This document will help to set up your Milvus development environment and to run tests. Please file an issue if there's a problem.
- Development
- Table of contents
Our official Milvus versions are releases as Docker images. To build Milvus Docker on your own, please follow these instructions.
The details below outline the hardware and software requirements for building on Linux and MacOS.
The following specification (either physical or virtual machine resources) is recommended for Milvus to build and run from source code.
- 8GB of RAM
- 50GB of free disk space
All Linux distributions are available for Milvus development. However a majority of our contributor worked with Ubuntu or CentOS systems, with a small portion of Mac (both x86_64 and Apple Silicon) contributors. If you would like Milvus to build and run on other distributions, you are more than welcome to file an issue and contribute!
Here's a list of verified OS types where Milvus can successfully build and run:
- Debian/Ubuntu
- Amazon Linux
- MacOS (x86_64)
- MacOS (Apple Silicon)
Linux systems (Recommend Ubuntu 20.04 or later):
go: >= 1.20
cmake: >= 3.18
gcc: 7.5
MacOS systems with x86_64 (Big Sur 11.5 or later recommended):
go: >= 1.20
cmake: >= 3.18
llvm: >= 15
MacOS systems with Apple Silicon (Monterey 12.0.1 or later recommended):
go: >= 1.20 (Arch=ARM64)
cmake: >= 3.18
llvm: >= 15
In the Milvus repository root, simply run:
$ ./scripts/install_deps.sh
- Google Test is automatically cloned from GitHub, which in some case could conflict with your local google test library.
Once you have finished, confirm that gcc
and make
are installed:
$ gcc --version
$ make --version
The algorithm library of Milvus, Knowhere is written in c++. CMake is required in the Milvus compilation. If you don't have it, please follow the instructions in the Installing CMake.
Confirm that cmake is available:
$ cmake --version
Note: 3.25 or higher cmake version is required to build Milvus.
Milvus uses Conan to manage third-party dependencies for c++.
Install Conan
pip install conan==1.61.0
Note: Conan version 2.x is not currently supported, please use version 1.61.
Milvus is written in Go. If you don't have a Go development environment, please follow the instructions in the Go Getting Started guide.
Confirm that your GOPATH
and GOBIN
environment variables are correctly set as detailed in How to Write Go Code before proceeding.
$ go version
Note: go >= 1.20 is required to build Milvus.
Milvus depends on etcd, Pulsar and MinIO. Using Docker Compose to manage these is an easy way in local development. To install Docker and Docker Compose in your development environment, follow the instructions from the Docker website below:
- Docker: https://docs.docker.com/get-docker/
- Docker Compose: https://docs.docker.com/compose/install/
To build the Milvus project, run the following command:
$ make
If this command succeed, you will now have an executable at bin/milvus
off of your Milvus project directory.
If you want to update proto file before make
, we can use the following command:
$ make generated-proto-go
If you want to know more, you can read Makefile.
The details below outline the software requirements for building on Ubuntu 20.04
apt update
wget https://github.com/Kitware/CMake/releases/download/v3.27.9/cmake-3.27.9-linux-aarch64.tar.gz
tar zxf cmake-3.27.9-linux-aarch64.tar.gz
mv cmake-3.27.9-linux-aarch64 /usr/local/cmake
vi /etc/profile
export PATH=$PATH:/usr/local/cmake/bin
source /etc/profile
cmake --version
sudo apt install -y clang-format clang-tidy ninja-build gcc g++ curl zip unzip tar
# Verify python3 version, need python3 version > 3.8
python3 --version
# pip install conan 1.61.0
pip3 install conan==1.61.0
wget https://go.dev/dl/go1.18.10.linux-arm64.tar.gz
tar zxf go1.18.10.linux-arm64.tar.gz
mv ./go /usr/local
vi /etc/profile
export PATH=$PATH:/usr/local/go/bin
source /etc/profile
go version
git clone https://github.com/milvus-io/milvus.git
git checkout v2.3.4
cd ./milvus
getconf PAGESIZE
The PAGESIZE for the ky10 SP3 operating system is 65536, which is 64KB.
The MILVUS_JEMALLOC_LG_PAGE
variable's primary function is to specify the size of large pages during the compilation of jemalloc. Jemalloc is a memory allocator designed to enhance the performance and efficiency of applications in a multi-threaded environment. By specifying the size of large pages, memory management and access can be optimized, thereby improving performance.
Large page support allows the operating system to manage and allocate memory in larger blocks, reducing the number of page table entries, thereby decreasing the time for page table lookups and improving the efficiency of memory access. This is particularly important when processing large amounts of data, as it can significantly reduce page faults and Translation Lookaside Buffer (TLB) misses, enhancing application performance.
On ARM64 architectures, different systems may support different page sizes, such as 4KB and 64KB. The MILVUS_JEMALLOC_LG_PAGE
setting allows developers to customize the compilation of jemalloc for the target platform, ensuring it can efficiently operate on systems with varying page sizes. By specifying the --with-lg-page
configuration option, jemalloc can utilize the optimal page size supported by the system when managing memory.
For example, if a system supports a 64KB page size, by setting MILVUS_JEMALLOC_LG_PAGE
to the corresponding value (the power of 2, 64KB is 2 to the 16th power, so the value is 16), jemalloc can allocate and manage memory in 64KB units, which can improve the performance of applications running on that system.
Modify the make configuration file, located at: ./milvus/scripts/core_build.sh
, with the following changes:
arch=$(uname -m)
CMAKE_CMD="cmake \
${CMAKE_EXTRA_ARGS} \
-DBUILD_UNIT_TEST=${BUILD_UNITTEST} \
-DCMAKE_INSTALL_PREFIX=${INSTALL_PREFIX}
-DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
-DCMAKE_CUDA_COMPILER=${CUDA_COMPILER} \
-DCMAKE_LIBRARY_ARCHITECTURE=${arch} \
-DBUILD_COVERAGE=${BUILD_COVERAGE} \
-DMILVUS_GPU_VERSION=${GPU_VERSION} \
-DMILVUS_CUDA_ARCH=${CUDA_ARCH} \
-DEMBEDDED_MILVUS=${EMBEDDED_MILVUS} \
-DBUILD_DISK_ANN=${BUILD_DISK_ANN} \
+ -DMILVUS_JEMALLOC_LG_PAGE=16 \
-DUSE_ASAN=${USE_ASAN} \
-DUSE_DYNAMIC_SIMD=${USE_DYNAMIC_SIMD} \
-DCPU_ARCH=${CPU_ARCH} \
-DINDEX_ENGINE=${INDEX_ENGINE} "
if [ -z "$BUILD_WITHOUT_AZURE" ]; then
CMAKE_CMD=${CMAKE_CMD}"-DAZURE_BUILD_DIR=${AZURE_BUILD_DIR} \
-DVCPKG_TARGET_TRIPLET=${VCPKG_TARGET_TRIPLET} "
fi
CMAKE_CMD=${CMAKE_CMD}"${CPP_SRC_DIR}"
Using -DMILVUS_JEMALLOC_LG_PAGE=16
as a compilation option for jemalloc is because it specifies the size
of "large pages" as 2 to the 16th power bytes, which equals 65536 bytes or 64KB. This value is set to optimize memory management and improve performance, especially on systems that support or prefer using large pages to reduce the overhead of page table management.
Specifying -DMILVUS_JEMALLOC_LG_PAGE=16
during the compilation of jemalloc informs jemalloc to assume the system's large page size is 64KB. This allows jemalloc to work more efficiently with the operating system's memory manager, using large pages to optimize performance. This is crucial for ensuring optimal performance on systems with different default page sizes, particularly in environments that might have different memory management needs due to varying hardware or system configurations.
cd ./milvus
cp build/docker/milvus/ubuntu20.04/Dockerfile .
Modify the Dockerfile as follows:
# Copyright (C) 2019-2022 Zilliz. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance
# with the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software distributed under the License
# is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
# or implied. See the License for the specific language governing permissions and limitations under the License.
FROM ubuntu:focal-20220426
ARG TARGETARCH
RUN apt-get update && \
apt-get install -y --no-install-recommends curl ca-certificates libaio-dev libgomp1 && \
apt-get remove --purge -y && \
rm -rf /var/lib/apt/lists/*
COPY ./bin/ /milvus/bin/
COPY ./configs/ /milvus/configs/
COPY ./internal/core/output/lib/ /milvus/lib/
ENV PATH=/milvus/bin:$PATH
ENV LD_LIBRARY_PATH=/milvus/lib:$LD_LIBRARY_PATH:/usr/lib
ENV LD_PRELOAD=/milvus/lib/libjemalloc.so
ENV MALLOC_CONF=background_thread:true
# Add Tini
ADD https://github.com/krallin/tini/releases/download/v0.19.0/tini-$TARGETARCH /tini
RUN chmod +x /tini
ENTRYPOINT ["/tini", "--"]
WORKDIR /milvus/
Build command: docker build -t ghostbaby/milvus:v2.3.4_arm64 .
Verify the image: docker run ghostbaby/milvus:v2.3.4_arm64 milvus run proxy
Pre-submission verification provides a battery of checks and tests to give your pull request the best chance of being accepted. Developers need to run as many verification tests as possible locally.
To run all pre-submission verification tests, use this command:
$ make verifiers
It is required that all pull request candidates should pass all Milvus unit tests.
Beforce running unit tests, you need to first bring up the Milvus deployment environment. You may set up a local docker environment with our docker compose yaml file to start unit testing. For Apple Silicon users (Apple M1):
$ cd deployments/docker/dev
$ docker-compose -f docker-compose-apple-silicon.yml up -d
$ cd ../../../
$ make unittest
For others:
$ cd deployments/docker/dev
$ docker-compose up -d
$ cd ../../../
$ make unittest
To run only cpp test:
$ make test-cpp
To run only go test:
$ make test-go
To run a single test case (TestSearchTask in /internal/proxy directory, for example):
$ source scripts/setenv.sh && go test -v ./internal/proxy/ -test.run TestSearchTask
If using Mac with M1 chip
$ source scripts/setenv.sh && go test -tags=dynamic -v ./internal/proxy/ -test.run TestSearchTask
Before submitting your pull request, make sure your code change is covered by unit test. Use the following commands to check code coverage rate:
Run unit test and generate code coverage report:
$ make codecov
This command will generate html report for Golang and C++ respectively.
For Golang report, open the go_coverage.html
under milvus project path.
For C++ report, open the cpp_coverage/index.html
under milvus project path.
You also can generate Golang coverage report by:
$ make codecov-go
Or C++ coverage report by:
$ make codecov-cpp
Milvus uses Python SDK to write test cases to verify the correctness of Milvus functions. Before running E2E tests, you need a running Milvus:
# Running Milvus cluster
$ cd deployments/docker/dev
$ docker-compose up -d
$ cd ../../../
$ ./scripts/start_cluster.sh
# Or running Milvus standalone
$ cd deployments/docker/dev
$ docker-compose up -d
$ cd ../../../
$ ./scripts/start_standalone.sh
To run E2E tests, use these commands:
$ cd tests/python_client
$ pip install -r requirements.txt
$ pytest --tags=L0 -n auto
After preparing deployment environment, we can start the cluster on your host machine
$ ./scripts/start_cluster.sh
start the cluster on your host machine
$ ./build/builder.sh make install // build milvus
$ ./build/build_image.sh // build milvus latest docker image
$ docker images // check if milvus latest image is ready
REPOSITORY TAG IMAGE ID CREATED SIZE
milvusdb/milvus latest 63c62ff7c1b7 52 minutes ago 570MB
$ install with docker compose
To check out code to work on, please refer to the GitHub Flow.
Q: The go building phase fails on Apple Silicon (Mac M1) machines.
A: Please double-check that you have right Go version installed, i.e. with OS=macOS and Arch=ARM64.
Q: "make" fails with "ld: library not found for -lSystem" on MacOS.
A: There are a couple of things you could try:
- Use Software Update (from About this Mac -> Overview) to install updates.
- Try the following commands:
sudo rm -rf /Library/Developer/CommandLineTools
sudo xcode-select --install
Q: Rocksdb fails to compile with "ld: warning: object file was built for newer macOS version (11.6) than being linked (11.0)." on MacOS.
A: Use Software Update (from About this Mac -> Overview) to install updates.
Q: Some Go unit tests failed.
A: We are aware that some tests can be flaky occasionally. If there's something you believe is abnormal (i.e. tests that fail every single time). You are more than welcome to file an issue!