- In case a lot of message were being processed, the heartbeat timer could be delayed for a long period of time.
- The old scheduler algorithm could be unfair when certain message priority were not available for processing. We now appropriatly handle repartition to others priority messages.
- Message of the same priority could be processed in the wrong order when on-disk buffers were used.
- No integrity check were performed on orphan on-disk buffer in case of an operating system crash. By using the prelude-failover API, we can now detect possibly corrupted disk buffer, or resume at the time we stopped recovering them.
- New sched-priority and sched-buffer-size configuration options.
- Fix a bug where several relaying plugin instance would only forward their message to a single Manager.