fedmsg-relay SOP

fedmsg-relay is running on app01, which is a bad choice. We should look to move it to a more isolated place in the future. busgateway01 would be a better choice.

"Ephemeral" scripts like pkgdb2branch.py, the post-receive git hook on pkgs01, and anywhere fedmsg-logger is used all depend on fedmsg-relay. Instead of emitting messages "directly" to the rest of the bus, they use fedmsg-relay as an intermediary.

Check that fedmsg-relay is running by looking for it in the process list. You can restart it in the standard way with sudo service fedmsg-relay restart. Check for its logs in /var/log/fedmsg/fedmsg-relay.log

Ephemeral scripts know where the fedmsg-relay is by looking for the relay_inbound and relay_outbound values in the global fedmsg config.

The fedmsg bus is designed to be "passive" in its normal operation. A mod_wsgi process under httpd sets up its fedmsg publisher socket to passively emit messages on a certain port. When some other service wants to receive these messages, it is up to that service to know where mod_wsgi is emitting and to actively connect there. In this way, emitting is passive and listening is active.

We get a problem when we have a one-off or "ephemeral" script that is not a long-running process — a script like pkgdb2branch which is run when a user runs it and which ends shortly after. Listeners who want these scripts messages will find that they are usually not available when they try to connect.

To solve this problem, we introduced the "fedmsg-relay" daemon which is a kind of "passive"-to-"passive" adaptor. It binds to an outbound port on one end where it will publish messages (like normal) but it also binds to an another port where it listens passively for inbound messages. Ephemeral scripts then actively connect to the passive inbound port of the fedmsg-relay to have their payloads echoed on the bus-proper.

See http://fedmsg.readthedocs.org/en/latest/topology/ for a diagram.