fix: fix missing multivector placeholder in local mode (#1399)

* fix: fix missing multivector placeholder in local mode

* fix: do not use deleted vectors in recommend, etc

* fix: fix mypy complaints in point id vector resolution

* fix: regen async

qdrant_client/local/async_qdrant_local.py

@@ -237,14 +237,6 @@ class AsyncQdrantLocal(AsyncQdrantBase):
             if lookup_from is not None and lookup_from.vector is not None
             else search_in_vector_name
         )
-        sparse = vector_name in collection.sparse_vectors
-        multi = vector_name in collection.multivectors
-        if sparse:
-            collection_vectors = collection.sparse_vectors
-        elif multi:
-            collection_vectors = collection.multivectors
-        else:
-            collection_vectors = collection.vectors
         mentioned_ids: set[types.PointId] = set()
 
         def input_into_vector(vector_input: types.VectorInput) -> types.VectorInput:
@@ -252,15 +244,7 @@ class AsyncQdrantLocal(AsyncQdrantBase):
                 if isinstance(vector_input, uuid.UUID):
                     vector_input = str(vector_input)
                 point_id = vector_input
-                if point_id not in collection.ids:
-                    raise ValueError(f"Point {point_id} is not found in the collection")
-                idx = collection.ids[point_id]
-                if vector_name in collection_vectors:
-                    vec = collection_vectors[vector_name][idx]
-                else:
-                    raise ValueError(f"Vector {vector_name} not found")
-                if isinstance(vec, np.ndarray):
-                    vec = vec.tolist()
+                vec = collection._vector_by_point_id(vector_name, point_id)
                 if collection_name == lookup_collection_name:
                     mentioned_ids.add(point_id)
                 return vec

qdrant_client/local/local_collection.py

@@ -307,7 +307,9 @@ class LocalCollection:
                     if v is not None:
                         multivectors[name].append(v)
                     else:
-                        multivectors[name].append(np.array([]))
+                        multivectors[name].append(
+                            np.ones((1, self.multivectors_config[name].size), dtype=np.float32)
+                        )
                         deleted_ids.append((idx, name))
 
             # setup dense vectors by name
@@ -1365,14 +1367,7 @@ class LocalCollection:
         ) -> None:
             for example in examples:
                 if isinstance(example, get_args(types.PointId)):
-                    if example not in collection.ids:
-                        raise ValueError(f"Point {example} is not found in the collection")
-
-                    idx = collection.ids[example]
-                    vec = collection_vectors[vector_name][idx]
-
-                    if isinstance(vec, np.ndarray):
-                        vec = vec.tolist()
+                    vec: Any = collection._vector_by_point_id(vector_name, example)
                     acc.append(vec)
                     if collection == self:
                         mentioned_ids.append(example)
@@ -1413,15 +1408,12 @@ class LocalCollection:
         sparse = vector_name in collection.sparse_vectors
         multi = vector_name in collection.multivectors
         if sparse:
-            collection_vectors = collection.sparse_vectors
             examples_into_vectors(positive, sparse_positive_vectors)
             examples_into_vectors(negative, sparse_negative_vectors)
         elif multi:
-            collection_vectors = collection.multivectors
             examples_into_vectors(positive, positive_multivectors)
             examples_into_vectors(negative, negative_multivectors)
         else:
-            collection_vectors = collection.vectors
             examples_into_vectors(positive, positive_vectors)
             examples_into_vectors(negative, negative_vectors)
 
@@ -1766,22 +1758,35 @@ class LocalCollection:
 
         return ids, scores
 
+    def _vector_by_point_id(
+        self, vector_name: str, point_id: types.PointId
+    ) -> list[float] | SparseVector | list[list[float]]:
+        if point_id not in self.ids:
+            raise ValueError(f"Point {point_id} is not found in the collection")
+
+        idx = self.ids[point_id]
+        if vector_name in self.vectors:
+            vector = self.vectors[vector_name][idx].tolist()
+        elif vector_name in self.sparse_vectors:
+            vector = self.sparse_vectors[vector_name][idx]
+        elif vector_name in self.multivectors:
+            vector = self.multivectors[vector_name][idx].tolist()
+        else:
+            raise ValueError(f"Vector {vector_name} not found")
+
+        # Absent vectors are kept as placeholders to keep the storage dense, they are
+        # filtered out of search results by the mask and must not be used as query vectors
+        if self.deleted_per_vector[vector_name][idx]:
+            raise ValueError(f"Vector with name {vector_name} for point {point_id} not found")
+
+        return vector
+
     @staticmethod
     def _preprocess_vector_input(
         target: models.VectorInput | None, collection: "LocalCollection", vector_name: str
     ) -> tuple[models.Vector, types.PointId | None]:
         if isinstance(target, get_args(types.PointId)):
-            if target not in collection.ids:
-                raise ValueError(f"Point {target} is not found in the collection")
-
-            idx = collection.ids[target]
-            if vector_name in collection.vectors:
-                target_vector = collection.vectors[vector_name][idx].tolist()
-            elif vector_name in collection.sparse_vectors:
-                target_vector = collection.sparse_vectors[vector_name][idx]
-            else:
-                target_vector = collection.multivectors[vector_name][idx].tolist()
-
+            target_vector = collection._vector_by_point_id(vector_name, target)
             return target_vector, target
 
         return target, None
@@ -1797,19 +1802,11 @@ class LocalCollection:
         multi_context_vectors = []
 
         for pair in context:
-            pair_vectors = []
+            # holds a dense, sparse or multi vector, dispatched on by type below
+            pair_vectors: list[Any] = []
             for example in [pair.positive, pair.negative]:
                 if isinstance(example, get_args(types.PointId)):
-                    if example not in collection.ids:
-                        raise ValueError(f"Point {example} is not found in the collection")
-
-                    idx = collection.ids[example]
-                    if vector_name in collection.vectors:
-                        vector = collection.vectors[vector_name][idx].tolist()
-                    elif vector_name in collection.sparse_vectors:
-                        vector = collection.sparse_vectors[vector_name][idx]
-                    else:
-                        vector = collection.multivectors[vector_name][idx].tolist()
+                    vector = collection._vector_by_point_id(vector_name, example)
 
                     pair_vectors.append(vector)
                     if collection == self:
@@ -2554,11 +2551,15 @@ class LocalCollection:
             if len(named_vectors) <= idx:
                 diff = idx - len(named_vectors) + 1
                 for _ in range(diff):
-                    named_vectors.append(np.array([]))
+                    named_vectors.append(
+                        np.ones((1, self.get_vector_params(vector_name).size), dtype=np.float32)
+                    )
 
             if vector is None:
                 # Add fake vector and mark as removed
-                named_vectors[idx] = np.array([])
+                named_vectors[idx] = np.ones(
+                    (1, self.get_vector_params(vector_name).size), dtype=np.float32
+                )
                 self.deleted_per_vector[vector_name] = np.append(
                     self.deleted_per_vector[vector_name], 1
                 )
@@ -2965,7 +2966,9 @@ class LocalCollection:
 
         if config.multivector_config is not None:
             self.multivectors_config[vector_name] = params
-            self.multivectors[vector_name] = [np.array([]) for _ in range(num_points)]
+            self.multivectors[vector_name] = [
+                np.ones((1, config.size), dtype=np.float32) for _ in range(num_points)
+            ]
         else:
             self.vectors_config[vector_name] = params
             self.vectors[vector_name] = np.zeros((num_points, config.size), dtype=np.float32)

qdrant_client/local/qdrant_local.py

@@ -257,15 +257,6 @@ class QdrantLocal(QdrantBase):
             else search_in_vector_name
         )
 
-        sparse = vector_name in collection.sparse_vectors
-        multi = vector_name in collection.multivectors
-        if sparse:
-            collection_vectors = collection.sparse_vectors
-        elif multi:
-            collection_vectors = collection.multivectors
-        else:
-            collection_vectors = collection.vectors
-
         # mentioned ids in the search collection which should be excluded from search
         mentioned_ids: set[types.PointId] = set()
 
@@ -276,16 +267,7 @@ class QdrantLocal(QdrantBase):
                 if isinstance(vector_input, uuid.UUID):
                     vector_input = str(vector_input)
                 point_id = vector_input  # rename for clarity
-                if point_id not in collection.ids:
-                    raise ValueError(f"Point {point_id} is not found in the collection")
-
-                idx = collection.ids[point_id]
-                if vector_name in collection_vectors:
-                    vec = collection_vectors[vector_name][idx]
-                else:
-                    raise ValueError(f"Vector {vector_name} not found")
-                if isinstance(vec, np.ndarray):
-                    vec = vec.tolist()
+                vec = collection._vector_by_point_id(vector_name, point_id)
                 if collection_name == lookup_collection_name:
                     mentioned_ids.add(point_id)
                 return vec

tests/congruence_tests/test_multivector_search_queries.py

@@ -157,6 +157,22 @@ def test_simple():
     compare_client_results(local_client, remote_client, searcher.simple_search_code)
 
 
+def test_simple_opt_vectors_search():
+    fixture_points = generate_multivector_fixtures(100, skip_vectors=True)
+
+    searcher = TestSimpleSearcher()
+
+    local_client = init_local()
+    init_client(local_client, fixture_points, vectors_config=multi_vector_config)
+
+    remote_client = init_remote()
+    init_client(remote_client, fixture_points, vectors_config=multi_vector_config)
+
+    compare_client_results(local_client, remote_client, searcher.simple_search_text)
+    compare_client_results(local_client, remote_client, searcher.simple_search_image)
+    compare_client_results(local_client, remote_client, searcher.simple_search_code)
+
+
 def test_mmr():
     fixture_points = generate_multivector_fixtures(10)
 

tests/congruence_tests/test_optional_vectors.py

@@ -1,5 +1,7 @@
 import numpy as np
+import pytest
 
+from qdrant_client.http.exceptions import UnexpectedResponse
 from qdrant_client.http.models import models
 from tests.congruence_tests.test_common import (
     COLLECTION_NAME,
@@ -169,3 +171,64 @@ def test_simple_opt_sparse_vectors_search():
             key=lambda x: x.id,
         ),
     )
+
+
+def test_point_id_input_with_missing_vector():
+    vectors_config = {
+        "dense": models.VectorParams(size=4, distance=models.Distance.COSINE),
+        "multi": models.VectorParams(
+            size=4,
+            distance=models.Distance.COSINE,
+            multivector_config=models.MultiVectorConfig(
+                comparator=models.MultiVectorComparator.MAX_SIM
+            ),
+        ),
+    }
+    # point 1 has no value for either named vector, it can neither be scored nor be used as a query
+    points = [
+        models.PointStruct(id=1, vector={}),
+        models.PointStruct(
+            id=2, vector={"dense": [1.0, 0.0, 0.0, 0.0], "multi": [[1.0, 0.0, 0.0, 0.0]]}
+        ),
+        models.PointStruct(
+            id=3, vector={"dense": [0.0, 1.0, 0.0, 0.0], "multi": [[0.0, 1.0, 0.0, 0.0]]}
+        ),
+    ]
+
+    local_client = init_local()
+    init_client(local_client, points, vectors_config=vectors_config)
+
+    remote_client = init_remote()
+    init_client(remote_client, points, vectors_config=vectors_config)
+
+    queries = [
+        1,
+        models.RecommendQuery(
+            recommend=models.RecommendInput(
+                positive=[1], strategy=models.RecommendStrategy.BEST_SCORE
+            )
+        ),
+        models.DiscoverQuery(
+            discover=models.DiscoverInput(
+                target=1, context=models.ContextPair(positive=2, negative=3)
+            )
+        ),
+        models.ContextQuery(context=models.ContextPair(positive=1, negative=3)),
+        models.RelevanceFeedbackQuery(
+            relevance_feedback=models.RelevanceFeedbackInput(
+                target=2,
+                feedback=[models.FeedbackItem(example=1, score=0.9)],
+                strategy=models.NaiveFeedbackStrategy(
+                    naive=models.NaiveFeedbackStrategyParams(a=0.5, b=1.0, c=0.7)
+                ),
+            )
+        ),
+    ]
+
+    for using in ("dense", "multi"):
+        for query in queries:
+            with pytest.raises(ValueError):
+                local_client.query_points(COLLECTION_NAME, query=query, using=using)
+
+            with pytest.raises(UnexpectedResponse):
+                remote_client.query_points(COLLECTION_NAME, query=query, using=using)