What two API protocols does Qdrant expose, and when would you choose REST versus gRPC?
REST and gRPC
Qdrant exposes both REST and gRPC APIs. REST is usually the easiest starting point because HTTP requests are simple to inspect with browsers, curl, API clients, and standard observability tooling.
gRPC uses Protocol Buffers and a binary transport, and it can be preferable for service-to-service workloads where lower protocol overhead, streaming capabilities, or high-throughput typed clients matter. The practical latency difference depends on the workload, network, payload size, and client implementation, so I would benchmark rather than assume gRPC is always faster.
The trade-off is debugging and ecosystem simplicity versus transport efficiency and strongly typed service integration. I often start with REST for administration or simple application paths and use gRPC when profiling shows transport overhead matters.
One misconception is that switching from REST to gRPC automatically fixes vector-search latency. If HNSW traversal, filtering, disk I/O, or embedding generation dominates the request, changing the API protocol may have little effect.
- 1
REST is easy to inspect and broadly compatible
- 2
gRPC provides a binary, strongly typed service interface
- 3
Benchmark transport changes against the actual workload
- 4
API protocol is only one component of end-to-end search latency
Scenario Questions
0-2 years experience
- 1You need to inspect a Qdrant request manually while debugging an integration. Which protocol would you start with and why?
- 2A teammate claims gRPC is mandatory for production Qdrant. What evidence would you ask for?
2-5 years experience
- 1Your service handles thousands of vector requests per second and network overhead is becoming measurable. How would you benchmark REST against gRPC?
- 2An application switches to gRPC but p99 latency does not improve. Which other parts of the retrieval path would you profile?
5-8 years experience
- 1Your platform has Java, Go, Python, and JavaScript services communicating with Qdrant. How would you choose a default API protocol while preserving operational simplicity?
- 2A high-throughput ingestion pipeline is saturating CPU with serialization overhead. How would you determine whether gRPC is the right optimization?
8+ years experience
- 1You are defining a platform API standard for Qdrant across many teams. How would you decide when REST versus gRPC is appropriate?
- 2A latency SLO is measured end-to-end and Qdrant protocol overhead is only 2% of request time. Would you migrate to gRPC? Explain the engineering trade-off.
Follow-up Questions
- What kind of workload would make gRPC worth introducing?
- How would you determine whether REST is actually a latency bottleneck?