wip

selfdrive/manager/process.py

@@ -47,13 +47,10 @@ def launcher(proc: str, name: str, log_path: str) -> None:
         raise
 
 def join_process(process: Process, timeout: float) -> None:
-  # Process().join(timeout) will hang due to a python 3 bug: https://bugs.python.org/issue28382
-  # We have to poll the exitcode instead
   t = time.monotonic()
   while time.monotonic() - t < timeout and process.exitcode is None:
     time.sleep(0.001)
 
-
 class ManagerProcess(ABC):
   daemon = False
   sigkill = False
@@ -61,7 +58,6 @@ class ManagerProcess(ABC):
   proc: Process | None = None
   enabled = True
   name = ""
-
   last_watchdog_time = 0
   watchdog_max_dt: int | None = None
   watchdog_seen = False
@@ -75,24 +71,22 @@ class ManagerProcess(ABC):
   def start(self) -> None:
     pass
 
+  
   def restart(self) -> None:
-    self.stop(sig=signal.SIGKILL)
+    if self.proc is not None and self.proc.exitcode is not None:
+      self.stop(sig=signal.SIGKILL, block=False)
     self.start()
 
   def check_watchdog(self, started: bool) -> None:
     if self.watchdog_max_dt is None or self.proc is None:
       return
-
     try:
       fn = WATCHDOG_FN + str(self.proc.pid)
       with open(fn, "rb") as f:
-        # TODO: why can't pylint find struct.unpack?
         self.last_watchdog_time = struct.unpack('Q', f.read())[0]
     except Exception:
       pass
-
     dt = time.monotonic() - self.last_watchdog_time / 1e9
-
     if dt > self.watchdog_max_dt:
       if (self.watchdog_seen or self.always_watchdog and self.proc.exitcode is not None) and ENABLE_WATCHDOG:
         cloudlog.error(f"Watchdog timeout for {self.name} (exitcode {self.proc.exitcode}) restarting ({started=})")
@@ -103,7 +97,6 @@ class ManagerProcess(ABC):
   def stop(self, retry: bool = True, block: bool = True, sig: signal.Signals = None) -> int | None:
     if self.proc is None:
       return None
-
     if self.proc.exitcode is None:
       if not self.shutting_down:
         cloudlog.info(f"killing {self.name}")
@@ -111,39 +104,23 @@ class ManagerProcess(ABC):
           sig = signal.SIGKILL if self.sigkill else signal.SIGINT
         self.signal(sig)
         self.shutting_down = True
-
         if not block:
           return None
-
       join_process(self.proc, 5)
-
-      # If process failed to die send SIGKILL
       if self.proc.exitcode is None and retry:
         cloudlog.info(f"killing {self.name} with SIGKILL")
         self.signal(signal.SIGKILL)
         self.proc.join()
-
     ret = self.proc.exitcode
     cloudlog.info(f"{self.name} is dead with {ret}")
-
     if self.proc.exitcode is not None:
       self.shutting_down = False
       self.proc = None
-
     return ret
 
   def signal(self, sig: int) -> None:
-    if self.proc is None:
+    if self.proc is None or self.proc.exitcode is not None or self.proc.pid is None:
       return
-
-    # Don't signal if already exited
-    if self.proc.exitcode is not None and self.proc.pid is not None:
-      return
-
-    # Can't signal if we don't have a pid
-    if self.proc.pid is None:
-      return
-
     cloudlog.info(f"sending signal {sig} to {self.name}")
     os.kill(self.proc.pid, sig)
 
@@ -243,6 +220,7 @@ class DaemonProcess(ManagerProcess):
   def stop(self, retry=True, block=True, sig=None) -> None:
     pass
 
+
 def ensure_running(procs: ValuesView[ManagerProcess], started: bool, params=None, CP: car.CarParams=None, not_run: list[str] | None=None, log_dir: str = None) -> list[ManagerProcess]:
     if not_run is None:
         not_run = []
@@ -251,11 +229,12 @@ def ensure_running(procs: ValuesView[ManagerProcess], started: bool, params=None
     for p in procs:
         log_path = log_dir+"/"+p.name+".log"
         if p.enabled and p.name not in not_run and p.should_run(started, params, CP):
-            p.start(log_path)
+            if p.proc is None or p.proc.exitcode is not None:
+                p.start(log_path)
             running.append(p)
         else:
             p.stop(block=False)
 
         p.check_watchdog(started)
 
-    return running
+    return running