With the switch to distributed-data we have much longer times in propagating, to the point of flushing out BGPCEP-983, where the topology disappears for 2+ seconds.

According to the documentation of Akka Distributed Data, this is due to us using writeLocal() and gossip dissemination. The latter is configured by default as:

akka.cluster.distributed-data {
  # How often the Replicator should send out gossip information
  gossip-interval = 2 s

  # How often the subscribers will be notified of changes, if any
  notify-subscribers-interval = 500 ms
}

These affect the reaction time we get in propagating changes – which are quite critical to our ability to converge.

Update these to accelerate propagation to at most a few tens of milliseconds.