sharding versus replication, which one should you use to scale?

Hey everyone, I've been thinking about how we can scale as more people use our app. Our current single-node relational database is starting to struggle. I know sharding and replication are the usual answers, but I'm trying to figure out the best way to choose between them. When should we split our data across multiple databases (sharding), and when should we just copy the data (replication)? Does it depend on how we query the data or what we need it for? I'm really interested in what you all have tried and what you think is most important when making this call.