- Replica Shards for Read op to load-balance.
- Replicas of Replicas are allowed, to have less load on primary for syncs as well.
- Default is async replication, non-blocking.
- Primary must have persistence on. Replica must have read-only.
- Redis Cluster provides a way for data automatically sharded for r/w, using hashtag buckets.
- Cluster nodes use a different ClusterPort for intra-comm.
- Re-sharding happens on add/del of nodes. Secondary shards can promote to primary on their failure.
- Clients can select correct master to write to, when sharded. Or can use Proxy assisted partitioning.