[EMAIL PROTECTED] wrote:
> Olivier,
>
> I did some small research for your question, and found three possible
> solutions, but, unfortunately, their working depends on the OS.
>
> the solutions are:
> 1. in-loop (my proposal) = wait for children in program loop
> 2. in sighandler = install handler for SIGCHLD and wait there for
> children
> 3. ignoring child (different from ignoring the signal, which is what
> SIPp does now)
>
> For number three, I found the initialisation
> "sigaction(SIGCHLD,SIG_NOWAIT)" should work on Solaris. Unfortunately I
> don't find any resembling definition *_NOWAIT on my linux box. There is
> a note on my linux man page for "wait", talking about SA_NOCLD-WAIT or
> setting SIGCHLD action to SIG_IGN, and different options/specifications
> for POSIX and later standards, and linux not supporting or conform to it
> (rather unclear text). General conclusion: for portable code, this
> doesn't seem the best way to go.
>
> The numbers one and two are both viable solutions, where for the first
> you need of course a recurring loop in the program, and SIPp has that.
> As there is also signal handling in the program, you could also opt for
> the second option. But don't forget, in both cases you will need the
> internal loop as you might need to get rid of multiple children at the
> same time.
>
> My personal solution would always be the first, as I've seen (and made)
> already many mistakes with signal handlers, trying to do much too much
> internally, and forgetting that the handler can be called anywhere in
> the program execution, also while in the middle of a recalculation of
> some variables. I checked the SIGUSRx in SIPp, the techique used is
> fine. You could indicate in the same way in a SIGCHLD handler that there
> is some "waiting" to do:
>
> child_caught=1;
>
> (as alternative for my child counter), and conditionally execute the
> wait loop in the main program loop (but don't forget to clear the flag
> BEFORE the wait loop):
>
>
> if (child_caught) {child_caught=0; while (waitpid(-1,(int
> *)NULL,WNOHANG)>0) ;}
>
> This would be a good implementation of point two, which is not much
> different from the first solution.
>
> I like the counter+in-loop more (it can help statistics), and you don't
> need to recharge the signal handler (some OSes require that) or watch
> out for nasty problems like if you would put the child_caught=0 after
> the while and see you last child still hanging around. But I leave the
> choice to you.
>
>
> So that's it for the system command. For the background mode, you don't
> need anything special at all, the code is just fine. In this case, the
> death of the parent SIPp process is handled by the process (normally
> your shell) used to launch the program, and as the parent SIPp had a
> child, that launching process will take over the care for that child. If
> you exit the shell, the process above it (e.g. telnetd, inetd or finally
> the init process) will continue the care job. That's why all <defunct>s
> disappear as soon as the main SIPp dies: the process above does the
> cleaning at that moment.
>
>
> In short, the background code is just fine, only the comment that the
> child gets the parent pid is not right: the child keeps its pid, but
> gets another parent (ppid). The code displaying the childs pid is
> correct.
>
Thanks for this detailed investigation.
You talk about solution 3 (ignoring child - which is different from
ignoring signal).
It seems that this modification (see below) is different from solution
3. What do you think of this one (which is really to ignore SIGCLD)?
+ struct sigaction action_ign;
+ memset(&action_ign, 0, sizeof(action_ign));
+ action_ign.sa_handler=SIG_IGN;
+ sigaction(SIGPIPE, &action_ign, NULL);
+ sigaction(SIGCLD, &action_ign, NULL);
I'm kind of sceptical with including new handling (taking care of
childs) in the main SIPp loop as it will have the bad side effect of
impacting everything...
Olivier.
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
Sipp-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/sipp-users
