in reply to Re^6: Killing on Windows(Updated)
in thread Killing on Windows
How do I have to understand "with prejudice" here? That all children are killed without any mercy?
Actually, you're right. The code is more convoluted than it first appears. Only -9 terminates with prejudice (using TerminateProcess() which is an unconditional, no chance of saving anything API).
See win32.c starting with
alien_process: if (my_kill((IsWin95() ? -pid : pid), sig)) return 0; } }
static int terminate_process(DWORD pid, HANDLE process_handle, int sig) { switch(sig) { case 0: /* "Does process exist?" use of kill */ return 1; case 2: if (GenerateConsoleCtrlEvent(CTRL_C_EVENT, pid)) return 1; break; case SIGBREAK: case SIGTERM: if (GenerateConsoleCtrlEvent(CTRL_BREAK_EVENT, pid)) return 1; break; default: /* For now be backwards compatible with perl 5.6 */ case 9: /* Note that we will only be able to kill processes owned by t +he * current process owner, even when we are running as an admin +istrator. * To kill processes of other owners we would need to set the * 'SeDebugPrivilege' privilege before obtaining the process h +andle. */ if (TerminateProcess(process_handle, sig)) return 1; break; } return 0; }
So, -9 seems to be right for your purpose, but if you want to give the processes at least the opportunity to exit with a little grace, you might try -2, -15 or -21 first. These are converted into trappable events that the process can choose to ignore, but will normally act upon by closing down as gracefully as possible.
The downside of this is that if a parent decides to respond to (say), the ControlBreakEvent, but one of its children chooses to ignore it, I think that path to that child would be severed and you would not be able to kill it with a subsequent -9.
I think Win32::Job is a safer bet and not much more complex than system 1, for the simple case. It also comes with some nice extras (timeouts) and from excellent heritage.
|
|---|