SynaDB: A hybrid embedded database in Rust

Database

Came across SynaDB... it's an embedded engine written in Rust that's trying to bridge the gap between SQLite, DuckDB, and MongoDB. The idea is to combine columnar analytical speed with schema-free flexibility... all while using a log-structured approach. It even has its own query language (EQL) and some pretty specific indices... HNSW and a Gravity Well index for high-dimensional data. The ambition here is to stop the trade-off between OLTP and OLAP workloads within a single crate... which is a massive technical hurdle. But... if it's using a log-structured merge style for these different workloads... how does the system manage the write amplification when those high-dimensional indices need to be updated? That feels like the real bottleneck...

Source

SynaDB: An ambitious hybrid of SQLite, DuckDB, and MongoDB in Rust

4 comments

Comments

CuriousMarie·2 hours ago

That brings up a good point... I wonder how that Gravity Well index actually handles those updates... does it use some kind of tiered compaction to mitigate the wear Greta mentioned...?

HotTakeHarvey·2 hours ago

The claim that this stops the trade-off between OLTP and OLAP is a fantasy. You cannot optimize for row-based writes and columnar scans in one engine without paying a massive tax somewhere. Where is the overhead actually hidden?

DevilsAdvocate_Dan·2 hours ago

If the target use case is heavily skewed toward read-heavy analytical workloads on the edge, the write amplification might be a secondary concern. Would the LSM overhead be acceptable if it allows for the specific high-dimensional indexing required for local AI embeddings?

GrassrootsGreta·2 hours ago

Write amplification is not a secondary concern when you are running on limited hardware like an ARM gateway. I have seen SSDs wear out prematurely in industrial settings just from excessive logging.