feat: Reset request cache before handling event (#201)

This is a just-in-case measure -- we don't know of any specific code that
handles events and relies on the cache. But it's likely to come up at some
point.

Ticket: openedx/openedx-events#235

CHANGELOG.rst

@@ -14,6 +14,12 @@ Change Log
 Unreleased
 **********
 
+[5.5.0] - 2023-09-21
+********************
+Changed
+=======
+* Reset edx-django-utils RequestCache before handling each event
+
 [5.4.0] - 2023-08-28
 ********************
 Changed

edx_event_bus_kafka/__init__.py

@@ -9,4 +9,4 @@
 from edx_event_bus_kafka.internal.consumer import KafkaEventConsumer
 from edx_event_bus_kafka.internal.producer import KafkaEventProducer, create_producer
 
-__version__ = '5.4.0'
+__version__ = '5.5.0'

edx_event_bus_kafka/internal/consumer.py

@@ -10,6 +10,7 @@
 from django.db import connection
 from django.dispatch import receiver
 from django.test.signals import setting_changed
+from edx_django_utils.cache import RequestCache
 from edx_django_utils.monitoring import function_trace, record_exception, set_custom_attribute
 from edx_toggles.toggles import SettingToggle
 from openedx_events.event_bus import EventBusConsumer
@@ -106,6 +107,31 @@ def _reconnect_to_db_if_needed():
         connection.connect()
 
 
+def _clear_request_cache():
+    """
+    Clear the RequestCache so that each event consumption starts fresh.
+
+    Signal handlers may be written with the assumption that they are called in the context
+    of a web request, so we clear the request cache just in case.
+    """
+    RequestCache.clear_all_namespaces()
+
+
+def _prepare_for_new_work_cycle():
+    """
+    Ensure that the application state is appropriate for performing a new unit of work.
+
+    This mimics some setup/teardown that is normally performed by Django in its
+    request/response based architecture and that is needed for ensuring a clean and
+    usable state in this worker-based application.
+    """
+    # Ensure that the database connection is active and usable.
+    _reconnect_to_db_if_needed()
+
+    # Clear the request cache, in case anything in the signal handlers rely on it.
+    _clear_request_cache()
+
+
 class KafkaEventConsumer(EventBusConsumer):
     """
     Construct consumer for the given topic and group. The consumer can then
@@ -278,8 +304,11 @@ def _consume_indefinitely(self):
                     msg = self.consumer.poll(timeout=CONSUMER_POLL_TIMEOUT)
                     if msg is not None:
                         with function_trace('_consume_indefinitely_consume_single_message'):
-                            # Before processing, make sure our db connection is still active
-                            _reconnect_to_db_if_needed()
+                            # Before processing, try to make sure our application state is cleaned
+                            # up as would happen at the start of a Django request/response cycle.
+                            # See https://github.com/openedx/openedx-events/issues/236 for details.
+                            _prepare_for_new_work_cycle()
+
                             signal = self.determine_signal(msg)
                             msg.set_value(self._deserialize_message_value(msg, signal))
                             self.emit_signals_from_message(msg, signal)