At scale, API communication overhead from serialization and network transfers significantly impacts OpenSearch performance.
Protobuf's compact binary format uses numerical identifiers to eliminate the repeated field names found in JSON, drastically reducing payload size.
Protobuf is a serialization format that can be used with transport protocols like REST or gRPC, which combines it with HTTP/2 for efficiency.
OpenSearch built native gRPC support and created an automated conversion pipeline to keep Protobuf definitions synchronized with OpenAPI specifications.
gRPC support for bulk and KNN search is available in OpenSearch 3.2 and can be enabled by adding a few lines to the configuration YAML file.
Official benchmarks show gRPC significantly improves throughput for KNN vector search and reduces latency by over 22% for bulk ingestion tasks.
Performance improvements come from Protobuf's compact payload reducing network transfer time and HTTP/2's multiplexing allowing for concurrent requests over a single connection.
Use gRPC for high-performance internal service-to-service communication, but stick with REST for simplicity, debuggability, and broader ecosystem compatibility.
The OpenSearch project actively seeks contributors and maintainers, encouraging community involvement and feedback on new features like gRPC.
Jobs that call for the skills explored in this talk.