weaviate/weaviate · error

failed to upsert new centroid %d

Error message

failed to upsert new centroid %d

What it means

When the first vector finds no candidate postings, HFresh creates an initial posting: it allocates a posting ID and inserts a Centroid (the vector itself) into the Centroids index. This error wraps a failure of h.Centroids.Insert for the new posting ID.

Source

Thrown at adapters/repos/db/vector/hfresh/insert.go:191

	targets, _, err := h.RNGSelect(v, 0)
	if err != nil {
		return nil, err
	}

	// if no postings were found, create a new posting while holding the lock
	if targets.Len() == 0 {
		postingID, err := h.IDs.Next()
		if err != nil {
			return nil, err
		}
		// use the vector as the centroid and register it in the SPTAG
		err = h.Centroids.Insert(postingID, &Centroid{
			Uncompressed: v,
			Compressed:   compressed,
			Deleted:      false,
		})
		if err != nil {
			return nil, errors.Wrapf(err, "failed to upsert new centroid %d", postingID)
		}
		// return the new posting ID
		targets = NewResultSet(1)
		targets.data = append(targets.data, Result{ID: postingID, Distance: 0})
	}

	return targets, nil
}

// Append adds a vector to the specified posting.
// It returns true if the vector was successfully added, false if the posting no longer exists.
// It is called synchronously during imports but also asynchronously by reassign operations.
func (h *HFresh) append(ctx context.Context, vector Vector, centroidID uint64, reassigned bool) (bool, error) {
	h.postingLocks.Lock(centroidID)

	// check if the posting still exists
	if !h.Centroids.Exists(centroidID) {
		// the posting might have been deleted concurrently,

View on GitHub (pinned to 75aa4b6d11)

Solutions

1. Inspect the wrapped Centroids.Insert error for the storage-level cause.
2. Check disk space and store health for the shard.
3. Retry the insert; ensureInitialPosting re-checks for concurrent postings under the lock before creating one.
4. If the centroid store is corrupted, restore the shard from backup.

if err := idx.Add(ctx, id, vec); err != nil && strings.Contains(err.Error(), "failed to upsert new centroid") {
    // first-posting creation failed at storage layer; check disk and retry
}

When it happens

Trigger: Inserting into an empty (or fully-deleted) HFresh index via Add/AddMulti where the centroid insert into the Centroids structure fails for the freshly allocated postingID.

Common situations: Underlying storage write failure (disk full, I/O error); ID generator (h.IDs.Next) succeeded but centroid store is corrupted or unavailable; concurrent index shutdown.

Related errors

• could not persist dimensions
• could not persist RQ data
• store muvera data in index metadata
• put muvera vector into bucket
• failed to get candidate centroid

AI-assisted analysis of weaviate/weaviate@75aa4b6d11 (2026-09-04). Data as JSON: /api/errors/a2ba099e4b1903b1. Report an issue: GitHub.