Internal Architecture Interview Questions | Qdrant | NextRound

1In a replicated cluster, what happens to search results if a query is served while one replica of a shard is temporarily out of sync?Advanced

2Walk through what happens internally, at a high level, when a client sends a filtered vector search request to a distributed Qdrant cluster.Advanced

3Why does Qdrant merge and re-rank results from multiple shards rather than simply concatenating each shard's top-k?Advanced

4How does a payload filter interact with segment selection during query execution - does Qdrant always scan every segment?Advanced

5What is the performance implication of running a query that touches every named vector on a point versus one that specifies using?Advanced

6What are Qdrant's tunable read/write consistency levels for distributed operations, and what trade-off do they represent?Advanced

7What is a segment in Qdrant's storage engine, and why does a collection consist of multiple segments rather than one monolithic index?Advanced

8What role does the Write-Ahead Log (WAL) play in Qdrant, and what failure scenario does it protect against?Advanced

9How does memory-mapped (mmap) storage allow Qdrant to serve a collection larger than available RAM?Advanced

10What does the background optimizer do in Qdrant, and why can too-aggressive optimization affect query latency?Advanced

11What is the practical difference between storing vectors in-memory versus on-disk in a collection configuration, and when would you choose on-disk?Advanced

12What is sharding in a distributed Qdrant cluster, and what determines which shard a given point is written to by default?Advanced

13What is custom sharding, and how does it change the way multitenant data is distributed across a cluster?Advanced

14What consensus algorithm does Qdrant use to keep cluster metadata consistent across nodes, and what does it coordinate?Advanced