Ack our inability to hard kill sub-procs

tractor/_portal.py

@@ -278,7 +278,8 @@ class Portal:
         try:
             # send cancel cmd - might not get response
             # XXX: sure would be nice to make this work with a proper shield
-            # with trio.CancelScope(shield=True):
+            # with trio.CancelScope() as cancel_scope:
+            # with trio.CancelScope(shield=True) as cancel_scope:
             with trio.move_on_after(0.5) as cancel_scope:
                 cancel_scope.shield = True
                 await self.run('self', 'cancel')

tractor/_spawn.py

@@ -188,8 +188,17 @@ async def spawn_subactor(
         # the outer scope since no actor zombies are
         # ever allowed. This ``__aexit__()`` also shields
         # internally.
-        async with proc:
-            log.debug(f"Terminating {proc}")
+        log.debug(f"Attempting to kill {proc}")
+
+        # NOTE: this timeout effectively does nothing right now since
+        # we are shielding the ``.wait()`` inside ``new_proc()`` which
+        # will pretty much never release until the process exits.
+        with trio.move_on_after(3) as cs:
+            async with proc:
+                log.debug(f"Terminating {proc}")
+        if cs.cancelled_caught:
+            log.critical(f"HARD KILLING {proc}")
+            proc.kill()
 
 
 async def new_proc(
@@ -354,6 +363,8 @@ async def new_proc(
                 await proc_waiter(proc)
             proc.join()
 
+        # This is again common logic for all backends:
+
         log.debug(f"Joined {proc}")
         # pop child entry to indicate we are no longer managing this subactor
         subactor, proc, portal = actor_nursery._children.pop(subactor.uid)