Question

我试图找出为什么我的子流程执行有时会挂起。我运行以下代码来执行具有权限丢弃的命令：

def safe_exec(q, uid, gid):
    try:
        os.setgroups([])
        os.setregid(gid, gid)
        os.setreuid(uid, uid)

        print("dropped")
        res = subprocess.check_output(['nm', '-D', '/lib/x86_64-linux-gnu/libc.so.6'])
        print("executed")
        q.put(res)
    except Exception as e:
        q.put(e)

if __name__ == "__main__":
    nobody = pwd.getpwnam('nobody')
    q = multiprocessing.Queue()
    p = multiprocessing.Process(
            target=safe_exec,
            args=(q, nobody.pw_uid, nobody.pw_gid))

    p.start()
    p.join(10)
    res = q.get(False)
    if isinstance(res, Exception):
        raise res
    else:
        print(res)

所有命令都不会发生挂起，但我可以使用nm -D .../libc.so.6在我的计算机上可靠地重现它。问题是挂起发生在safe_exec的末尾 - 我可以看到＆＃34;掉线＆＃34;并且＆＃34;执行＆＃34;打印出来的时候，这个过程仍然悬而未决。

Strace显示以下内容（缩写）。孩子按预期开始和进入：

48858 clone(child_stack=0, flags=CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIGCHLD, child_tidptr=0x7f9aa3eab9d0) = 48859
48859 setgroups(0, [])                  = 0
48859 setregid(65534, 65534)            = 0
48859 setreuid(65534, 65534)            = 0
48859 fstat(1, {st_mode=S_IFCHR|0600, st_rdev=makedev(136, 6), ...}) = 0
48859 mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f9aa3ec2000
48859 write(1, "dropped\n", 8)          = 8

父母现在开始尝试加入：

48858 wait4(48859,  <unfinished ...>
...
48858 wait4(48859, 0x7fffa52cc07c, WNOHANG, NULL) = 0
48859 <... wait4 resumed> [{WIFEXITED(s) && WEXITSTATUS(s) == 0}], 0, NULL) = 48860

subprocess.check_output()结束，＆＃34;执行＆＃34;据报道：

48859 --- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=48860, si_status=0, si_utime=0, si_stime=0} ---
48859 write(1, "executed\n", 9)         = 9

主进程开始增加超时并等待。 select / wait4周期为10秒。

48858 wait4(48859, 0x7fffa52cc07c, WNOHANG, NULL) = 0
48858 select(0, NULL, NULL, NULL, {0, 16000}) = 0 (Timeout)
48858 wait4(48859, 0x7fffa52cc07c, WNOHANG, NULL) = 0
48858 select(0, NULL, NULL, NULL, {0, 32000}) = 0 (Timeout)
48858 wait4(48859, 0x7fffa52cc07c, WNOHANG, NULL) = 0
48858 select(0, NULL, NULL, NULL, {0, 50000}) = 0 (Timeout)
48858 wait4(48859, 0x7fffa52cc07c, WNOHANG, NULL) = 0
48858 select(0, NULL, NULL, NULL, {0, 50000}) = 0 (Timeout)
[ + many more ]

直到safe_exec完成，在主要流程放弃.join(10)之后：

48858 --- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=48859, si_status=0, si_utime=0, si_stime=0} ---
48858 wait4(48859, [{WIFEXITED(s) && WEXITSTATUS(s) == 0}], WNOHANG, NULL) = 48859

这就是它 - 正常的流程退出。在这种情况下我无法理解错误。似乎没有什么能明显失败。

使用gdb查看正在运行的进程，父进程在wait内join停留：

#0  0x00007fde4486b6d3 in __select_nocancel () at ../sysdeps/unix/syscall-template.S:81
#1  0x000000000062dbd6 in floatsleep (secs=0) at ../Modules/timemodule.c:948
#2  0x000000000062c843 in time_sleep (self=0x0, args=(<float at remote 0x1e15620>,)) at ../Modules/timemodule.c:206
#3  0x00000000004896f9 in PyCFunction_Call (func=<built-in function sleep>, arg=(<float at remote 0x1e15620>,), kw=0x0) at ../Objects/methodobject.c:81
#4  0x000000000052ff77 in call_function (pp_stack=0x7fff39cdc9f0, oparg=1) at ../Python/ceval.c:4356
#5  0x000000000052a7f7 in PyEval_EvalFrameEx (
    f=Frame 0x1f643d0, for file /usr/lib/python2.7/multiprocessing/forking.py, line 165, in wait (self=<Popen(returncode=None, pid=67389) at remote 0x7fde43a83ed0>, timeout=20, deadline=<float at remote 0x1e15558>, delay=<float at remote 0x1e15620>, res=None, remaining=<float at remote 0x1e154e0>), throwflag=0) at ../Python/ceval.c:2993

并且孩子也在等待最终确定加入：

#0  0x00007f96798520c9 in futex_abstimed_wait (cancel=true, private=<optimised out>, abstime=0x0, expected=0, futex=0x1c109c0) at sem_waitcommon.c:42
#1  do_futex_wait (sem=sem@entry=0x1c109c0, abstime=0x0) at sem_waitcommon.c:208
#2  0x00007f9679852164 in __new_sem_wait_slow (sem=0x1c109c0, abstime=0x0) at sem_waitcommon.c:277
#3  0x00007f967985220a in __new_sem_wait (sem=<optimised out>) at sem_wait.c:28
#4  0x00000000005787f5 in PyThread_acquire_lock (lock=0x1c109c0, waitflag=1) at ../Python/thread_pthread.h:324
#5  0x000000000062aa9f in lock_PyThread_acquire_lock (self=0x7f9679c21c70, args=()) at ../Modules/threadmodule.c:52
#6  0x00000000004896f9 in PyCFunction_Call (func=<built-in method acquire of thread.lock object at remote 0x7f9679c21c70>, arg=(), kw=0x0) at ../Objects/methodobject.c:81
#7  0x000000000052ff77 in call_function (pp_stack=0x7ffda823dce0, oparg=0) at ../Python/ceval.c:4356
#8  0x000000000052a7f7 in PyEval_EvalFrameEx (
    f=Frame 0x1c107b0, for file /usr/lib/python2.7/threading.py, line 340, in wait (self=<_Condition(_Verbose__verbose=False, _Condition__lock=<thread.lock at remote 0x7f9679c21c20>, acquire=<built-in method acquire of thread.lock object at remote 0x7f9679c21c20>, _Condition__waiters=[<thread.lock at remote 0x7f9679c21c70>], release=<built-in method release of thread.lock object at remote 0x7f9679c21c20>) at remote 0x7f9676cec1b0>, timeout=None, waiter=<thread.lock at remote 0x7f9679c21c70>, saved_state=None), throwflag=0)
    at ../Python/ceval.c:2993

孩子的python堆栈是：

_Condition.wait
Thread.join
Queue._finalize_join
Finalize.__call__

我很感激任何关于为什么这个过程停滞不前的想法。

代码在python 2.7和3.4上运行

Answer 1

问题是队列上发送的小结果或空结果是正确处理的，但是长期结果并没有阻止进程加入。

通过在加入之前将结果从队列中删除来解决/解决该问题。

在多处理中挂exec

1 个答案: