Garnet speaks the RESP protocol, but it is not guaranteed to behave exactly like Redis for every command or atomicity assumption. Capture the commands your application really sends and replay them against a test Garnet instance. Check transactions, scripts, expiration, pub/sub, and failure recovery where relevant.
Measure latency and throughput under your own key sizes and concurrency. Practice persistence and restore before using it for anything beyond disposable cache data. A compatibility claim is a starting point; your workload and consistency expectations decide the migration.
Akousa