
Rajat Jain developed a configurable blob compression feature for the facebook/rocksdb repository, enabling users to fine-tune compression settings for blob files through a new blob_compression_opts field. He integrated this option into AdvancedColumnFamilyOptions and MutableCFOptions, ensuring it could be adjusted at runtime via SetOptions. The implementation, written in C++, focused on database optimization and performance tuning, allowing for workload-driven adjustments such as ZSTD level selection to reduce CPU usage and L3 cache pressure. By porting and wiring this feature, Rajat delivered a flexible, production-safe solution that improves throughput and latency for flush-heavy blob workloads without introducing regressions.
April 2026: Delivered Blob Compression Options for Blob Files in RocksDB, enabling configurable compression for blob data via a new blob_compression_opts field. The option is added to AdvancedColumnFamilyOptions and MutableCFOptions and wired through BlobFileBuilder::GetCompressor, with mutable exposure and runtime mutability via SetOptions. Default behavior remains dFast to preserve stability. This work ported from the external PR D98326754 (PR #14568) and integrated under differential revision D99478562 (reviewed by xingbowang). Technical impact includes enabling workload-driven tuning (e.g., ZSTD level=1 with fast path) to reduce CPU usage and L3 cache pressure on flush-heavy workloads. Business value centers on higher blob workload throughput, lower latency, and flexible, safe configuration for production deployments.
April 2026: Delivered Blob Compression Options for Blob Files in RocksDB, enabling configurable compression for blob data via a new blob_compression_opts field. The option is added to AdvancedColumnFamilyOptions and MutableCFOptions and wired through BlobFileBuilder::GetCompressor, with mutable exposure and runtime mutability via SetOptions. Default behavior remains dFast to preserve stability. This work ported from the external PR D98326754 (PR #14568) and integrated under differential revision D99478562 (reviewed by xingbowang). Technical impact includes enabling workload-driven tuning (e.g., ZSTD level=1 with fast path) to reduce CPU usage and L3 cache pressure on flush-heavy workloads. Business value centers on higher blob workload throughput, lower latency, and flexible, safe configuration for production deployments.

Overview of all repositories you've contributed to across your timeline