GRPC Search API benchmark

[amberzsy]

Please describe the end goal of this project

based on [Feature] GRPC Document APIs #16784 implementation, perform benchmark on varies query types and search/indexSearch api with different scenarios.

Supporting References

#15190

Issues

• with [Feature] GRPC Search APIs #16783 PoC implementation, initial benchmark on TermQuery/MatchAllQuery with large payload for request and response.
• Evaluate/Research and benchmark Search grpc stream api.
• Benchmark KnnQuery with large dimension and vector sizes.

Related component

Search

[finnegancarroll]

Thanks for creating this issue! Do you think OSB can be leveraged for benchmarking or will a custom solution be needed? I notice both the existing queries and documents in OSB workloads are pretty simplistic so serialization/deserialization of the query itself will be minimal.

[amberzsy]

I notice both the existing queries and documents in OSB workloads are pretty simplistic so serialization/deserialization of the query itself will be minimal.

correct. i think it still have some value to have some basic testing or validation. but to see the real performance gain, will need more complicated/prod-like test case.

Thanks for creating this issue! Do you think OSB can be leveraged for benchmarking or will a custom solution be needed?

for short term, we will do a custom solution which leverage Uber internal benchmark framework and tooling to generate large payload(request and response) with high qps.
for long term, imo, we should add have grpc support in OSB and add more benchmark use cases. which i created issue opensearch-project/opensearch-benchmark#702 for tracking.

[amberzsy]

assigned to @amberzsy