Feast Benchmark for Async SDK

The data from this benchmark is generated based on data generator from https://github.com/feast-dev/feast-benchmarks.

Setup

Installation

To install requirements:

poetry install

To activate shell:

poetry shell

Run Redis instance using a Docker container:

docker-compose up -d

Feast

To apply changes to registry:

cd registry

feast apply

After running this command, the file registry/registry.db is updated.

To materialize data to Redis instance:

feast materialize 2024-05-03 2024-05-04

Running Server

Go back to root folder cd .. so you can start the server.

uvicorn server.server:app --workers 2 --log-level critical

Some example commands to request the server:

Sync endpoint

curl "http://localhost:8000/get_online_features"

Changing the feature service:

curl "http://localhost:8000/get_online_features?feature_service=feature_service_1"

Changing the batch size (number of entity rows):

curl "http://localhost:8000/get_online_features?feature_service=feature_service_1&batch_size=100"

Async endpoint

curl "http://localhost:8000/get_online_features_async"

Changing the feature service:

curl "http://localhost:8000/get_online_features_async?feature_service=feature_service_1"

Benchmarking Server

You can use wrk or any other load test tool to benchmark the server.

brew install wrk

To start the benchmark execution for sync endpoint:

wrk --threads 4 --latency "http://localhost:8000/get_online_features?feature_service=feature_service_1"

To start the benchmark execution for async endpoint:

wrk --threads 4 --latency "http://localhost:8000/get_online_features_async?feature_service=feature_service_1"