = query_results[query].copy()
-
             if all(col in aggr_query_data.columns for col in cols_to_use):
                 if use_vm_metrics:
                     aggr_query_data = aggr_query_data.loc[aggr_query_data["job"] == VM_JOB_NAME]
@@ -256,9 +255,9 @@ def get_power_data(self, query_results, energy_components, source, use_vm_metric
                 return None
             aggr_query_data = query_results[query].copy()
             if not use_vm_metrics:
-                aggr_query_data = aggr_query_data.loc[aggr_query_data["job"] != VM_JOB_NAME]
-            # filter source
-            aggr_query_data = aggr_query_data[aggr_query_data[SOURCE_COL] == source]
+                # aggr_query_data = aggr_query_data.loc[aggr_query_data["job"] != VM_JOB_NAME]
+                # filter source
+                aggr_query_data = aggr_query_data[aggr_query_data[SOURCE_COL] == source]
             if len(aggr_query_data) == 0:
                 return None
             if unit_col is not None: