Update process.py

openpilot/system/manager/process.py

@@ -105,6 +105,12 @@ class ManagerProcess(ABC):
 
     return ret
 
+  def reap(self) -> None:
+    # a process that exited on its own is otherwise never restarted: start() early-returns while proc is
+    # set, and only stop() clears it. modeld dying on a bad eGPU load left it dead for the whole drive.
+    if self.proc is not None and self.proc.exitcode is not None:
+      self.stop()
+
   def signal(self, sig: int) -> None:
     if self.proc is None:
       return
@@ -145,6 +151,7 @@ class NativeProcess(ManagerProcess):
     # In case we only tried a non blocking stop we need to stop it before restarting
     if self.shutting_down:
       self.stop()
+    self.reap()
 
     if self.proc is not None:
       return
@@ -169,6 +176,7 @@ class PythonProcess(ManagerProcess):
     # In case we only tried a non blocking stop we need to stop it before restarting
     if self.shutting_down:
       self.stop()
+    self.reap()
 
     if self.proc is not None:
       return