Merge pull request #2155 from FedML-AI/alexleung/dev_v070_for_refactor

[CoreEngine] 1. fixed the issue that the fork method is not support i…

Author: fedml-alex

python/fedml/__init__.py

@@ -1,4 +1,5 @@
 import logging
+import platform
 
 import multiprocess as multiprocessing
 import os
@@ -92,13 +93,7 @@ def init(args=None, check_env=True, should_init_logs=True):
     # Windows/Linux/MacOS compatability issues on multi-processing
     # https://github.com/pytorch/pytorch/issues/3492
     """
-    if multiprocessing.get_start_method() != "fork":
-        # force all platforms (Windows/Linux/macOS) to use the same way (fork) for multiprocessing
-        multiprocessing.set_start_method("fork", force=True)
-
-    # if multiprocessing.get_start_method() != "spawn":
-    #     # force all platforms (Windows/Linux/MacOS) to use the same way (spawn) for multiprocessing
-    #     multiprocessing.set_start_method("spawn", force=True)
+    _init_multiprocessing()
 
     """
     # https://stackoverflow.com/questions/53014306/error-15-initializing-libiomp5-dylib-but-found-libiomp5-dylib-already-initial
@@ -450,6 +445,21 @@ def _run_distributed():
     pass
 
 
+def _init_multiprocessing():
+    """
+    # Windows/Linux/MacOS compatability issues on multi-processing
+    # https://github.com/pytorch/pytorch/issues/3492
+    """
+    if platform.system() == "Windows":
+        if multiprocessing.get_start_method() != "spawn":
+            # force all platforms (Windows/Linux/macOS) to use the same way (spawn) for multiprocessing
+            multiprocessing.set_start_method("spawn", force=True)
+    else:
+        if multiprocessing.get_start_method() != "fork":
+            # force all platforms (Windows/Linux/macOS) to use the same way (fork) for multiprocessing
+            multiprocessing.set_start_method("fork", force=True)
+
+
 def set_env_version(version):
     set_env_kv("FEDML_ENV_VERSION", version)
     load_env()

python/fedml/computing/scheduler/comm_utils/hardware_utils.py

@@ -27,7 +27,7 @@ def __get_util(cls) -> Optional[GPUCardUtil]:
             except Exception as e:
                 pass
 
-        logging.error("No GPU card detected")
+        # logging.error("No GPU card detected")
         return None
 
     @staticmethod

python/fedml/computing/scheduler/model_scheduler/device_model_db.py

@@ -1,6 +1,7 @@
 import json
 import logging
 import os
+import platform
 import time
 
 from fedml.computing.scheduler.model_scheduler.device_server_constants import ServerConstants
@@ -261,7 +262,10 @@ def open_job_db(self):
             self.db_base_dir = ServerConstants.get_database_dir()
 
         job_db_path = os.path.join(self.db_base_dir, FedMLModelDatabase.MODEL_DEPLOYMENT_DB)
-        self.db_engine = create_engine('sqlite:////{}'.format(job_db_path), echo=False)
+        if platform.system() == "Windows":
+            self.db_engine = create_engine('sqlite:///{}'.format(job_db_path), echo=False)
+        else:
+            self.db_engine = create_engine('sqlite:////{}'.format(job_db_path), echo=False)
 
         db_session_class = sessionmaker(bind=self.db_engine)
         self.db_connection = db_session_class()

python/fedml/computing/scheduler/scheduler_core/base_db.py

@@ -1,5 +1,6 @@
 import json
 import os
+import platform
 import time
 
 from sqlalchemy import Column, String, TEXT, Integer, Float, create_engine, and_
@@ -25,7 +26,10 @@ def open_job_db(self):
         if self.db_connection is not None:
             return
 
-        self.db_engine = create_engine('sqlite:////{}'.format(self.db_path), echo=False)
+        if platform.system() == "Windows":
+            self.db_engine = create_engine('sqlite:///{}'.format(self.db_path), echo=False)
+        else:
+            self.db_engine = create_engine('sqlite:////{}'.format(self.db_path), echo=False)
 
         db_session_class = sessionmaker(bind=self.db_engine)
         self.db_connection = db_session_class()

python/fedml/computing/scheduler/scheduler_core/compute_gpu_db.py

@@ -8,6 +8,7 @@
 from fedml.core.common.singleton import Singleton
 from .base_db import FedMLBaseDb
 from .compute_utils import ComputeUtils
+from ..master.server_constants import ServerConstants
 
 Base = declarative_base()
 

python/fedml/computing/scheduler/scheduler_core/scheduler_base_protocol_manager.py

@@ -46,9 +46,7 @@ def __init__(self, args, agent_config=None, is_master=False):
         self.status_reporter = None
         self.user_name = args.user_name
 
-        if multiprocessing.get_start_method() != "fork":
-            # force all platforms (Windows/Linux/macOS) to use the same way (fork) for multiprocessing
-            multiprocessing.set_start_method("fork", force=True)
+        fedml._init_multiprocessing()
 
     def generate_topics(self):
         # generate the subscribed topics.