René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Poor thread scaling when constructing instances or accessing attributes","articleBody":"# Bug report\n\n### Remaining scaling bugs\n\n- [x] dataclass\n- [x] namedtuple\n- [ ] enum\n\n### Bug description:\n\nWhen constructing dataclass or NamedTuple instances on multiple threads (on a free threading build), or accessing enum class attributes, performance doesn't scale when using multiple threads.\n\nRegular class example (scales well):\n```python\n# b_regular_class.py\nfrom threading import Thread\nfrom time import time\nimport sys\n\nclass Foo:\n    def __init__(self, x):\n        self.x = x\n\nniter = 5 * 1000 * 1000\n\ndef benchmark(n):\n    for i in range(n):\n        Foo(x=1)\n\nfor nth in (1, 4):\n    t0 = time()\n    threads = [Thread(target=benchmark, args=(niter,)) for _ in range(nth)]\n    for t in threads:\n        t.start()\n    for t in threads:\n        t.join()\n    print(f\"{nth=} {(time() - t0) / nth}\")\n```\n\nDataclass example (doesn't scale well):\n```py\n# b_dataclass.py\nfrom threading import Thread\nfrom dataclasses import dataclass\nfrom time import time\nimport sys\n\n@dataclass\nclass Foo:\n    x: int\n\nniter = 5 * 1000 * 1000\n\ndef benchmark(n):\n    for i in range(n):\n        Foo(x=1)\n\nfor nth in (1, 4):\n    t0 = time()\n    threads = [Thread(target=benchmark, args=(niter,)) for _ in range(nth)]\n    for t in threads:\n        t.start()\n    for t in threads:\n        t.join()\n    print(f\"{nth=} {(time() - t0) / nth}\")\n```\n\nNamed tuple example (doesn't scale well):\n```py\n# b_namedtuple.py\nfrom threading import Thread\nfrom typing import NamedTuple\nfrom time import time\nimport sys\n\nclass Foo(NamedTuple):\n    x: int\n\nniter = 5 * 1000 * 1000\n\ndef benchmark(n):\n    for i in range(n):\n        Foo(x=1)\n\nfor nth in (1, 4):\n    t0 = time()\n    threads = [Thread(target=benchmark, args=(niter,)) for _ in range(nth)]\n    for t in threads:\n        t.start()\n    for t in threads:\n        t.join()\n    print(f\"{nth=} {(time() - t0) / nth}\")\n```\n\nEnum example (doesn't scale well):\n```py\n# b_enum.py\nfrom threading import Thread\nfrom time import time\nfrom enum import Enum\nimport sys\n\nclass Foo(Enum):\n    X = 1\n    Y = 2\n\nniter = 5 * 1000 * 1000\n\ndef benchmark(n):\n    for i in range(n):\n        Foo.X\n        Foo.Y.value\n\nfor nth in (1, 4):\n    t0 = time()\n    threads = [Thread(target=benchmark, args=(niter,)) for _ in range(nth)]\n    for t in threads:\n        t.start()\n    for t in threads:\n        t.join()\n    print(f\"{nth=} {(time() - t0) / nth}\")\n```\nResults on recent main branch (running on an EC2 instance):\n```\n(cpython-dev) jukka@jukka-coder-dbx free-threading-benchmarks $ py b_regular_class.py\nnth=1 1.1085155010223389\nnth=4 0.2796591520309448\n(cpython-dev) jukka@jukka-coder-dbx free-threading-benchmarks $ py b_dataclass.py\nnth=1 1.1910037994384766\nnth=4 1.0931583642959595\n(cpython-dev) jukka@jukka-coder-dbx free-threading-benchmarks $ py b_namedtuple.py\nnth=1 1.5688557624816895\nnth=4 2.0257126092910767\n(cpython-dev) jukka@jukka-coder-dbx free-threading-benchmarks $ py b_enum.py\nnth=1 0.9439797401428223\nnth=4 2.272495985031128\n```\nThe expected behavior is that when using 4 threads (`nth=4`), the elapsed time per benchmark iteration (the second printed value) goes down significantly compared to when using a single thread (`nth=1`), which happens with the first benchmark (`b_regular_class.py`) but not the others.\n\ncc @colesbury (we discussed this at CPython Core Dev Sprint in person)\n\n### CPython versions tested on:\n\nCPython main branch\n\n### Operating systems tested on:\n\nLinux\n\n\u003c!-- gh-linked-prs --\u003e\n### Linked PRs\n* gh-141596\n* gh-141603\n* gh-141750\n* gh-144332\n* gh-144406\n* gh-144407\n* gh-144977\n\u003c!-- /gh-linked-prs --\u003e\n","author":{"url":"https://github.com/JukkaL","@type":"Person","name":"JukkaL"},"datePublished":"2025-09-18T09:29:23.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":12},"url":"https://github.com/139103/cpython/issues/139103"}