We have a need for highly reliable CDC events to be sent from maxscale to kafka (for billing reasons).

Currently, to protect against maxscale crashes (either the process or the whole machine), we would have to run multiple instances of maxscale, each of which would process the binlogs and send them to kafka, leading to duplicate events.

I believe that Marcus mentioned the possibility to have maxscale's cluster (which is aware of active/passive maxscale instances) be able to failover the processing. This would ideally tie in with the direct kafka integration.