Handle failure to publish message

Author: mvdbeek

pulsar/client/amqp_exchange.py

@@ -233,15 +233,20 @@ def publish(self, name, payload):
             with pools.producers[connection].acquire() as producer:
                 log.debug("%sHave producer for publishing to key %s", publish_log_prefix, key)
                 publish_kwds = self.__prepare_publish_kwds(publish_log_prefix)
-                producer.publish(
-                    payload,
-                    serializer='json',
-                    exchange=self.__exchange,
-                    declare=[self.__exchange],
-                    routing_key=key,
-                    **publish_kwds
-                )
-                log.debug("%sPublished to key %s", publish_log_prefix, key)
+                try:
+                    producer.publish(
+                        payload,
+                        serializer='json',
+                        exchange=self.__exchange,
+                        declare=[self.__exchange],
+                        routing_key=key,
+                        **publish_kwds
+                    )
+                    log.debug("%sPublished to key %s", publish_log_prefix, key)
+                    return True
+                except Exception as e:
+                    log.error("%sFailed to publish to key %s: %s", publish_log_prefix, key, str(e))
+                    self.__fail_publish(name, payload, e)
 
     def ack_manager(self):
         log.debug('Acknowledgement manager thread alive')
@@ -261,15 +266,37 @@ def ack_manager(self):
                                       'republishing original message on queue %s',
                                       unack_uuid, resubmit_queue)
                             try:
-                                self.publish(resubmit_queue, payload)
-                                self.publish_uuid_store.set_time(unack_uuid)
+                                if self.publish(resubmit_queue, payload):
+                                    self.publish_uuid_store.set_time(unack_uuid)
+                                else:
+                                    # If we fail to publish, we need to remove the uuid from the store
+                                    # so it doesn't get republished again.
+                                    self.__discard_publish_uuid(unack_uuid, failed)
                             except self.recoverable_exceptions as e:
                                 self.__handle_io_error(e)
                                 continue
         except Exception:
-            log.exception("Problem with acknowledgement manager, leaving ack_manager method in problematic state!")
+            log.exception("Problem with acknowledgement manager, leaving ack manager in problematic state!")
             raise
-        log.debug('Acknowledgement manager thread exiting')
+
+    def __fail_publish(self, name, payload, exception):
+        # Send just a few safe keys if we have them:
+        keys_to_send = [
+            "job_id",
+            "returncode",
+            "stdout",
+            "stderr",
+            "job_stdout",
+            "job_stderr",
+        ]
+        new_payload = {}
+        for key in keys_to_send:
+            if key in payload:
+                new_payload[key] = payload[key]
+        # Add the original payload to the new payload
+        new_payload["exception"] = str(exception)
+        new_payload["status"] = "failed"
+        self.publish(name, new_payload)
 
     def __get_payload(self, uuid, failed):
         """Retry reading a message from the publish_uuid_store once, delete on the second failure."""