Lakegres brings vector search straight to your existing Hudi and Iceberg tables
Lakegres: Vector search on existing Open/Hudi/Iceberg tables
Onehouse Lakegres now runs vector search directly on existing and new Hudi and Iceberg tables, with the index stored as derived metadata inside the open table itself, so there's no need to copy data into a separate vector database. The index groups vectors into clusters, keeps compact quantized codes, and updates with the same commits that write the table. On 247 million vectors, warm queries hit a 63.5 ms median, roughly 10 times faster than Databricks AI Search.
On 247 million vectors, warm queries return in 63.5 ms at the median. Lakegres is about 10 times faster than Databricks AI Search at the warm median.
- sherlockbeard
any measured commit latency overhead from the vector index on a streaming job, once it's built? post benchmarkss build time + query latency but not the write path stuff.
- rjshmhndr
vector search with < 100msecs on a lake is unheard of - pretty impressive
- jauntywundrkind
Pretty good write up, talking to the problem, surveying what and how other players do things?
Went looking to see if there was going to be anything open (mostly it's product pitch), found:
> Iceberg support builds on the Hudi-backed indexing work in Apache XTable (Incubating). We developed the design in the open with engineers from Uber and Walmart to support their planet-scale tables. The open RFC documents the storage layout, query path and consistency model.
> The design is settled and ships in Lakegres today; the Hudi implementation is being upstreamed and XTable is in release-candidate testing ahead of OSS releases that other engines can integrate.
With two links inline, https://github.com/apache/incubator-xtable/pull/935 https://github.com/apache/hudi/blob/master/rfc/rfc-109/rfc-1...