Daniel Doubrovkine bio photo

Daniel Doubrovkine

aka dB., @awscloud, former CTO @artsy, +@vestris, NYC

Email Twitter LinkedIn Github Strava
Creative Commons License

The OpenSearch API specification is authored in OpenAPI and used to auto-generate OpenSearch language clients. I wanted to know how much of the API was described in it vs. the actual API implemented in the default distribution of OpenSearch that includes all plugins. To do so, I have exposed an iterator over REST handlers in OpenSearch core, and wrote a plugin that rendered a very minimal OpenAPI spec at runtime. All that was left was to compare the manually authored OpenAPI spec in opensearch-api-specification to the runtime one, added in opensearch-api-specification#179. The comparison workflow output a total and relative number of APIs described.

To surface this information in pull requests I wanted to add a comment in the API coverage workflow. This required a PAT token, so I initially authored that workflow with pull_request_target in opensearch-api-specification#196.

- name: Gather Coverage
  id: coverage
  shell: bash
  run: |
    ...
    cat >>"$GITHUB_OUTPUT" <<EOL
    current=$current
    total=$total
    percent=$percent
    EOL
- uses: peter-evans/create-or-update-comment@v4
  if: github.event_name == 'pull_request_target'
  with:
    issue-number: $
    body: |
        API specs implemented for $/$ ($%) APIs.

This is actually insecure because we run a JavaScript tool to generate the API spec with source code coming from the pull request, and a user can execute arbitrary code this way and gain access to the secure token. One solution is to run the tool from main, but we can do better.

In opensearch-api-specification#251 I split the coverage workflow in one that gathered information into a JSON file and uploaded it as an artifact of the pull request workflow, and another that downloaded the artifact and commented on the pull request.

name: Gather API Coverage
on: [push, pull_request]

- name: Gather Coverage
  id: coverage
  shell: bash
  run: |
    ...
    cat >>"coverage.json" <<EOL
    {
      "pull_request":$,
      "current":$current,
      "total":$total,
      "percent":$percent
    }
    EOL
- uses: actions/upload-artifact@v4
  with:
    name: coverage
    path: coverage.json
name: Comment with API Coverage

on:
  workflow_run:
    workflows: ["Gather API Coverage"]
    types:
      - completed

jobs:
  comment:
    runs-on: ubuntu-latest
    if: >
      github.event.workflow_run.event == 'pull_request' &&
      github.event.workflow_run.conclusion == 'success'
    steps:
      - name: Download Coverage Report
        uses: actions/download-artifact@v4
        with:
          github-token: $
          name: coverage
          run-id: $

      - name: 'Comment on PR'
        uses: actions/github-script@v3
        with:
          github-token: $
          script: |
            const fs = require('fs');
            var data = JSON.parse(fs.readFileSync('./coverage.json'));
            console.log(data);
            await github.issues.createComment({
              owner: context.repo.owner,
              repo: context.repo.repo,
              issue_number: data.pull_request,
              body: `API specs implemented for ${data.current}/${data.total} (${data.percent}%) APIs.`
            });

This is a very convenient pattern of passing structured data from a pull request workflow to one that has access to write data in a GitHub repository. Check out the latest versions of the coverage workflows if you are going to use them.