update log format

intelligent-machine-learning · Jan 16, 2025 · b648ea4 · b648ea4
1 parent f350f4b
commit b648ea4
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 4 deletions.
diff --git a/dlrover/python/elastic_agent/diagnosis/diagnosis_agent.py b/dlrover/python/elastic_agent/diagnosis/diagnosis_agent.py
@@ -264,7 +264,7 @@ def diagnose_training_failure(self) -> NodeAction:
     def _report_failure_to_master(self, failures, restart_count):
         errors = {}
         if len(failures) == 0:
-            logger.info("there is no failures. skip failures report")
+            logger.info("Skip failure report due to empty failures")
             return
         for rank, failure in failures.items():
             dt = str(datetime.fromtimestamp(int(failure.timestamp)))
@@ -288,7 +288,7 @@ def send_heartbeat(self):
             action = self._client.report_heart_beat(ts)
             self._agent_context.enqueue_diagnosis_action(action)
         except Exception as e:
-            logger.warning(f"fail to report a heartbeat: {e}")
+            logger.warning(f"Fail to report a heartbeat: {e}")
 
     def _periodically_report(self):
         logger.info("Start diagnosis agent periodically reporter.")

diff --git a/dlrover/python/master/node/dist_job_manager.py b/dlrover/python/master/node/dist_job_manager.py
@@ -1198,9 +1198,8 @@ def handle_training_failure(
     ):
         """Process the training failure reported by the node."""
         node = self._job_context.job_node(node_type, node_id)
-        logger.info(f"handle failed node: {node}")
+        logger.info(f"Handle failed node: {node}")
         if node.is_released:
-            logger.info(f"The node {node.name} has been released.")
             return
         relaunch_node = self._process_error(
             node, restart_count, error_data, level