Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

ci: add a github bot to support advanced PR review workflows #3037

Open
wants to merge 19 commits into
base: master
Choose a base branch
from

Conversation

aeddi
Copy link
Contributor

@aeddi aeddi commented Oct 28, 2024

This pull request aims to add a bot that extends GitHub's functionalities like codeowners file and other merge protection mechanisms. Interaction with the bot is done via a comment. You can test it on the demo repo here : GnoCheckBot/demo#1

Fixes #1007
Related to #1466, #2788

  • The config.go file contains all the conditions and requirements in an 'If - Then' format.
// Automatic check

{
  Description: "Changes to 'tm2' folder should be reviewed/authored by at least one member of both EU and US teams",
  If: c.And(
    c.FileChanged(gh, "tm2"),
    c.BaseBranch("main"),
  ),
  Then: r.And(
    r.Or(
      r.ReviewByTeamMembers(gh, "eu", 1),
      r.AuthorInTeam(gh, "eu"),
    ),
    r.Or(
      r.ReviewByTeamMembers(gh, "us", 1),
      r.AuthorInTeam(gh, "us"),
    ),
  ),

}
  • There are two types of checks: some are automatic and managed by the bot (like the one above), while others are manual and need to be verified by a specific org team member (like the one below). If no team is specified, anyone with comment editing permission can check it.
// Manual check
{
  Description: "The documentation is accurate and relevant",
  If:          c.FileChanged(gh, `.*\.md`),
  Teams: []string{
    "tech-staff",
    "devrels",
  },
},
  • The conditions (If) allow checking, among other things, who the author is, who is assigned, what labels are applied, the modified files, etc. The list is available in the condition folder.
  • The requirements (Then) allow, among other things, assigning a member, verifying that a review is done by a specific user, applying a label, etc. (List in requirement folder).
  • A PR Check (the icon at the bottom with all the CI checks) will remain orange/pending until all checks are validated, after which it will turn green.
Screenshot 2024-11-05 at 18 37 34
  • The Github Actions workflow associated with the bot ensures that PRs are processed concurrently, while ensuring that the same PR is not processed by two runners at the same time.
  • We can manually process a PR by launching the workflow directly from the GitHub Actions interface.
Screenshot 2024-11-06 at 01 36 42

To do

  • implement base version of the bot
  • cleanup code / comments
  • setup a demo repo
  • add debug printing on dry run
  • add some tests on requirements and conditions
Contributors' checklist...
  • Added new tests, or not needed, or not feasible
  • Provided an example (e.g. screenshot) to aid review or the PR is self-explanatory
  • Updated the official documentation or not needed
  • No breaking changes were made, or a BREAKING CHANGE: xxx message was included in the description
  • Added references to related issues and PRs
  • Provided any useful hints for running manual tests

@aeddi aeddi self-assigned this Oct 28, 2024
@aeddi aeddi marked this pull request as draft October 28, 2024 10:00
Copy link

codecov bot commented Oct 28, 2024

Codecov Report

Attention: Patch coverage is 60.94840% with 280 lines in your changes missing coverage. Please review.

Project coverage is 63.79%. Comparing base (4f27a57) to head (b12be4d).
Report is 9 commits behind head on master.

Files with missing lines Patch % Lines
contribs/github-bot/client/client.go 35.32% 116 Missing and 3 partials ⚠️
contribs/github-bot/param/param.go 0.00% 70 Missing ⚠️
contribs/github-bot/logger/terminal.go 0.00% 23 Missing ⚠️
contribs/github-bot/param/prlist.go 0.00% 18 Missing ⚠️
contribs/github-bot/logger/action.go 0.00% 14 Missing ⚠️
contribs/github-bot/requirement/branch.go 69.23% 6 Missing and 2 partials ⚠️
contribs/github-bot/requirement/reviewer.go 91.78% 4 Missing and 2 partials ⚠️
contribs/github-bot/logger/logger.go 28.57% 5 Missing ⚠️
contribs/github-bot/logger/noop.go 37.50% 5 Missing ⚠️
contribs/github-bot/condition/file.go 85.18% 3 Missing and 1 partial ⚠️
... and 2 more
Additional details and impacted files
@@            Coverage Diff             @@
##           master    #3037      +/-   ##
==========================================
+ Coverage   63.77%   63.79%   +0.01%     
==========================================
  Files         548      573      +25     
  Lines       78681    79587     +906     
==========================================
+ Hits        50180    50773     +593     
- Misses      25117    25407     +290     
- Partials     3384     3407      +23     
Flag Coverage Δ
contribs/github-bot 60.94% <60.94%> (?)
contribs/gnodev 60.54% <ø> (ø)
contribs/gnofaucet 14.82% <ø> (ø)
gno.land 73.70% <ø> (+0.07%) ⬆️
gnovm 67.93% <ø> (+<0.01%) ⬆️
misc/genstd 79.72% <ø> (ø)
tm2 62.49% <ø> (+0.02%) ⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@zivkovicmilos zivkovicmilos self-requested a review October 28, 2024 10:04
@thehowl
Copy link
Member

thehowl commented Oct 30, 2024

@aeddi would you like a preliminary review?

@ltzmaxwell
Copy link
Contributor

could you please add a brief description to give people an early preview, even if it's draft?

@aeddi
Copy link
Contributor Author

aeddi commented Nov 5, 2024

@thehowl Sorry for the delayed response. I hadn't seen your message. I still have some cleanup to do and need to figure out how to write some tests, but I think the PR is pre-reviewable now.
Thanks!

@ltzmaxwell Done ;)

@aeddi aeddi marked this pull request as ready for review November 6, 2024 00:51
@ltzmaxwell
Copy link
Contributor

I haven't looked into this in depth yet, but I'm considering whether we could introduce an LLM, like ChatGPT, to assist with code review. It could help summarize PR modifications and potentially identify issues in the code. This is outside the current scope, but I wanted to leave some thoughts here for discussion.

@aeddi
Copy link
Contributor Author

aeddi commented Nov 6, 2024

Yes, why not have this kind of tool just to detect details we might have missed?
I know it's a controversial topic, but personally, knowing that our code is completely open-source, I wouldn't mind having a report from an LLM to pick out only the relevant suggestions.

Just quickly checking, I see there are Actions on the Github marketplace dedicated to this.

@moul
Copy link
Member

moul commented Nov 12, 2024

When do you think we should start using it and providing feedback for iteration?

@aeddi
Copy link
Contributor Author

aeddi commented Nov 12, 2024

@moul As soon as it's reviewed and merged, I think we can start by setting up a few simple rules to make sure everything is working well, then iterate from there.
The problem is that it's tricky to test a bot based on GitHub Actions. I've done a lot of testing on two private repos, but I hope there aren't any issues I haven't anticipated.
For example, we can test safe operations like requesting reviews from certain users + adding a manual check, or something like this.

BTW, the failing check is because the bot workflow added to this PR is trying to run while the repo isn't configured yet (specifically, the workflow requires setting a GitHub Actions secret and variable).

Copy link
Contributor

@ajnavarro ajnavarro left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this way of implementing a bot is not secure. Giving random access to code execution to anyone creating a PR is not a good idea IMO.

Comment on lines +7 to +15
- assigned
- unassigned
- labeled
- unlabeled
- opened
- reopened
- synchronize # PR head updated
- converted_to_draft
- ready_for_review
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All existing types for reference:

- assigned
- unassigned
- labeled
- unlabeled
- opened
- edited
- closed
- reopened
- synchronize
- converted_to_draft
- locked
- unlocked
- enqueued
- dequeued
- milestoned
- demilestoned
- ready_for_review
- review_requested
- review_request_removed
- auto_merge_enabled
- auto_merge_disabled

Why don't just run on all types?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I only included the ones that were used by the existing conditions/requirements of the bot. It seems better to me than subscribing to all possible events for several reasons :

  • It makes it clearer which events the bot uses
  • in term of security, it follows the PoLP (even though a trigger is not exactly a privilege)
  • it prevents the workflow being triggered and launching two runners for nothing

With the idea that if we added, for example, a condition/requirement based on adding to/removing from a milestone, we would just add the corresponding trigger event to the workflow.


# Watch for changes on PR comment
issue_comment:
types: [created, edited, deleted]
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The same, why don't just run on all types and let the bot decide what to do?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same as above.

go-version-file: go.mod

- name: Run GitHub Bot
working-directory: contribs/github-bot
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

More than a bot is a script, right? I was expecting a server handling events through the GitHub API with a state.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

When we defined the specifications with @moul and @thehowl (I can't remember who else was on the call we had in Turin about it), one of the requirements was that the bot shouldn't run on a server but that everything should be processed through GitHub Actions.

It makes things more complicated to test and the manual checks more complicated to implement using the comment body as their state instead of a DB or similar, but it has the advantage of having a piece of code that runs in a serverless way.

But if the name is misleading, we can rename it.

# This job creates a matrix of PR numbers based on the inputs from the various
# events that can trigger this workflow so that the process-pr job below can
# handle the parallel processing of the pull-requests
define-prs-matrix:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't fully understand this job. Do you mind to explain it?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes sorry if it's not clear enough, so there are two constraints:

  • First, we need to prevent multiple runners from processing the same PR at the same time (to avoid any bug)
  • Second, there are different events that can trigger this workflow, and not all of them provide the PR number(s) to process the same way.

So the goal of this job (define-prs-matrix) is first to establish a list of PR numbers to process from the different possible events, then to format this list as a matrix so that the next job (process-pr) can launch a runner for each PR number in the matrix while ensuring (through the 'concurrency' feature of GitHub Actions) that no PR is processed concurrently.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@aeddi some additional questions here:

  • Is there a scenario where one PR can block another PR? Or concurrency only happen when manually triggered?
  • How much time does a job usually take to complete? If one PR can block another PR and take 10 minutes (e.g., long process, network issue, ...) to complete, it might start to be a problem.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  • The concurrency parameter of the job is really set to prevent a PR X from being processed by two runners in parallel. For example, if PR 1111 is already being processed by one runner and a new event (any event, manual or not) triggers the job to also process PR 1111, the new runner will be queued and will wait for the first runner to finish execution before starting the job. However, the processing of PR 2222 can occur without any issues in parallel with that of 1111.
  • Never saw a run last longer than 1 min, on average the execution time is around 40 secs (and one PR can't block another one anyway ;) )

@aeddi
Copy link
Contributor Author

aeddi commented Nov 13, 2024

I'm not sur to understand what differ from any other action. I mean to run code when opening a PR or any other GitHub event, we already have more than 130 checks that run for each PR we open.
But maybe I didn't fully understand what you meant?

Also in the case of this bot, we own its code in our repo and have total control over it. Which looks more secure to me than using third-party actions.

Still following the PoLP, the only permissions the bot requires right now are those listed in the README, which are not very sensitive IMO. The write permissions are for the scopes Pull Requests (managing comments, labels, milestones, etc.) and Commit Statuses (showing a check at the bottom of the PR). It can't change any code on the repo, for instance.

Do you have another approach to suggest or specific guideline to follow?

@Kouteki Kouteki added the in focus Core team is prioritizing this work label Nov 13, 2024
Copy link
Member

@gfanton gfanton left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

overall LGTM, my only requirement is to get ride of this // nolint:govet and fix cancel leaking.

I'm not sure about @ajnavarro's concern; as long as the token is correctly scoped, it should be fine IMO.

To me, the main issue is that if the PR comes from a fork, the workflow could not be able to grab the bot secret and therefore cannot access the token. Can you (have you?) check the scenario where you trigger this workflow from a branch of an external fork?

from: https://docs.github.com/en/actions/security-for-github-actions/security-guides/using-secrets-in-github-actions#using-secrets-in-a-workflow

With the exception of GITHUB_TOKEN, secrets are not passed to the runner when a workflow is triggered from a forked repository.

contribs/github-bot/client/client.go Outdated Show resolved Hide resolved
contribs/github-bot/client/client.go Outdated Show resolved Hide resolved
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: for consistency's sake among the project, I would prefer that you use ffcli here. You can check other main files in the project; most of them are using the internal command package, which is actually a fork of ffcli.

# This job creates a matrix of PR numbers based on the inputs from the various
# events that can trigger this workflow so that the process-pr job below can
# handle the parallel processing of the pull-requests
define-prs-matrix:
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@aeddi some additional questions here:

  • Is there a scenario where one PR can block another PR? Or concurrency only happen when manually triggered?
  • How much time does a job usually take to complete? If one PR can block another PR and take 10 minutes (e.g., long process, network issue, ...) to complete, it might start to be a problem.

contribs/github-bot/comment.go Outdated Show resolved Hide resolved
This was referenced Nov 14, 2024
@aeddi
Copy link
Contributor Author

aeddi commented Nov 14, 2024

About the PR opened from a fork : I remember we discussed it together, but I totally forgot to take it into account. The workaround is to use the pull_request_target trigger event to enforce the use of the workflow from the base branch instead of the version from the head branch of the PR.

In this context, GitHub allows the workflow to access to GitHub Actions secrets and also allows GITHUB_TOKEN to request for write permission.

I just tested it on the demo repo, and it works. I will make the correction, thanks for the reminder.

Edit : done fb598ee

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
in focus Core team is prioritizing this work
Projects
Status: In Progress
Status: Done
Status: In Review
Development

Successfully merging this pull request may close these issues.

Proposal: improving review process with bot-assisted checkboxes
7 participants