Part 2 of the OpenObserve performance engineering series. A transposed bloom filter layer cuts random trace_id lookups from 2,584ms to 89ms by collapsing 170 S3 round trips into a single 5,440-byte row read.
Same ~2TB of data, same count query, same querier config — two parameter changes took a Tantivy query from 49 seconds to 2 seconds. Learn how raising compact file size and enabling footer cache drove a 25× speedup by slashing S3 requests from 10,000+ to ~600.
