Load a production-shaped subset of relationships into Dgraph and write the traversals your application actually needs. Include high-degree nodes and filtering that can expose poor query plans. Compare correctness and latency with the current data store under concurrent reads and writes.
Graph modeling and operations matter as much as query syntax. Rehearse schema evolution, access control, backup, and restore. Choose a graph database only when relationships are central enough to justify a different storage and operational model.
Akousa