"manually" running innocent-looking mapreduce code causes erlang crash

As a way to try some alternative approach figuring out our erlang VM crashes I've created the following simple test: http://paste.debian.net/201623/

And run it via /diag/eval:

wget O --user=Administrator --password=asdasd --post-data="ns_orchestrator:mapreduce_test_loop()." http://lh:9000/diag/eval

I started about 8-10 of such requests (they never return) and left office.

This morning I found erlang VM's core dump. But notably crash was within V8. Looking at core dump mtime I conculded that it happened about 2 hours after I started this v8/mapreduce "torture".

All VM settings are our defaults (i.e. async threads on).