Add dsr1 and gpt-oss test cases

Signed-off-by: Chenfei Zhang <[email protected]>

Author: chenfeiz0326

tests/integration/defs/perf/open_search_db_utils.py

@@ -20,6 +20,7 @@
 import re
 import sys
 import time
+from datetime import datetime
 
 from defs.trt_test_alternative import print_info
 
@@ -198,7 +199,7 @@ def get_job_info():
     }
 
 
-def query_history_data():
+def query_history_data(gpu_type):
     """
     Query post-merge data with specific gpu type and model name
     """
@@ -218,6 +219,16 @@ def query_history_data():
                             "b_is_post_merge": True
                         }
                     },
+                    {
+                        "term": {
+                            "b_is_regression": False
+                        }
+                    },
+                    {
+                        "term": {
+                            "s_gpu_type": gpu_type
+                        }
+                    },
                     {
                         "range": {
                             "ts_created": {
@@ -348,27 +359,42 @@ def calculate_best_perf_result(history_data_list, new_data):
     return best_metrics
 
 
-def get_history_data(new_data_dict):
+def get_history_data(new_data_dict, gpu_type):
     """
     Query history post-merge data for each cmd_idx
     """
+
+    def get_latest_data(data_list):
+        if not data_list:
+            return None
+        time_format = "%b %d, %Y @ %H:%M:%S.%f"
+        # Find the item with the maximum ts_created value
+        latest_data = max(
+            data_list,
+            key=lambda x: datetime.strptime(x["ts_created"], time_format))
+        return latest_data
+
     history_baseline_dict = {}
     history_data_dict = {}
     cmd_idxs = new_data_dict.keys()
     for cmd_idx in cmd_idxs:
         history_data_dict[cmd_idx] = []
-        history_baseline_dict[cmd_idx] = None
-    history_data_list = query_history_data()
+        history_baseline_dict[cmd_idx] = []
+    history_data_list = query_history_data(gpu_type)
     if history_data_list:
         for history_data in history_data_list:
             for cmd_idx in cmd_idxs:
                 if match(history_data, new_data_dict[cmd_idx]):
                     if history_data.get("b_is_baseline") and history_data.get(
                             "b_is_baseline") == True:
-                        history_baseline_dict[cmd_idx] = history_data
+                        history_baseline_dict[cmd_idx].append(history_data)
                     else:
                         history_data_dict[cmd_idx].append(history_data)
                     break
+    # Sometime database has several baselines and we only use the latest baseline one
+    for cmd_idx, baseline_list in history_baseline_dict:
+        latest_baseline = get_latest_data(baseline_list)
+        history_baseline_dict[cmd_idx] = latest_baseline
     return history_baseline_dict, history_data_dict
 
 

tests/integration/defs/perf/test_perf.py

@@ -2194,7 +2194,7 @@ def upload_test_results_to_database(self):
 
             # Get history data for each cmd_idx
             history_baseline_dict, history_data_dict = get_history_data(
-                new_data_dict)
+                new_data_dict, self._config.gpu_type)
             # Prepare regressive test cases
             regressive_data_list = prepare_regressive_test_cases(
                 history_baseline_dict, new_data_dict)