Python gRPC server for apploximate nearest neighbor by Faiss.

You can install as package by pip install on repository root.

pip install .

You can start gRPC server by running following command.

python python/faiss_grpc/main.py

Python Faiss gRPC server has some environment variables starts with prefix FAISS_GRPC_.

Python Faiss gRPC server supports .env file, above environment variables can specified like

# .env
FAISS_GRPC_INDEX_PATH=/path/to/index
FAISS_GRPC_NORMALIZE_QUERY=True
FAISS_GRPC_NPROBE=10

FAISS_GRPC_HOST=[::]
FAISS_GRPC_PORT=50051
FAISS_GRPC_MAX_WORKERS=2

Client side code is under the examples/client.py. You can run following command or directly running on interpreter.

cd examples

# show usage of client example
python client.py -h

# show heatbeat message
python client.py heatbeat

# search by query, get numer of neighbors given value (query is auto generated in command as identity vector)
python client.py search 10

# search by specified id, get numer of neighbors given value
python client.py search-by-id 0 10

Following command will generate two python grpc code faiss_pb2.py and faiss_pb2_grpc.py under the python/faiss_grpc/proto. If you update grpc definition proto/faiss.proto, you should regenerate at first.

make codegen

Then you have to fix import path in faiss_pb2_grpc.py.

# Before
import faiss_pb2 as faiss__pb2
# After
import faiss_grpc.proto.faiss_pb2 as faiss__pb2

• Avoid to use SearchById on the index built by add_with_ids (can use from IndexIDMap, IndexIVF etc.). These index does not keep id complicatedly so reconstruct_n method may do unexpected behavior.
• Support only CPU index.

• Prepare docker image
• Auto download index from remote location if FAISS_GRPC_INDEX_PATH was specified remote path