[R-package] [ci] Add test on R package with sanitizers

[jameslamb]

This PR adds tests with the sanitizers run by CRAN on package submissions. See this blog post for a lot more background.

The tests take 22 minutes to run, so in this PR I'm proposing that we add them as a manual test that can be triggered by a comment (copying @StrikerRUS 's great work on #3424 ).

This can be triggered by commenting /gha run r-sanitizers-check on a PR.

How this makes LightGBM better

• catches issues like memory violations in lib_lightgbm
• allows us to catch issues in CI to improve the likelihood of CRAN accepting submissions of the R package

[StrikerRUS]

@jameslamb

The tests take 22 minutes to run, so in this PR I'm proposing that we add them as a manual test that can be triggered by a comment

I don't think that 22 min is something unacceptable for us. For example, right now duration of some tests is about 17 min.

So, given that we have 20 free parallel builds for GitHub Actions, I believe we can run new test just as a normal check.

I remember old times before rebalancing CI jobs when we had to wait about 40min-1h 😄 .

[jameslamb]

haha ok! I'll move it back to the main R GitHub Actions then

[StrikerRUS]

@jameslamb BTW, haven't you find a way to make one particular check "optional, but if run then required to be passed"?

[jameslamb]

@jameslamb BTW, haven't you find a way to make one particular check "optional, but if run then required to be passed"?

no, I'm not sure how to do that, sorry!

[StrikerRUS]

Wow! Amazing job as always! I'm pleasantly surprised that we can use container and write just few lines to have a sanitizers checks. Looks great overall, just check my minor comments below.

[StrikerRUS]

Was it left intentionally or it was for debugging purposes?

[jameslamb]

I left it in intentionally, so that if the job fails you have enough logs to be able to tell what went wrong

[StrikerRUS]

OK.

[StrikerRUS]

@jameslamb

no, I'm not sure how to do that, sorry!

No problem! Just wanted to be sure that I'm not discovering something that is already known 🙂 .

Looks like it is impossible to do such things with standard GitHub Actions API. But I have an idea to use REST API and query a status of required workflow via it in our cool all-successful job:

LightGBM/.github/workflows/r_package.yml

Lines 162 to 168 in 186711d

	all-successful:
	# https://git.luolix.topmunity/t/is-it-possible-to-require-all-github-actions-tasks-to-pass-without-enumerating-them/117957/4?u=graingert
	runs-on: ubuntu-latest
	needs: [test]
	steps:
	- name: Note that all tests succeeded
	run: echo "🎉"

https://docs.github.com/en/free-pro-team@latest/rest/reference/actions#get-a-workflow-run

[StrikerRUS]

+1 step towards CRAN, nice!

[StrikerRUS]

@jameslamb What do you think about moving all out R jobs into containers? It looks quite comfortable to have all R infrastructure and do not waste CI time to install it (very often with network errors, BTW) by our own. 🙂

[StrikerRUS]

But I have an idea to use REST API and query a status of required workflow via it in our cool all-successful job:

OK, today I went ahead and prepared a draft for this feature. I tested it a little bit and it looks like it can work 👍 .
But it requires two actions from maintainers: drop a comment to trigger optional workflow and re-run (if already completed) main workflow to fetch the latest status of optional one.

# .ci/check_workflow_status.py
import json
from os import environ
from sys import exit
from time import sleep
from urllib import request


def get_runs():
    with request.urlopen("https://api.github.com/repos/microsoft/LightGBM/actions/workflows/test_1.yml/runs") as url:
        data = json.loads(url.read().decode())
    pr_runs = []
    if environ.get("GITHUB_EVENT_NAME", "") == "pull_request":
        pr_runs = [i for i in data['workflow_runs']
                   if i['event'] == 'pull_request_review_comment' and
                   (i.get('pull_requests') and
                    i['pull_requests'][0]['number'] == int(environ.get("GITHUB_REF").split('/')[-2]) or
                    i['head_branch'] == environ.get("GITHUB_HEAD_REF").split('/')[-1])]
    return sorted(pr_runs, key=lambda i: i['run_number'], reverse=True)


def get_status(runs):
    status = 'ok'
    for run in runs:
        if run['status'] == 'completed':
            if run['conclusion'] in {'cancelled', 'skipped'}:
                continue
            if run['conclusion'] in {'failure', 'timed_out'}:
                status = 'fail'
                break
            if run['conclusion'] == 'success':
                break
        if run['status'] in {'in_progress', 'queued'}:
            status = 'rerun'
            break
    return status


if __name__ == "__main__":
    while True:
        status = get_status(get_runs())
        if status != 'rerun':
            break
        sleep(60)
    if status == 'fail':
        exit(1)

# .github/workflows/test_1.yml
name: Test 1

on:
  pull_request_review_comment:
    types: [created]

jobs:
  test:
    name: Test 1
    runs-on: ubuntu-latest
    if: github.event.comment.body == '/gha run' && contains('OWNER,MEMBER,COLLABORATOR', github.event.comment.author_association)
    timeout-minutes: 60
    strategy:
      fail-fast: false
    steps:
      - name: Checkout repository
        uses: actions/checkout@v1
        with:
          fetch-depth: 5
          submodules: true
      - name: Test
        run: |
            sleep 2m
            exit -1

# .github/workflows/r_package.yml

...

  all-successful:
    # https://git.luolix.topmunity/t/is-it-possible-to-require-all-github-actions-tasks-to-pass-without-enumerating-them/117957/4?u=graingert
    runs-on: ubuntu-latest
    needs: [test]
    steps:
      - name: Checkout repository
        uses: actions/checkout@v1
        with:
          fetch-depth: 5
          submodules: false
      - name: Install Python
        uses: actions/setup-python@v2
        with:
          python-version: '3.x'
      - name: Note that all tests succeeded
        run: python "$GITHUB_WORKSPACE/.ci/check_workflow_status.py"

[StrikerRUS]

But it requires two actions from maintainers: ...

Probably it is possible to simplify it to just one comment with the help of https://docs.github.com/en/free-pro-team@latest/rest/reference/actions#re-run-a-workflow. But I'm not sure about the required permissions though.

[jameslamb]

@jameslamb What do you think about moving all out R jobs into containers? It looks quite comfortable to have all R infrastructure and do not waste CI time to install it (very often with network errors, BTW) by our own. 🙂

I think that would only work for Linux builds, and that that wouldn't help much unfortunately...since most of the temporary errors in CI have been on Mac and Windows.

[jameslamb]

But I have an idea to use REST API and query a status of required workflow via it in our cool all-successful job:

OK, today I went ahead and prepared a draft for this feature. I tested it a little bit and it looks like it can work 👍 .
But it requires two actions from maintainers: drop a comment to trigger optional workflow and re-run (if already completed) main workflow to fetch the latest status of optional one.

# .ci/check_workflow_status.py
import json
from os import environ
from sys import exit
from time import sleep
from urllib import request


def get_runs():
    with request.urlopen("https://api.github.com/repos/microsoft/LightGBM/actions/workflows/test_1.yml/runs") as url:
        data = json.loads(url.read().decode())
    pr_runs = []
    if environ.get("GITHUB_EVENT_NAME", "") == "pull_request":
        pr_runs = [i for i in data['workflow_runs']
                   if i['event'] == 'pull_request_review_comment' and
                   (i.get('pull_requests') and
                    i['pull_requests'][0]['number'] == int(environ.get("GITHUB_REF").split('/')[-2]) or
                    i['head_branch'] == environ.get("GITHUB_HEAD_REF").split('/')[-1])]
    return sorted(pr_runs, key=lambda i: i['run_number'], reverse=True)


def get_status(runs):
    status = 'ok'
    for run in runs:
        if run['status'] == 'completed':
            if run['conclusion'] in {'cancelled', 'skipped'}:
                continue
            if run['conclusion'] in {'failure', 'timed_out'}:
                status = 'fail'
                break
            if run['conclusion'] == 'success':
                break
        if run['status'] in {'in_progress', 'queued'}:
            status = 'rerun'
            break
    return status


if __name__ == "__main__":
    while True:
        status = get_status(get_runs())
        if status != 'rerun':
            break
        sleep(60)
    if status == 'fail':
        exit(1)

# .github/workflows/test_1.yml
name: Test 1

on:
  pull_request_review_comment:
    types: [created]

jobs:
  test:
    name: Test 1
    runs-on: ubuntu-latest
    if: github.event.comment.body == '/gha run' && contains('OWNER,MEMBER,COLLABORATOR', github.event.comment.author_association)
    timeout-minutes: 60
    strategy:
      fail-fast: false
    steps:
      - name: Checkout repository
        uses: actions/checkout@v1
        with:
          fetch-depth: 5
          submodules: true
      - name: Test
        run: |
            sleep 2m
            exit -1

# .github/workflows/r_package.yml

...

  all-successful:
    # https://git.luolix.topmunity/t/is-it-possible-to-require-all-github-actions-tasks-to-pass-without-enumerating-them/117957/4?u=graingert
    runs-on: ubuntu-latest
    needs: [test]
    steps:
      - name: Checkout repository
        uses: actions/checkout@v1
        with:
          fetch-depth: 5
          submodules: false
      - name: Install Python
        uses: actions/setup-python@v2
        with:
          python-version: '3.x'
      - name: Note that all tests succeeded
        run: python "$GITHUB_WORKSPACE/.ci/check_workflow_status.py"

oooo interesting! Are you thinking about something like this for the checks in #3424 and #3443 ?

[StrikerRUS]

@jameslamb

Are you thinking about something like this for the checks in #3424 and #3443 ?

Yeah, definitely!

[StrikerRUS]

@guolinke Could you please generate a secret access token for LightGBM repository with public_repo and workflows permissions? I'm working on allowing optional checks to be in either skipped or succeeded status (#3439 (comment)). And I need that token to re-trigger our main GitHub Actions workflow from optional one to workaround additional manual step for maintainer.
Steps 1 and 2 from this guide: https://stevenmortimer.com/running-github-actions-sequentially/#step-1---create-a-personal-access-token-pat.

[guolinke]

@StrikerRUS you can check WORKFLOW in https://github.com/microsoft/LightGBM/settings/secrets

[StrikerRUS]

@guolinke Thanks a lot! Unfortunately, I don't have an access to this tab of LightGBM repo settings:

But I think it shouldn't block me from using WORKFLOW name as a secret value in appropriate place in the workflow 🙂 .

[guolinke]

yeah, you can use it as ${{ secrets.WORKFLOW }}

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.