I said:
> More to the point, what have you got SORT_MEM set to?
Or maybe that's not the point. A quick look through the sources didn't
turn up any repalloc calls that look like they could make requests that
large (at least not before you'd run out of memory from other causes).
Could you attach to the backend process with gdb, set a breakpoint
at elog, and get a backtrace, so we can see where the oversize
request is coming from?
regards, tom lane