pub const EMBEDDINGS: usize = 30;Expand description
Embedding vectors (upsert_embeddings).
The smallest by a wide margin, because DiskANN index maintenance makes an
embedding the most expensive row in the system. That cost grows with the
corpus, not the chunk (D-059): a fixed 30-vector chunk costs 49 µs per
vector into an empty corpus and 224 µs into an 8,000-vector one. Graph
insertion getting dearer as the graph grows is what DiskANN is, so unlike
EDGES there is nothing here to fix — but it does mean this size buys
latency at some throughput, not for free.