Skip to content
Switch branches/tags
Go to file
In aarch64 machines, the bazelisk installation command fails due to binary format mismatch. We need to take care of the architecture running on.

Signed-off-by: Kotaro Inoue <>
83 contributors

Users who have contributed to this file

@lizan @htuch @PiotrSikora @sunjayBhatia @mattklein123 @alyssawilk @wrowe @davinci26 @cmluciano @zuercher @dnoe @snowp

Building Envoy with Bazel

Installing Bazelisk as Bazel

It is recommended to use Bazelisk installed as bazel, to avoid Bazel compatibility issues.

On Linux, run the following commands:

sudo wget -O /usr/local/bin/bazel$([ $(uname -m) = "aarch64" ] && echo "arm64" || echo "amd64")
sudo chmod +x /usr/local/bin/bazel

On macOS, run the following command:

brew install bazelisk

On Windows, run the following commands:

mkdir %USERPROFILE%\bazel
powershell Invoke-WebRequest -OutFile %USERPROFILE%\bazel\bazel.exe

Production environments

To build Envoy with Bazel in a production environment, where the Envoy dependencies are typically independently sourced, the following steps should be followed:

  1. Configure, build and/or install the Envoy dependencies.
  2. bazel build -c opt //source/exe:envoy-static from the repository root.

Quick start Bazel build for developers

This section describes how to and what dependencies to install to get started building Envoy with Bazel. If you would rather use a pre-build Docker image with required tools installed, skip to this section.

As a developer convenience, a WORKSPACE and rules for building a recent version of the various Envoy dependencies are provided. These are provided as is, they are only suitable for development and testing purposes. The specific versions of the Envoy dependencies used in this build may not be up-to-date with the latest security patches. See this doc for how to update or override dependencies.

  1. Install external dependencies.


    On Ubuntu, run the following:

    sudo apt-get install \
       autoconf \
       automake \
       cmake \
       curl \
       libtool \
       make \
       ninja-build \
       patch \
       python3-pip \
       unzip \


    On Fedora (maybe also other red hat distros), run the following:

    dnf install \
        aspell-en \
        cmake \
        libatomic \
        libstdc++ \
        libstdc++-static \
        libtool \
        lld \
        ninja-build \
        patch \


    On Linux, we recommend using the prebuilt Clang+LLVM package from LLVM official site. Extract the tar.xz and run the following:


    This will setup a clang.bazelrc file in Envoy source root. If you want to make clang as default, run the following:

    echo "build --config=clang" >> user.bazelrc

    Note: Either libc++ or libstdc++-7-dev (or higher) must be installed.

    Config Flag Choices

    Different config flags specify the compiler libraries:

    • --config=libc++ means using clang + libc++
    • --config=clang means using clang + libstdc++
    • no config flag means using gcc + libstdc++


    On macOS, you'll need to install several dependencies. This can be accomplished via Homebrew:

    brew install coreutils wget cmake libtool go bazel automake ninja clang-format autoconf aspell

    notes: coreutils is used for realpath, gmd5sum and gsha256sum

    The full version of Xcode (not just Command Line Tools) is also required to build Envoy on macOS. Envoy compiles and passes tests with the version of clang installed by Xcode 11.1: Apple clang version 11.0.0 (clang-1100.0.33.8).

    Having the binutils keg installed in Brew is known to cause issues due to putting an incompatible version of ar on the PATH, so if you run into issues building third party code like luajit consider uninstalling binutils.


    Note: These instructions apply to Windows 10 SDK, version 1803 (10.0.17134.12). Earlier versions will not compile because the afunix.h header is not available. The recommended Windows version is equal or later than Windows 10 SDK, version 1903 (10.0.18362.1)

    Install bazelisk in the PATH using the bazel.exe executable name as described above in the first section.

    When building Envoy, Bazel creates very long path names. One way to work around these excessive path lengths is to change the output base directory for bazel to a very short root path. An example Bazel configuration to help with this is to use C:\_eb as the bazel base path. This and other preferences should be set up by placing the following bazelrc configuration line in a system %ProgramData%\bazel.bazelrc file or the individual user's %USERPROFILE%\.bazelrc file (rather than including it on every bazel command line):

    startup --output_base=C:/_eb

    Another option to shorten the output root for Bazel is to set the USERNAME environment variable in your shell session to a short value. Bazel uses this value when constructing its output root path if no explicit --output_base is set.

    Bazel also creates file symlinks when building Envoy. It's strongly recommended to enable file symlink support using Bazel's instructions. For other common issues, see the Using Bazel on Windows page.

    The paths in this document are given as examples, make sure to verify you are using the correct paths for your environment. Also note that these examples assume using a cmd.exe shell to set environment variables etc., be sure to do the equivalent if using a different shell.

    python3: Specifically, the Windows-native flavor distributed by The POSIX flavor available via MSYS2, the Windows Store flavor and other distributions will not work. Add a symlink for python3.exe pointing to the installed python.exe for Envoy scripts and Bazel rules which follow POSIX python conventions. Add pip.exe to the PATH and install the wheel package.

    mklink %USERPROFILE%\Python39\python3.exe %USERPROFILE%\Python39\python.exe
    set PATH=%PATH%;%USERPROFILE%\Python39
    set PATH=%PATH%;%USERPROFILE%\Python39\Scripts
    pip install wheel

    Build Tools for Visual Studio 2019: For building with MSVC, you must install at least the VC++ workload. You may alternately install the entire Visual Studio 2019 and use the Build Tools installed in that package. Earlier versions of VC++ Build Tools/Visual Studio are not recommended or supported. If installed in a non-standard filesystem location, be sure to set the BAZEL_VC environment variable to the path of the VC++ package to allow Bazel to find your installation of VC++. NOTE: ensure that the link.exe that resolves on your PATH is from VC++ Build Tools and not /usr/bin/link.exe from MSYS2, which is determined by their relative ordering in your PATH.

    set PATH=%PATH%;%USERPROFILE%\VSBT2019\VC\Tools\MSVC\14.26.28801\bin\Hostx64\x64

    The Windows SDK contains header files and libraries you need when building Windows applications. Bazel always uses the latest, but you can specify a different version by setting the environment variable BAZEL_WINSDK_FULL_VERSION. See bazel/windows

    Ensure CMake and ninja binaries are on the PATH. The versions packaged with VC++ Build Tools are sufficient in most cases, but are 32 bit binaries. These flavors will not run in the project's GCP CI remote build environment, so 64 bit builds from the CMake and ninja projects are used instead.

    set PATH=%PATH%;%USERPROFILE%\VSBT2019\Common7\IDE\CommonExtensions\Microsoft\CMake\CMake\bin
    set PATH=%PATH%;%USERPROFILE%\VSBT2019\Common7\IDE\CommonExtensions\Microsoft\CMake\Ninja

    MSYS2 shell: Install to a path with no spaces, e.g. C:\msys64.

    Set the BAZEL_SH environment variable to the path of the installed MSYS2 bash.exe executable. Additionally, setting the MSYS2_ARG_CONV_EXCL environment variable to a value of * is often advisable to ensure argument parsing in the MSYS2 shell behaves as expected.

    set PATH=%PATH%;%USERPROFILE%\msys64\usr\bin
    set BAZEL_SH=%USERPROFILE%\msys64\usr\bin\bash.exe
    set MSYS2_PATH_TYPE=inherit

    Set the TMPDIR environment variable to a path usable as a temporary directory (e.g. C:\Windows\TEMP), and create a directory symlink C:\c to C:\, so that the MSYS2 path /c/Windows/TEMP is equivalent to the Windows path C:\Windows\TEMP:

    set TMPDIR=C:\Windows\TEMP
    mklink /d C:\c C:\

    The TMPDIR path and MSYS2 mktemp command are used frequently by the rules_foreign_cc component of Bazel as well as Envoy's test scripts, causing problems if not set to a path accessible to both Windows and msys commands. [Note the ci/ script which builds envoy and run tests in CI creates this symlink automatically.]

    In the MSYS2 shell, install additional packages via pacman:

    pacman -S diffutils patch unzip zip

    Git: This version from the Git project, or the version distributed using pacman under MSYS2 will both work, ensure one is on the PATH:.

    set PATH=%PATH%;%USERPROFILE%\Git\bin

    Lastly, persist environment variable changes.

    setx PATH "%PATH%"
    setx BAZEL_SH "%BAZEL_SH%"
    setx BAZEL_VC "%BAZEL_VC%"
    setx TMPDIR "%TMPDIR%"

    On Windows the supported/recommended shell to interact with bazel is MSYS2. This means that all the bazel commands (i.e. build, test) should be executed from MSYS2.

  2. Install Golang on your machine. This is required as part of building BoringSSL and also for Buildifer which is used for formatting bazel BUILD files.

  3. go get -u to install buildifier. You may need to set BUILDIFIER_BIN to $GOPATH/bin/buildifier in your shell for buildifier to work.

  4. go get -u to install buildozer. You may need to set BUILDOZER_BIN to $GOPATH/bin/buildozer in your shell for buildozer to work.

  5. bazel build //source/exe:envoy-static from the Envoy source directory. Add -c opt for an optimized release build or -c dbg for an unoptimized, fully instrumented debugging build.

Building Envoy with the CI Docker image

Envoy can also be built with the Docker image used for CI, by installing Docker and executing the following.

On Linux, run:

./ci/ './ci/'

From a Windows host with Docker installed, the Windows containers feature enabled, and bash (installed via MSYS2 or Git bash), run:

Note: the command below executes the whole Windows CI and unlike Linux you are not able to set specific build targets. You can modify ./ci/ to modify bazel arguments, tests to run, etc. as well as set environment variables to adjust your container build environment.

./ci/ './ci/'

See also the documentation for developer use of the CI Docker image.

Building Envoy with Remote Execution

Envoy can also be built with Bazel Remote Execution, part of the CI is running with the hosted GCP RBE service.

To build Envoy with a remote build services, run Bazel with your remote build service flags and with --config=remote-clang. For example the following command runs build with the GCP RBE service used in CI:

bazel build //source/exe:envoy-static --config=remote-clang \
    --remote_cache=grpcs:// \
    --remote_executor=grpcs:// \

Change the value of --remote_cache, --remote_executor and --remote_instance_name for your remote build services. Tests can be run in remote execution too.

Note: Currently the test run configuration in .bazelrc doesn't download test binaries and test logs, to override the behavior set --experimental_remote_download_outputs accordingly.

Building Envoy with Docker sandbox

Building Envoy with Docker sandbox uses the same Docker image used in CI with fixed C++ toolchain configuration. It produces more consistent output which is not depending on your local C++ toolchain. It can also help debugging issues with RBE. To build Envoy with Docker sandbox:

bazel build //source/exe:envoy-static --config=docker-clang

Tests can be run in docker sandbox too. Note that the network environment, such as IPv6, may be different in the docker sandbox so you may want set different options. See below to configure test IP versions.

Linking against libc++ on Linux

To link Envoy against libc++, follow the quick start to setup Clang+LLVM and run:

bazel build --config=libc++ //source/exe:envoy-static

Or use our configuration with Remote Execution or Docker sandbox, pass --config=remote-clang-libc++ or --config=docker-clang-libc++ respectively.

If you want to make libc++ as default, add a line build --config=libc++ to the user.bazelrc file in Envoy source root.

Using a compiler toolchain in a non-standard location

By setting the CC and LD_LIBRARY_PATH in the environment that Bazel executes from as appropriate, an arbitrary compiler toolchain and standard library location can be specified. One slight caveat is that (at the time of writing), Bazel expects the binutils in $(dirname $CC) to be unprefixed, e.g. as instead of x86_64-linux-gnu-as.

Note: this configuration currently doesn't work with Remote Execution or Docker sandbox, you have to generate a custom toolchains configuration for them. See bazelbuild/bazel-toolchains for more details.

Supported compiler versions

We now require Clang >= 5.0 due to known issues with std::string thread safety and C++14 support. GCC >= 7 is also known to work. Currently the CI is running with Clang 10.

Clang STL debug symbols

By default Clang drops some debug symbols that are required for pretty printing to work correctly. More information can be found here. The easy solution is to set --copt=-fno-limit-debug-info on the CLI or in your .bazelrc file.

Removing debug info

If you don't want your debug or release binaries to contain debug info to reduce binary size, pass --define=no_debug_info=1 when building. This is primarily useful when building envoy as a static library. When building a linked envoy binary you can build the implicit .stripped target from cc_binary or pass --strip=always instead.

Testing Envoy with Bazel

All the Envoy tests can be built and run with:

bazel test //test/...

An individual test target can be run with a more specific Bazel label, e.g. to build and run only the units tests in test/common/http/

bazel test //test/common/http:async_client_impl_test

To observe more verbose test output:

bazel test --test_output=streamed //test/common/http:async_client_impl_test

It's also possible to pass into an Envoy test additional command-line args via --test_arg. For example, for extremely verbose test debugging:

bazel test --test_output=streamed //test/common/http:async_client_impl_test --test_arg="-l trace"

By default, testing exercises both IPv4 and IPv6 address connections. In IPv4 or IPv6 only environments, set the environment variable ENVOY_IP_TEST_VERSIONS to "v4only" or "v6only", respectively.

bazel test //test/... --test_env=ENVOY_IP_TEST_VERSIONS=v4only
bazel test //test/... --test_env=ENVOY_IP_TEST_VERSIONS=v6only

By default, tests are run with the gperftools heap checker enabled in "normal" mode to detect leaks. For other mode options, see the gperftools heap checker documentation. To disable the heap checker or change the mode, set the HEAPCHECK environment variable:

# Disables the heap checker
bazel test //test/... --test_env=HEAPCHECK=
# Changes the heap checker to "minimal" mode
bazel test //test/... --test_env=HEAPCHECK=minimal

If you see a leak detected, by default the reported offsets will require addr2line interpretation. You can run under --config=clang-asan to have this automatically applied.

Bazel will by default cache successful test results. To force it to rerun tests:

bazel test //test/common/http:async_client_impl_test --cache_test_results=no

Bazel will by default run all tests inside a sandbox, which disallows access to the local filesystem. If you need to break out of the sandbox (for example to run under a local script or tool with --run_under), you can run the test with --strategy=TestRunner=local, e.g.:

bazel test //test/common/http:async_client_impl_test --strategy=TestRunner=local --run_under=/some/path/

Stack trace symbol resolution

Envoy can produce backtraces on demand and from assertions and other fatal actions like segfaults. Where supported, stack traces will contain resolved symbols, though not include line numbers. On systems where absl::Symbolization is not supported, the stack traces written in the log or to stderr contain addresses rather than resolved symbols. If the symbols were resolved, the address is also included at the end of the line.

The tools/ script exists to process the output and do additional symbol resolution including file names and line numbers. It requires the addr2line program be installed and in your path. Any log lines not relevant to the backtrace capability are passed through the script unchanged (it acts like a filter). File and line information is appended to the stack trace lines.

The script runs in one of two modes. To process log input from stdin, pass -s as the first argument, followed by the executable file path. You can postprocess a log or pipe the output of an Envoy process. If you do not specify the -s argument it runs the arguments as a child process. This enables you to run a test with backtrace post processing. Bazel sandboxing must be disabled by specifying local execution. Example command line with run_under:

bazel test -c dbg //test/server:backtrace_test
--run_under=`pwd`/tools/ --strategy=TestRunner=local
--cache_test_results=no --test_output=all

Example using input on stdin:

bazel test -c dbg //test/server:backtrace_test --cache_test_results=no --test_output=streamed |& tools/ -s bazel-bin/test/server/backtrace_test

You will need to use either a dbg build type or the opt build type to get file and line symbol information in the binaries.

By default will install signal handlers to print backtraces at the location where a fatal signal occurred. The signal handler will re-raise the fatal signal with the default handler so a core file will still be dumped after the stack trace is logged. To inhibit this behavior use --define=signal_trace=disabled on the Bazel command line. No signal handlers will be installed.

Running a single Bazel test under GDB

bazel build -c dbg //test/common/http:async_client_impl_test
bazel build -c dbg //test/common/http:async_client_impl_test.dwp
gdb bazel-bin/test/common/http/async_client_impl_test

We need to use -c dbg Bazel option to generate debugging symbols and without that GDB will not be very useful. The debugging symbols are stored as separate debugging information files (.dwo files) and we can build a DWARF package file with .dwp target. The .dwp file need to be presented in the same folder with the binary for a full debugging experience.

Running Bazel tests requiring privileges

Some tests may require privileges (e.g. CAP_NET_ADMIN) in order to execute. One option is to run them with elevated privileges, e.g. sudo test. However, that may not always be possible, particularly if the test needs to run in a CI pipeline. tools/ may be used in such situations to run the tests in a privileged docker container.

The script works by wrapping the test execution in the current repository's circle ci build container, then executing it either locally or on a remote docker container. In both cases, the container runs with the --privileged flag, allowing it to execute operations which would otherwise be restricted.

The command line format is: tools/ <bazel-test-target> [optional-flags-to-bazel]

The script uses two optional environment variables to control its behaviour:

  • RUN_REMOTE=<yes|no>: chooses whether to run on a remote docker server.
  • LOCAL_MOUNT=<yes|no>: copy/mount local libraries onto the docker container.

Use RUN_REMOTE=yes when you don't want to run against your local docker instance. Note that you will need to override a few environment variables to set up the remote docker. The list of variables can be found in the Documentation.

Use LOCAL_MOUNT=yes when you are not building with the Envoy build container. This will ensure that the libraries against which the tests dynamically link will be available and of the correct version.


Running the http integration test in a privileged container:

tools/  //test/integration:integration_test --jobs=4 -c dbg

Running the http integration test compiled locally against a privileged remote container:

RUN_REMOTE=yes MOUNT_LOCAL=yes tools/  //test/integration:integration_test \
  --jobs=4 -c dbg

Additional Envoy build and test options

In general, there are 3 compilation modes that Bazel supports:

  • fastbuild: -O0, aimed at developer speed (default).
  • opt: -O2 -DNDEBUG -ggdb3 -gsplit-dwarf, for production builds and performance benchmarking.
  • dbg: -O0 -ggdb3 -gsplit-dwarf, no optimization and debug symbols.

You can use the -c <compilation_mode> flag to control this, e.g.

bazel build -c opt //source/exe:envoy-static

To override the compilation mode and optimize the build for binary size, you can use the sizeopt configuration:

bazel build //source/exe:envoy-static --config=sizeopt


To build and run tests with the gcc compiler's address sanitizer (ASAN) and undefined behavior (UBSAN) sanitizer enabled:

bazel test -c dbg --config=asan //test/...

The ASAN failure stack traces include line numbers as a result of running ASAN with a dbg build above. If the stack trace is not symbolized, try setting the ASAN_SYMBOLIZER_PATH environment variable to point to the llvm-symbolizer binary (or make sure the llvm-symbolizer is in your $PATH).

If you have clang-5.0 or newer, additional checks are provided with:

bazel test -c dbg --config=clang-asan //test/...

Thread sanitizer (TSAN) tests rely on a TSAN-instrumented version of libc++ and can be run under the docker sandbox:

bazel test -c dbg --config=docker-tsan //test/...

Alternatively, you can build a local copy of TSAN-instrumented libc++. Follow the quick start instruction to setup Clang+LLVM environment. Download LLVM sources from the LLVM official site

curl -sSfL "" | tar zx

Configure and build a TSAN-instrumented libc++. Please note that LLVM_USE_SANITIZER=Thread preprocessor definition is used to enable TSAN instrumentation, and CMAKE_INSTALL_PREFIX="/opt/libcxx_tsan" defines the installation directory path.

mkdir tsan
pushd tsan

  -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_INSTALL_PREFIX="/opt/libcxx_tsan" "../llvm-project-llvmorg-11.0.1/llvm"
ninja install-cxx install-cxxabi

rm -rf /opt/libcxx_tsan/include

Generate local_tsan.bazelrc containing bazel configuration for tsan tests:

bazel/ </path/to/instrumented/libc++/home>

To execute TSAN tests using the local instrumented libc++ library pass --config=local-tsan to bazel:

bazel test --config=local-tsan //test/...

For memory sanitizer (MSAN) testing, it has to be run under the docker sandbox which comes with MSAN instrumented libc++:

bazel test -c dbg --config=docker-msan //test/...

To run the sanitizers on OS X, prefix macos- to the config option, e.g.:

bazel test -c dbg --config=macos-asan //test/...

Log Verbosity

Log verbosity is controlled at runtime in all builds.

To obtain nghttp2 traces, you can set ENVOY_NGHTTP2_TRACE in the environment for enhanced logging at -l trace. For example, in tests:

bazel test //test/integration:protocol_integration_test --test_output=streamed \
  --test_arg="-l trace" --test_env="ENVOY_NGHTTP2_TRACE="

Similarly, QUICHE verbose logs can be enabled by setting ENVOY_QUICHE_VERBOSITY=n in the environment where n is the desired verbosity level (e.g. --test_env="ENVOY_QUICHE_VERBOSITY=2".

Disabling optional features

The following optional features can be disabled on the Bazel build command-line:

  • Hot restart with --define hot_restart=disabled
  • Google C++ gRPC client with --define google_grpc=disabled
  • Backtracing on signals with --define signal_trace=disabled
  • Active stream state dump on signals with --define signal_trace=disabled or --define disable_object_dump_on_signal_trace=disabled
  • tcmalloc with --define tcmalloc=disabled. Also you can choose Gperftools' implementation of tcmalloc with --define tcmalloc=gperftools which is the default for builds other than x86_64 and aarch64.
  • deprecated features with --define deprecated_features=disabled
  • http3/quic with --//bazel:http3=False

Enabling optional features

The following optional features can be enabled on the Bazel build command-line:

  • Exported symbols during linking with --define exported_symbols=enabled. This is useful in cases where you have a lua script that loads shared object libraries, such as those installed via luarocks.
  • Perf annotation with --define perf_annotation=enabled (see source/common/common/perf_annotation.h for details).
  • BoringSSL can be built in a FIPS-compliant mode with --define boringssl=fips (see FIPS 140-2 for details).
  • ASSERT() can be configured to log failures and increment a stat counter in a release build with --define log_fast_debug_assert_in_release=enabled. SLOW_ASSERT()s can be included with --define log_debug_assert_in_release=enabled. The default behavior is to compile all debug assertions out of release builds so that the condition is not evaluated. This option has no effect in debug builds.
  • memory-debugging (scribbling over memory after allocation and before freeing) with --define tcmalloc=debug. Note this option cannot be used with FIPS-compliant mode BoringSSL and tcmalloc is built from the sources of Gperftools.
  • Default path normalization with --define path_normalization_by_default=true. Note this still could be disable by explicit xDS config.
  • Manual stamping via VersionInfo with --define manual_stamp=manual_stamp. This is needed if the version_info_lib is compiled via a non-binary bazel rules, e.g envoy_cc_library. Otherwise, the linker will fail to resolve symbols that are included via the linktamp rule, which is only available to binary targets. This is being tracked as a feature in:
  • Process logging for Android applications can be enabled with --define logger=android.
  • Excluding assertions for known issues with --define disable_known_issue_asserts=true. A KNOWN_ISSUE_ASSERT is an assertion that should pass (like all assertions), but sometimes fails for some as-yet unidentified or unresolved reason. Because it is known to potentially fail, it can be compiled out even when DEBUG is true, when this flag is set. This allows Envoy to be run in production with assertions generally enabled, without crashing for known issues. KNOWN_ISSUE_ASSERT should only be used for newly-discovered issues that represent benign violations of expectations.
  • Envoy can be linked to zlib-ng instead of zlib with --define zlib=ng.

Enabling and disabling extensions

Envoy uses a modular build which allows extensions to be removed if they are not needed or desired. Extensions that can be removed are contained in extensions_build_config.bzl. Contrib build extensions are contained in contrib_build_config.bzl. Note that contrib extensions are only included by default when building the contrib executable and in the default contrib images pushed to Docker Hub.

The extensions disabled by default can be enabled by adding the following parameter to Bazel, for example to enable envoy.filters.http.kill_request extension, add --//source/extensions/filters/http/kill_request:enabled. The extensions enabled by default can be disabled by adding the following parameter to Bazel, for example to disable envoy.wasm.runtime.v8 extension, add --//source/extensions/wasm_runtime/v8:enabled=false. Note not all extensions can be disabled.

If you're building from a custom build repository, the parameters need to prefixed with @envoy, for example --@envoy//source/extensions/filters/http/kill_request:enabled.

You may persist those options in user.bazelrc in Envoy repo or your .bazelrc.

Contrib extensions can be enabled and disabled similarly to above when building the contrib executable. For example:

bazel build //contrib/exe:envoy-static --//contrib/squash/filters/http/source:enabled=false

Will disable the squash extension when building the contrib executable.

Customize extension build config

You can also use the following procedure to customize the extensions for your build:

  • The Envoy build assumes that a Bazel repository named @envoy_build_config exists which contains the file @envoy_build_config//:extensions_build_config.bzl. In the default build, a synthetic repository is created containing extensions_build_config.bzl.
  • Start by creating a new Bazel workspace somewhere in the filesystem that your build can access. This workspace should contain:
    • Empty WORKSPACE file.
    • Empty BUILD file.
    • A copy of extensions_build_config.bzl.
    • Comment out any extensions that you don't want to build in your file copy.

To have your local build use your overridden configuration repository there are two options:

  1. Use the --override_repository CLI option to override the @envoy_build_config repo.
  2. Use the following snippet in your WORKSPACE before you load the Envoy repository. E.g.,
workspace(name = "envoy_filter_example")

    name = "envoy_build_config",
    # Relative paths are also supported.
    path = "/somewhere/on/filesystem/envoy_build_config",

    name = "envoy",
    # Relative paths are also supported.
    path = "/somewhere/on/filesystem/envoy",


When performing custom builds, it is acceptable to include contrib extensions as well. This can be done by including the desired Bazel paths from contrib_build_config.bzl into the overriden extensions_build_config.bzl. (There is no need to specifically perform a contrib build to include a contrib extension.)

Extra extensions

If you are building your own Envoy extensions or custom Envoy builds and encounter visibility problems with, you may need to adjust the default visibility rules to be public, as documented in extensions_build_config.bzl. See the instructions above about how to create your own custom version of extensions_build_config.bzl.

Release builds

Release builds should be built in opt mode, processed with strip and have a section with the Git SHA1 at which the build took place. They should also ignore any local .bazelrc for reproducibility. This can be achieved with:

bazel --bazelrc=/dev/null build -c opt //source/exe:envoy-static.stripped

One caveat to note is that the Git SHA1 is truncated to 16 bytes today as a result of the workaround in place for

Coverage builds

To generate coverage results, make sure you are using a Clang toolchain and have llvm-cov and llvm-profdata in your PATH. Then run:


Note that it is important to ensure that the versions of clang, llvm-cov and llvm-profdata are consistent and that they match the most recent Clang/LLVM toolchain version in use by Envoy (see the build container toolchain for reference).

The summary results are printed to the standard output and the full coverage report is available in generated/coverage/coverage.html.

To generate coverage results for fuzz targets, use the FUZZ_COVERAGE environment variable, e.g.:


This generates a coverage report for fuzz targets after running the target for one minute against fuzzing engine libfuzzer using its coprus as initial seed inputs. The full coverage report will be available in generated/fuzz_coverage/coverage.html.

Coverage for every PR is available in Circle in the "artifacts" tab of the coverage job. You will need to navigate down and open "coverage.html" but then you can navigate per normal. NOTE: We have seen some issues with seeing the artifacts tab. If you can't see it, log out of Circle, and then log back in and it should start working.

The latest coverage report for main is available here. The latest fuzz coverage report for main is available here.

It's also possible to specialize the coverage build to a specified test or test dir. This is useful when doing things like exploring the coverage of a fuzzer over its corpus. This can be done by passing coverage targets as the command-line arguments and using the VALIDATE_COVERAGE environment variable, e.g. for a fuzz test:

FUZZ_COVERAGE=true VALIDATE_COVERAGE=false test/ //test/common/common:base64_fuzz_test

Cleaning the build and test artifacts

bazel clean will nuke all the build/test artifacts from the Bazel cache for Envoy proper. To remove the artifacts for the external dependencies run bazel clean --expunge.

If something goes really wrong and none of the above work to resolve a stale build issue, you can always remove your Bazel cache completely. It is likely located in ~/.cache/bazel.

Adding or maintaining Envoy build rules

See the developer guide for writing Envoy Bazel rules.

Bazel performance on (virtual) machines with low resources

If the (virtual) machine that is performing the build is low on memory or CPU resources, you can override Bazel's default job parallelism determination with --jobs=N to restrict the build to at most N simultaneous jobs, e.g.:

bazel build --jobs=2 //source/exe:envoy-static

Debugging the Bazel build

When trying to understand what Bazel is doing, the -s and --explain options are useful. To have Bazel provide verbose output on which commands it is executing:

bazel build -s //source/exe:envoy-static

To have Bazel emit to a text file the rationale for rebuilding a target:

bazel build --explain=file.txt //source/exe:envoy-static

To get more verbose explanations:

bazel build --explain=file.txt --verbose_explanations //source/exe:envoy-static

Resolving paths in bazel build output

Sometimes it's useful to see real system paths in bazel error message output (vs. symbolic links). tools/ is provided to help with this. See the comments in that file.

Compilation database

Run tools/ to generate a JSON Compilation Database. This could be used with any tools (e.g. clang-tidy) compatible with the format. It is recommended to run this script with TEST_TMPDIR set, so the Bazel artifacts doesn't get cleaned up in next bazel build or bazel test.

The compilation database could also be used to setup editors with cross reference, code completion. For example, you can use You Complete Me or clangd with supported editors.

This requires Python 3.8.0+, download from here if you do not have it installed already.

Use the following command to prepare a compilation database:

TEST_TMPDIR=/tmp tools/

Running format linting without docker

The easiest way to run the clang-format check/fix commands is to run them via docker, which helps ensure the right toolchain is set up. However you may prefer to run clang-format scripts on your workstation directly:

  • It's possible there is a speed advantage
  • Docker itself can sometimes go awry and you then have to deal with that
  • Type-ahead doesn't always work when waiting running a command through docker

To run the tools directly, you must install the correct version of clang. This may change over time, check the version of clang in the docker image. You must also have 'buildifier' installed from the bazel distribution.

Note that if you run the script you will need to have aspell installed.

Edit the paths shown here to reflect the installation locations on your system:

export CLANG_FORMAT="$HOME/ext/clang+llvm-11.0.1-x86_64-linux-gnu-ubuntu-20.04/bin/clang-format"
export BUILDIFIER_BIN="/usr/bin/buildifier"

Once this is set up, you can run clang-format without docker:

./tools/code_format/ check
./tools/spelling/ check
./tools/code_format/ fix
./tools/spelling/ fix

Advanced caching setup

Setting up an HTTP cache for Bazel output helps optimize Bazel performance and resource usage when using multiple compilation modes or multiple trees.

Setup local cache

You may use any Remote Caching backend as an alternative to this.

This requires Go 1.11+, follow the instructions to install if you don't have one. To start the cache, run the following from the root of the Envoy repository (or anywhere else that the Go toolchain can find the necessary dependencies):

go run --dir ${HOME}/bazel_cache --host --port 28080 --max_size 64

See Bazel remote cache for more information on the parameters. The command above will setup a maximum 64 GiB cache at ~/bazel_cache on port 28080. You might want to setup a larger cache if you run ASAN builds.

NOTE: Using docker to run remote cache server described in remote cache docs will likely have slower cache performance on macOS due to slow disk performance on Docker for Mac.

Adding the following parameter to Bazel everytime or persist them in .bazelrc.