Revert "improve lanuch"

RickyLoveFreedom · Jan 5, 2023 · 640d03e · 640d03e
1 parent 177f13d
commit 640d03e
Showing 1 changed file with 4 additions and 26 deletions.
diff --git a/flagai/launch.py b/flagai/launch.py
@@ -17,7 +17,6 @@
 import socket
 from flagai.logger import log_dist
 
-import signal
 
 def fetch_hostfile(hostfile_path):
     if not os.path.isfile(hostfile_path):
@@ -129,12 +128,7 @@ def launch_dist(launcher='distributed_deepspeed',
             cmd_launch.append('"')
             run_cmd = ' '.join(cmd_launch)
             log_dist(run_cmd)
-            p = subprocess.Popen(run_cmd, shell=True, preexec_fn=os.setsid)
-            def signal_handler(signal, frame):
-                os.killpg(os.getpgid(p.pid), 9)
-            signal.signal(signal.SIGINT, signal_handler)
-            p.wait()
-            # subprocess.Popen(run_cmd, shell=True)
+            subprocess.Popen(run_cmd, shell=True)
             node_rank += 1
 
     elif num_nodes == 1 and launcher == 'distributed_torch':
@@ -164,13 +158,7 @@ def signal_handler(signal, frame):
         cmd_launch.append('--not_call_launch')
         run_cmd = ' '.join(cmd_launch)
         log_dist(run_cmd)
-        # subprocess.Popen(run_cmd, shell=True)
-
-        p = subprocess.Popen(run_cmd, shell=True, preexec_fn=os.setsid)
-        def signal_handler(signal, frame):
-            os.killpg(os.getpgid(p.pid), 9)
-        signal.signal(signal.SIGINT, signal_handler)
-        p.wait()
+        subprocess.Popen(run_cmd, shell=True)
 
     elif launcher == 'distributed_deepspeed':
         if hostfile is None:
@@ -218,12 +206,7 @@ def signal_handler(signal, frame):
         cmd_launch.append('--not_call_launch')
         run_cmd = ' '.join(cmd_launch)
         log_dist(run_cmd)
-        # subprocess.Popen(run_cmd, shell=True)
-        p = subprocess.Popen(run_cmd, shell=True, preexec_fn=os.setsid)
-        def signal_handler(signal, frame):
-            os.killpg(os.getpgid(p.pid), 9)
-        signal.signal(signal.SIGINT, signal_handler)
-        p.wait()
+        subprocess.Popen(run_cmd, shell=True)
 
     elif num_nodes == 1 and launcher == 'simple_torch':
         # This launcher
@@ -255,11 +238,6 @@ def signal_handler(signal, frame):
 
             run_cmd = ' '.join(cmd_launch)
             log_dist(run_cmd)
-            # subprocess.Popen(run_cmd, shell=True)
-            p = subprocess.Popen(run_cmd, shell=True, preexec_fn=os.setsid)
-            def signal_handler(signal, frame):
-                os.killpg(os.getpgid(p.pid), 9)
-            signal.signal(signal.SIGINT, signal_handler)
-            p.wait()
+            subprocess.Popen(run_cmd, shell=True)
     else:
         raise Exception('No aviable launcher')