From ef445ec8f8c30894333506790f1c324bcb52ff77 Mon Sep 17 00:00:00 2001
From: Chitsing KUI <kuizhiqing@msn.com>
Date: Sun, 25 Jun 2023 16:30:24 +0800
Subject: [PATCH] add flag to disable gpu capture (#54762)

---
 python/paddle/distributed/launch/context/args_envs.py   | 7 +++++++
 python/paddle/distributed/launch/controllers/watcher.py | 3 +++
 2 files changed, 10 insertions(+)

diff --git a/python/paddle/distributed/launch/context/args_envs.py b/python/paddle/distributed/launch/context/args_envs.py
index 8026ef3d0fb..df81569be95 100644
--- a/python/paddle/distributed/launch/context/args_envs.py
+++ b/python/paddle/distributed/launch/context/args_envs.py
@@ -89,6 +89,13 @@ def parse_args():
         help="rank node by ip. Default False",
     )
 
+    base_group.add_argument(
+        "--enable_gpu_log",
+        type=strtobool,
+        default=True,
+        help="enable capture gpu log while running. Default True",
+    )
+
     base_group.add_argument(
         "--nnodes",
         type=str,
diff --git a/python/paddle/distributed/launch/controllers/watcher.py b/python/paddle/distributed/launch/controllers/watcher.py
index ad7bc5a84b8..25855572620 100644
--- a/python/paddle/distributed/launch/controllers/watcher.py
+++ b/python/paddle/distributed/launch/controllers/watcher.py
@@ -27,6 +27,9 @@ class Watcher:
 
         self.gpu_util = []
 
+        if not self.ctx.args.enable_gpu_log:
+            return
+
         # gpu log file
         self.gpus = self.ctx.args.devices or self.ctx.node.device.labels
         if len(self.gpus) > 0:
-- 
GitLab