diff --git a/tests/army/frame/common.py b/tests/army/frame/common.py
index 913e88a7ad..b816095817 100644
--- a/tests/army/frame/common.py
+++ b/tests/army/frame/common.py
@@ -18,6 +18,7 @@ import time
 import socket
 import json
 import toml
+import subprocess
 from frame.boundary import DataBoundary
 import taos
 from frame.log import *
@@ -1830,6 +1831,51 @@ class TDCom:
                 if i == 1:
                     self.record_history_ts = ts_value
 
+    def generate_query_result(self, inputfile, test_case):
+        if not os.path.exists(inputfile):
+            tdLog.exit(f"Input file '{inputfile}' does not exist.")
+        else:
+            self.query_result_file = f"./temp_{test_case}.result"
+            os.system(f"taos -f {inputfile} | grep -v 'Query OK'|grep -v 'Copyright'| grep -v 'Welcome to the TDengine Command' > {self.query_result_file}  ")
+            return self.query_result_file
+
+    def compare_result_files(self, file1, file2):
+
+        try:
+            # use subprocess.run to execute  diff/fc commands
+            # print(file1, file2)
+            if platform.system().lower() != 'windows':
+                cmd='diff'
+                result = subprocess.run([cmd, "-u", "--color", file1, file2], text=True, capture_output=True)
+            else:
+                cmd='fc'
+                result = subprocess.run([cmd, file1, file2], text=True, capture_output=True)
+            # if result is not empty, print the differences and files name. Otherwise, the files are identical.
+            if result.stdout:
+                tdLog.debug(f"Differences between {file1} and {file2}")
+                tdLog.notice(f"\r\n{result.stdout}")
+                return False
+            else:
+                return True
+        except FileNotFoundError:
+            tdLog.debug("The 'diff' command is not found. Please make sure it's installed and available in your PATH.")
+        except Exception as e:
+            tdLog.debug(f"An error occurred: {e}")
+
+
+    def compare_testcase_result(self, inputfile,expected_file,test_case):
+        test_reulst_file = self.generate_query_result(inputfile,test_case)
+
+        if self.compare_result_files(expected_file, test_reulst_file ):
+            tdLog.info("Test passed: Result files are identical.")
+            os.system(f"rm -f {test_reulst_file}")
+        else:
+            caller = inspect.getframeinfo(inspect.stack()[1][0])
+            tdLog.exit(f"{caller.lineno}(line:{caller.lineno}) failed: sqlfile is {inputfile}, expect_file:{expected_file}  != reult_file:{test_reulst_file} ")
+
+            tdLog.exit("Test failed: Result files are different.")
+
+
 def is_json(msg):
     if isinstance(msg, str):
         try:
@@ -1864,4 +1910,6 @@ def dict2toml(in_dict: dict, file:str):
     with open(file, 'w') as f:
         toml.dump(in_dict, f)
 
+
+
 tdCom = TDCom()
diff --git a/tests/army/query/function/ans/pi_1.csv b/tests/army/query/function/ans/pi_1.csv
new file mode 100644
index 0000000000..6f9baf1a71
--- /dev/null
+++ b/tests/army/query/function/ans/pi_1.csv
@@ -0,0 +1,114 @@
+
+taos> select pi()
+           pi()            |
+============================
+         3.141592653589793 |
+
+taos> select pi() + 1
+         pi() + 1          |
+============================
+         4.141592653589793 |
+
+taos> select pi() - 1
+         pi() - 1          |
+============================
+         2.141592653589793 |
+
+taos> select pi() * 2
+         pi() * 2          |
+============================
+         6.283185307179586 |
+
+taos> select pi() / 2
+         pi() / 2          |
+============================
+         1.570796326794897 |
+
+taos> select pi() from ts_4893.meters limit 5
+           pi()            |
+============================
+         3.141592653589793 |
+         3.141592653589793 |
+         3.141592653589793 |
+         3.141592653589793 |
+         3.141592653589793 |
+
+taos> select pi() + 1 from ts_4893.meters limit 1
+         pi() + 1          |
+============================
+         4.141592653589793 |
+
+taos> select pi() - 1 from ts_4893.meters limit 1
+         pi() - 1          |
+============================
+         2.141592653589793 |
+
+taos> select pi() * 2 from ts_4893.meters limit 1
+         pi() * 2          |
+============================
+         6.283185307179586 |
+
+taos> select pi() / 2 from ts_4893.meters limit 1
+         pi() / 2          |
+============================
+         1.570796326794897 |
+
+taos> select pi() + pi() from ts_4893.meters limit 1
+        pi() + pi()        |
+============================
+         6.283185307179586 |
+
+taos> select pi() - pi() from ts_4893.meters limit 1
+        pi() - pi()        |
+============================
+         0.000000000000000 |
+
+taos> select pi() * pi() from ts_4893.meters limit 1
+        pi() * pi()        |
+============================
+         9.869604401089358 |
+
+taos> select pi() / pi() from ts_4893.meters limit 1
+        pi() / pi()        |
+============================
+         1.000000000000000 |
+
+taos> select pi() + id from ts_4893.meters order by ts limit 5
+         pi() + id         |
+============================
+         3.141592653589793 |
+         4.141592653589793 |
+         5.141592653589793 |
+         6.141592653589793 |
+         7.141592653589793 |
+
+taos> select abs(pi())
+         abs(pi())         |
+============================
+         3.141592653589793 |
+
+taos> select pow(pi(), 2)
+       pow(pi(), 2)        |
+============================
+         9.869604401089358 |
+
+taos> select sqrt(pi())
+        sqrt(pi())         |
+============================
+         1.772453850905516 |
+
+taos> select cast(pi() as int)
+ cast(pi() as int) |
+====================
+                 3 |
+
+taos> select pi()
+           pi()            |
+============================
+         3.141592653589793 |
+
+taos> select substring_index(null, '.', 2)
+ substring_index(null, '.', 2) |
+================================
+ NULL                          |
+
diff --git a/tests/army/query/function/in/pi.in b/tests/army/query/function/in/pi.in
index c0ccc0b079..dc5d24b655 100644
--- a/tests/army/query/function/in/pi.in
+++ b/tests/army/query/function/in/pi.in
@@ -1,20 +1,21 @@
-select pi();
-select pi() + 1;
-select pi() - 1;
-select pi() * 2;
-select pi() / 2;
-select pi() from ts_4893.meters limit 5;
-select pi() + 1 from ts_4893.meters limit 1;
-select pi() - 1 from ts_4893.meters limit 1;
-select pi() * 2 from ts_4893.meters limit 1;
-select pi() / 2 from ts_4893.meters limit 1;
-select pi() + pi() from ts_4893.meters limit 1;
-select pi() - pi() from ts_4893.meters limit 1;
-select pi() * pi() from ts_4893.meters limit 1;
-select pi() / pi() from ts_4893.meters limit 1;
-select pi() + id from ts_4893.meters order by ts limit 5;
-select abs(pi());
-select pow(pi(), 2);
-select sqrt(pi());
-select cast(pi() as int);
-select pi();
+select pi()
+select pi() + 1
+select pi() - 1
+select pi() * 2
+select pi() / 2
+select pi() from ts_4893.meters limit 5
+select pi() + 1 from ts_4893.meters limit 1
+select pi() - 1 from ts_4893.meters limit 1
+select pi() * 2 from ts_4893.meters limit 1
+select pi() / 2 from ts_4893.meters limit 1
+select pi() + pi() from ts_4893.meters limit 1
+select pi() - pi() from ts_4893.meters limit 1
+select pi() * pi() from ts_4893.meters limit 1
+select pi() / pi() from ts_4893.meters limit 1
+select pi() + id from ts_4893.meters order by ts limit 5
+select abs(pi())
+select pow(pi(), 2)
+select sqrt(pi())
+select cast(pi() as int)
+select pi()
+select substring_index(null, '.', 2)
diff --git a/tests/army/query/function/test_function.py b/tests/army/query/function/test_function.py
index 18a0d46711..aae0cf6eee 100644
--- a/tests/army/query/function/test_function.py
+++ b/tests/army/query/function/test_function.py
@@ -17,14 +17,15 @@ import random
 
 import taos
 import frame
-import frame.etool
 
+from frame.etool import *
 from frame.log import *
 from frame.cases import *
 from frame.sql import *
 from frame.caseBase import *
-from frame import *
 
+from frame import etool
+from frame.common import *
 
 class TDTestCase(TBase):
     updatecfgDict = {
@@ -84,8 +85,16 @@ class TDTestCase(TBase):
                     tdSql.error(err_statement)
                     err_statement = ''
 
+    def test_normal_query_new(self, testCase):
+        # read sql from .sql file and execute
+        tdLog.info(f"test normal query.")
+        self.sqlFile = etool.curFile(__file__, f"in/{testCase}.in")
+        self.ansFile = etool.curFile(__file__, f"ans/{testCase}_1.csv")
+
+        tdCom.compare_testcase_result(self.sqlFile, self.ansFile, testCase)
+
     def test_pi(self):
-        self.test_normal_query("pi")
+        self.test_normal_query_new("pi")
 
     def test_round(self):
         self.test_normal_query("round")
diff --git a/tests/system-test/7-tmq/ts-4674.py b/tests/system-test/7-tmq/ts-4674.py
index 709debaef1..0b3dc1b077 100644
--- a/tests/system-test/7-tmq/ts-4674.py
+++ b/tests/system-test/7-tmq/ts-4674.py
@@ -24,45 +24,6 @@ class TDTestCase:
         tdLog.debug(f"start to excute {__file__}")
         tdSql.init(conn.cursor())
 
-    #tdSql.init(conn.cursor(), logSql)  # output sql.txt file
-
-    # def consume_TS_4674_Test(self):
-    #
-    #     os.system("nohup taosBenchmark -y -B 1 -t 4 -S 1000 -n 1000000 -i 1000 -v 1 -a 3 > /dev/null 2>&1 &")
-    #     time.sleep()
-    #     tdSql.execute(f'create topic topic_all with meta as database test')
-    #     consumer_dict = {
-    #         "group.id": "g1",
-    #         "td.connect.user": "root",
-    #         "td.connect.pass": "taosdata",
-    #         "auto.offset.reset": "earliest",
-    #     }
-    #     consumer = Consumer(consumer_dict)
-    #
-    #     try:
-    #         consumer.subscribe(["topic_all"])
-    #     except TmqError:
-    #         tdLog.exit(f"subscribe error")
-    #
-    #     try:
-    #         while True:
-    #             res = consumer.poll(5)
-    #             if not res:
-    #                 print(f"null")
-    #                 continue
-    #             val = res.value()
-    #             if val is None:
-    #                 print(f"null")
-    #                 continue
-    #             cnt = 0;
-    #             for block in val:
-    #                 cnt += len(block.fetchall())
-    #
-    #             print(f"block {cnt} rows")
-    #
-    #     finally:
-    #         consumer.close()
-
     def get_leader(self):
         tdLog.debug("get leader")
         tdSql.query("show vnodes")
@@ -74,19 +35,20 @@ class TDTestCase:
 
     def balance_vnode(self):
         leader_before = self.get_leader()
-
+        tdSql.query("balance vgroup leader")
         while True:
             leader_after = -1
-            tdSql.query("balance vgroup leader")
+            tdLog.debug("balancing vgroup leader")
             while True:
+                tdLog.debug("get new vgroup leader")
                 leader_after = self.get_leader()
                 if leader_after != -1 :
-                    break;
+                    break
                 else:
                     time.sleep(1)
             if leader_after != leader_before:
                 tdLog.debug("leader changed")
-                break;
+                break
             else :
                 time.sleep(1)
 
@@ -115,7 +77,7 @@ class TDTestCase:
         except TmqError:
             tdLog.exit(f"subscribe error")
 
-        cnt = 0;
+        cnt = 0
         balance = False
         try:
             while True: