"Andrew Badr" <andrewbadr@gmail.com> writes:
> ECC memory, RAID 10 w/ adaptec 3405 hardware controller. Period between
> crashes ranged from about 1min-5mins. Just switched to a new box so the
> problem is "gone". In the original email I meant ubuntu 6.06.
Well, that could mean it's a hardware problem. Or it could be a libm
problem --- is the new box running exactly the same system software?
The third likely possibility is that cost_sort() is passing some invalid
value to log(), and you're not seeing that happen on the new box because
of slightly different statistics or some such. This would still imply
a bug in libm --- it shouldn't be possible to make log segfault --- but
there could also be a corner case that cost_sort is mishandling.
If you can still reproduce the problem on the old box, please rebuild
Postgres with --enable-debug, and/or put some debug logging into
cost_sort to see what it's passing to log(), and see what you can learn.
regards, tom lane