[Coin-discuss] workers segfault

[Laszlo Ladanyi]

Hi Paul,

I like your boss' attitude :-).

The segfault symptom you describe can happen when the lp process has finished
processing a node and receives a new UB before it receives a new node. In that
case no LP solver exists, so you get a segfault. The simplest (and
correct) fix is to test the existence of an LP solver before setting the dual
objective limit. I have changed the code accordingly and committed the change.

--Laci

On Fri, 17 Oct 2003, Paul Pacheco wrote:

> I finally got the parallel solver working. 
> 
> I want to thank you for such a great framework. My boss just loves sitting 
> with me and watching this thing working with xpvm, and seing the parallel 
> algorithm in action. He has called everybody in the office to show them how 
> this thing works.
> 
> 
> 
> 
> 
> I found one last bug before I could get this to work, I attach a patch for 
> that.
> 
> The symthom was that the slaves were seg faulting and dying after a few 
> minutes, and the master didn't realize that, and it eventually hung waiting 
> for the dead slaves.
> 
> The problem seems to happen when a upper bound is found and sent to all the 
> slaves. a this point, the method BCP_lp_process_ub_message is called in the 
> slaves, and it tries to access p.lp_solver which is null.
> 
> If the patch solves the problem the wrong way, please let me know.
>