From cc12d27f3cca4e19a8d7014ea5c59842aa4e9739 Mon Sep 17 00:00:00 2001
From: kuizhiqing <kuizhiqing@baidu.com>
Date: Wed, 17 Aug 2022 11:05:39 +0800
Subject: [PATCH] fix tcp LINGER (#45190)

---
 python/paddle/distributed/launch/context/node.py           | 2 ++
 python/paddle/distributed/launch/controllers/collective.py | 2 +-
 python/paddle/distributed/launch/controllers/master.py     | 2 +-
 python/paddle/distributed/launch/controllers/ps.py         | 2 +-
 4 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/python/paddle/distributed/launch/context/node.py b/python/paddle/distributed/launch/context/node.py
index 39f42d0210..6ee8fa6d10 100644
--- a/python/paddle/distributed/launch/context/node.py
+++ b/python/paddle/distributed/launch/context/node.py
@@ -49,6 +49,8 @@ class Node(object):
         for _ in range(100):
             with closing(socket.socket(socket.AF_INET,
                                        socket.SOCK_STREAM)) as s:
+                s.setsockopt(socket.SOL_SOCKET, socket.SO_LINGER,
+                             struct.pack('ii', 1, 0))
                 s.bind(('', 0))
                 port = s.getsockname()[1]
                 if port in self._allocated_ports:
diff --git a/python/paddle/distributed/launch/controllers/collective.py b/python/paddle/distributed/launch/controllers/collective.py
index e155d31b45..302070e55c 100644
--- a/python/paddle/distributed/launch/controllers/collective.py
+++ b/python/paddle/distributed/launch/controllers/collective.py
@@ -93,7 +93,7 @@ class CollectiveController(Controller):
         self.pod.replicas = self.pod_replicas()
 
         # rank will be reset when restart
-        self.pod.rank = self.ctx.args.rank
+        self.pod.rank = int(self.ctx.args.rank)
 
         port = self.ctx.node.get_free_port()
 
diff --git a/python/paddle/distributed/launch/controllers/master.py b/python/paddle/distributed/launch/controllers/master.py
index 825be9c368..c71d0890f1 100644
--- a/python/paddle/distributed/launch/controllers/master.py
+++ b/python/paddle/distributed/launch/controllers/master.py
@@ -102,7 +102,7 @@ class HTTPMaster(Master):
             print(" ".join(cmd))
             print("-" * 80)
 
-            if self.ctx.args.rank >= 0:
+            if int(self.ctx.args.rank) >= 0:
                 self.ctx.logger.warning(
                     "--rank set in the command may not compatible in auto mode")
 
diff --git a/python/paddle/distributed/launch/controllers/ps.py b/python/paddle/distributed/launch/controllers/ps.py
index 573f578d24..f785311a52 100644
--- a/python/paddle/distributed/launch/controllers/ps.py
+++ b/python/paddle/distributed/launch/controllers/ps.py
@@ -111,7 +111,7 @@ class PSController(Controller):
 
     def _build_pod_with_master(self):
 
-        self.pod.rank = self.ctx.args.rank
+        self.pod.rank = int(self.ctx.args.rank)
 
         server_num = self.ctx.args.server_num or 1
         servers = [
-- 
GitLab