Row Cache Invalidation and Tunable Consistency Alignment

Row Cache Invalidation with Tunable Consistency

Cassandra row cache keeps frequently accessed partitions in memory to reduce disk I/O. Invalidation is handled locally on the node that receives the write, and a QUORUM write does not trigger a cluster-wide invalidation of row caches on all replicas.

A subsequent read at consistency ONE can be routed to a replica whose local row cache still holds the previous value, resulting in a stale read after a successful QUORUM write. There is no native mechanism to force a synchronous, global cache purge for a specific partition key during a write operation.

• What configuration controls remote row cache invalidation for a partition key after a QUORUM write?
• Can consistency level be aligned with row cache behavior to prevent stale reads without disabling the cache or using ALL for reads?