SCTP Failover/Resilience Scenarios

[buko]

At this point the benefits of SCTP are small (multi-homing is TBD), but one aspect can be seen in that, if you run the date-demo project, you can stop and restart the server while the client is running, without the client either failing or needing to do anything to reconnect.

Will admit, don't completely understand all the intricacies of SCTP but we're very interested in its support for failover/resilience. The scenario we have is a micro-service architecture where services may disappear and reappear and where services may have primaries (that write/mutate data) and secondaries/replicas (that can service queries and possibly be elected to become the new primaries).

SCTP interests us because it seems to support these critical use cases... though it's not clear exactly how to make it happen. Maybe some light can be shed here? Playing around with Scamper it seems to support basic point-to-point well enough but we're again interested in the failover/reslience servers. We did observe the behavior quoted above but it's not clear how exactly this works. What is the metric that sctp might use to determine the remote endpoint is gone? At some point this behavior becomes a bit pathological.

As far as failover we understand this functionality doesn't exist yet but if it did how would it work? How does the client discover the replicas? Does it reply on the primary being aware? (That would actually be ideal.)