On Monday, 18 October at around 4pm, a problem occurred on one of our two ejabberd instances that manages the proper functioning of XMPP (real-time message exchange protocol). Normally, this should not be a problem because the second instance takes over until the first one restarts. However, in this case, the restart was not done correctly, which generated presence desynchronization at the end of the day when the users disconnected. The disconnection was not always taken into account correctly, which meant that we considered users to be connected when they were not and vice versa. As a consequence, we were able to engage some visitors but the conversations could not be distributed due to the lack of real availability of agents. These conversations appear this morning for the concerned agents.
Our technical team is currently investigating why the server restart did not take place as planned. In parallel, we are resynchronizing the presence of users who have not yet reconnected.