I have done exactly that in the past. The biggest problem with implementing
something like this is you have to open a pseudo tty (pty) and run the
program attached to the pty. Some programs behave differently if they are
attached to a tty. Also another issue is there is some overhead with the
"rush -notes" command. When 100's of cpus are all calling "rush -notes" you
do feel it on the job server.
> My thinking here would be that Rush would require some kind of wrapper
> around the application to convert the application's progress reports
> into something that makes more sense to Rush. Pretty much in the same
> way that Alfred does it. I work mainly with Shake, and if I've, say, got
> 5 frames rendering on each render machine, the wrapper script would
> translate Shake saying that frame 1/5 is 50% complete to telling rush
> that the whole box is 10% complete. Alfred ignores any output from the
> application except for lines that look something like "ALF_PROGRESS 23%"
>
|