The Streaming Reactive Framework (SRF) library (proununced "surf") is a reactive, network-aware, flexible, and performance-oriented streaming data framework that standardizes building modular and reusable pipelines with both C++ and Python​. The goal of SRF is to serve as a common high-performance streaming data layer in which all personas of developers - ranging from Data Scientists to DevOps and Performance Engineers can find value.

• Built in C++ for performance, with Python bindings for ease of use and rapid prototyping, with options for maximizing performance
• Distributed computation with message transfers over RMDA using UCX
• Dynamic reconfiguration to scale up and out at runtime​; requires no changes to pipeline configuration
• Unopinionated data model: messages of any type can be used in the pipeline
• Built from the ground up with asynchronous computation for mitigation of I/O and GPU blocking
• Automatically handles backpressure (when the sender is operating faster than the receiver can keep up) and reschedules computation as needed

• Installation
  • Prerequisites
  • Conda Installation
    • Python Bindings
    • C++ Bindings
    • Full SRF Library
    • Optional Settings
  • Source Installation
  • Docker Installation
• Quickstart Guide
• Contributing
  • Third-Party code

SRF includes both Python and C++ bindings and supports installation via conda, Docker, or from source.

• Pascal architecture (Compute capability 6.0) or better
• NVIDIA driver 450.80.02 or higher
• conda or miniconda
• If using Docker:
  • Docker
  • The NVIDIA container toolkit

Installing via Conda is the easiest method for getting the SRF components and supports both the Python and C++ bindings. To install the SRF conda package and build the C++ and hybrid components, follow the steps below. Please note, Conda performance can be greatly increased via Mamba. To install Mamba in an existing Conda environment, simply run:

conda install -c conda-forge mamba

If you choose to use Mamba rather than Conda, simply replace conda with mamba in the instructions below.

# If needed, create a new conda environment
conda create --name srf python=3.8

# Activate the newly created conda environment
conda activate srf

# Install SRF Python bindings
conda install -c rapidsai -c nvidia -c conda-forge srf

# If needed, create a new conda environment
conda create --name srf python=3.8

# Activate the newly created conda environment
conda activate srf

# Install SRF Python bindings
conda install -c rapidsai -c nvidia -c conda-forge libsrf

# If needed, create a new conda environment
conda create --name srf python=3.8

# Activate the newly created conda environment
conda activate srf

# Install SRF Python bindings
conda install -c rapidsai -c nvidia -c conda-forge srf libsrf

To avoid specifying the channels in the Conda commands above:

conda config --env --add channels conda-forge &&\
conda config --env --add channels nvidia &&\
conda config --env --add channels rapidsai

And to opt-in to nightly releases:

conda config --env --add channels nvidia/label/dev &&
conda config --env --add channels rapidsai-nightly

Installing via source is for more advanced users and is necessary to try SRF features before an official release.

export $SRF_HOME=$(pwd)/srf
git clone git@github.com:nv-morpheus/srf.git $SRF_HOME
cd $SRF_HOME

# note: `mamba` may be used in place of `conda` for better performance.
conda env create -n srf-dev --file $SRF_HOME/ci/conda/environments/dev_env.yml
conda activate srf-dev

mkdir $SRF_HOME/build
cd $SRF_HOME/build
cmake ..
make -j $(nproc)

export SRF_TEST_INTERNAL_DATA_PATH=$SRF_HOME/src/tests
$SRF_HOME/build/src/tests/test_srf_private.x
$SRF_HOME/build/tests/test_srf.x
$SRF_HOME/build/tests/logging/test_srf_logging.x

pip install -e $SRF_HOME/python

pytest $SRF_HOME/python

A Dockerfile is provided at $SRF_HOME and can be built with

docker build -t srf:latest .

To run the container

docker run --gpus all --rm -it srf:latest /bin/bash

To quickly learn about both the C++ and Python SRF APIs, including following along with various complexity examples, we recommend following the SRF Quickstart Repository located here. This tutorial walks new users through topics like

• Creating a simple SRF pipeline
• Using a custom datatype between SRF nodes
• Using Reactive-style operators inside nodes for complex functionality
• Understand how threads and buffers can effect performance
• Mixing C++ and Python, for example - defining compute nodes in C++ and calling them from Python

To learn how to contribute to SRF, please read the Contribution Guide.

SRF is licensed under the Apache v2.0 license. All new source files including CMake and other build scripts should contain the Apache v2.0 license header. Any edits to existing source code should update the date range of the copyright to the current year. The format for the license header is:

/*
 * SPDX-FileCopyrightText: Copyright (c) <year>, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 * SPDX-License-Identifier: Apache-2.0
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 * http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

Thirdparty code included in the source tree (that is not pulled in as an external dependency) must be compatible with the Apache v2.0 license and should retain the original license along with a url to the source. If this code is modified, it should contain both the Apache v2.0 license followed by the original license of the code and the url to the original code.

Ex:

/**
 * SPDX-FileCopyrightText: Copyright (c) 2018-2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 * SPDX-License-Identifier: Apache-2.0
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 * http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

//
// Original Source: https://github.com/org/other_project
//
// Original License:
// ...