torchci

pytorch-probot

A GitHub App built with Probot that implements bot actions for PyTorch

This bot implements a few behaviors. This bot currently only implements idempotent behaviors (i.e., it is harmless if the bot process events multiple times. If you add support for non-idempotent behaviors, you need to make sure only the GitHub Action or AWS Lambda is enabled.

auto-cc-bot

Add an issue to your project like https://github.com/pytorch/pytorch/issues/24422 and add a .github/pytorch-probot.yml file with:

tracking_issue: 24422

Based on who is listed in the tracking issue, the bot will automatically CC people when labels are added to an issue.

auto-label-bot

If an issue is labeled high priority, also label it triage review
If an issue is labeled topic: flaky-tests, also label it high priority and triage review
If an issue or pull request contains a regex in its title, label it accordingly, e.g., a title containing 'ROCm' would yield the module: rocm label.

trigger-circleci-workflows

Trigger circleci workflows based off of labeling events / push events

Configuration (.github/pytorch-circleci-labels.yml) should look similar to this:

labels_to_circle_params:
  # Refers to github labels
  ci/binaries:
    # Refers to circleci parameters
    # For circleci documentation on pipeline parameters check:
    #      https://circleci.com/docs/2.0/pipeline-variables/#pipeline-parameters-in-configuration
    parameter: run_binaries_tests
    # [[optional]] Automatically trigger workflows with parameters on push
    default_true_on:
      branches:
        - nightly
        # Regex is allowed as well
        - ci-all/.*
      # Even works on tags!
      tags:
        - v[0-9]+(\.[0-9]+)*-rc[0-9]+
  # Multiple label / parameters can be defined
  ci/bleh:
    parameter: run_bleh_tests
  ci/foo:
    parameter: run_foo_tests

Setup

# Install dependencies
yarn install

# Run the tests
yarn test

# Run the bot
yarn start

Live testing as a GitHub App

If you want to smoketest the bot on a test repository, you'll need to create a GitHub app. Go to the webpage from probot; it will walk through the process.

Deploying GitHub Actions

Although a GitHub App is convenient for testing, it requires an actual server to deploy in prod. Previously we ran the server on AWS, but this deployment process was substantially more involved. GitHub Actions deployment is simpler. Follow the instructions at https://github.com/actions/toolkit/blob/master/docs/action-versioning.md

Right now the GitHub Actions deployment is a little rocky because massive queueing in the PyTorch repository means it takes something like 30min before actions are run. So we are also running AWS side-by-side.

Deploying to AWS

.github/workflows/build.yml will build and deploy the code on every push to main.

Contributing

If you have suggestions for how pytorchbot could be improved, or want to report a bug, open an issue! We'd love all and any contributions.

For more, check out the Contributing Guide.

pytorch / probot

readme