Arvados-DCO-1.1-Signed-off-by: Radhika Chippada <radhika@curoverse.com>

[arvados.git] / services / nodemanager / arvnodeman / baseactor.py

diff --git a/services/nodemanager/arvnodeman/baseactor.py b/services/nodemanager/arvnodeman/baseactor.py
index 840ba4c931ad856ecb1de94e3cc1aa48518867fd..988b83c142b15d6a62058ec05ee0863d03476503 100644 (file)
--- a/services/nodemanager/arvnodeman/baseactor.py
+++ b/services/nodemanager/arvnodeman/baseactor.py
@@ -1,3 +1,7 @@
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
 from __future__ import absolute_import, print_function
 
 import errno
@@ -84,7 +88,7 @@ class BaseNodeManagerActor(pykka.ThreadingActor):
         if (exception_type in (threading.ThreadError, MemoryError) or
             exception_type is OSError and exception_value.errno == errno.ENOMEM):
             lg.critical("Unhandled exception is a fatal error, killing Node Manager")
-            os.kill(os.getpid(), signal.SIGQUIT)
+            os.kill(os.getpid(), signal.SIGKILL)
 
     def ping(self):
         return True
@@ -94,29 +98,25 @@ class WatchdogActor(pykka.ThreadingActor):
     def __init__(self, timeout, *args, **kwargs):
          super(pykka.ThreadingActor, self).__init__(*args, **kwargs)
          self.timeout = timeout
+         self.actors = [a.proxy() for a in args]
          self.actor_ref = TellableActorRef(self)
          self._later = self.actor_ref.tell_proxy()
 
-    def kill_self(self, act):
+    def kill_self(self, e, act):
         lg = getattr(self, "_logger", logging)
+        lg.critical("Watchdog exception", exc_info=e)
         lg.critical("Actor %s watchdog ping time out, killing Node Manager", act)
-        os.kill(os.getpid(), signal.SIGQUIT)
+        os.kill(os.getpid(), signal.SIGKILL)
 
     def on_start(self):
         self._later.run()
 
     def run(self):
-        actors = pykka.ActorRegistry.get_all()
-        for a in actors:
-            if a.actor_class is WatchdogActor:
-                continue
-            try:
-                a.proxy().ping().get(self.timeout)
-            except pykka.ActorDeadError:
-                pass
-            except pykka.Timeout:
-                self.kill_self(a)
-                return
-
-        time.sleep(20)
-        self._later.run()
+        a = None
+        try:
+            for a in self.actors:
+                a.ping().get(self.timeout)
+            time.sleep(20)
+            self._later.run()
+        except Exception as e:
+            self.kill_self(e, a)