Re: sophie: sig_exit() callable by child, causes weird hangs

On Tue, 10 Aug 2004, David Birnbaum wrote:

> Chris,
>
> That certainly explains the symptoms of our various accept() problems. 
> Good catch.  3.04rc2 didn't work for me at all on Solaris, tho, so I 
> will probably back-port this to 3.03.

Looking at the code, the actual changes between 3.03 and rc2 are really 
small.  What problems were you having with Solaris?  Have you tried 
reverting individual bits?  The only major changes that I can see are the 
setsockopt() calls and the nanosleep stuff.

> As to why it was SEGV'ing in the first place, though....

Yup.  I had a few groups of segfaults yesterday, prompting the bughunt. 
Usually they're far more sporadic.  I'm waiting to see if they start up 
again so I can try to figure out what's causing them.


-Chris


> -----
>
> On Mon, 9 Aug 2004, Chris Stromsoe wrote:
>
>> While trying to figure out why sophie was inexplicably hanging on me 
>> earlier today, I found a problem.  If a child process segfaults, the 
>> sig_exit handler gets called and will close and unlink the sophie 
>> socket, leaving the parent to sit in accept() forever.  I added a test 
>> to sig_exit() to check for child status.  This may fix some of the 
>> other random "hang" issues that have come up over the last few months 
>> that sound sort of similar to what I was seeing earlier.  Patch is 
>> attached.
>> 
>> 
>> -Chris
> _______________________________________________
> vtools mailing list
> [email protected]
> http://www.vanja.com/list/listinfo.cgi/vtools
>