Program terminated with signal SIGABRT

• uvloop version: uvloop 0.16.0
• Python version: 3.9.5
• Platform: Ubuntu 5.11.0-31-generic
• Can you reproduce the bug with PYTHONASYNCIODEBUG in env?: Yes. No suspicious message other than "Executing <Task pending name='Task-60' ...> took 0.578 seconds
• Does uvloop behave differently from vanilla asyncio? How?: Yes. Python process does not crash

gdb:

Program terminated with signal SIGABRT, Aborted.

bt
#0  __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:49
49      ../sysdeps/unix/sysv/linux/raise.c: No such file or directory.
[Current thread is 1 (Thread 0x7fc465c0c640 (LWP 11019))]
(gdb) bt
#0  __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:49
#1  0x00007fc477a83864 in __GI_abort () at abort.c:79
#2  0x00007fc46c33eef5 in uv__io_poll (loop=loop@entry=0x7fc45003f9f0, timeout=<optimized out>) at src/unix/epoll.c:166
#3  0x00007fc46c43d794 in uv_run (loop=0x7fc45003f9f0, mode=mode@entry=UV_RUN_DEFAULT) at src/unix/core.c:389
#4  0x00007fc46c39bc50 in __pyx_f_6uvloop_4loop_4Loop___run (__pyx_v_self=0x7fc450000fe0, __pyx_v_mode=UV_RUN_DEFAULT) at uvloop/loop.c:14642
#5  0x00007fc46c39e1b7 in __pyx_f_6uvloop_4loop_4Loop__run (__pyx_v_self=0x7fc450000fe0, __pyx_v_mode=UV_RUN_DEFAULT) at uvloop/loop.c:15030
#6  0x00007fc46c3c5f3d in __pyx_pf_6uvloop_4loop_4Loop_24run_forever (__pyx_v_self=0x7fc450000fe0) at uvloop/loop.c:27698
#7  __pyx_pw_6uvloop_4loop_4Loop_25run_forever (__pyx_v_self=<Loop at remote 0x7fc450000fe0>, unused=<optimized out>) at uvloop/loop.c:27519
#8  0x00007fc46c373d66 in __Pyx_PyObject_CallMethO (arg=0x0, func=<built-in method run_forever of Loop object at remote 0x7fc450000fe0>)
    at uvloop/loop.c:167062
#9  __Pyx_PyObject_CallNoArg (func=<built-in method run_forever of Loop object at remote 0x7fc450000fe0>) at uvloop/loop.c:36016
#10 0x00007fc46c3dceeb in __pyx_pf_6uvloop_4loop_4Loop_44run_until_complete (__pyx_v_future=<_asyncio.Task at remote 0x7fc466423380>,
    __pyx_v_self=<optimized out>) at uvloop/loop.c:29299
#11 __pyx_pw_6uvloop_4loop_4Loop_45run_until_complete (__pyx_v_self=<optimized out>, __pyx_v_future=<optimized out>) at uvloop/loop.c:28897
#12 0x000000000053d245 in method_vectorcall_O (func=<optimized out>, args=0x7fc450000f48, nargsf=<optimized out>, kwnames=<optimized out>)
    at ../Objects/descrobject.c:462
#13 0x0000000000516543 in _PyObject_VectorcallTstate (kwnames=0x0, nargsf=<optimized out>, args=0x7fc450000f48,
    callable=<method_descriptor at remote 0x7fc476cfff90>, tstate=0x2f350b0) at ../Include/cpython/abstract.h:118
#14 PyObject_Vectorcall (kwnames=0x0, nargsf=<optimized out>, args=0x7fc450000f48, callable=<method_descriptor at remote 0x7fc476cfff90>)
    at ../Include/cpython/abstract.h:127
#15 call_function (kwnames=0x0, oparg=<optimized out>, pp_stack=<synthetic pointer>, tstate=0x2f350b0) at ../Python/ceval.c:5072
#16 _PyEval_EvalFrameDefault (tstate=<optimized out>, f=<optimized out>, throwflag=<optimized out>) at ../Python/ceval.c:3504
#17 0x0000000000515377 in _PyEval_EvalFrame (throwflag=0,
    f=Frame 0x7fc450000dc0, for file /usr/lib/python3.9/asyncio/runners.py, line 300, in run (main=<coroutine at remote 0x7fc4664264c0>, debug=None, loop=<Loop at remote 0x7fc450000fe0>), tstate=0x2f350b0) at ../Include/internal/pycore_ceval.h:40

py-bt
Traceback (most recent call first):
  File "/usr/lib/python3.9/asyncio/runners.py", line 300, in run
  File "main.py", line 70, in worker
    asyncio.run(async_worker(q_tasks, fetcher))
  File "/usr/lib/python3.9/threading.py", line 892, in run
    self._target(*self._args, **self._kwargs)
  File "/usr/lib/python3.9/threading.py", line 954, in _bootstrap_inner
    self.run()
  File "/usr/lib/python3.9/threading.py", line 912, in _bootstrap
    self._bootstrap_inner()

Start from main.py:70 in worker and reproduce the SIGABRT with uvloop 0.16.0, Python 3.9.5, Ubuntu, and PYTHONASYNCIODEBUG enabled. Inspect the gdb backtrace through uvloop/loop.c and libuv's src/unix/epoll.c, then compare the behavior with vanilla asyncio. Done means the reported workload no longer aborts and the cause or a reliable fix is documented.