mongodb-mongoose-patterns

Cursor-based pagination filters records with `createdAt: { $lt: cursor }` and sorts descending, fetching limit+1 items to detect hasMore. This avoids the O(N) scan cost of `skip()` and works with a compound index on `{ conversationId: 1, createdAt: -1 }`.

Embed data that is 1-to-few, queried together most of the time, and bounded under the 16MB document limit, such as message attachments. Reference data that is 1-to-many or unbounded, like messages in a conversation, and index the reference fields.

The `.lean()` method returns plain JavaScript objects instead of hydrated Mongoose documents, skipping change tracking and getters. For read-only GET endpoints this is roughly 5x to 10x faster and uses significantly less memory.

Create a TTL index with `schema.index({ createdAt: 1 }, { expireAfterSeconds: 3600 })` so MongoDB automatically deletes documents after the specified time. This suits OTPs, password reset tokens, and cache entries.

The `skip()` operation forces MongoDB to scan and discard every preceding document, giving O(N) cost that grows with page depth. Keyset pagination using an indexed cursor field keeps lookups at O(log N) regardless of position.