TiFlash

TiFlash is a columnar storage component of TiDB and TiDB Cloud, the fully-managed service of TiDB. It mainly plays the role of Analytical Processing (AP) in the Hybrid Transactional/Analytical Processing (HTAP) architecture of TiDB.

TiFlash stores data in columnar format and synchronizes data updates in real-time from TiKV by Raft logs with sub-second latency. Reads in TiFlash are guaranteed transactionally consistent with Snapshot Isolation level. TiFlash utilizes Massively Parallel Processing (MPP) computing architecture to accelerate the analytical workloads.

TiFlash repository is based on ClickHouse. We appreciate the excellent work of the ClickHouse team.

Quick Start

Start with TiDB Cloud

Quickly explore TiFlash with a free trial of TiDB Cloud.

See TiDB Cloud Quick Start Guide.

Start with TiDB

See Quick Start with HTAP and Use TiFlash.

Build TiFlash

TiFlash can be built on the following hardware architectures:

• x86-64 / amd64
• aarch64

And the following operating systems:

• Linux
• MacOS

1. Prepare Prerequisites

The following packages are required:

• CMake 3.21.0+
• Clang 14.0.0+
• Rust
• Python 3.0+
• Ninja-Build or GNU Make
• Ccache (not necessary but highly recommended to reduce rebuild time)

Detailed steps for each platform are listed below.

Ubuntu / Debian

sudo apt update

# Install Rust toolchain, see https://rustup.rs for details
curl https://sh.rustup.rs -sSf | sh -s -- --default-toolchain none
source $HOME/.cargo/env

# Install LLVM, see https://apt.llvm.org for details
# Clang will be available as /usr/bin/clang++-15
wget https://apt.llvm.org/llvm.sh
chmod +x llvm.sh
sudo ./llvm.sh 15 all

# Install other dependencies
sudo apt install -y cmake ninja-build zlib1g-dev libcurl4-openssl-dev ccache

Note for Ubuntu 18.04 and Ubuntu 20.04:

The default installed cmake may be not recent enough. You can install a newer cmake from the Kitware APT Repository:

sudo apt install -y software-properties-common lsb-release
wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | gpg --dearmor - | sudo tee /etc/apt/trusted.gpg.d/kitware.gpg >/dev/null
sudo apt-add-repository "deb https://apt.kitware.com/ubuntu/ $(lsb_release -cs) main"
sudo apt update
sudo apt install -y cmake

If you are facing "ld.lld: error: duplicate symbol: ssl3_cbc_digest_record":

It is likely because you have a pre-installed libssl3 where TiFlash prefers libssl1. TiFlash has vendored libssl1, so that you can simply remove the one in the system to make compiling work:

sudo apt remove libssl-dev

If this doesn't work, please file an issue.

Archlinux

# Install Rust toolchain, see https://rustup.rs for details
curl https://sh.rustup.rs -sSf | sh -s -- --default-toolchain none
source $HOME/.cargo/env

# Install compilers and dependencies
sudo pacman -S clang lld libc++ libc++abi compiler-rt openmp lcov cmake ninja curl openssl zlib llvm ccache

CentOS 7

Please refer to release-centos7-llvm/env/prepare-sysroot.sh

MacOS

# Install Rust toolchain, see https://rustup.rs for details
curl https://sh.rustup.rs -sSf | sh -s -- --default-toolchain none
source $HOME/.cargo/env

# Install compilers
xcode-select --install

# Install other dependencies
brew install ninja cmake [email protected] ccache

If your MacOS is higher or equal to 13.0, it should work out of the box because by default Apple clang is 14.0.0. But if your MacOS is lower than 13.0, you should install llvm clang manually.

brew install llvm@15

# check llvm version
clang --version # should be 15.0.0 or higher

2. Checkout Source Code

git clone https://github.com/pingcap/tiflash.git --recursive -j 20
cd tiflash

3. Build

To build TiFlash for development:

# In the TiFlash repository root:
mkdir cmake-build-debug  # The directory name can be customized
cd cmake-build-debug

cmake .. -GNinja -DCMAKE_BUILD_TYPE=DEBUG

ninja tiflash

Note: In Linux, usually you need to explicitly specify to use LLVM.

# In cmake-build-debug directory:
cmake .. -GNinja -DCMAKE_BUILD_TYPE=DEBUG \
  -DCMAKE_C_COMPILER=/usr/bin/clang-14 \
  -DCMAKE_CXX_COMPILER=/usr/bin/clang++-14

In MacOS, if you install llvm clang, you need to explicitly specify to use llvm clang.

Add the following lines to your shell environment, e.g. ~/.bash_profile.

export PATH="/opt/homebrew/opt/llvm/bin:$PATH"
export CC="/opt/homebrew/opt/llvm/bin/clang"
export CXX="/opt/homebrew/opt/llvm/bin/clang++"

Or use CMAKE_C_COMPILER and CMAKE_CXX_COMPILER to specify the compiler, like this:

cmake .. -GNinja -DCMAKE_BUILD_TYPE=DEBUG -DCMAKE_C_COMPILER=/opt/homebrew/opt/llvm/bin/clang -DCMAKE_CXX_COMPILER=/opt/homebrew/opt/llvm/bin/clang++

After building, you can get TiFlash binary in dbms/src/Server/tiflash in the cmake-build-debug directory.

Build Options

TiFlash has several CMake build options to tweak for development purposes. These options SHOULD NOT be changed for production usage, as they may introduce unexpected build errors and unpredictable runtime behaviors.

To tweak options, pass one or multiple -D...=... args when invoking CMake, for example:

cd cmake-build-debug
cmake .. -GNinja -DCMAKE_BUILD_TYPE=DEBUG -DFOO=BAR
                                          ^^^^^^^^^

• Build Type:

  • -DCMAKE_BUILD_TYPE=RELWITHDEBINFO: Release build with debug info (default)

  • -DCMAKE_BUILD_TYPE=DEBUG: Debug build

  • -DCMAKE_BUILD_TYPE=RELEASE: Release build

  Usually you may want to use different build directories for different build types, e.g. a new build directory named cmake-build-release for the release build, so that compile unit cache will not be invalidated when you switch between different build types.

• Build with Unit Tests:

  • -DENABLE_TESTS=ON: Enable unit tests (enabled by default in debug profile)

  • -DENABLE_TESTS=OFF: Disable unit tests (default in release profile)

• Build using GNU Make instead of ninja-build:

  Click to expand instructions

  To use GNU Make, simply don't pass -GNinja to cmake:

  cd cmake-build-debug
  cmake .. -DCMAKE_BUILD_TYPE=DEBUG
  make tiflash -j

  NOTE: Option -j (defaults to your system CPU core count, otherwise you can optionally specify a number) is used to control the build parallelism. Higher parallelism consumes more memory. If you encounter compiler OOM or hang, try to lower the parallelism by specifying a reasonable number, e.g., half of your system CPU core count or even smaller, after -j, depending on the available memory in your system.

• Build with System Libraries:

  Click to expand instructions

  For local development, it is sometimes handy to use pre-installed third-party libraries in the system, rather than to compile them from sources of the bundled (internal) submodules.

  Options are supplied to control whether to use internal third-party libraries (bundled in TiFlash) or to try using the pre-installed system ones.

  WARNING: It is NOT guaranteed that TiFlash would still build if any of the system libraries are used. Build errors are very likely to happen, almost all the time.

  You can view these options along with their descriptions by running:

  cd cmake-build-debug
  cmake -LH | grep "USE_INTERNAL" -A3

  All of these options are default as ON, as the names tell, using the internal libraries and build from sources.

  There is another option to append extra paths for CMake to find system libraries:

  • PREBUILT_LIBS_ROOT: Default as empty, can be specified with multiple values, seperated by ;

• Build for AMD64 Architecture:

  Click to expand instructions

  To deploy TiFlash under the Linux AMD64 architecture, the CPU must support the AVX2 instruction set. Ensure that cat /proc/cpuinfo | grep avx2 has output.

  If need to build TiFlash for AMD64 architecture without such instruction set, please use cmake option -DNO_AVX_OR_HIGHER=ON.

Run Unit Tests

Unit tests are automatically enabled in debug profile. To build these unit tests:

cd cmake-build-debug
cmake .. -GNinja -DCMAKE_BUILD_TYPE=DEBUG
ninja gtests_dbms       # Most TiFlash unit tests
ninja gtests_libdaemon  # Settings related tests
ninja gtests_libcommon

Then, to run these unit tests:

cd cmake-build-debug
./dbms/gtests_dbms
./libs/libdaemon/src/tests/gtests_libdaemon
./libs/libcommon/src/tests/gtests_libcommon

More usages are available via ./dbms/gtests_dbms --help.

Run Sanitizer Tests

TiFlash supports testing with thread sanitizer and address sanitizer.

To build unit test executables with sanitizer enabled:

# In the TiFlash repository root:
mkdir cmake-build-sanitizer
cd cmake-build-sanitizer
cmake .. -GNinja -DENABLE_TESTS=ON -DCMAKE_BUILD_TYPE=ASan # or TSan
ninja gtests_dbms
ninja gtests_libdaemon
ninja gtests_libcommon

There are known false positives reported from leak sanitizer (which is included in address sanitizer). To suppress these errors, set the following environment variables before running the executables:

LSAN_OPTIONS=suppressions=test/sanitize/asan.suppression

Run Integration Tests

1. Build your own TiFlash binary using debug profile:

   cd cmake-build-debug
   cmake .. -GNinja -DCMAKE_BUILD_TYPE=DEBUG
   ninja tiflash

2. Start a local TiDB cluster with your own TiFlash binary using TiUP:

   cd cmake-build-debug
   tiup playground nightly --tiflash.binpath ./dbms/src/Server/tiflash
   
   # Or using a more stable cluster version:
   # tiup playground v6.1.0 --tiflash.binpath ./dbms/src/Server/tiflash

   TiUP is the TiDB component manager. If you don't have one, you can install it via:

   curl --proto '=https' --tlsv1.2 -sSf https://tiup-mirrors.pingcap.com/install.sh | sh

   If you are not running the cluster using the default port (for example, you run multiple clusters), make sure that the port and build directory in tests/_env.sh are correct.

3. Run integration tests:

   # In the TiFlash repository root:
   cd tests
   ./run-test.sh
   
   # Or run specific integration test:
   # ./run-test.sh fullstack-test2/ddl

Note: some integration tests (namely, tests under delta-merge-test) requires a standalone TiFlash service without a TiDB cluster, otherwise they will fail. To run these integration tests: TBD

Run MicroBenchmark Tests

To build micro benchmark tests, you need release profile and tests enabled:

# In the TiFlash repository root:
mkdir cmake-build-release
cd cmake-build-release
cmake .. -GNinja -DCMAKE_BUILD_TYPE=RELEASE -DENABLE_TESTS=ON
ninja bench_dbms

Then, to run these micro benchmarks:

cd cmake-build-release
./dbms/bench_dbms

# Or run with filter:
# ./dbms/bench_dbms --benchmark_filter=xxx

More usages are available via ./dbms/bench_dbms --help.

Generate LLVM Coverage Report

TBD.

Contributing

Here is the overview of TiFlash architecture The architecture of TiFlash's distributed storage engine and transaction layer.

See TiFlash Development Guide and TiFlash Design documents.

Before submitting a pull request, please resolve clang-tidy errors and use format-diff.py to format source code, otherwise CI build may raise error.

NOTE: It is required to use clang-format 12.0.0+.

# In the TiFlash repository root:
merge_base=$(git merge-base upstream/master HEAD)
python3 release-centos7-llvm/scripts/run-clang-tidy.py -p cmake-build-debug -j 20 --files `git diff $merge_base --name-only`
# if there are too much errors, you can try to run the script again with `-fix`
python3 format-diff.py --diff_from $merge_base

License

TiFlash is under the Apache 2.0 license. See the LICENSE file for details.