profit/golangLAKEHOUSE
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

vectord: BatchAdd — single-lock variadic batch (Option A)

Browse Source 
Replaces the per-item Add loop in the HTTP handler with one call to
Index.BatchAdd, which acquires the write-lock once and pushes the
whole batch through coder/hnsw's variadic Graph.Add. Pre-validation
stays in the handler so per-item error messages keep their item-index
precision.

Microbench (internal/vectord/batch_bench_test.go) at d=768 cosine:

  N=16    SingleAdd 283µs/op  →  BatchAdd 170µs/op   1.66×
  N=128   SingleAdd 7.9ms/op  →  BatchAdd 7.5ms/op   1.05×
  N=1024  SingleAdd 87.5ms/op →  BatchAdd 83.4ms/op  1.05×

Win is biggest at staffing-driver batch sizes (N=16) where
per-call lock + validation overhead is a meaningful fraction. At
larger N the inner HNSW neighborhood search per insert dominates,
which is the load-bearing finding for Option B (sharded indexes):
the throughput ceiling lives inside the library, not at the lock,
so sharding to N parallel Graphs is the only path to true
concurrent-Add throughput.

g1, g1p, g2 smokes all PASS post-change.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
root
2026-04-29 18:05:48 -05:00
co-authored by Claude Opus 4.7
parent afbb506dbc
commit f1c188323c
3 changed files with 140 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
cmd/vectord/main.go
+11 -14
View File
@@ -274,21 +274,18 @@ func (h *handlers) handleAdd(w http.ResponseWriter, r *http.Request) {
return
}
}
// Pre-validation above is exhaustive (id, dim, finite, zero-norm),
// so BatchAdd takes the write-lock once and pushes the whole batch
// into coder/hnsw via one variadic Graph.Add. Saves N-1 lock
// acquisitions per HTTP batch.
batch := make([]vectord.BatchItem, len(req.Items))
for j, it := range req.Items {
if err := idx.Add(it.ID, it.Vector, it.Metadata); err != nil {
// Vector-validation errors (NaN/Inf, zero-norm under
// cosine) only surface here; pre-validation is intentional
// minimal scope (id + dim only).
if errors.Is(err, vectord.ErrDimensionMismatch) ||
strings.Contains(err.Error(), "non-finite") ||
strings.Contains(err.Error(), "zero-norm") {
http.Error(w, "items["+strconv.Itoa(j)+"]: "+err.Error(), http.StatusBadRequest)
return
}
slog.Error("add", "name", name, "id", it.ID, "err", err)
http.Error(w, "internal", http.StatusInternalServerError)
return
}
batch[j] = vectord.BatchItem{ID: it.ID, Vector: it.Vector, Metadata: it.Metadata}
}
if err := idx.BatchAdd(batch); err != nil {
slog.Error("batch add", "name", name, "err", err)
http.Error(w, "internal", http.StatusInternalServerError)
return
}
// One save per batch (post-loop), not per item. Per scrum
// O-W4-style discipline: HTTP-batch boundary is the natural unit.