chore: Version bump qdrant to v1.3.0

- bump qdrant-client to v1.3.0
- minor improvements during vector loading
- cleanup dockerfile

ann_benchmarks/algorithms/qdrant/Dockerfile

@@ -1,22 +1,23 @@
-ARG QDRANT_VERSION=1.1.1
-ARG QDRANT_CLIENT_VERSION=1.1.6
+ARG QDRANT_VERSION=1.3.0
+ARG QDRANT_CLIENT_VERSION=1.3.1
 FROM qdrant/qdrant:v${QDRANT_VERSION}
 
-RUN apt-get update
-RUN apt-get install -y python3-pip build-essential python3-h5py
-RUN python3 -m pip install --upgrade pip setuptools wheel
-
+ARG QDRANT_CLIENT_VERSION
 WORKDIR /home/app
+COPY requirements.txt .
+
+RUN apt-get update \
+ && apt-get install -y python3-pip build-essential python3-h5py \
+ && python3 -m pip install --break-system-packages --upgrade pip setuptools wheel \
+ && python3 -m pip install --break-system-packages -r requirements.txt \
+ && python3 -m pip install --break-system-packages qdrant-client==${QDRANT_CLIENT_VERSION}
 
-COPY requirements.txt run_algorithm.py ./
+COPY run_algorithm.py .
 
-ARG QDRANT_VERSION
-RUN python3 -m pip install -r requirements.txt
-RUN python3 -m pip install qdrant-client==${QDRANT_VERSION}
+RUN printf '#!/bin/bash\n\
+cd /qdrant && ./qdrant &\n\
+sleep 5\n\
+python3 -u run_algorithm.py "$@"' > entrypoint.sh \
+ && chmod u+x entrypoint.sh
 
-RUN echo '#!/bin/bash' >> entrypoint.sh
-RUN echo 'cd /qdrant && ./qdrant &' >> entrypoint.sh
-RUN echo 'sleep 5' >> entrypoint.sh
-RUN echo 'python3 -u run_algorithm.py "$@"' >> entrypoint.sh
-RUN chmod u+x entrypoint.sh
-ENTRYPOINT ["/home/app/entrypoint.sh"]
+ENTRYPOINT ["/home/app/entrypoint.sh"]

ann_benchmarks/algorithms/qdrant/module.py

@@ -4,9 +4,16 @@
 import numpy as np
 from qdrant_client import QdrantClient
 from qdrant_client import grpc
-from qdrant_client.http.models import (CollectionStatus, Distance,
-                                       VectorParams, OptimizersConfigDiff, ScalarQuantization,
-                                       ScalarQuantizationConfig, ScalarType, HnswConfigDiff)
+from qdrant_client.http.models import (
+    CollectionStatus,
+    Distance,
+    VectorParams,
+    OptimizersConfigDiff,
+    ScalarQuantization,
+    ScalarQuantizationConfig,
+    ScalarType,
+    HnswConfigDiff,
+)
 
 from ..base.module import BaseANN
 
@@ -50,13 +57,18 @@ def fit(self, X):
                 )
             )
 
+        # Disabling indexing during bulk upload
+        # https://qdrant.tech/documentation/tutorials/bulk-upload/#disable-indexing-during-upload
+        # Uploading to multiple shards
+        # https://qdrant.tech/documentation/tutorials/bulk-upload/#parallel-upload-into-multiple-shards
         self._client.recreate_collection(
             collection_name=self._collection_name,
+            shard_number=2,
             vectors_config=VectorParams(size=X.shape[1], distance=self._distances_mapping[self._metric]),
             optimizers_config=OptimizersConfigDiff(
                 default_segment_number=2,
-                max_segment_size=100000000,
-                indexing_threshold=1000,
+                memmap_threshold=20000,
+                indexing_threshold=0,
             ),
             quantization_config=quantization_config,
             # TODO: benchmark this as well
@@ -68,7 +80,20 @@ def fit(self, X):
         )
 
         self._client.upload_collection(
-            collection_name=self._collection_name, vectors=X, ids=list(range(X.shape[0])), parallel=1
+            collection_name=self._collection_name,
+            vectors=X,
+            ids=list(range(X.shape[0])),
+            batch_size=BATCH_SIZE,
+            parallel=1,
+        )
+
+        # Re-enabling indexing
+        self._client.update_collection(
+            collection_name=self._collection_name,
+            optimizers_config=OptimizersConfigDiff(
+                indexing_threshold=20000,
+            ),
+            timeout=TIMEOUT,
         )
 
         # wait for vectors to be fully indexed
@@ -92,20 +117,19 @@ def set_query_arguments(self, hnsw_ef, rescore):
         self._search_params["rescore"] = rescore
 
     def query(self, q, n):
-        quantization_search_params = grpc.QuantizationSearchParams(
-            ignore=False,
-            rescore=self._search_params["rescore"],
-        )
-
         search_request = grpc.SearchPoints(
             collection_name=self._collection_name,
             vector=q.tolist(),
             limit=n,
             with_payload=grpc.WithPayloadSelector(enable=False),
+            with_vectors=grpc.WithVectorsSelector(enable=False),
             params=grpc.SearchParams(
                 hnsw_ef=self._search_params["hnsw_ef"],
-                quantization=quantization_search_params,
-            )
+                quantization=grpc.QuantizationSearchParams(
+                    ignore=False,
+                    rescore=self._search_params["rescore"],
+                ),
+            ),
         )
 
         search_result = self._client.grpc_points.Search(search_request, timeout=TIMEOUT)
@@ -135,11 +159,13 @@ def iter_batches(iterable, batch_size) -> Iterable[List[Any]]:
                 vector=q.tolist(),
                 limit=n,
                 with_payload=grpc.WithPayloadSelector(enable=False),
+                with_vectors=grpc.WithVectorsSelector(enable=False),
                 params=grpc.SearchParams(
                     hnsw_ef=self._search_params["hnsw_ef"],
                     quantization=quantization_search_params,
-                )
-            ) for q in X
+                ),
+            )
+            for q in X
         ]
 
         self.batch_results = []
@@ -162,4 +188,4 @@ def get_batch_results(self):
 
     def __str__(self):
         hnsw_ef = self._search_params["hnsw_ef"]
-        return f"Qdrant(quantization={self._quantization}, hnsw_ef={hnsw_ef})"
+        return f"Qdrant(quantization={self._quantization}, hnsw_ef={hnsw_ef})"