Tair (Redis OSS-compatible) uses asynchronous replication by default: the master node sends write operations to replica nodes in the background without waiting for each replica to acknowledge receipt. Replica nodes are eventually consistent with the master — at any given moment, a replica may not reflect the most recent writes.
How replication works
When a write completes on the master, the master immediately returns a success response to the client, then forwards the operation to replica nodes asynchronously. Until a replica applies that operation, reads from that replica return the previous value.
This design prioritizes write performance over read consistency — asynchronous replication keeps write latency low and throughput high.
Causes of replication lag
Two conditions increase replication lag:
-
High write throughput on the master node: When the master processes write operations faster than replicas can apply them, replicas fall behind. The higher the write volume, the larger the lag window.
-
Network latency between nodes: Replication depends on the network path between master and replica nodes. Elevated network latency delays when replicas receive and apply changes from the master.
Applicable scope
-
Tair