Solved A problem with unexpected termination

[Харбин Хэйлунцзян]

I am developing a certain service and these days have been testing it using a stress Windows application. Sometimes the service terminates without any message. No core dump. I can reproduce the conditions under which it happens and when I run the service under GDB the latter reports:

Thread 3 received signal SIGPIPE, Broken pipe.
Sent by kernel.
[Switching to LWP 100162 of process 1359]
_sendto () at _sendto.S:4
warning: 4      _sendto.S: No such file or directory

There isn't a single call to sendto in the service. How do I deal with that? What am I expected to do next?

 

[SirDice]

Hard to tell without the source code. You might be using some library function that itself uses sendto(2) to send the data. And it might be sending data to a socket that got closed for whatever reason.

     [EPIPE]            The socket is unable to send anymore data
                        (SBS_CANTSENDMORE has been set on the socket).  This
                        typically means that the socket is not connected.

 

[cracauer@]

That's probably a red Herring. SIGPIPE is usually not caught for exiting. But if you run a program in gdb it will trigger the debugger on SIGPIPE.

handle SIGPIPE nostop pass noprint

 

[Харбин Хэйлунцзян]

You might be using some library function that itself uses sendto(2) to send the data.

I only explicitly specify pthread and rt. The service only calls send on accepted connections. For example:

    else {
        iResult = send(this->kk, http404, sizeof(http404) - 1, 0);
        if (iResult == -1) printf("%x : %d : send 404 failed: %d\n", this->kk, this->pSess->kk, errno);
        else printf("%x : %d : Sent 404\n", this->kk, this->pSess->kk);
    }

    return iResult;
}

I would expect to get -1, not termination.

 

[Харбин Хэйлунцзян]

But if you run a program in gdb it will trigger the debugger on SIGPIPE.

That's exactly what I do as is written in my original posting.

And there's worse news. If at this point I shortly press the power button, which usually shuts down the operating system and computer, I get:

Stopping cron.
Waiting for PIDS: 1274.
Stopping sshd.
Waiting for PIDS: 1299.
Stopping devd.
Waiting for PIDS: 1002.
Writing entropy file: .
Writing early boot entropy file: .
Terminated
.
Nov 12 16:46:15 s17 syslogd: exiting on signal 15

And it hangs...

 

[Харбин Хэйлунцзян]

handle SIGPIPE nostop pass noprint

With this I get lots of

2ed : 7669 : send 200 failed: 32

As expected. But that's under the control of GDB. Without GDB it just terminates. Probably the library is lousy. I guess it's because they stint the 32bit subsystem as a stepdaughter.

 

[cracauer@]

Well, the default signal handler for SIGPIPE will exit, and silently so. You can install your own which doesn't exit, or prints a message or whatever you want.

 

[Харбин Хэйлунцзян]

Well, the default signal handler for SIGPIPE will exit, and silently so. You can install your own which doesn't exit, or prints a message or whatever you want.

This call at the start solves the problem:

signal(SIGPIPE, SIG_IGN);

The prospect doesn't look as gloomy as it did yesterday.
Thank you.