Faiss gRPC Server

Faiss

A library for efficient similarity search and clustering of dense vectors.

References:

• https://github.com/facebookresearch/faiss
• https://github.com/facebookresearch/faiss/wiki/Threads-and-asynchronous-calls
• https://github.com/daangn/faiss-server

Protocol buffer compile for gRPC

Install gRPC tool first.

pip3 install grpcio==1.15.0 grpcio-tools==1.15.0

Compile Protocol buffer.

$ python3 -m grpc_tools.protoc -I. --python_out=. --grpc_python_out=. faiss.proto

Set application configuration

1. Copy conf.yaml.sample to conf.yaml
2. Modify conf.yaml with your local development environment.

Build docker image

$ docker build -t [YOUR_DOCKER_IMAGE_NAME or YOUR_DOCKER_IMAGE_NAME:VERSION] .

or

$ ./build.sh [YOUR_DOCKER_IMAGE_NAME or YOUR_DOCKER_IMAGE_NAME:VERSION]

eg.

$ ./build.sh nextmining/faiss-server:latest

Run the faiss server

$ ./run.sh [YOUR_DOCKER_IMAGE_NAME or YOUR_DOCKER_IMAGE_NAME:VERSION] [YOUR_DOCKER_CONTAINER_NAME] [VECTOR_DIMENSION_SIZE] [MAX_WORKERS] [NUM_THREADS] [PO
RT]

eg.

$ ./run.sh nextmining/faiss-server:latest faiss-server 200 5 10 50051

Stop the faiss server

Stop and delete docker container.

$ ./stop.sh faiss-server

Enter the docker container with shell command and test.

$ docker exec -it faiss-server /bin/bash

$ python client_sample.py test --dim 200 --host localhost --port 50051

Test client samples.

$ python client_sample.py import data/embeds.csv data/ids.csv data/keys.csv --host localhost:50051
$ python client_sample.py search-by-key a2 --host localhost:50051 --count 2
$ python client_sample.py get-embedding 1 --host localhost:50051
$ python client_sample.py search-by-embedding 1 --host localhost:50051 --count 2

Import index files from local to faiss.

$ python client_sample.py import ./data/embeds.csv ./data/ids.csv ./data/keys.csv --host localhost:50051

Import index files from Azure Blobs to faiss.

$ python client_sample.py import blobs://recommendation/item_embeddings/all/embeds.csv blobs://recommendation/item_embeddings/all/ids.csv blobs://recommendation/item_embeddings/all/keys.csv --host localhost:50051

Deploy to Kubernetes

Push docker image to your Container Registries Service

eg. Azure Container Registries(ACR)

$ docker image tag nextmining/faiss-server:latest YOUR_CONTAINER_REGISTRIES_SERVICE/faiss-server:1.0.0
$ docker image push YOUR_CONTAINER_REGISTRIES_SERVICE/faiss-server:1.0.0

Deploy to Kubernetes Service

1. Copy faiss-server.yaml.new.sample to faiss-server.yaml
2. Modify faiss-server.yaml with your Kubernetes environment.

$ kubectl apply -f /YOUR_LOCAL_PATH/faiss-server.yaml