[racket] Getting unusual behavior from racket/sandbox: not truly capping memory usage?!

From: Danny Yoo (dyoo at hashcollision.org)
Date: Sat Apr 20 00:25:17 EDT 2013

Followup on the sandbox issue: Eli suggested that I push move of the
evaluation stuff into the sandbox.  I had assumed that there was some
parameterization that controlled values coming out of eval, but
perhaps not!

So I made the following change:
    https://github.com/dyoo/whalesong/commit/f55723b3324c9a1aa554440cc6fe2487e4744b1d#whalesong/repl-prototype/sandboxed-server.rkt


I think that finally did the trick: it looks like evaluation is all
being wrapped by the sandbox: I watch the output from top and it does
seem like it's GCing whenever it hits the memory ceiling I impose.


The bad news is that I'm seeing a memory leak in my web application
now, and I deliberately tried not to leak memory.  :P  I'm not exactly
sure how to trace it yet, but I'll start by turning parts of my web
service off and see if I can isolate the offending component.

Posted on the users mailing list.