If a listener is started or stopped via ejabberd_listener:add_listener/3
or ejabberd_listener:delete_listener/3, the configuration for all
listener modules is updated using the Module:transform_listen_option/2
function for each listener module that exports such a function.
However, for listener modules that don't provide that function (such as
ejabberd_stun), all but one option was dropped. This is now fixed.
The issue could be triggered e.g. by enabling mod_proxy65 in the modules
section.
Use dynamic Rebar configuration
Make iconv dependency optional
Disable transient_supervisors compile option
Add hipe compilation support
Only compile ibrowse and lhttpc when needed
Make it possible to generate an OTP application release
Add --enable-debug compile option
Add --enable-all compiler option
Add --enable-tools configure option
Add --with-erlang configure option.
Add --enable-erlang-version-check configure option.
Add lager support
Improve the test suite
If the callback module has a socket type of independent and needs to
create the listener itself, do not pre-bind the port. The same holds if
there are errors in the listener configuration.
It may happen that auth or rdbms client tcp connections bind a local
socket to a port number required by a configered listener. The ejabberd
applications fails to start up and needs to be restarted.
In plain C you would bind(2) the listener port and listen(2) later on.
gen_tcp:listen/2 does not allow to separate these two steps though, so
another way is not to accept connections while start up. OTOH, the
kernel will syn/ack incoming connections and receive data, leaving them
in a buffer for the ejabberd to read from. If this is unwanted, a load
balancer would need to receive data from the ejabberd server before
adding the node to its pool.
This patch binds tcp ports while initializing the ejabberd_listener
process, storing ListenSockets in an ets table. start_listeners/0 will
reuse these ports later on.
Evgeniy Khramtsov