In cases where clients reject messages and get repeated redeliveries, the broker's journals can come to contain a huge number of update records. In the present journal compaction impleementation, these records cannot be removed so long as the message has not been delivered. So if a client is badly-behaved, journals will be added until storage is exhausted. The situation is made worse by the fact that the broker does not (perhaps can not) respect the max-disk-usage configuration in a situation like this.