Enable and stabilize MetaDrive simulation tests in CI

.github/workflows/tests.yaml

@@ -196,19 +196,33 @@ jobs:
       (github.event.pull_request.head.repo.full_name == 'commaai/openpilot'))
       && fromJSON('["namespace-profile-amd64-8x16"]')
       || fromJSON('["ubuntu-24.04"]') }}
-    if: false  # FIXME: Started to timeout recently
+    env:
+      ONNXCPU: 1
+      SIM_LOGS: 1
+      CI: 1
     steps:
     - uses: actions/checkout@v6
       with:
         submodules: true
     - run: ./tools/op.sh setup
     - name: Build openpilot
       run: scons
+    - name: Install MetaDrive
+      run: uv pip install "metadrive-simulator @ git+https://github.com/commaai/metadrive.git@minimal"
     - name: Driving test
-      timeout-minutes: 2
+      timeout-minutes: 10
       run: |
         source openpilot/selfdrive/test/setup_xvfb.sh
         pytest -s openpilot/tools/sim/tests/test_metadrive_bridge.py
+    - name: Upload simulator logs
+      uses: actions/upload-artifact@v6
+      if: always()
+      with:
+        name: simulator-logs-${{ github.run_id }}
+        path: |
+          /tmp/comma_download/**/*.bz2
+          /tmp/comma_download/**/*.hevc
+          /tmp/comma_download/**/*.zst
 
   create_ui_report:
     name: Create UI Report

openpilot/tools/sim/bridge/common.py

@@ -1,4 +1,6 @@
+import os
 import signal
+import time
 import threading
 import functools
 import numpy as np
@@ -41,7 +43,8 @@ def __init__(self, dual_camera, high_quality):
     self.params = Params()
     self.params.put_bool("AlphaLongitudinalEnabled", True, block=True)
 
-    self.rk = Ratekeeper(100, None)
+    self._ci = os.getenv("CI") is not None
+    self.rk = Ratekeeper(20 if self._ci else 100, None)
 
     self.dual_camera = dual_camera
     self.high_quality = high_quality
@@ -203,3 +206,5 @@ def _run(self, q: Queue):
       self.started.value = True
 
       self.rk.keep_time()
+      if self._ci:
+        time.sleep(0.01)

openpilot/tools/sim/bridge/metadrive/metadrive_process.py

@@ -1,4 +1,5 @@
 import math
+import os
 import time
 import numpy as np
 
@@ -93,7 +94,8 @@ def get_cam_as_rgb(cam):
       img = img.get() # convert cupy array to numpy
     return img
 
-  rk = Ratekeeper(100, None)
+  _ci = os.getenv("CI") is not None
+  rk = Ratekeeper(20 if _ci else 100, None)
 
   steer_ratio = 8
   vc = [0,0]

openpilot/tools/sim/launch_openpilot.sh

@@ -6,7 +6,12 @@ export SIMULATION="1"
 export SKIP_FW_QUERY="1"
 export FINGERPRINT="HONDA_CIVIC_2022"
 
-export BLOCK="${BLOCK},camerad,loggerd,encoderd,micd,logmessaged,manage_athenad"
+# keep camera process blocked (simulator publishes frames), but optionally allow logging
+block_list="camerad,micd,logmessaged,manage_athenad"
+if [[ -z "$SIM_LOGS" ]]; then
+  block_list="$block_list,loggerd,encoderd"
+fi
+export BLOCK="${BLOCK},${block_list}"
 if [[ "$CI" ]]; then
   # TODO: offscreen UI should work
   export BLOCK="${BLOCK},ui"

openpilot/tools/sim/tests/test_sim_bridge.py

@@ -1,4 +1,5 @@
 import os
+import signal
 import subprocess
 import time
 import pytest
@@ -22,7 +23,7 @@ def setup_method(self):
 
   def test_driving(self):
     # Startup manager and bridge.py. Check processes are running, then engage and verify.
-    p_manager = subprocess.Popen("./launch_openpilot.sh", cwd=SIM_DIR)
+    p_manager = subprocess.Popen("./launch_openpilot.sh", cwd=SIM_DIR, start_new_session=True)
     self.processes.append(p_manager)
 
     sm = messaging.SubMaster(['selfdriveState', 'onroadEvents', 'managerState'])
@@ -31,7 +32,7 @@ def test_driving(self):
     p_bridge = bridge.run(q, retries=10)
     self.processes.append(p_bridge)
 
-    max_time_per_step = 60
+    max_time_per_step = 180 if os.getenv("CI") else 60
 
     # Wait for bridge to startup
     start_waiting = time.monotonic()
@@ -86,7 +87,13 @@ def test_driving(self):
   def teardown_method(self):
     print("Test shutting down. CommIssues are acceptable")
     for p in reversed(self.processes):
-      p.terminate()
+      try:
+        os.killpg(os.getpgid(p.pid), signal.SIGKILL)
+      except (ProcessLookupError, PermissionError, OSError):
+        pass
 
     for p in reversed(self.processes):
-      p.kill()
+      try:
+        p.kill()
+      except (ProcessLookupError, OSError):
+        pass