Radu Greab wrote:
>
> The patch to qpsmtpd-async also closes the config socket in child. I
> think the child does not need that socket.
I used the config port for the first time today and I found that, if
my patch is applied, the config port becomes useless: you can connect
to it but there is nothing interesting to see because the clients are
connected to the child, not to the parent.
I'm submitting an updated patch that keeps the original functionality of
the config port. With the updated patch, the config socket is kept open
in both parent and child (or children) and all processes will monitor
it. However it will be a matter of luck which process will accept the
connection, either the parent or a child.
Am I right to suppose that a better option would be to have the config
port open and watched only in the parent and when someone connects to it
and issues commands the parent will relay the commands to _all_
children, will wait for all results and will send back the combined
output from all children?
Thank you,
Radu Greab
=== qpsmtpd-async
==================================================================
--- qpsmtpd-async (revision 111)
+++ qpsmtpd-async (local)
@@ -163,9 +163,12 @@
$SIG{PIPE} = 'IGNORE';
$SIG{HUP} = 'IGNORE';
+ # close $CONFIG_SERVER;
+ Qpsmtpd::PollServer->Reset;
Qpsmtpd::PollServer->OtherFds(
- fileno($reader) => sub { command_handler($reader) },
- fileno($SERVER) => \&accept_handler,
+ fileno($reader) => sub { command_handler($reader) },
+ fileno($SERVER) => \&accept_handler,
+ fileno($CONFIG_SERVER) => \&config_handler,
);
$plugin_loader->run_hooks('post-fork');
