FPv2: add timing unit test (#27753)

* add timing unit test

* formatting

* not clean

* Revert "not clean"

This reverts commit dab35d582c274efc32d693fc0a2773d8727be2bd.

* Revert "formatting"

This reverts commit 56a1ae4bb721594cc68952e5747a3ebc652f62d9.

* need to consider subaddrs (0.3 to 0.7s for toyota)

* clean up

* move to test_fw_fingerprint

* use a thread to get real time

* lower threshold

* revert that

* ?

* some sleep

* at top

* use the config

* function to measure query time, similar to opendbc's convention

* revert that

* avg

* Update selfdrive/car/tests/test_fw_fingerprint.py

* draft

* starting a thread takes a long time!

* refs

* uncomment

selfdrive/car/tests/test_fw_fingerprint.py

@@ -1,13 +1,16 @@
 #!/usr/bin/env python3
 import random
+import time
 import unittest
 from collections import defaultdict
 from parameterized import parameterized
+import threading
 
 from cereal import car
+from common.params import Params
 from selfdrive.car.car_helpers import interfaces
 from selfdrive.car.fingerprints import FW_VERSIONS
-from selfdrive.car.fw_versions import FW_QUERY_CONFIGS, VERSIONS, match_fw_to_car
+from selfdrive.car.fw_versions import FW_QUERY_CONFIGS, VERSIONS, match_fw_to_car, get_fw_versions
 
 CarFw = car.CarParams.CarFw
 Ecu = car.CarParams.Ecu
@@ -15,6 +18,14 @@ Ecu = car.CarParams.Ecu
 ECU_NAME = {v: k for k, v in Ecu.schema.enumerants.items()}
 
 
+class FakeSocket:
+  def receive(self, non_blocking=False):
+    pass
+
+  def send(self, msg):
+    pass
+
+
 class TestFwFingerprint(unittest.TestCase):
   def assertFingerprints(self, candidates, expected):
     candidates = list(candidates)
@@ -103,5 +114,58 @@ class TestFwFingerprint(unittest.TestCase):
                          f'{brand.title()}: FW query whitelist missing ecus: {ecu_strings}')
 
 
+class TestFwFingerprintTiming(unittest.TestCase):
+  def _benchmark(self, brand, num_pandas, ref_time, tol, n):
+    params = Params()
+    fake_socket = FakeSocket()
+
+    times = []
+    for _ in range(n):
+      params.put_bool("ObdMultiplexingEnabled", True)
+      thread = threading.Thread(target=get_fw_versions, args=(fake_socket, fake_socket, brand), kwargs=dict(num_pandas=num_pandas))
+      thread.start()
+      t = time.perf_counter()
+      while thread.is_alive():
+        time.sleep(0.02)
+        if not params.get_bool("ObdMultiplexingChanged"):
+          params.put_bool("ObdMultiplexingChanged", True)
+      times.append(time.perf_counter() - t)
+
+    avg_time = round(sum(times) / len(times), 2)
+    self.assertLess(avg_time, ref_time + tol)
+    self.assertGreater(avg_time, ref_time - tol, "Performance seems to have improved, update test refs.")
+    return avg_time
+
+  @parameterized.expand([(1,), (2,), ])
+  def test_fw_query_timing(self, num_pandas):
+    brand_ref_times = {
+      1: {
+        'body': 0.1,
+        'chrysler': 0.3,
+        'ford': 0.2,
+        'honda': 0.5,
+        'hyundai': 0.7,
+        'mazda': 0.1,
+        'nissan': 0.3,
+        'subaru': 0.1,
+        'tesla': 0.2,
+        'toyota': 0.7,
+        'volkswagen': 0.2,
+      },
+      2: {
+        'hyundai': 1.1,
+      }
+    }
+
+    for brand, config in FW_QUERY_CONFIGS.items():
+      with self.subTest(brand=brand, num_pandas=num_pandas):
+        multi_panda_requests = [r for r in config.requests if r.bus > 3]
+        if not len(multi_panda_requests) and num_pandas > 1:
+          raise unittest.SkipTest("No multi-panda FW queries")
+
+        avg_time = self._benchmark(brand, num_pandas, brand_ref_times[num_pandas][brand], 0.1, 10)
+        print(f'{brand=}, {num_pandas=}, {len(config.requests)=}, avg FW query time={avg_time} seconds')
+
+
 if __name__ == "__main__":
   unittest.main()