This is a typical GR fail safe scenario. Two different hosts having independent datomic running. A channel is established between host for replication of data.
One approach to replay all the transaction of one instance to other instance. But this will give performance issue for large data
Is this use case addressed yet ?
In the (on prem) Datomic data modell and architecture I argue that a replication as you describe it is unnescessary. There is only one transactor (where you can have a hot transactor waiting to take over).
If you want to read from the data you can read from the backstore. No need to replicate the data first. If you want higher availiability of the data you can use a more distributed backstore, and also use caching functionality as memcached and valcache.