On Mon, 11 Dec 2006, Philip M. Gollucci wrote:
Grant wrote:
> > Also, I tried restarting the interchange daemon with
> > PERL_SIGNALS=unsafe and the ALERT/segfaults came MUCH MUCH more
> > frequently. Does that tell us anything?
>
>
> It would make sense that, when you have high load, there is a problem
> processing many concurrent requests which triggers the PIPE signal, so
> you should find out what the error is, and handle it more gracefully.
>
> You might want to change the die sub to print out $! and $? - that may
> give you a bit of a clue as to what caused the PIPE signal.
>
> I'm guessing (and it is a guess) that the segfaults may be caused
> because the die sub sends a web response, but that sub could be called
> while your server is busy doing something else, and the two actions
> collide.
Very good guess. Commenting out the web response stuff seems to have
eliminated the segfaults. Adding $! and $? to the warn line, I'm
getting one of these two bits along with each ALERT now:
Broken pipe 0
Inappropriate ioctl for device 0
This could happen from a Cntrl-C or stop in a browser.
Add
require Carp;
Carp::cluck() to your die() function.
Is this someplace that checking $r->connection->aborted() would be useful?
