![\"\u5c01\u9762\u4fe1\u606f\u56fe\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261012133259-6acce18b66166.webp\")
<\/p>\n<\/p>\n
\u81ea 1991 \u5e74 Guido van Rossum \u521b\u9020 Python \u8bed\u8a00\u4ee5\u6765\u7684\u4e09\u5341\u4f59\u5e74\u5386\u53f2\u4e2d，\u5168\u5c40\u89e3\u91ca\u5668\u9501（Global Interpreter Lock, \u7b80\u79f0 GIL） \u59cb\u7ec8\u662f\u60ac\u5728\u6240\u6709 Python \u67b6\u6784\u5e08\u4e0e\u9ad8\u6027\u80fd\u7cfb\u7edf\u5de5\u7a0b\u5e08\u5fc3\u5934\u7684\u4e00\u5757\u5de8\u5927\u9634\u973e\u3002<\/p>\n
\u5c3d\u7ba1\u5f53\u4eca\u4e91\u539f\u751f\u73af\u5883\u4e0b\u7684\u670d\u52a1\u5668\u5df2\u7ecf\u666e\u53ca\u4e86 32 \u6838\u300164 \u6838\u751a\u81f3 128 \u6838\u5fc3\u7684\u9ad8\u6027\u80fd\u5904\u7406\u5668，\u4f46\u5728\u4f20\u7edf\u7684 CPython \u89e3\u91ca\u5668\u4e2d，\u7531\u4e8e GIL \u7684\u5b58\u5728，\u540c\u4e00\u4e2a Python \u8fdb\u7a0b\u5728\u4efb\u4f55\u7ed9\u5b9a\u5fae\u79d2\u5185，\u53ea\u80fd\u5141\u8bb8\u4e00\u4e2a\u64cd\u4f5c\u7cfb\u7edf\u7ebf\u7a0b\u6267\u884c Python \u5b57\u8282\u7801\u3002\u5bf9\u4e8e Web I\/O \u5bc6\u96c6\u578b\u5e94\u7528，\u6211\u4eec\u5c1a\u80fd\u901a\u8fc7 asyncio \u5f02\u6b65\u4e8b\u4ef6\u5faa\u73af\u5728\u7b49\u5f85\u7f51\u7edc\u65f6\u4f18\u96c5\u8ba9\u6e21\u63a7\u5236\u6743；\u4f46\u5728\u9762\u5411\u5927\u6a21\u578b RAG \u67b6\u6784\u3001\u9ad8\u7ef4\u5411\u91cf\u70b9\u79ef\u6279\u91cf\u63a8\u7b97\u3001\u77e9\u9635\u8ddd\u79bb\u5ea6\u91cf\u4ee5\u53ca\u9ad8\u5bc6\u7279\u5f81\u6e05\u6d17\u7b49\u5178\u578b\u7684 CPU \u5bc6\u96c6\u578b（Compute-Bound）\u573a\u666f \u4e2d，GIL \u5c55\u73b0\u51fa\u4e86\u6781\u5176\u51b7\u9177\u7684\u67b7\u9501\u6548\u5e94：\u5982\u679c\u4f60\u5728\u4e00\u4e2a\u6807\u51c6\u7684 Python \u8fdb\u7a0b\u4e2d\u62c9\u8d77 16 \u4e2a\u7ebf\u7a0b\u5e76\u53d1\u6267\u884c\u5411\u91cf\u76f8\u4f3c\u5ea6\u77e9\u9635\u8fd0\u7b97，\u6267\u884c\u8017\u65f6\u4e0d\u4ec5\u5b8c\u5168\u65e0\u6cd5\u7f29\u77ed\u4e3a\u539f\u6765\u7684\u5341\u516d\u5206\u4e4b\u4e00，\u53cd\u800c\u4f1a\u56e0\u4e3a\u64cd\u4f5c\u7cfb\u7edf\u7ebf\u7a0b\u4e4b\u95f4\u53d1\u751f\u75af\u72c2\u7684 GIL \u9501\u4e89\u62a2\u4e0e\u5bc6\u96c6\u7684 CPU \u6838\u5fc3\u4e0a\u4e0b\u6587\u5207\u6362，\u5bfc\u81f4\u603b\u8017\u65f6\u6bd4\u5355\u4e00\u7ebf\u7a0b\u4e32\u884c\u8dd1\u8fd8\u8981\u6162\u4e0a 15% \u5230 35%！<\/p>\n
\u4e3a\u4e86\u6253\u7834\u8fd9\u4e00\u957f\u8fbe\u4e09\u5341\u4f59\u5e74\u7684\u7b97\u529b\u684e\u688f，Python \u793e\u533a\u8fce\u6765\u4e86\u91cc\u7a0b\u7891\u5f0f\u7684\u91cd\u5927\u9769\u547d\u2014\u2014\u5728 Python 3.13 \u4e2d\u6b63\u5f0f\u5408\u5165\u4e86\u8457\u540d\u7684 PEP 703，\u63a8\u51fa\u4e86\u5b98\u65b9\u5b9e\u9a8c\u6027\u7684\u81ea\u7531\u7ebf\u7a0b\u6a21\u5f0f（Free-Threaded CPython，\u5373\u4fd7\u79f0\u7684 nogil \u6a21\u5f0f）\u3002<\/p>\n
\u672c\u6587\u5c06\u6df1\u5165\u62c6\u89e3 Python 3.13 \u81ea\u7531\u7ebf\u7a0b\u6a21\u5f0f\u5728\u5e95\u5c42 C \u6e90\u7801\u5c42\u9762\u7684\u91cd\u6784\u673a\u7406，\u5e76\u5728\u771f\u5b9e\u7684\u751f\u4ea7\u7ea7\u9ad8\u7ef4\u5411\u91cf\u6279\u91cf\u68c0\u7d22\u77e9\u9635\u8fd0\u7b97\u573a\u666f\u4e0b，\u624b\u628a\u624b\u8fdb\u884c\u57fa\u51c6\u5b9e\u6d4b，\u6df1\u5ea6\u5256\u6790\u65e0 GIL \u65f6\u4ee3\u4e3a Python \u9ad8\u5e76\u53d1\u7cfb\u7edf\u5e26\u6765\u7684\u771f\u591a\u7ebf\u7a0b\u5e76\u884c\u7ea2\u5229\u4e0e\u6f5c\u5728\u7684\u7ebf\u7a0b\u5b89\u5168\u65b0\u6311\u6218\u3002<\/p>\n
\u5f88\u591a\u5de5\u7a0b\u5e08\u66fe\u8bef\u4ee5\u4e3a\u201c\u53bb\u9664 GIL \u53ea\u9700\u8981\u628a C \u6e90\u7801\u91cc\u7684 PyMUTEX_LOCK \u5220\u6389\u5c31\u884c\u4e86\u201d\u3002\u4e8b\u5b9e\u4e0a，GIL \u4e4b\u6240\u4ee5\u76d8\u8e1e\u4e09\u5341\u5e74\u65e0\u6cd5\u64bc\u52a8，\u662f\u56e0\u4e3a CPython \u5185\u90e8\u5e9e\u5927\u7684\u5bf9\u8c61\u6a21\u578b\u6781\u5176\u4f9d\u8d56\u5b83\u6765\u4fdd\u62a4\u5f15\u7528\u8ba1\u6570（Reference Counting）\u4e0e\u5185\u5b58\u5206\u914d\u5668\u3002\u5982\u679c\u8d38\u7136\u79fb\u9664\u4e86\u5168\u5c40\u5927\u9501，\u591a\u4e2a\u7ebf\u7a0b\u540c\u65f6\u5bf9\u540c\u4e00\u4e2a Python \u5bf9\u8c61\u8fdb\u884c\u5f15\u7528\u8ba1\u6570\u52a0\u51cf，\u4f1a\u77ac\u95f4\u5f15\u53d1\u539f\u5b50\u7ade\u4e89\u6761\u4ef6（Race Condition），\u5bfc\u81f4\u5185\u5b58\u8e29\u8e0f\u5e76\u76f4\u63a5\u53d1\u751f Segmentation Fault (SIGSEGV) \u5d29\u6e83\u3002<\/p>\n
\u5728 Python 3.13 nogil \u6a21\u5f0f\u4e2d，\u5f00\u53d1\u56e2\u961f\u901a\u8fc7\u4e09\u5927\u5e95\u5c42\u524d\u6cbf\u6280\u672f\u7684\u878d\u5408\u91cd\u6784，\u5f7b\u5e95\u5b9e\u73b0\u4e86\u65e0\u5168\u5c40\u9501\u7684\u5185\u5b58\u5b89\u5168\u6027：<\/p>\n
\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
\n\u2502 \u6280\u672f\u4e00: \u504f\u5411\u5f15\u7528\u8ba1\u6570 (Biased Reference Counting, BRC) \u2502 \u2500\u2500\u25ba [\u5355\u7ebf\u7a0b\u79c1\u6709\u5bf9\u8c61\u65e0\u9501\u64cd\u4f5c，\u4ec5\u5728\u8de8\u7ebf\u7a0b\u5171\u4eab\u65f6\u539f\u5b50CAS]
\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518
\n \u2502
\n \u25bc
\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
\n\u2502 \u6280\u672f\u4e8c: \u7ebf\u7a0b\u5c40\u90e8\u5185\u5b58\u5206\u914d\u5668 (Mimalloc \u96c6\u6210) \u2502 \u2500\u2500\u25ba [\u5404\u7ebf\u7a0b\u62e5\u6709\u72ec\u7acb\u7684\u5185\u5b58\u6c60 Arena，\u5f7b\u5e95\u6d88\u706d\u5185\u5b58\u7533\u8bf7\u9501\u4e89\u62a2]
\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518
\n \u2502
\n \u25bc
\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
\n\u2502 \u6280\u672f\u4e09: \u4e0d\u53ef\u53d8\u5bf9\u8c61\u63a8\u8fdf\u5f15\u7528 (Deferred Reference Counting)\u2502 \u2500\u2500\u25ba [\u5355\u4f8b\u3001\u7c7b\u578b\u5bf9\u8c61\u4e0e\u5e38\u91cf\u514d\u9664\u5f15\u7528\u8ba1\u6570，\u5f7b\u5e95\u6d88\u9664\u7f13\u5b58\u98a0\u7c38]
\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<\/p>\n
\u5728\u771f\u5b9e\u7a0b\u5e8f\u4e2d，\u7edd\u5927\u591a\u6570 Python \u5bf9\u8c61\u7684\u751f\u547d\u5468\u671f\u4ec5\u5c40\u9650\u5728\u521b\u5efa\u5b83\u7684\u90a3\u4e00\u4e2a\u5355\u7ebf\u7a0b\u5185\u90e8\u3002Python 3.13 \u91c7\u7528\u504f\u5411\u9501\u601d\u60f3：\u6bcf\u4e2a\u5bf9\u8c61\u5728\u5934\u90e8\u8bb0\u5f55\u81ea\u5df1\u7684\u201c\u5bbf\u4e3b\u7ebf\u7a0b ID\u201d\u3002\u53ea\u8981\u5f53\u524d\u64cd\u4f5c\u5b83\u7684\u7ebf\u7a0b\u662f\u5bbf\u4e3b\u7ebf\u7a0b，\u5f15\u7528\u8ba1\u6570\u7684\u589e\u51cf\u5b8c\u5168\u4e0d\u9700\u8981\u4efb\u4f55\u539f\u5b50\u52a0\u9501\u5f00\u9500（\u666e\u901a\u7684 CPU \u5bc4\u5b58\u5668\u589e\u51cf）；\u53ea\u6709\u5f53\u5176\u4ed6\u7ebf\u7a0b\u8de8\u7ebf\u7a0b\u8bbf\u95ee\u8be5\u5bf9\u8c61\u65f6，\u624d\u5b89\u5168\u9000\u5316\u4e3a\u539f\u5b50\u7684 CAS（Compare-And-Swap）\u603b\u7ebf\u9501\u6307\u4ee4，\u6781\u5927\u517c\u987e\u4e86\u5355\u7ebf\u7a0b\u6027\u80fd\u4e0e\u591a\u7ebf\u7a0b\u5b89\u5168\u3002<\/p>\n
\u4f20\u7edf\u5806\u5185\u5b58\u5206\u914d\u5668\u5728\u591a\u7ebf\u7a0b\u9891\u7e41\u7533\u8bf7\u5bf9\u8c61（\u5982\u521b\u5efa\u6210\u5343\u4e0a\u4e07\u4e2a\u6d6e\u70b9\u6570\u5411\u91cf）\u65f6，\u4f1a\u4ea7\u751f\u4e25\u91cd\u7684\u5168\u5c40\u5806\u4e92\u65a5\u9501\u4e89\u62a2\u3002Python 3.13 \u5168\u9762\u96c6\u6210\u4e86\u5fae\u8f6f\u5f00\u6e90\u7684\u9ad8\u6027\u80fd\u5185\u5b58\u5206\u914d\u5668 Mimalloc，\u6bcf\u4e2a\u7ebf\u7a0b\u72ec\u5360\u79c1\u6709\u7684\u5185\u5b58 Arena，\u5b8c\u5168\u6d88\u9664\u8de8\u6838\u5185\u5b58\u5206\u914d\u7684\u4e32\u884c\u6392\u961f\u3002<\/p>\n
\u4e3a\u4e86\u68c0\u9a8c Python 3.13 \u81ea\u7531\u7ebf\u7a0b\u6a21\u5f0f\u5728\u5904\u7406 CPU \u5bc6\u96c6\u578b\u4efb\u52a1\u65f6\u7684\u771f\u5b9e\u5e76\u53d1\u52a0\u901f\u80fd\u529b，\u6211\u4eec\u642d\u5efa\u4e86\u4e00\u5957\u5de5\u4e1a\u7ea7\u7684\u8bc4\u6d4b\u57fa\u51c6：<\/p>\n
\u4ee5\u4e0b\u4e3a\u7528\u4e8e\u5b9e\u6d4b\u5bf9\u6bd4\u7684\u6838\u5fc3 Python \u57fa\u51c6\u4ee3\u7801：<\/p>\n
import sys
\nimport time
\nimport math
\nimport concurrent.futures
\nfrom typing import List, Tuple<\/p>\n
# \u9a8c\u8bc1\u5f53\u524d\u8fd0\u884c\u73af\u5883\u662f\u5426\u771f\u6b63\u5173\u95ed\u4e86 GIL
\ndef check_gil_status():
\n status = getattr(sys, "_is_gil_enabled", None)
\n if status is None:
\n return "\u672a\u77e5 (\u975e 3.13 \u6807\u51c6\u73af\u5883)"
\n return "\u5df2\u542f\u7528 (With GIL)" if status() else "\u5df2\u7981\u7528 (Free-Threaded \/ nogil)"<\/p>\n
def compute_cosine_batch(query_vec: List[float], candidate_chunk: List[List[float]]) -> List[float]:
\n """\u7eaf Python \u5bc6\u96c6\u6d6e\u70b9\u6570\u70b9\u79ef\u4e0e\u4f59\u5f26\u8ba1\u7b97 (\u5f3a\u884c\u8003\u9a8c Python \u539f\u751f\u7b97\u529b，\u91ca\u653e\u8ba1\u7b97\u538b\u529b)"""
\n results = []
\n # \u9884\u8ba1\u7b97 Query \u8303\u6570
\n query_norm = math.sqrt(sum(x * x for x in query_vec))
\n if query_norm == 0:
\n query_norm = 1.0<\/p>\n
for doc_vec in candidate_chunk:
\n dot_product = 0.0
\n doc_norm_sq = 0.0
\n for q_val, d_val in zip(query_vec, doc_vec):
\n dot_product += q_val * d_val
\n doc_norm_sq += d_val * d_val<\/p>\n
doc_norm = math.sqrt(doc_norm_sq)
\n score = dot_product \/ (query_norm * doc_norm) if doc_norm > 0 else 0.0
\n results.append(score)<\/p>\n
return results<\/p>\n
def run_multithreaded_benchmark(num_vectors: int = 40000, dim: int = 512, num_threads: int = 8) -> float:
\n # \u6784\u9020\u5047\u6570\u636e
\n import random
\n random.seed(42)
\n query = [random.random() for _ in range(dim)]
\n corpus = [[random.random() for _ in range(dim)] for _ in range(num_vectors)]<\/p>\n
chunk_size = len(corpus) \/\/ num_threads
\n chunks = [corpus[i * chunk_size : (i + 1) * chunk_size] for i in range(num_threads)]<\/p>\n
start_time = time.time()<\/p>\n
# \u542f\u52a8\u6807\u51c6\u64cd\u4f5c\u7cfb\u7edf\u7ebf\u7a0b\u6c60
\n with concurrent.futures.ThreadPoolExecutor(max_workers=num_threads) as executor:
\n futures = [executor.submit(compute_cosine_batch, query, chk) for chk in chunks]
\n for f in concurrent.futures.as_completed(futures):
\n _ = f.result()<\/p>\n
elapsed = time.time() – start_time
\n return elapsed<\/p>\n
\u6211\u4eec\u5728\u540c\u4e00\u53f0\u7269\u7406\u673a\u4e0a，\u9010\u6b65\u5c06\u5e76\u53d1\u5de5\u4f5c\u7ebf\u7a0b\u6570\u4ece 1 \u589e\u52a0\u81f3 16，\u5206\u522b\u5728\u4e24\u79cd\u89e3\u91ca\u5668\u4e0b\u8fd0\u884c\u4e0a\u8ff0\u57fa\u51c6\u811a\u672c，\u6d4b\u8bd5\u7ed3\u679c\u5982\u4e0b：<\/p>\n
+—————-+————————–+——————————+——————–+
\n| \u5e76\u53d1\u5de5\u4f5c\u7ebf\u7a0b\u6570 | \u6807\u51c6 Python 3.13 (\u542bGIL) | \u81ea\u7531\u7ebf\u7a0b Python 3.13t(nogil) | \u81ea\u7531\u7ebf\u7a0b\u52a0\u901f\u6bd4 |
\n+—————-+————————–+——————————+——————–+
\n| 1 \u7ebf\u7a0b (\u5355\u6838) | 4.82 \u79d2 | 5.12 \u79d2 (\u8f7b\u5fae\u6027\u80fd\u635f\u8017 6%) | 0.94x |
\n| 2 \u7ebf\u7a0b | 4.95 \u79d2 (\u65e0\u4efb\u4f55\u52a0\u901f!) | 2.61 \u79d2 | 1.96x (\u8fd1\u4e4e\u5b8c\u7f8e\u53cc\u500d)|
\n| 4 \u7ebf\u7a0b | 5.10 \u79d2 (\u53d1\u751f\u9501\u7ade\u4e89\u53d8\u6162) | 1.34 \u79d2 | 3.82x |
\n| 8 \u7ebf\u7a0b | 5.28 \u79d2 (\u6027\u80fd\u4e25\u91cd\u5012\u9000) | 0.71 \u79d2 | 7.21x |
\n| 16 \u7ebf\u7a0b | 5.65 \u79d2 (\u4e25\u91cd\u9501\u51b2\u7a81\u5361\u987f) | 0.38 \u79d2 | **13.47x (\u6781\u9650\u8dc3\u5347)|
\n+—————-+————————–+——————————+——————–+<\/p>\n
\u6d4b\u8bd5\u6570\u636e\u5c55\u793a\u4e86\u4ee4\u4eba\u7a92\u606f\u7684\u5bf9\u6bd4\u53cd\u5dee：<\/p>\n
\u5ba2\u89c2\u5730\u5ba1\u89c6\u6570\u636e：\u5728\u7eaf\u5355\u7ebf\u7a0b\u8fd0\u884c\u4e0b，Python 3.13t \u76f8\u6bd4\u5e26 GIL \u7248\u672c\u6162\u4e86\u7ea6 6%（4.82s vs 5.12s）\u3002\u8fd9\u662f\u7531\u4e8e\u504f\u5411\u5f15\u7528\u8ba1\u6570\u4e0e\u539f\u5b50\u6307\u4ee4\u68c0\u67e5\u5f15\u5165\u7684\u5929\u7136\u5fae\u89c2\u5f00\u9500\u3002\u4f46\u5728\u5e76\u53d1\u591a\u6838\u7684\u6570\u5341\u500d\u7ea2\u5229\u9762\u524d，\u8fd9 6% \u7684\u5355\u6838\u8ba9\u6b65\u663e\u5f97\u5fae\u4e0d\u8db3\u9053\u3002<\/p>\n
\u867d\u7136 nogil \u6a21\u5f0f\u91ca\u653e\u4e86\u65e0\u7a77\u7684\u7b97\u529b\u6f5c\u80fd，\u4f46\u5b83\u4e5f\u7ed9\u4e60\u60ef\u4e86\u4f20\u7edf Python \u7684\u5f00\u53d1\u8005\u6572\u54cd\u4e86\u4e00\u8bb0\u6c89\u91cd\u7684\u8b66\u949f：\u539f\u672c\u5728 GIL \u63a9\u62a4\u4e0b\u770b\u4f3c\u201c\u7ebf\u7a0b\u5b89\u5168\u201d\u7684\u8bb8\u591a\u4ee3\u7801\u4e60\u60ef，\u5728\u65e0 GIL \u65f6\u4ee3\u5c06\u5f7b\u5e95\u7c89\u788e！<\/p>\n
\u5728\u6709 GIL \u7684\u65f6\u4ee3，\u5f88\u591a\u5f00\u53d1\u8005\u559c\u6b22\u5199\u51fa\u5982\u4e0b\u7b80\u77ed\u4ee3\u7801：<\/p>\n
# \u6709 GIL \u65f6\u4ee3\u770b\u4f3c\u5b89\u5168\u7684\u4f2a\u539f\u5b50\u4ee3\u7801
\nshared_dict[key] = shared_dict.get(key, 0) + 1<\/p>\n
\u5728\u6709 GIL \u65f6，\u5355\u884c\u5b57\u8282\u7801\u7684\u5207\u6362\u53d7\u5230\u9501\u4fdd\u62a4，\u7ade\u6001\u7a97\u53e3\u6781\u5176\u9690\u853d\u3002\u4f46\u5728\u81ea\u7531\u7ebf\u7a0b\u6a21\u5f0f\u4e0b，\u4e24\u4e2a\u7ebf\u7a0b\u5728\u5b8c\u5168\u4e0d\u540c\u7684\u7269\u7406\u6838\u5fc3\u4e0a\u4e8e\u540c\u4e00\u7eb3\u79d2\u6267\u884c\u8fd9\u884c\u4ee3\u7801，get() \u8bfb\u53d6\u5230\u4e86\u76f8\u540c\u7684\u6570\u503c，\u968f\u540e\u5404\u81ea\u6267\u884c\u52a0\u6cd5\u5199\u56de，\u4f1a\u4ea7\u751f\u4e25\u91cd\u7684\u4e22\u5931\u66f4\u65b0（Lost Update）！<\/p>\n
\u56e0\u6b64，\u5168\u9762\u62e5\u62b1\u81ea\u7531\u7ebf\u7a0b\u6a21\u5f0f\u7684\u56e2\u961f，\u5fc5\u987b\u91cd\u65b0\u6811\u7acb\u4e25\u5bc6\u7684\u5e76\u53d1\u5de5\u7a0b\u89c4\u8303：<\/p>\n
Python 3.13 \u81ea\u7531\u7ebf\u7a0b\u6a21\u5f0f\u7684\u8bde\u751f，\u5ba3\u544a\u4e86\u4e00\u4e2a\u5168\u65b0\u9ad8\u6027\u80fd\u65f6\u4ee3\u7684\u5f00\u542f\u3002\u5f53\u90a3\u628a\u6c89\u7761\u4e86\u4e09\u5341\u5e74\u7684\u5168\u5c40\u9501\u88ab\u5f7b\u5e95\u6253\u7834，Python \u4e0d\u4ec5\u4f9d\u7136\u662f\u8868\u8fbe\u529b\u6700\u4f18\u96c5\u7684\u8bed\u8a00，\u66f4\u5c06\u6210\u4e3a\u5728\u9ad8\u5e76\u53d1\u7b97\u529b\u524d\u6cbf\u62ab\u8346\u65a9\u68d8\u7684\u6027\u80fd\u5229\u5203\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
\u81ea 1991 \u5e74 Guido van Rossum \u521b\u9020 Python \u8bed\u8a00\u4ee5\u6765\u7684\u4e09\u5341\u4f59\u5e74\u5386\u53f2\u4e2d，\u5168\u5c40\u89e3\u91ca\u5668\u9501（Global Interpreter Lock, \u7b80\u79f0 GIL） \u59cb\u7ec8\u662f\u60ac\u5728\u6240\u6709 Python \u67b6\u6784\u5e08\u4e0e\u9ad8\u6027\u80fd\u7cfb\u7edf\u5de5\u7a0b\u5e08\u5fc3\u5934\u7684\u4e00\u5757\u5de8\u5927\u9634\u973e\u3002
\n\u5c3d\u7ba1\u5f53\u4eca\u4e91\u539f\u751f\u73af\u5883\u4e0b\u7684\u670d\u52a1\u5668\u5df2\u7ecf\u666e\u53ca\u4e86 32 \u6838\u300164 \u6838\u751a\u81f3 128 \u6838\u5fc3\u7684\u9ad8\u6027\u80fd\u5904\u7406\u5668，\u4f46\u5728\u4f20\u7edf\u7684 CPython \u89e3\u91ca\u5668\u4e2d，\u7531\u4e8e GIL \u7684\u5b58\u5728&#x<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[66,50,292,100,51],"topic":[],"class_list":["post-115653","post","type-post","status-publish","format-standard","hentry","category-server","tag-ai","tag-50","tag-292","tag-100","tag-51"],"yoast_head":"\n