Database replication provides a mechanism to ensure high availability. When replication is enabled, your dataset is replicated to a slave shard, which is constantly synchronized with the master shard. If the master shard fails, an automatic failover happens and the slave shard is promoted to be the new master shard. When the old master shard recovers, it becomes the slave shard of the new master shard. This auto-failover mechanism guarantees that data is served with minimal to no interruption.

You can tune your high availability configuration with:

  • Rack/Zone Awareness - When rack-zone awareness is used additional logic ensures that master and slave shards never share the same rack, thus ensuring availability even under loss of an entire rack.
  • High Availability for Slave Shards - When high availability for slave shards is used, the slave shard is automatically migrated on node failover to maintain high availability.

Note: Enabling replication has implications for the total database size, as explained in Database memory limit.

Redis on Flash Replication Considerations

We recommend that you set the sequential replication feature using rladmin. This is due to the potential for relatively slow replication times that can occur with Redis on Flash enabled databases. In some cases, if sequential replication is not set up, there is a risk of an Out Of Memory (OOM) situation. While it does not cause data loss on the master shards, the replication to slave shards may not succeed as long as there is high write-rate traffic on the master and multiple replications at the same time.

The rladmin command below sets the number of master shards eligible to be replicated from the same cluster node, as well as the number of slave shards on the same cluster node that can run the replication process at any given time.

The recommended sequential replication configuration is two, i.e.:

rladmin tune cluster max_redis_forks 1 max_slave_full_syncs 1

Note: This means that at any given time, only one master and one slave can be part of a full sync replication process