Database / Milvus Vector database Interview questions
Why use Partitions instead of separate Collections for data isolation?
Both provide a way to logically separate data, but they carry different overhead. A Collection has its own independent schema, indexes, and load/release lifecycle; creating many Collections means managing many independent sets of that metadata and configuration. A Partition shares its parent Collection's schema and index configuration, adding a much lighter-weight subdivision within a single, shared metadata and index setup.
| Partitions | Separate Collections |
| Share one schema and index configuration. | Each has its own independent schema and indexes. |
| Lower metadata and management overhead at scale. | Higher overhead when managing large numbers of tenants/groups. |
| Can be selectively loaded/released together with siblings or independently. | Loaded/released fully independently, with no shared configuration. |
| Best when all groups share the same data shape. | Best when groups genuinely need different schemas. |
The practical rule of thumb: reach for Partitions when isolating groups of data that share the same structure (most multi-tenant SaaS cases), and reach for separate Collections specifically when different groups genuinely need different schemas or fully independent index configurations, since that's the scenario Partitions can't accommodate.
More Related questions...
