Searching a large collection

I start with the real query and its execution plan. The index should support its filters and sorting.

db.orders.createIndex({ customerId: 1, createdAt: -1 });

db.orders.find({ customerId: 'u1' })
  .sort({ createdAt: -1 }).limit(20)
  .explain('executionStats');

I check: the number of returned results versus documents and keys examined, collection scans and the cost of sorting.

In practice: I limit the returned fields and use sensible pagination; with large offsets I consider a cursor on an indexed key.

Pitfall: indexes cost memory, space and write time. I don't create an index on every field, nor assume every regex will use one efficiently.