weaviate/weaviate · error

dist between %d and %d

Error message

dist between %d and %d

What it means

Wraps a distance-calculation error between the inserted node and one of its neighbors inside connectNeighborAtLevel. Deleted objects (storobj.ErrNotFound) are handled gracefully; any other distancer error is wrapped here and aborts neighbor connection for that node.

Source

Thrown at adapters/repos/db/vector/hnsw/neighbor_connections.go:366

		// we can simply append
		// updatedConnections = append(currentConnections, n.node.id)
		neighbor.appendConnectionAtLevelNoLock(level, n.node.id, maximumConnections)
		if err := n.graph.commitLog.AddLinkAtLevel(neighbor.id, level, n.node.id); err != nil {
			return err
		}
	} else {
		// we need to run the heuristic

		dist, err := n.graph.distBetweenNodes(n.node.id, neighborID)
		if err != nil {
			var e storobj.ErrNotFound
			if errors.As(err, &e) {
				n.graph.handleDeletedNode(e.DocID, "connectNeighborAtLevel")
				// it seems either the node or the neighbor were deleted in the meantime,
				// there is nothing we can do now
				return nil
			}
			return errors.Wrapf(err, "dist between %d and %d", n.node.id, neighborID)
		}

		candidates := priorityqueue.NewMax[any](len(currentConnections) + 1)
		candidates.Insert(n.node.id, dist)

		for _, existingConnection := range currentConnections {
			dist, err := n.graph.distBetweenNodes(existingConnection, neighborID)
			if err != nil {
				var e storobj.ErrNotFound
				if errors.As(err, &e) {
					n.graph.handleDeletedNode(e.DocID, "connectNeighborAtLevel")
					// was deleted in the meantime
					continue
				}
				return errors.Wrapf(err, "dist between %d and %d", existingConnection, neighborID)
			}

			candidates.Insert(existingConnection, dist)

View on GitHub (pinned to 75aa4b6d11)

Solutions

1. Identify the object IDs from the wrapped error and validate/repair their vectors (re-export and re-import the affected objects)
2. Check for NaN/Inf values in vectors supplied at import time and fix client-side normalization
3. Restore the shard from backup if vectors on disk are corrupt
4. Keep weaviate upgraded — several distancer error-handling fixes landed in later versions

// validate vectors client-side before batch import
for _, o := range objects {
  if len(o.Vector) != expectedDim { return fmt.Errorf("dim mismatch for %s", o.ID) }
  for _, x := range o.Vector { if math.IsNaN(float64(x)) || math.IsInf(float64(x), 0) { return fmt.Errorf("non-finite vector for %s", o.ID) } }
}

func validVector(v []float32, dim int) bool {
  if len(v) != dim { return false }
  for _, x := range v { if math.IsNaN(float64(x)) || math.IsInf(float64(x), 0) { return false } }
  return true
}

When it happens

Trigger: connectNeighborAtLevel calls the distancer for (node, neighborID) and gets a non-ErrNotFound error — e.g. the neighbor object exists but its vector is unreadable/corrupt, or a quantization/PQ codec errors on the stored bytes.

Common situations: Corrupted or truncated vector data after an unclean shutdown; wrong distance metric configured for imported vectors (NaN/Inf vectors); module-serialized vectors failing to deserialize.

Related errors

• calculate distance of current last result
• calculate distance between candidate and query
• calculated distance between worst result and query
• index is not hnsw
• at least one iterator required

AI-assisted analysis of weaviate/weaviate@75aa4b6d11 (2026-09-04). Data as JSON: /api/errors/dd5bb46adeded6c5. Report an issue: GitHub.