remove unused logging

axlearn/cloud/common/bastion.py

@@ -714,7 +714,6 @@ def _append_to_job_history(self, job: Job, *, msg: str, state: JobLifecycleState
         with tf_io.gfile.GFile(os.path.join(self._job_history_dir, f"{job.spec.name}"), "a") as f:
             curr_time = datetime.now(timezone.utc).strftime("%m%d %H:%M:%S")
             f.write(f"{curr_time} {msg}\n")
-        logging.info("DEBUGG 1")
         # Publish event into queue.
         if self._event_publisher:
             self._event_publisher.publish(

axlearn/cloud/common/event_queue.py

@@ -187,7 +187,7 @@ def publish(self, event: Event):
         while attempt <= self._num_tries:
             try:
                 # Ensure connection is established before publishing.
-                if not self._channel:
+                if not self._channel or not self._connection:
                     logging.error("RabbitMQ publisher channel is closed, reconnecting...")
                     self.connect()
 
@@ -228,6 +228,7 @@ def publish(self, event: Event):
             except Exception as e:  # pylint: disable=broad-except
                 # Unknown errors. Don't retry. Log to avoid crashing clients.
                 logging.error("Failed to publish event: %s. Error: %s.", message, str(e))
+                return
 
     def _handle_publish_error(self):
         """Handle publish errors with retrying on connection issue."""