Signal SIGTERM handling of multi-thread process under Linux

Question

I was debugging the signal handling of one multi-thread process under Linux, and found one strange appearance. In a sentence of this problem is “SIGTERM handler is empty and is registered to system successfully, but the process may still be killed by SIGTERM when process starting”.

The detailed description as below:

• This process is forked by parent process, and started by execve

• Process fake operation

  void sighdr(int sig) {      
      Trace();  
  }     
  main() {     
      sigaction(); // Register SIGTERM handler  
      prog_init(); // Very complex operation, read/write UNIX socket, 
                   // starting new thread, and so on. The whole   operation 
                   // will take almost 12 seconds.     
      prog_loop(); // loop of main thread 
  }
  

• If signal SIGTERM fired in the course of prog_init() operation

  The sighdr() will be invoked by kernel, and the trace file can record the related information OK. But then the process stopped. As the amount of code of second thread is very large, so I couldn't position the last instruction. And a very few times (1/60), the process will not be stopped, but the running status is not correct, it's hanged in recvmsg() of Linux, the prog_loop() will never be executed.

• If signal SIGTERM fired in prog_loop()

  The process runs fine, the SIGTERM will be ignored.

• After running, the /proc/pid/status

Threads: 2
SigQ: 0/124393
SigPnd: 0000000000000000
ShdPnd: 0000000000000000
SigBlk: 0000000000000000
SigIgn: 0000000000000005
SigCgt: 0000000180005000

So my question:
1) Is there any possible reason of this phenomenon?
2) Is there any debug method to find what the process happened? I have tried the valgrind, but no valuable found.