Google GitGoogle Git
Sign in
llvm/llvm-zorg/b6512fa9e017a3958ec7a976e304d5f4ef6cfea5/./google-bazel-bot
tree: b2512f98b248719efb1da77e63136fd2c57e954f
  1. bazel-bot/
  2. docker/
  3. backend.tf
  4. bazel-buildkite.yaml
  5. bazel-fixer-bot.yaml
  6. main.tf
  7. README.md
google-bazel-bot/README.md

Google Bazel Bot

This folder contains all of the source code for the AI bazel build fixer bot. The bot uses a combination of deterministic tooling and LLMs to automatically generate PRs that fix the bazel build.

Components

  1. Terraform - The GCP resources that are needed to run the bot. This includes a GKE cluster and Kubernetes resources.
  2. Dockerfile - The container definition used for running both the postcommit bazel CI and the fixer bot.
  3. Fixer bot source code - The actual source code that drives the fix loop and interactes with Github to post a PR if it has generated a fix successfully.

Running locally

The bazel bot fixer that runs in GCP uses GitHub app. It uses a secret to generate short-lived tokens which are used to interact with fork repository at google-bazel-bot/llvm-project and upstream LLVM repository at llvm/llvm-project where PRs are created. For running the bot locally, one must use their own personal fork to act as both fork and PR repository. Fine-grained personal access tokens (PATs) must be generated for the personal repository with read/write permission to Contents, Pull request, Workflow.

By default, running the bot on test commits as follows disables usage of GitHub app token and relies on user-provided environment variable, which must be provided in test mode. For example, bot can be run on test commits as follows:

GOOGLE_CLOUD_PROJECT=llvm-bazel GOOGLE_CLOUD_LOCATION=global GOOGLE_GENAI_USE_VERTEXAI=1 BUILDKITE_API_TOKEN=<yourkey> GITHUB_FORK_USER=<you> GITHUB_FORK_USER_TOKEN=<yourforktoken> python bazelbot_server_main.py --test_commits=test.commits.log --llvm_git_repo=/tmp/bazelbot-llvm-git-repo if using VertexAI

or

GEMINI_API_KEY=<yourkey> GOOGLE_GENAI_USE_VERTEXAI=0 BUILDKITE_API_TOKEN=<yourkey> GITHUB_FORK_USER=<you> GITHUB_FORK_USER_TOKEN=<yourforktoken> python bazelbot_server_main.py --test_commits=test.commits.log --llvm_git_repo=/tmp/bazelbot-llvm-git-repo if using Gemini API.