Skip to main content

Workflows configuration

Once you've linked your repo to BuildBuddy via BuildBuddy workflows, BuildBuddy will automatically run bazel test //... on each push to your repo, reporting results to the BuildBuddy UI.

But you may wish to configure multiple test commands with different test tag filters, or run the same tests on multiple different platform configurations (running some tests on Linux, and some on macOS, for example).

This page describes how to configure your workflows beyond the default configuration.

Configuring workflow actions and triggers​

BuildBuddy workflows can be configured using a file called buildbuddy.yaml, which can be placed at the root of your git repo.

buildbuddy.yaml consists of multiple actions. Each action describes a list of commands to be run in order, as well as the set of git events that should trigger these commands.

note

The configuration in buildbuddy.yaml only takes effect after you enable workflows for the repo.

Example config​

You can copy this example config as a starting point for your own buildbuddy.yaml:

buildbuddy.yaml
actions:
  - name: "Test all targets"
    triggers:
      push:
        branches:
          - "main" # <-- replace "main" with your main branch name
      pull_request:
        branches:
          - "*"
    steps:
      - run: "bazel test //..."

This config is equivalent to the default config that we use if you do not have a buildbuddy.yaml file at the root of your repo.

Running bash commands​

Each step can run arbitrary bash code, which may be useful for running Bazel commands conditionally, or for installing system dependencies that aren't available in BuildBuddy's available workflow images.

Because workflows are run in snapshotted microVMs, system dependencies will be persisted across workflow runs. However, we recommend fetching dependencies with Bazel whenever possible, rather than relying on system dependencies.

To specify multiple bash commands, you can either specify a block of bash code within a single step:

buildbuddy.yaml
# ...
steps:
  - run: |
      sudo apt-get update && sudo apt-get install -y my-lib
      bazel test //...

Or you can specify one command per step. Note that each step is run in a separate bash process, so locally initialized variables will not persist across steps:

buildbuddy.yaml
# ...
steps:
  - run: sudo apt-get update && sudo apt-get install -y my-lib
  - run: bazel test //...

Bazel configuration​

Bazel version​

BuildBuddy runs each bazel command in your workflow with a bazelisk-compatible wrapper so that your .bazelversion file is respected.

If .bazelversion is missing, the latest version of Bazel is used. We always recommend including a .bazelversion in your repo to prevent problems caused by using conflicting versions of Bazel in different build environments.

bazelrc​

BuildBuddy runs each bazel command directly in your workspace, which means that your .bazelrc is respected. If you have lots of flags, we recommend adding them to your .bazelrc instead of adding them to your buildbuddy.yaml.

BuildBuddy also provides a bazelrc file which passes these default options to each bazel invocation listed in steps:

  • --bes_backend and --bes_results_url, so that the results from each Bazel command are viewable with BuildBuddy
  • --remote_header=x-buildbuddy-api-key=YOUR_API_KEY, so that invocations are authenticated by default
  • --build_metadata=ROLE=CI, so that workflow invocations are tagged as CI invocations, and so that workflow tests are viewable in the test grid

BuildBuddy's bazelrc takes lower precedence than your workspace .bazelrc. You can view the exact flags provided by this bazelrc by inspecting the command line details in the invocation page (look for buildbuddy.bazelrc).

note

BuildBuddy remote cache and remote execution (RBE) are not enabled by default for workflows, and require additional configuration. The configuration steps are the same as when running Bazel locally. See the Quickstart page in the BuildBuddy UI.

Secrets​

Trusted workflow executions can access secrets using environment variables.

For example, if we have a secret named REGISTRY_TOKEN and we want to set the remote header x-buildbuddy-platform.container-registry-password to the value of that secret, we can get the secret value using $REGISTRY_TOKEN, as in the following example:

buildbuddy.yaml
# ...
steps:
  - run: "bazel test ... --remote_exec_header=x-buildbuddy-platform.container-registry-password=$REGISTRY_TOKEN"

To access the environment variables within build or test actions, you may need to explicitly expose the environment variable to the actions by using a bazel flag like --action_env or --test_env:

buildbuddy.yaml
# ...
steps:
  - run: "bazel test ... --test_env=REGISTRY_TOKEN"

Merge queue support​

BuildBuddy workflows are compatible with GitHub's merge queues.

To ensure that workflows are run as part of merge queue CI, configure a push trigger that runs whenever GitHub pushes its temporary merge queue branch, as described in Triggering merge group checks with third-party CI providers.

Example buildbuddy.yaml file:

buildbuddy.yaml
- action: Test
  triggers:
    push:
      # Run when a merge queue branch is pushed or the main branch is
      # pushed.
      branches: ["main", "gh-readonly-queue/*"]
  # ...

Linux image configuration​

By default, workflows run on an Ubuntu 18.04-based image. You can use a newer, Ubuntu 20.04-based image using the container_image action setting:

buildbuddy.yaml
actions:
  - name: "Test all targets"
    container_image: "ubuntu-20.04" # <-- add this line
    steps:
      - run: "bazel test //..."

The supported values for container_image are "ubuntu-18.04" (default) or "ubuntu-20.04".

By default, workflow VMs have the following resources available:

  • 3 CPU
  • 8 GB of RAM
  • 20 GB of disk space

These values are configurable using resource requests.

Mac configuration​

By default, workflows will execute on BuildBuddy's Linux executors, but it is also possible to run workflows on macOS by using self-hosted executors.

  1. Set up one or more Mac executors that will be dedicated to running workflows, following the steps in the Enterprise Mac RBE Setup guide.

    Then, in your buildbuddy-executor.plist file, find the EnvironmentVariables section and set MY_POOL to workflows. You'll also need to set SYS_MEMORY_BYTES to allow enough memory to be used for workflows (a minimum of 8GB is required).

buildbuddy-executor.plist
        ...
        <key>EnvironmentVariables</key>
        <dict>
            ...
            <!-- Set the required executor pool name for workflows -->
            <key>MY_POOL</key>
            <string>workflows</string>
            <!-- Allocate 16GB of memory to workflows (8GB minimum) -->
            <key>SYS_MEMORY_BYTES</key>
            <string>16000000000</string>
        </dict>
        ...

  1. If you haven't already, enable workflows for your repo, then create a file called buildbuddy.yaml at the root of your repo. See the Example config for a starting point.

  2. Set os: "darwin" on the workflow action that you would like to build on macOS. For Apple silicon (ARM-based) Macs, add arch: "arm64" as well. Note: if you copy another action as a starting point, be sure to give the new action a unique name:

buildbuddy.yaml
actions:
  - name: "Test all targets (Mac)"
    os: "darwin" # <-- add this line
    arch: "arm64" # <-- add this line for Apple silicon (ARM-based) Macs only
    triggers:
      push:
        branches:
          - "main"
      pull_request:
        branches:
          - "*"
    steps:
      - run: "bazel test //..."

That's it! Whenever any of the configured triggers are matched, one of the Mac executors in the workflows pool should execute the workflow, and BuildBuddy will publish the results to your branch.

Attaching Bazel artifacts to workflows​

Bazel supports several flags such as --remote_grpc_log that allow writing additional debug logs and metadata files associated with an invocation.

To provide easy access to these files, BuildBuddy supports a special directory called the workflow artifacts directory. If you write files to this directory, BuildBuddy will automatically upload those files and show them in the UI for the workflow. You can get the path to the workflow artifacts directory using the environment variable $BUILDBUDDY_ARTIFACTS_DIRECTORY.

Example buildbuddy.yaml configuration:

buildbuddy.yaml
actions:
  - name: "Test"
    # ...
    steps:
      - run: "bazel test //... --remote_grpc_log=$BUILDBUDDY_ARTIFACTS_DIRECTORY/grpc.log"

BuildBuddy creates a new artifacts directory for each Bazel command, and recursively uploads all files in the directory after the command exits.

buildbuddy.yaml schema​

BuildBuddyConfig​

The top-level BuildBuddy workflow config, which specifies bazel commands that can be run on a repo, as well as the events that trigger those commands.

Fields:

  • actions (Action list): List of actions that can be triggered by BuildBuddy. Each action corresponds to a separate check on GitHub. If multiple actions are matched for a given event, the actions are run in order. If an action fails, subsequent actions will still be executed.

Action​

A named group of Bazel commands that run when triggered.

Fields:

  • name (string): A name unique to this config, which shows up as the name of the check in GitHub.
  • triggers (Triggers): The triggers that should cause this action to be run.
  • os (string): The operating system on which to run the workflow. Defaults to "linux". "darwin" (macOS) is also supported, but requires using self-hosted Mac executors running on a dedicated workflows pool.
  • arch (string): The CPU architecture of the workflow runner. Defaults to "amd64". "arm64" is also supported when running under os: "darwin", but requires using self-hosted Apple silicon (ARM-based) Mac executors running on a dedicated workflows pool.
  • pool (string): The executor pool name for running workflows. This option has no effect unless self_hosted: true is also specified.
  • self_hosted (boolean): Whether to run the workflow on self-hosted executors. The executor's default isolation type will be used to run workflows. Unless pool is also specified, the configured pool name for the self-hosted workflow executors must be "workflows". This option is ignored for macOS workflows, since macOS workflows are always required to be self-hosted.
  • container_image (string): The Linux container image to use (has no effect for Mac workflows). Supported values are "ubuntu-18.04" and "ubuntu-20.04". Defaults to "ubuntu-18.04".
  • resource_requests (ResourceRequests): the requested resources for this action.
  • user (string): User to run the workflow as. This can be set to "root" to run the workflow as root, but it is recommended to keep the default value, which is a non-root user provisioned in the CI environment (usually named "buildbuddy"). Note: some legacy workflows might still have "root" as the default user, but we are in the process of migrating all users to non-root by default.
  • env (map with string values): Map of static environment variables and their values.
  • git_fetch_filters (string list): list of --filter option values to the git fetch command used when fetching the git commits to build. Defaults to ["blob:none"].
  • git_fetch_depth (int): --depth option value used when fetching the git commits to build. When using this option in combination with a pull_request trigger, it's recommended to set merge_with_base: false in the pull_request trigger, since the limited fetch depth might prevent the merge-base commit from being fetched. Defaults to 0 (unset).
  • git_clean_exclude (string list): List of directories within the workspace that are excluded when running git clean across actions that are executed in the same runner instance. This is an advanced option and is not recommended for most users.
  • bazel_workspace_dir (string): A subdirectory within the repo containing the bazel workspace for this action. By default, this is assumed to be the repo root directory.
  • steps (list): Bash commands to be run in order. If a command fails, subsequent ones are not run, and the action is reported as failed. Otherwise, the action is reported as succeeded. Environment variables are expanded, which means that the commands can reference secrets if the workflow execution is trusted.
  • timeout (duration string, e.g. '30m', '1h'): If set, workflow actions that have been running for longer than this duration will be canceled automatically. This only applies to a single invocation, and does not include multiple retry attempts.

Triggers​

Defines whether an action should run when a branch is pushed to the repo.

Fields:

  • push (PushTrigger): Configuration for push events associated with the repo. This is mostly useful for reporting commit statuses that show up on the home page of the repo.
  • pull_request (PullRequestTrigger): Configuration for pull request events associated with the repo. This is required if you want to use BuildBuddy to report the status of this action on pull requests, and optionally prevent pull requests from being merged if the action fails.

PushTrigger​

Defines whether an action should execute when a branch is pushed.

Fields:

  • branches (string list): The branches that, when pushed to, will trigger the action. This field accepts a simple wildcard character ("*") as a possible value, which will match any branch, as well as "gh-readonly-queue/*", which matches GitHub's merge queue branches.

PullRequestTrigger​

Defines whether an action should execute when a pull request (PR) branch is pushed.

Fields:

  • branches (string list): The base branches of a pull request. For example, if this is set to [ "v1", "v2" ], then the associated action is only run when a PR wants to merge a branch into the v1 branch or the v2 branch. This field accepts a simple wildcard character ("*") as a possible value, which will match any branch.
  • merge_with_base (boolean, default: true): Whether to merge the base branch into the PR branch before running the workflow action. This can help ensure that the changes in the PR branch do not conflict with the main branch. However, the action will not be continuously re-run as changes are pushed to the base branch. For stronger protection against breaking the main branch, you may wish to use merge queues.

ResourceRequests​

Defines the requested resources for a workflow action.

Fields:

  • memory (string or number): Requested amount of memory for the workflow action. Can be specified as an exact number of bytes, or a numeric string containing an IEC unit abbreviation. For example: 8GB represents 8 * (1024)^3 bytes of memory.
  • cpu (string or number): Requested amount of CPU for the workflow action. Can be specified as a number of CPU cores, or a numeric string containing an m suffix to represent thousandths of a CPU core. For example: 8000m represents 8 CPU cores.
  • disk (string or number): Requested amount of disk space for the workflow action. Can be specified as a number of bytes, or a numeric string containing an IEC unit abbreviation. For example: 8GB represents 8 * (1024)^3 bytes of disk space.