This page provides an overview of persistence for Memorystore for Redis Cluster.
Memorystore for Redis Cluster supports two types of persistence:
For the best instance availability and data durability, we recommend enabling both persistence and high availability. While persistence and high availability overlap in the protection they offer, they have different strengths that complement each other. HA can be considered the first line of defense against individual node failures and zonal outages. Persistence provides disaster recovery for rare events when all nodes in a shard fail and HA is unable to help.
Choosing a persistence type
Choosing between AOF and RDB persistence comes down to a choice between performance and data durability. These tradeoffs are generally true of all managed Redis services because of the nature of AOF and RDB persistence in OSS Redis.
If your top priority is data durability and preserving the highest percentage of writes sent to your Redis server, we recommend choosing AOF persistence. AOF's ability to persist data every second (or for every write) offers superior data durability compared to RDB snapshots which save data on an hourly basis.
If performance is your top priority, we recommend using RDB persistence because it usually places lower performance demands on your instance than AOF persistence does for a similarly sized instance. If you are comfortable with some staleness in your data on recovery and you want the performance that RDB persistence provides for a similarly sized instance, then RDB persistence is the better choice for you.
It is possible to have both high throughput with the durability of AOF persistence. To get both you can enable AOF persistence and also choose a high shard count. A high shard count provides more vCPUs which improves performance. However, some latency increase is expected for AOF in any configuration.