Merge branch '11807-yaml-to-json'

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

diff --git a/services/nodemanager/arvnodeman/clientactor.py b/services/nodemanager/arvnodeman/clientactor.py
index 77d85d640ca68ac61cfde028ac37dd82ea7c88bb..afc4f1cb5845d3e73e8d65dbb6658659594a686b 100644 (file)
--- a/services/nodemanager/arvnodeman/clientactor.py
+++ b/services/nodemanager/arvnodeman/clientactor.py
@@ -1,4 +1,7 @@
 #!/usr/bin/env python
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
 
 from __future__ import absolute_import, print_function
 
@@ -30,40 +33,42 @@ class RemotePollLoopActor(actor_class):
     response to subscribers.  It takes care of error handling, and retrying
     requests with exponential backoff.
 
-    To use this actor, define CLIENT_ERRORS and the _send_request method.
-    If you also define an _item_key method, this class will support
-    subscribing to a specific item by key in responses.
+    To use this actor, define the _send_request method.  If you also
+    define an _item_key method, this class will support subscribing to
+    a specific item by key in responses.
     """
     def __init__(self, client, timer_actor, poll_wait=60, max_poll_wait=180):
         super(RemotePollLoopActor, self).__init__()
         self._client = client
         self._timer = timer_actor
-        self._logger = logging.getLogger(self.LOGGER_NAME)
-        self._later = self.actor_ref.proxy()
+        self._later = self.actor_ref.tell_proxy()
+        self._polling_started = False
         self.min_poll_wait = poll_wait
         self.max_poll_wait = max_poll_wait
         self.poll_wait = self.min_poll_wait
-        self.last_poll_time = None
         self.all_subscribers = set()
         self.key_subscribers = {}
         if hasattr(self, '_item_key'):
             self.subscribe_to = self._subscribe_to
 
+    def on_start(self):
+        self._logger = logging.getLogger("%s.%s" % (self.__class__.__name__, id(self.actor_urn[9:])))
+
     def _start_polling(self):
-        if self.last_poll_time is None:
-            self.last_poll_time = time.time()
+        if not self._polling_started:
+            self._polling_started = True
             self._later.poll()
 
     def subscribe(self, subscriber):
         self.all_subscribers.add(subscriber)
-        self._logger.debug("%r subscribed to all events", subscriber)
+        self._logger.debug("%s subscribed to all events", subscriber.actor_ref.actor_urn)
         self._start_polling()
 
     # __init__ exposes this method to the proxy if the subclass defines
     # _item_key.
     def _subscribe_to(self, key, subscriber):
         self.key_subscribers.setdefault(key, set()).add(subscriber)
-        self._logger.debug("%r subscribed to events for '%s'", subscriber, key)
+        self._logger.debug("%s subscribed to events for '%s'", subscriber.actor_ref.actor_urn, key)
         self._start_polling()
 
     def _send_request(self):
@@ -79,18 +84,33 @@ class RemotePollLoopActor(actor_class):
 
     def _got_error(self, error):
         self.poll_wait = min(self.poll_wait * 2, self.max_poll_wait)
-        self._logger.warning("Client error: %s - waiting %s seconds",
-                             error, self.poll_wait)
+        return "got error: {} - will try again in {} seconds".format(
+            error, self.poll_wait)
+
+    def is_common_error(self, exception):
+        return False
 
-    def poll(self):
+    def poll(self, scheduled_start=None):
+        self._logger.debug("sending request")
         start_time = time.time()
+        if scheduled_start is None:
+            scheduled_start = start_time
         try:
             response = self._send_request()
-        except self.CLIENT_ERRORS as error:
-            self.last_poll_time = start_time
-            self._got_error(error)
+        except Exception as error:
+            errmsg = self._got_error(error)
+            if self.is_common_error(error):
+                self._logger.warning(errmsg)
+            else:
+                self._logger.exception(errmsg)
+            next_poll = start_time + self.poll_wait
         else:
-            self.last_poll_time += self.poll_wait
             self._got_response(response)
-        self._timer.schedule(self.last_poll_time + self.poll_wait,
-                             self._later.poll)
+            next_poll = scheduled_start + self.poll_wait
+            self._logger.info("got response with %d items in %s seconds, next poll at %s",
+                              len(response), (time.time() - scheduled_start),
+                              time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(next_poll)))
+        end_time = time.time()
+        if next_poll < end_time:  # We've drifted too much; start fresh.
+            next_poll = end_time + self.poll_wait
+        self._timer.schedule(next_poll, self._later.poll, next_poll)