Connection polling errors -- log more information and attempt to recover gracefully.

kombu/transport/redis.py

@@ -529,10 +529,25 @@ class Channel(virtual.Channel):
 
     def _poll_error(self, type, **options):
         try:
+            # This might error out if the (redis)client.connection is already disconnected
             self.client.parse_response(self.client.connection, type)
         except self.connection_errors:
             warn("Connection poll error on command type {}".format(type), exc_info=True)
-            
+        except AttributeError as ex:
+            # This is definitely hacky, but currently fixes a problem where Redis connection has already been closed
+            # when we call parse_response. Hoping for suggestions on the best way to handle this.
+            if not "object has no attribute 'readline'" in ex.message:
+                raise
+            # Maybe we can get more information on the root cause from this logging.
+            warn("Unable to read from channel {} of type {}. Err: {}".format(self, type, ex.message))
+            # Disconnect so that we will automatically reconnect in the next polling cycle
+            #   Someone may want to verify this and determine if there's a better way
+            #   Used _brpop_read and close methods as examples to deduce what should work
+            for attr in 'client', 'subclient':
+                try:
+                    self.__dict__[attr].connection.disconnect()
+                except (KeyError, AttributeError, self.ResponseError):
+                    pass
 
     def _get(self, queue):
         with self.conn_or_acquire() as client: