[Fix]: implement thread-safety singleton to avoid deadlock for very large-scale training scenarios (#5625)

* implement thread-safety singleton

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* refactor singleton implementation

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

colossalai/context/singleton_meta.py

@@ -1,22 +1,27 @@
+import threading
+
+
 class SingletonMeta(type):
     """
-    The Singleton class can be implemented in different ways in Python. Some
+    Thread-safe Singleton Meta with double-checked locking.
-    possible methods include: base class, decorator, metaclass. We will use the
+    Reference: https://en.wikipedia.org/wiki/Double-checked_locking
-    metaclass because it is best suited for this purpose.
     """
 
     _instances = {}
+    _lock = threading.Lock()
 
     def __call__(cls, *args, **kwargs):
-        """
+        # First check (without locking) for performance reasons
-        Possible changes to the value of the `__init__` argument do not affect
+        if cls not in cls._instances:
-        the returned instance.
+            # Acquire a lock before proceeding to the second check
-        """
+            with cls._lock:
+                # Second check with lock held to ensure thread safety
                 if cls not in cls._instances:
                     instance = super().__call__(*args, **kwargs)
                     cls._instances[cls] = instance
         else:
             assert (
                 len(args) == 0 and len(kwargs) == 0
-            ), f"{cls.__name__} is a singleton class and a instance has been created."
+            ), f"{cls.__name__} is a singleton class and an instance has been created."
+
         return cls._instances[cls]