We can better parallelize the workload for keyspace events where we subscribe to all of the upstream nodes.
Ideally, we'd split the work into Kafka Connect tasks and use the max task count as a limiter. Each task would subscribe to an upstream node and then reconfigure upon topology changes.
This strategy doesn't work for regular pub/sub though, since regular pub/sub is cluster-wide (apart from the new sharding pub/sub logic that was recently introduced and not yet supported). As such, https://github.com/jaredpetersen/kafka-connect-redis/issues/27 is a blocker.
We can better parallelize the workload for keyspace events where we subscribe to all of the upstream nodes.
Ideally, we'd split the work into Kafka Connect tasks and use the max task count as a limiter. Each task would subscribe to an upstream node and then reconfigure upon topology changes.
This strategy doesn't work for regular pub/sub though, since regular pub/sub is cluster-wide (apart from the new sharding pub/sub logic that was recently introduced and not yet supported). As such, https://github.com/jaredpetersen/kafka-connect-redis/issues/27 is a blocker.