Bazel Integration with Mergify

This guide shows how to upload the JUnit reports of the tests Bazel runs to Test Engine from your CI workflow.

bazel test writes a JUnit report for every test target, without any configuration:

bazel test //...

Each target’s report is bazel-testlogs/<package>/<target>/test.xml: the target //cart:cart_test writes bazel-testlogs/cart/cart_test/test.xml. bazel-testlogs is a link into Bazel’s output directory, and the upload follows it. Upload every report with the pattern bazel-testlogs/**/test.xml.

What a report holds depends on the test:

• A test whose framework writes JUnit XML to the path Bazel gives it in XML_OUTPUT_FILE reports each of its test cases. GoogleTest does this, in a cc_test that depends on @googletest//:gtest_main.

• Any other test gets a report that Bazel writes for it: one test case, named after the target, which passes or fails with the target as a whole.

Run the tests in a step with an id, and let the upload step decide the job’s result:

- name: Run Bazel Tests
  id: tests
  continue-on-error: true
  run: bazel test //...

- name: Mergify CI Upload
  if: success() || failure()
  uses: mergifyio/gha-mergify-ci@v27
  with:
    token: ${{ secrets.MERGIFY_TOKEN }}
    report_path: bazel-testlogs/**/test.xml
    test_step_outcome: ${{ steps.tests.outcome }}

Key Points:

• if: success() || failure(): Runs the upload step even if tests fail, ensuring Test Engine has the full report.
• report_path: bazel-testlogs/**/test.xml: Points to where your JUnit file is located. Make sure it matches the path you set in your CI job.
• test_step_outcome: ${{ steps.tests.outcome }}: Passes the test runner step's outcome so Mergify can detect silent failures where the runner crashed but the JUnit report appears clean. Add an id (such as tests) to your test runner step and update the steps.<id>.outcome reference to match.
• On pull requests from forks and from Dependabot, MERGIFY_TOKEN is empty and this step fails: skip it as shown in Pull Requests Without the Secret.

If you use a job matrix in your workflow (e.g., to test across multiple versions), every leg reports the same job name, because GitHub Actions gives them all one GITHUB_JOB. Set the job_name input so Test Engine can tell the reports apart:

jobs:
  example_matrix:
    strategy:
      matrix:
        version: [10, 12, 14]

Your upload step should look like:

- name: Mergify CI Upload
  if: success() || failure()
  uses: mergifyio/gha-mergify-ci@v27
  with:
    job_name: example_matrix (${{ matrix.version }})
    token: ${{ secrets.MERGIFY_TOKEN }}
    report_path: bazel-testlogs/**/test.xml
    test_step_outcome: ${{ steps.tests.outcome }}

To benefit from Test Engine quarantine, you need to add continue-on-error: true in your GitHub Actions step that executes your tests and generates the JUnit file. The step running the gha-mergify-ci action then decides whether the job passes or fails, taking quarantined tests into account.

You should also pass test_step_outcome: ${{ steps.tests.outcome }} to the step that runs mergifyio/gha-mergify-ci (where tests is the id of your test runner step) to detect silent failures where the test runner crashed but the JUnit report appears clean. Without this input, a crash that produces a partial or empty report could be mistakenly treated as a success.

With continue-on-error: true, the job’s result comes from the upload step. On pull requests from forks and from Dependabot, MERGIFY_TOKEN is empty and the upload step has to be skipped: add the step that fails the job when the tests failed and nothing was uploaded, as shown in pull requests without the secret.

steps:
  - label: "Run tests"
    command: <your test command>
    plugins:
      - mergifyio/mergify-ci#v8:
          action: junit-process
          report_path: bazel-testlogs/**/test.xml

Key Points:

• The plugin runs in the post-command hook, so it uploads results after your tests finish, even if they fail.
• report_path: bazel-testlogs/**/test.xml: Points to where your JUnit file is located. Make sure it matches the path you set in your test configuration.
• Silent failure detection is automatic: the plugin reads the step's exit code to detect cases where the test runner crashed but the JUnit report appears clean.

If you use a build matrix in your pipeline, each variation reports its own step label as its job name, so the variations are already distinct. Set the job_name plugin property when the label is not the name you want Test Engine to file the reports under:

steps:
  - label: "Flaky detection ({{matrix}})"
    matrix:
      - "3.10"
      - "3.11"
      - "3.12"
    command: <your test command>
    plugins:
      - mergifyio/mergify-ci#v8:
          action: junit-process
          job_name: "Tests ({{matrix}})"
          report_path: bazel-testlogs/**/test.xml

When using the Buildkite plugin for Test Engine quarantine, the plugin automatically detects test failures using the step’s exit code. No extra configuration is needed.

The mergifyio/mergify-ci plugin runs in the post-command hook and reads BUILDKITE_COMMAND_EXIT_STATUS to detect silent failures where the test runner crashed but the JUnit report appears clean.

Quote the pattern, so that mergify ci junit-process expands it rather than the shell: without globstar, bash reads ** as * and misses the reports at other depths.

Install the Mergify CLI in your pipeline and export MERGIFY_TOKEN. Run mergify ci junit-process after your tests:

set -o pipefail
exit_code=0
bazel test //... || exit_code=$?
mergify ci junit-process \
  --test-exit-code "$exit_code" \
  "bazel-testlogs/**/test.xml"

Key Points:

• MERGIFY_TOKEN: Export this environment variable with your Mergify application key so junit-process can authenticate to the API.
• || exit_code=$?: CI shells run with -e, so a failing test command on its own line would end the script before the upload. This form keeps going and records the code. The script needs bash for set -o pipefail: on Jenkins, whose sh steps often run dash, start it with #!/bin/bash.
• --test-exit-code "$exit_code": Passes the test runner's exit code so Mergify can detect silent failures where the runner crashed but the JUnit report appears clean.
• junit-process exits last, so its status becomes the step's. Quarantine depends on that: it exits 0 when every failure it found is quarantined, and 1 when one is not. Re-exiting with the test runner's own code would fail the step whatever quarantine decided.
• The branch quarantine is evaluated against is detected from your CI provider, falling back to the checked-out branch. Pass --tests-target-branch <branch> only to override it.

A test that reports its own test cases is named the way its framework names them: the GoogleTest test TEST(CartTest, AddsAnItem) is reported as CartTest.AddsAnItem, with its source file. These names do not carry the target, so give tests in two targets distinct names to keep their histories apart.

A test that Bazel reports for is named after its target’s package and name: the sh_test //cart:smoke_test is reported as cart/smoke_test. Moving or renaming the target starts a new history for it.

After pushing these changes:

1. Your CI workflow runs bazel test.
2. Bazel writes a report per test target under bazel-testlogs/.
3. Mergify uploads the reports to Test Engine.

You can then review your test results, including any failures or flaky tests, directly in the Test Engine dashboard.

• One test case per target: the target’s test framework does not write to XML_OUTPUT_FILE, so Bazel reports the target as a single test.

• The CLI provides information about the upload. Check the logs in GitHub Actions.
• File paths: Double-check that the output file matches the path used in report_path.
• Permissions: Make sure the MERGIFY_TOKEN is valid and setup in your GitHub Actions secrets as explained in the docs.
• Workflow conditions: If your step is not running, confirm the if condition is actually triggered in your job.