Mad Science
summary
Production geospatial service uses LevelDB ecosystem (levelup/leveldown) plus a level-mysql fork and DynamoDB for horizontal partitioning, adds a buffered bulk-import mode and in-memory batching to avoid compaction stalls, and encodes locations with a quadkey quadtree prefix to partition by precision. Naive MySQL-style writes jump from ~100 TPS to about 800 TPS after changes, peak raw transactions hit ~1,600, and point/radius queries return in roughly 20 ms per block.
AI-written summary · about
