Vector Search easy
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?
- A Propagate deletes from the source to the index promptly, e.g., via change data capture events.
- B Switch the index from HNSW to IVF, since IVF automatically drops vectors removed from the source.
- C Increase the recall target so the index finds the newer documents ahead of the deleted ones.
- D Move the vector index into memory so that it reads the latest data directly from the database.
Answer: A. Propagate deletes from the source to the index promptly, e.g., via change data capture events.
The index is **derived** from the source, so it stays stale until changes reach it. Streaming deletes through change data capture keeps it fresh but adds pipeline complexity, and many indexes store deletes as tombstones that need periodic compaction. No index type watches the source on its own, so switching to IVF does nothing. Nightly rebuilds are fine for static corpora where hours of staleness are acceptable.
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?
- 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?
- Your HNSW index (M=16) reaches recall@10 of 91% at ef_search=64 with a p99 of 6 ms. The target is 97% recall within a 15 ms p99. A full rebuild takes 14 hours. What is the right tuning order?