N-way replication, multiple masters

[Raymond Sellars]

Hi

Looking for some guidance on how to setup a dovecot cluster with more than
2 servers. Ideally as multiple masters so I don't have the complexity of
back recovery/synchronisation in failover situations. And I can utilised
upstream load balancers (director or other) to automatically move between
available nodes.

Current pattern is:
High availability at one site (AWS availability zone) with two active
servers behind dovecot directors and an AWS network load balancer in front
of the directors.

A third dovecot server as a disaster recovery instance (another
availability zone or region), with ideally real time and bidirectional
replication. Behind a different network and NLB access path.
Dovecot --version = 2.3.14
Centos 7

Previously for the HA pair we utilised NFS but the AWS EFS equivalent
doesn't perform  ( very high latency and NFS meta data caching required).
And then for DR we utilised the dovecot replication. 2 way so I have
automated recovery and could do active active testing/verification as
required with no down time. This help verify recovery pathways.

https://doc.dovecot.org/configuration_manual/replication/# - reading the
documentation is seems clear that replication only supports a PAIR of
masters. Any ideas on how I can add in a 3rd replication node? Two pairs?
I'm happy to utilise a star pattern. HA 1 <-> HA2, HA 1 <-> DR.

Not possible?
Can I run more than one replicator?
Or should I utilise doveadm sync via cron jobs and 2-way sync.

Any thoughts, suggestions most welcome. How are others achieving > 2
clusters for dovecot servers supporting the same domain? Most of this is
for availability rather than purely performance scaling.

Thanks
Raymond