[R] Passing arguments to forked children

[Godmar Back]

On Fri, Jul 10, 2009 at 8:16 PM, <nabble.30.miller_2555 at spamgourmet.com> wrote:
> Hi -
>
>    I have attempted to use the fork::fork() function to perform
> parallel processing. However, the child R function called needs to
> know a given set of parameters to complete its task. Specifically, I
> iterate through a vector, and output values based on the elements of
> that vector to a database. The output strings contain elements of the
> iterated vector. I mocked-up the following code as an example (NOTE:
> WHILE NOT SPECIFICALLY DANGEROUS, THIS CODE MAXED OUT THE LIMIT OF MY
> SYSTEMS FORKS -- this means that, if you run this code, no additional
> processes on your system may start until you kill the parent R
> session! BE VERY CAREFUL IF YOU DECIDE TO EXECUTE THIS CODE --
> obviously, I do not recommend it. Presumably, an infinite recursion
> scenario arose so  you are just left with a *lot* of R sessions. Also,
> as each R session has equal access to stdin, you cannot reliably type
> commands into a given R session to terminate it -- so definitely don't
> run in a CLI environment -- at least you can kill the parent window
> running R in a GUI environment). In any case, here is the code:
>
> # -- BEGIN CODE
> library("fork");
>
> myforksub <- function(mymsg='default') {
>        cat(mymsg,sep='\n');
>        exit();
> }
>
> myforkparent <- function(n=10, mymsg='') {
>        mypid <- c();
>        for (i in 1:n) {
>                mypid <- c(mypid, fork(myforksub(mymsg)));
>        }
>        # wait(NULL) apparently does not wait for all children to finish
>        for (i in 1:n) {
>                wait(mypid[i]);
>        }
> }
>
> myforkparent(mymsg='new');
>
> # -- END CODE
>
> Obviously, 'fork(myforksub)' will work fine, but myforksub cannot
> access the mymsg variable containing the 'new' value. How can I amend
> the above without having to resort to socket connections to pass
> information?


Like so:
               mypid <- c(mypid, fork(function () {
                    myforksub(sprintf("Hi, I'm child %d", i))
               }));

BTW:
       # wait(NULL) apparently does not wait for all children to finish

wait(NULL) traditionally waits for any single child to finish, not for all.

 - Godmar