when should you think about getting a database read replica? like, how do you know its time?

hey folks, ive been running a pretty successful app for a while now, and were starting to see some performance hiccups during peak hours. our primary database is feeling the strain, specifically on read operations. were not at a point where were looking at full sharding, but the thought of a read replica has crossed my mind. what are your experiences with introducing read replicas? what kind of read/write ratios or latency issues usually signal its time to go down this path? any gotchas to watch out for, especially around replication lag