ArangoDB may hold document and graph workloads in the same system. A replacement that handles one side well can still make the application slower or more complex. Build a staging dataset with realistic vertex and edge counts, document shapes, and indexes. Replay representative reads and writes while monitoring latency and resource use.
Check transactional boundaries, query semantics, backups, and how permission rules move. Export and re-import enough data to expose type or relationship mismatches. Include a failure-and-restore drill, since a fast query benchmark says little about recoverability. Select an alternative only after the application behavior—not just a schema diagram—matches.
Akousa