[racket] large raco parallel build takes down computer?

I'm not sure it is the same, but I've seen bad performance when memory
is exhausted, which can happen with the parallel build. (My machines
have swap, tho, which maybe is why I don't see the machine crashing.)

Robby

On Thu, Oct 14, 2010 at 4:06 AM, Neil Van Dyke <neil at neilvandyke.org> wrote:
> Has anyone seen recent "raco setup" parallel builds take down a Linux
> machine?  Just wondering, for when I try to debug a problem I'm seeing.
>
> The last day, I've been moving a large app from 4.2.5 to 5.0.1, and on two
> occasions, clean builds have mostly locked up my 2-core 32-bit X86 Linux
> workstation, to the point that I have to reboot.
>
> It appears that perhaps the scheduler is starving other processes, including
> X, in favor of giving all cores almost fully to "raco setup".  Or, perhaps
> "raco setup" has somehow exhausted the RAM (this machine does not use swap
> space) and for some unknown reason is not getting OOM-killed.
>
> I don't have time to debug this right now, so I temporarily made the build
> process always leave a free CPU core, which has worked on two clean builds
> thus far, using only 1/3 of the available RAM.  (Note: The earlier,
> system-crashing builds had build errors from code that I later fixed, so I
> suppose it's possible that "raco setup" used a lot more memory because of
> the errors.)
>
> LD_LIBRARY_PATH='/usr/lib/xxx' PLTCOLLECTS='/some/dir/collects:'
> /usr/local/plt-5.0.1/bin/racocgc setup --workers `racket -e "(require
> racket/future) (max 1 (- (processor-count) 1))"` -D  -l foo bar baz
>
> --
> http://www.neilvandyke.org/
> _________________________________________________
>  For list-related administrative tasks:
>  http://lists.racket-lang.org/listinfo/users
>