Vector Search easy
A startup proposes storing all product data only in a vector database and dropping its relational database to simplify the stack. What is the main problem with this plan?
- A Vector databases cannot join across collections, so product-to-order lookups need a second index.
- B The vector index is derived data; you still need a source of truth for rebuilds and transactions.
- C Similarity scores shift after every reindex, so product rankings would be unstable for users.
- D Most vector databases rebuild the whole index on each write, so frequent price updates would be slow.
Answer: B. The vector index is derived data; you still need a source of truth for rebuilds and transactions.
A vector index is **derived** data. Reindexing after a model change needs the original source data, and orders, prices, and inventory need transactional guarantees that a similarity index does not give. Joins and score drift are real limits, but they are workarounds away; the missing source of truth is not. Dropping the primary store is only acceptable for a prototype or for a corpus you can fully regenerate.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- You are building an IVF index on 20 million vectors. A teammate asks why the build needs a 'training' step before any vectors can be added. What is the reason?
- An HNSW index serves a marketplace where about 30% of listings are updated or removed each week. Each update is applied as a delete plus an insert, and deletes mark tombstones. Over a month, recall@10 drifts from 97% to 89% and p99 latency rises at the same ef_search. What is happening, and what should you do?
- Users delete documents in your primary database, but the deleted documents keep appearing in vector search results for hours. The vector index is rebuilt nightly. What is the best fix?
- Your embedding model outputs unnormalized vectors whose magnitudes vary widely. A teammate wants to switch the index from cosine similarity to dot product because it is cheaper to compute. What is the consequence?
- You are setting an SLO for a new ANN-backed recommendation search over 50 million vectors. A product manager asks for 100% recall@10 with single-digit millisecond latency. What is a realistic response?
- A B2B SaaS app serves 500 customers, and every query searches only one customer's documents. One large customer runs heavy batch queries that slow down search for everyone. Which sharding approach fits best?