Skip to content

Latest commit

 

History

History
221 lines (187 loc) · 8.3 KB

README.md

File metadata and controls

221 lines (187 loc) · 8.3 KB
Raw

Morpheus Runtime Core (MRC)

The Morpheus Runtime Core (MRC) library is a reactive, network-aware, flexible, and performance-oriented streaming data library that standardizes building modular and reusable pipelines with both C++ and Python. Designed to efficiently handle non-deterministic volumes and highly variable data streams, MRC is the backbone for NVIDIA's Morpheus cyber security library. It's goal 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, find value.

Major features and differentiators

  • 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 RDMA 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

Anatomy of a MRC Pipeline

MRC Pipeline

Table of Contents

Installation

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

Prerequisites

Conda Installation

Installing via Conda is the easiest method for getting the MRC components and supports both the Python and C++ bindings. To install the MRC 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.

Python Bindings

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

# Activate the newly created conda environment
conda activate mrc

# Install MRC Python bindings
conda install -c rapidsai -c nvidia -c conda-forge mrc

C++ Bindings

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

# Activate the newly created conda environment
conda activate mrc

# Install MRC Python bindings
conda install -c rapidsai -c nvidia -c conda-forge libmrc

Full MRC Library

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

# Activate the newly created conda environment
conda activate mrc

# Install MRC Python bindings
conda install -c rapidsai -c nvidia -c conda-forge mrc libmrc

Optional Settings

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

Source Installation

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

Clone MRC repository

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

Create MRC Conda Environment

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

Build MRC

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

Run MRC C++ Tests

export MRC_TEST_INTERNAL_DATA_PATH=$MRC_HOME/src/tests
$MRC_HOME/build/src/tests/test_mrc_private.x
$MRC_HOME/build/tests/test_mrc.x
$MRC_HOME/build/tests/logging/test_mrc_logging.x

Install MRC Python Bindings

pip install -e $MRC_HOME/build/python

Run MRC Python Tests

pytest $MRC_HOME/python

Docker Installation

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

docker build -t mrc:latest .

To run the container

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

Quickstart Guide

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

  • Creating a simple MRC pipeline
  • Using a custom datatype between MRC 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

Contributing

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

MRC 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.
 */

Third-Party code

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:
// ...