Bug report
Bug description:
gc.get_stats() (gc_get_stats_impl) plain-copies the per-generation stats structs
gcstate->generation_stats->{young,old}.items[index] without any lock or atomics,
| stats[0] = gcstate->generation_stats->young.items[gcstate->generation_stats->young.index]; | |
| stats[1] = gcstate->generation_stats->old[0].items[gcstate->generation_stats->old[0].index]; | |
| stats[2] = gcstate->generation_stats->old[1].items[gcstate->generation_stats->old[1].index]; |
while a collection on another thread writes those same items[index] fields (collections++, collected += m, ...) at the end of gc_collect_main,
| /* Update stats */ | |
| struct gc_generation_stats *stats = get_stats(gcstate, generation); | |
| stats->ts_start = start; | |
| stats->ts_stop = stop; | |
| stats->collections++; | |
| stats->collected += m; | |
| stats->uncollectable += n; | |
| stats->duration += duration; | |
| stats->candidates += state.candidates; |
so gc.get_stats() races the collector.
Reproducer:
import gc from threading import Thread def collector(): for _ in range(50000): a = {}; a['self'] = a gc.collect() def reader(): for _ in range(50000): gc.get_stats() threads = [Thread(target=collector) for _ in range(1)] threads += [Thread(target=reader) for _ in range(8)] for t in threads: t.start() for t in threads: t.join()
TSAN Report:
==================
WARNING: ThreadSanitizer: data race (pid=2301505)
Read of size 8 at 0x7fffb41a0180 by thread T2:
#0 gc_get_stats_impl /cpython/Modules/gcmodule.c:379:16
#1 gc_get_stats /cpython/Modules/clinic/gcmodule.c.h:456:12
#2 cfunction_vectorcall_NOARGS /cpython/Objects/methodobject.c:508:24
#3 _PyObject_VectorcallTstate /cpython/./Include/internal/pycore_call.h:144:11
#4 PyObject_Vectorcall /cpython/Objects/call.c:327:12
#5 _Py_VectorCallInstrumentation_StackRefSteal /cpython/Python/ceval.c:768:11
#6 _PyEval_EvalFrameDefault /cpython/Python/generated_cases.c.h:1846:35
...
Previous write of size 8 at 0x7fffb41a0180 by thread T1:
#0 gc_collect_main /cpython/Python/gc_free_threading.c:2288:23
#1 _PyGC_Collect /cpython/Python/gc_free_threading.c:2568:12
#2 gc_collect_impl /cpython/Modules/gcmodule.c:93:12
#3 gc_collect /cpython/Modules/clinic/gcmodule.c.h:143:21
#4 _Py_BuiltinCallFastWithKeywords_StackRef /cpython/Python/ceval.c:841:11
#5 _PyEval_EvalFrameDefault /cpython/Python/generated_cases.c.h:2508:35
...
SUMMARY: ThreadSanitizer: data race /cpython/Modules/gcmodule.c:379:16 in gc_get_stats_impl
==================
CPython versions tested on:
CPython main branch
Operating systems tested on:
Linux