Re: Milter hangs, revisited

[Terry Kennedy]

> >   Yup, no invocation of spamc.
>
> Ok, maybe the other end isn't attached to spamc :)  Do that "lsof |
> grep" bit and see what process is on the other end of the pipe.
 
  I think we're past this. It looks like something in the process was
getting messed up by the closeall(). I've seen incredibly weird things
happen in the thread environment.

> >   I'm a bit nervous running with the closeall() commented out, though.
> >
> >   Any ideas on why this is happening and a better workaround?
>
> Actually, try this.  Replace the close(fd++) call in closeall() with
> fcntl(fd++, F_SETFD, FD_CLOEXEC), which should end up doing the same
> thing in a different way.  Hopefully the process won't just end up
> hanging in execvp() :)

  That seems to be working. I'll run this in production for a day or so
and see what happens.

> Makes me want to build a non-threaded milter API, it does.

  Really. I actually looked at a bunch of the "fake SMTP server on port 25
which washes mail through spamd and then gives it to the real mailer" pack-
ages, but I didn't want to give up the stuff I'm doing with early-reject in
sendmail.

  Fortunately, it looks like the milter is now working as desired.

        Thanks!
        Terry