c – 冻结__write_nocancel

2023年6月30日 229次阅读

所以我有一个非常重要的程序,它偶尔会冻结.

该程序使用Qt,开放场景图和谷歌日志记录.这种冻结发生在谷歌日志记录中.该程序本身正在打印大量的调试信息.我能够通过gdb-server连接到程序,这是堆栈跟踪：

#0  0x000000397ac0e030 in __write_nocancel () at ../sysdeps/unix/syscall-template.S:82
#1  0x00007f5eecb74aeb in google::LogMessage::SendToLog() () from /lib64/libglog.so.0
#2  0x00007f5eecb71fc7 in google::LogMessage::Flush() () from /lib64/libglog.so.0
#3  0x00007f5eecb721a9 in google::LogMessage::~LogMessage() () from /lib64/libglog.so.0
#4  0x00000000004874a6 in LSDB::process (this=0x242d918, lsp=0x25f9200, circuit=0x24c7af0) at ../src/model/trill/LSDB.cpp:481
#5  0x00000000004a0f6f in Circuit::rx (this=0x24c7af0, eth=0x246fdf0) at ../src/model/trill/Circuit.cpp:355
#6  0x000000000045c950 in Circuit::qt_static_metacall (_o=0x24c7af0, _c=QMetaObject::InvokeMetaMethod, _id=0, _a=0x7fffaade95a0)
    at ../src/model/trill/Circuit.moc.cpp:55
#7  0x000000398798cb9f in QMetaObject::activate (sender=0x2470140, m=<optimized out>, local_signal_index=<optimized out>, argv=0x7fffaade95a0)
    at kernel/qobject.cpp:3547
#8  0x0000000000459610 in FramePublisher::subscription (this=0x2470140, _t1=0x246fdf0) at ../src/model/system/FramePublisher.moc.cpp:98
#9  0x000000000047c0d6 in FramePublisher::rx (this=0x2470140, frame=0x246fdf0) at ../src/model/system/FramePublisher.hpp:21
#10 0x000000000047bedb in EthernetPort::rx (this=0x246d7a0, frame=0x25a4180) at ../src/model/system/EthernetPort.cpp:81
#11 0x000000000045a208 in EthernetPort::qt_static_metacall (_o=0x246d7a0, _c=QMetaObject::InvokeMetaMethod, _id=1, _a=0x7fffaade9810)
    at ../src/model/system/EthernetPort.moc.cpp:51
#12 0x000000398798cb9f in QMetaObject::activate (sender=0x246d7a0, m=<optimized out>, local_signal_index=<optimized out>, argv=0x7fffaade9810)
    at kernel/qobject.cpp:3547
#13 0x0000000000459ddc in Port::rx (this=0x246d7a0, _t1=0x25a4180) at ../src/model/system/Port.moc.cpp:110
#14 0x00000000004803a6 in Port::inject (this=0x246d7a0, frame=0x25a4180) at ../src/model/system/Port.cpp:25

…

请注意冻结本身发生在__write_nocancel中.只有一个线程运行…

(gdb) info threads Id Target Id Frame
* 1 Thread 21507 0x000000397ac0e030 in __write_nocancel () at ../sysdeps/unix/syscall-template.S:82

关于什么导致冻结的任何想法？让我知道其他信息可能会有所帮助.

最佳答案 你说输出正在写入stderr.如果没有通过单独的进程不断读取stderr,则I / O缓冲区可能已满,因此它在写入更多内容之前等待stderr被读取.

如果进程A写入stderr并且进程B通常从stderr读取但当前正在等待进程A,则可能导致死锁.在你的情况下,如果写入stderr太大,它只会导致死锁,因为较小的写入会立即成功并释放进程A,从而释放进程B从stderr读取.

这是我的一些猜测,但总的来说我会假设你的问题是你正在等待写入一个完整的缓冲区.

点赞