Just found this article last night on circular replication with
auto-recovery from single node in chain failures:
4 nodes lost in a row would break circle, a circle of 4 or less will
not have circular replication broken unless N-1 nodes go offline ..
after 7 servers the worst case min number of nodes that could be lost
and cause circular replication to fail is 4 in two consecutive pairs
with servers that are up in-between .. so risk of circular replication
failure increases noticably with each node added after node 8.
Looks very promising.
Uses federated tables for 'is my peer alive' along with SQL scheduler
and a stored procedure to do the alive check and then reconnect a peer
to the peer+1 server if a replication peer goes away.
Will be implementing this within the next few weeks as a POC and then
to production if it indeed behaves the way the article claims it will.
Has anyone else on this list seen that article or implemented that