Prevent synchronized crash-session uploads during driving

openpilot/system/sentry.py

@@ -58,7 +58,7 @@ def init(project: SentryProject = SentryProject.SELFDRIVE) -> bool:
       sdk.init(dsn=project.value, default_integrations=False, auto_enabling_integrations=False,
                integrations=[threading_module.ThreadingIntegration(propagate_scope=True)],
                transport=BoundedHttpTransport, transport_queue_size=TRANSPORT_QUEUE_SIZE,
-               shutdown_timeout=EXIT_FLUSH_TIMEOUT, enable_backpressure_handling=False,
+               shutdown_timeout=EXIT_FLUSH_TIMEOUT, enable_backpressure_handling=False, auto_session_tracking=False,
                max_value_length=8192, traces_sample_rate=0.0,
                release=os.getenv("GIT_COMMIT", "unknown"), environment="Release")
       _initialized_pid = os.getpid()
@@ -86,11 +86,9 @@ def init(project: SentryProject = SentryProject.SELFDRIVE) -> bool:
       sdk.set_tag(key, str(value) if value is not None else "unknown")
   except Exception:
     cloudlog.warning("Crash reporting metadata unavailable")
-  if fresh:
-    try:
-      sdk.start_session()
-    except Exception:
-      cloudlog.warning("Crash reporting session unavailable")
+  # Forked daemons must not close and upload an inherited manager session.
+  # Its deferred 60-second flush synchronizes background TLS work across them.
+  # Explicit exceptions retain the bounded transport and threading integration.
   return True
 
 

openpilot/system/tests/test_sentry.py

@@ -98,7 +98,6 @@ def test_child_import_and_main_failure_after_real_fork(reporting, monkeypatch, d
     writer.send((sentry._initialized_pid, event_list(reporting)))
 
   monkeypatch.setattr(reporting.sdk, 'capture_exception', capture)
-  monkeypatch.setattr(process.messaging, 'reset_context', Mock())
   monkeypatch.setattr(process, 'setproctitle', Mock())
   monkeypatch.setattr(sentry, 'flush', Mock())
 
@@ -179,7 +178,9 @@ def test_native_apport_keeps_local_copy_and_queues_header_stack_qlog(reporting,
     monkeypatch.setattr(sentry, 'sdk', None)
   tombstoned.report_tombstone_apport(str(source))
   assert not source.exists()
-  copies = list((reporting.tmp_path / 'crash').iterdir())
+  crash_dir = reporting.tmp_path / 'crash'
+  assert (crash_dir / '.lock').is_file()
+  copies = [path for path in crash_dir.iterdir() if path.name != '.lock']
   assert len(copies) == 1 and copies[0].read_text() == payload
   if not report_available:
     assert event_list(reporting) == []
@@ -279,3 +280,40 @@ def test_metadata_or_attachment_error_does_not_disable_capture(reporting, monkey
   sentry.capture_exception(ValueError('metadata and attachment unavailable'))
   event, = event_list(reporting)
   assert event['exception']['values'][0]['type'] == 'ValueError'
+
+
+def test_forked_daemon_has_no_deferred_session_upload(reporting, monkeypatch):
+  ctx = multiprocessing.get_context('fork')
+  reader, writer = ctx.Pipe(duplex=False)
+
+  def child():
+    try:
+      assert sentry.init()
+      client = reporting.sdk.get_client()
+      pending = list(client.session_flusher.pending_sessions)
+      try:
+        raise RuntimeError('child crash still reported without session telemetry')
+      except RuntimeError:
+        reporting.sdk.capture_exception()
+      writer.send((pending, event_list(reporting), client.options['auto_session_tracking']))
+    finally:
+      writer.close()
+
+  process = ctx.Process(target=child)
+  process.start()
+  writer.close()
+  try:
+    assert reader.poll(5), 'child did not report its SDK state'
+    pending, events, automatic = reader.recv()
+    process.join(5)
+    assert not process.is_alive() and process.exitcode == 0
+    assert pending == [], 'inherited manager session schedules a 60-second upload'
+    assert automatic is False
+    event, = events
+    assert event['exception']['values'][0]['value'] == 'child crash still reported without session telemetry'
+    assert event['user']['id'] == 'test-device'
+  finally:
+    if process.is_alive():
+      process.kill()
+      process.join(5)
+    reader.close()

prebuilt.json

@@ -1 +1 @@
-{"protected_inventory_sha256": "b519174b880231cedd1dd6d1c29066db4523a5c72efe4317660eba7a9f340bd3", "schema_version": 1}
+{"protected_inventory_sha256": "158cf8d45beb165a7fb5d2d5e2fc84c85b565fc7809078ecf6236c4acae21fee", "schema_version": 1}

tools/ci/run_host_tests.py

@@ -12,7 +12,8 @@ import tempfile
 
 ROOT = Path(__file__).resolve().parents[2]
 RUNNER = ROOT / 'tools/test_runner.py'
-UI_PYTEST_FILES = (
+EXTRA_PYTEST_FILES = (
+  'openpilot/system/tests/test_sentry.py',
   'openpilot/system/ui/lib/tests/test_text_caches.py',
   'openpilot/system/ui/widgets/tests/test_label_cache.py',
   'openpilot/selfdrive/ui/tests/test_onboarding_textures.py',
@@ -28,7 +29,7 @@ def pytest_files(root=ROOT):
            (isinstance(node, ast.ClassDef) and node.name.startswith('Test') and not node.bases)
            for node in module.body):
       files.append(path.relative_to(root).as_posix())
-  return files + [path for path in UI_PYTEST_FILES if (root / path).is_file()]
+  return files + [path for path in EXTRA_PYTEST_FILES if (root / path).is_file()]
 
 
 def main(arguments=None):

tools/ci/tests/test_feature_test_discovery.py

@@ -37,6 +37,7 @@ class TestFeatureTestDiscovery(unittest.TestCase):
     workflow = (ROOT / '.github/workflows/tests.yaml').read_text()
     self.assert_hosted_pytest_contract(workflow)
     selected = pytest_files()
+    self.assertIn('openpilot/system/tests/test_sentry.py', selected)
     self.assertTrue(selected)
     for package in ('flm', 'spot_monitor', 'sentry_mode', 'galaxy', 'ui'):
       self.assertTrue(any(path.startswith(f'openpilot/starpilot/{package}/') for path in selected), package)