Prev Next

Database / Qdrant Vector DB Interview questions

What is the purpose of a payload index?

A payload index is a dedicated index built on a specific payload field, letting Qdrant quickly find points matching a filter condition on that field without scanning every point's payload for every query — the payload equivalent of the HNSW index's role for vector similarity.

client.create_payload_index(
    collection_name="documents",
    field_name="category",
    field_schema=models.PayloadSchemaType.KEYWORD,
)

Without an index, filtering on a field requires checking every point's payload individually to see whether it satisfies the filter condition, which becomes noticeably slower as a collection grows into the millions of points; with an index, Qdrant can directly look up which points satisfy a condition, similar in spirit to an index in a relational database.

Qdrant recommends indexing fields that are frequently filtered on and that meaningfully narrow down the result set — a highly selective field like a unique object ID benefits more from indexing than a low-cardinality field with only a handful of possible values — and it automatically factors available indexes and filter selectivity into how it plans and executes a query.

What does a payload index avoid needing to do for every filtered query?
What kind of field benefits most from indexing, according to Qdrant's guidance?

More Related questions...

What is Qdrant? What is the purpose of Qdrant? What are the key features of Qdrant? What is a collection in Qdrant? What is a point in Qdrant? What is a payload in Qdrant? What is the HNSW algorithm? What distance metrics does Qdrant support? How do you create a collection in Qdrant? How do you insert/upsert points into a collection? What is the difference between REST and gRPC APIs in Qdrant? What client libraries are available for Qdrant? What is payload filtering in Qdrant? Define scalar quantization in Qdrant? What is a segment in Qdrant? How do you perform a similarity search in Qdrant? What is the purpose of a payload index? List the supported field types for payload indexing? What is Qdrant Cloud? What is memmap storage in Qdrant? What is the difference between Qdrant and Pinecone? What is the difference between Qdrant and Weaviate? Why is Qdrant implemented in Rust? How does the HNSW graph work internally in Qdrant? What is the difference between scalar, binary, and product quantization? Explain the internal working of binary quantization and why it's fast? What is oversampling and rescoring in quantized search? How does Qdrant implement filtering during HNSW traversal? Explain the internal working of Qdrant's sharding and replication? What consensus protocol does Qdrant use for distributed clusters, and how does it work? What are named vectors, and when should you use them? What are sparse vectors in Qdrant? Explain hybrid search using the Query API and Prefetch? What is Reciprocal Rank Fusion (RRF) versus Distribution-Based Score Fusion (DBSF)? How do you implement multitenancy in Qdrant? Explain the lifecycle of a write operation in Qdrant (WAL, segments, optimizers)? What is the role of the Write-Ahead Log (WAL) in Qdrant? How do you take and restore snapshots in Qdrant? What is the difference between keeping vectors on-disk versus the HNSW index in RAM? Explain the execution flow of a filtered vector search query in Qdrant? When should you choose binary quantization versus scalar quantization? How do you optimize Qdrant for high-throughput production workloads? What is the ACORN-1 method, and why does it matter for filtered search? Explain the internal working of Qdrant's segment optimizer/merging? How do you implement multi-vector (late interaction / ColBERT-style) search in Qdrant? What is the role of the payload index in query planning? How does Qdrant handle consistency during a node failure? Explain the execution flow of a RAG pipeline built with Qdrant as the retrieval layer? What are the trade-offs of self-hosting Qdrant versus using Qdrant Cloud? What is Qdrant's Discovery/Recommendation API used for?
Show more question and Answers...


Comments & Discussions