This repository contains a set of scripts and configuration files that can be used to deploy Buildbarn on various platforms. Buildbarn is pretty flexible, in that it can both be used for single-node remote caching setups and large-scale remote execution setups. Unless noted otherwise, the configurations in this repository all use assume the following setup:
- Sharded storage, using the Buildbarn storage daemon. To apply the sharding to client RPCs, a separate set of stateless frontend servers is used to fan out requests.
- Remote execution of build actions, using container images from Act, distributed by catthehacker, used for running GitHub Actions locally under Ubuntu 22.04.
- An installation of the Buildbarn Browser.
Below is a diagram of what this setup Buildbarn looks like. In this diagram, the arrows represent the direction in which network connections are established.
This repository currently demonstrates the following versions. Binaries can be found under the CI Build link, at the top right in the GitHub Actions page.
| Repository | Container images and binaries |
|---|---|
bb-browser af89e4bc662023-09-17 04:20:39 UTC |
ghcr.io/buildbarn/bb-browser:20240930T111151Z-af89e4b CI artifacts |
bb-remote-execution d03d5e37082023-10-04 10:23:25 UTC |
ghcr.io/buildbarn/bb-runner-installer:20241215T094620Z-d03d5e3 ghcr.io/buildbarn/bb-scheduler:20241215T094620Z-d03d5e3 ghcr.io/buildbarn/bb-worker:20241215T094620Z-d03d5e3 CI artifacts |
bb-storage 078d9d76e02023-10-08 11:11:12 UTC |
ghcr.io/buildbarn/bb-storage:20241212T082716Z-078d9d7 CI artifacts |
A changelog for each update to bb-deployments is maintained in changelog.md.
The Meroton blog describes some of the changes and features in more detail and includes Jsonnet configuration migration examples.
This example aims to showcase a very simple build and test with remote execution using docker-compose as the deployment for Buildbarn. We will be compiling examples from the abseil-hello project using Bazel.
First clone the repo and start up a docker-compose example:
git clone https://github.com/buildbarn/bb-deployments.git
cd bb-deployments/docker-compose
./run.sh
You may see initially see an error message along the lines of:
worker-ubuntu22-04_1 | xxxx/xx/xx xx:xx:xx rpc error: code = Unavailable desc = Failed to ...: connection error: desc = "transport: Error while dialing dial tcp xxx.xx.x.x:xxxx: connect: connection refused"
This is usually because container of the worker has started before the scheduler or runner and so it cannot connect to them. After a second or so, this error message should stop.
Bazel can perform remote builds against these deployments by using toolchains
adapted to the remote environment. The script
tools/remote-toolchains/extract-bazel-auto-toolchains.sh has been used to
construct such a C++ toolchain which is activated using
--config=remote-ubuntu-22-04, see .bazelrc and WORKSPACE for the exact
setup.
Note that the name and SHA of the container image is configured in multiple
places: BUILD.bazel, .jsonnet configuration for Buildbarn and for the actual
runner container (docker compose, kubernetes yaml, etc...)
Now try a build (using bazel or bazelisk):
bazel build --config=remote-ubuntu-22-04 @abseil-hello//:hello_main
The output should look something like:
INFO: 33 processes: 4 internal, 29 remote.
INFO: Build completed successfully, 33 total actions
You can check to see if the binary has built successfully by trying:
bazel run --config=remote-ubuntu-22-04 @abseil-hello//:hello_main
You may experience problems with wrong version of glibc compared to what the remote is building for.
Equally, you can try to execute a test remotely:
bazel test --config=remote-ubuntu-22-04 @abseil-hello//:hello_test
Which will give you an output containing something like:
INFO: 49 processes: 4 internal, 45 remote.
INFO: Build completed successfully, 49 total actions
@abseil-hello//:hello_test PASSED in 0.1s
Executed 1 out of 1 test: 1 test passes.
You might experience problems loading abseil-hellos/libhello.so, in which case
you can link statically by using --dynamic_mode=off.
Next, we will try out the remote caching capability. If you clean your local build cache and then rerun a build:
bazel clean
bazel build --config=remote-ubuntu-22-04 @abseil-hello//:hello_main
You'll see an output containing information that we hit the remote cache instead of executing on a worker.
There is a Buildbarn example in the Buck2 repository: here Platform properties are defined in platform/defs.bzl and the rpc endpoints are set in .buckconfig.
Pants defines the endpoint and the properties in the main configuration file pants.toml.
[GLOBAL]
remote_cache_read = true
remote_cache_write = true
remote_store_address = "grpc://localhost:8980"
remote_execution_address = "grpc://localhost:8980"
remote_execution = true
remote_instance_name = "fuse"
remote_execution_extra_platform_properties = [
"OSFamily=linux",
"container-image=docker://ghcr.io/catthehacker/ubuntu:act-22.04@sha256:5f9c35c25db1d51a8ddaae5c0ba8d3c163c5e9a4a6cc97acd409ac7eae239448",
]
It is possible to use goma, the buildsystem in the chromium project with Buildbarn.
The instructions are available here: docs/goma.md.
You do not need to define a toolchain for remote execution, like this repository does. For simple projects where all actions can build with the same executors you can use set the platform properties as command line arguments.
bazel build \
--remote_executor=grpc://localhost:8980 \
--remote_instance_name=fuse \
--remote_default_exec_properties OSFamily=linux \
--remote_default_exec_properties container-image="docker://ghcr.io/catthehacker/ubuntu:act-22.04@sha256:5f9c35c25db1d51a8ddaae5c0ba8d3c163c5e9a4a6cc97acd409ac7eae239448" \
@abseil-hello//:hello_main
There is a #buildbarn channel on buildteamworld.slack.com that you can join to get in touch with other people who use and hack on Buildbarn.
Via our partners commercial support & hosting can be procured.
First make sure the different Buildbarn components are in sync. Then perform:
# Update go.mod.
go mod tidy -e
# Regenerate go_dependencies.bzl.
bazel run //:gazelle -- update-repos -from_file=go.mod -to_macro go_dependencies.bzl%go_dependencies -prune
# Format go_dependencies.bzl according to GitHub Actions.
sed -i '/^$/d' go_dependencies.bzl
bazelisk run //:buildifier.check
# Update the Kubernetes and Docker compose deployments.
./tools/update-container-image-versions.shYou might have to update the WORKSPACE file as well,
until MODULE.bazel is in place.
A number of linting and formatting steps are performed in the GitHub Actions flow. Some of the steps are:
# Gazelle
bazel run //:gazelle -- update-repos -from_file=go.mod -to_macro go_dependencies.bzl%go_dependencies -prune
bazel run //:gazelle
# Buildifier
sed '/^$/d' go_dependencies.bzl > go_dependencies.bzl.new
mv go_dependencies.bzl.new go_dependencies.bzl
bazel run //:buildifier.check
# Gofmt
bazel run @cc_mvdan_gofumpt//:gofumpt -- -w -extra $PWD
# Golint
bazel run @org_golang_x_lint//golint -- -set_exit_status $PWD/...It is annoying to keep it up-to date. It seems to depend on the host system setup. The best way to solve CI errors is just to download the suggested patch from the action and apply it directly.
- Open the failing Github action.
- Click 'Download log archive' in the cog menu in the upper right.
- Unpack the archive.
- Format a useful patch from
logs/build_and_test/13_Test style conformance.txt.
Rough inspiration:
cp ~/Downloads/logs_*.zip .
aunpack logs_*.zip
cp logs_*/build_and_test/*'_Test style conformance.txt' lockfile.patch
# Remove git action metadata
sed -i -e '1,4d' -e '$d' lockfile.patch
# Remove timestamps
sed -i -E 's/^.{29}//' lockfile.patch
git apply lockfile.patch
git add MODULE.bazel.lock