On Thu, 2010-09-16 at 17:20 -0400, Wietse Venema wrote:
> Vernon A. Fort:
> > using versions 2.8-20100913 and 2.8-20100915 - when i shutdown postfix,
> > i sometimes (more often that not) see the postscreen daemon hanging out
> > around 10 seconds after the master process terminates.  Same results on
> > two different machines (intel/amd).
> 
> postscreen cannot be terminated immediately by "postfix stop" as
> that would corrupt the database and you would never receive mail
> anymore.
> 
> When you run an experimental release, then you must accept that
> not all the features are finished.
> 
> Currently, postscreen does not distinguish between "postfix stop"
> and "postfix reload". It always goes into the background, waits
> until all existing clients close the connection, and then exits.
> 
> Fixing the "restart" may be as simple as closing the listen socket
> when postscreen goes off into the background. At least that seems
> to work with FreeBSD 8. You can try this on your own system.
> 
> *** ./event_server.c- Fri Sep 10 11:22:11 2010
> --- ./event_server.c  Thu Sep 16 17:18:01 2010
> ***************
> *** 295,302 ****
>       case 0:
>       (void) msg_cleanup((MSG_CLEANUP_FN) 0);
>       event_fork();
> !     for (fd = MASTER_LISTEN_FD; fd < MASTER_LISTEN_FD + socket_count; fd++)
>           event_disable_readwrite(fd);
>       var_use_limit = 1;
>       return (0);
>       /* Let the master start a new process. */
> --- 295,304 ----
>       case 0:
>       (void) msg_cleanup((MSG_CLEANUP_FN) 0);
>       event_fork();
> !     for (fd = MASTER_LISTEN_FD; fd < MASTER_LISTEN_FD + socket_count; fd++) 
> {
>           event_disable_readwrite(fd);
> +         (void) close(fd);
> +     }
>       var_use_limit = 1;
>       return (0);
>       /* Let the master start a new process. */
> 
> > This is only a problem when i do a /etc/init.d/postfix restart.  It's
> > likely a problem with the rc scripts (gentoo here) but i thought i would
> > post this observation anyway.
> 
> The "restart" feature is added by your OS distribution, so I can't
> promise that it will ever be a supported feature.
> 
>       Wietse

I agree with the restart - O/S dependent (not postfix).  The patch seems
to work but i will need to observer under heavy load.  However, i don't
really restart postfix much - doing a stop {pause} start is not a
problem since it not needed (restart) often.

Thanks - i do like this postscreen.  My overall observation is the load
on the servers are down as well as the spam.

Vernon

Reply via email to