rust-vmm-ci

The rust-vmm-ci repository contains integration tests and Buildkite pipeline definitions that are used for running the CI for all rust-vmm crates.

Having a centralized place for the tests is one of the enablers for keeping the same quality standard for all crates in rust-vmm.

Getting Started with rust-vmm-ci

To run the integration tests defined in the pipeline as part of the CI:

1. Add rust-vmm-ci as a git submodule to your repository

# Add rust-vmm-ci as a submodule. This will point to the latest rust-vmm-ci
# commit from the master branch. The following command will also add a
# `.gitmodules` file and the `rust-vmm-ci` to the index.
git submodule add https://github.com/rust-vmm/rust-vmm-ci.git
# Commit the changes to your repository so that the CI can run using the
# rust-vmm-ci pipeline and tests.
git commit -s -m "Added rust-vmm-ci as submodule"

2. Create the coverage test configuration file named coverage_config_ARCH.json in the root of the repository, where ARCH is the architecture of the machine. There are two coverage test configuration files, one per each platform. The example of the configuration file for x86_64 architecture can be found in coverage_config_x86_64.json.sample, and the example of the configuration file for aarch64 architecture can be found in coverage_config_aarch64.json.sample.

The json must have the following fields:

• coverage_score: The coverage of the repository.
• exclude_path: This field is used for excluding files from the report. It should be used to exclude autogenerated files. Files in exclude_path are separated by one coma. If the repository does not have any autogenerated files, exclude_path should be an empty string.
• crate_features: cargo kcov does not build crate features by default. To get the coverage report including optional features these need to be specified in crate_features separated by coma. If the crate does not have any features, this field should be empty.

This file is required for the coverage integration so it needs to be added to the repository as well.

3. Create a new pipeline definition in Buildkite. For this step ask one of the rust-vmm Buildkite admins to create one for you. Add a pipeline step that is uploading the rust-vmm-ci pipeline:

buildkite-agent pipeline upload rust-vmm-ci/.buildkite/pipeline.yml

4. The code owner of the repository will have to setup a WebHook for triggering the CI on Pull Requests and pushes.

Buildkite Pipeline

The Buildkite pipeline is the definition of tests to be run as part of the CI. It includes steps for running unit tests and linters (including coding style checks), and computing the coverage.

Currently the tests can run on Linux x86_64 and aarch64 hosts.

Example of step that checks the build:

steps:
  - label: "build-gnu-x86"
    commands:
     - cargo build --release
    retry:
      automatic: false
    agents:
      platform: x86_64.metal
    plugins:
      - docker#v3.0.1:
          image: "rustvmm/dev:v5"
          always-pull: true

To see all steps in the pipeline check the .buikite/pipeline.yml file.

Custom Pipeline

Some crates might need to test functionality that is specific to that particular component and thus cannot be added to the common pipeline.

In this situation, the repositories need to create a custom pipeline (besides the rust-vmm-ci pipeline) and add it in the repository. The preferred path for the custom pipeline is .buildkite/pipeline.yml.

For example to test the build with one non-default feature enabled, the following step can be added in the custom pipeline under .buildkite/pipeline.yml.

steps:
  - label: "build-gnu-x86-bzimage"
    commands:
     - cargo build --release --features bzimage
    retry:
      automatic: false
    agents:
      platform: x86_64.metal
    plugins:
      - docker#v3.0.1:
          image: "rustvmm/dev:v5"
always-pull: true

Custom Repository Hooks

The integration tests of some repositories have dependencies on external resources. One example is linux-loader which needs to download a bzImage before running the unit tests. Because this is specific to the linux-loader crate, the logic for downloading the required resources cannot be part of the common pipeline. The mechanism used here is Repository Hooks. The hooks are defined per repository and live in the crate repository under .buildkite/hooks.

Example of post-checkout hook that downloads and extracts a bzImage:

#!/bin/bash

DEB_NAME="linux-image-4.9.0-9-amd64_4.9.168-1_amd64.deb"
DEB_URL="http://ftp.debian.org/debian/pool/main/l/linux/${DEB_NAME}"

REPO_PATH="${BUILDKITE_BUILD_CHECKOUT_PATH}"
DEB_PATH="${REPO_PATH}/${DEB_NAME}"
EXTRACT_PATH="${REPO_PATH}/src/bzimage-archive"
BZIMAGE_PATH="${EXTRACT_PATH}/boot/vmlinuz-4.9.0-9-amd64"

mkdir -p ${EXTRACT_PATH}

wget ${DEB_URL} -P ${REPO_PATH}
dpkg-deb -x ${DEB_PATH} ${EXTRACT_PATH}

mv ${BZIMAGE_PATH} ${REPO_PATH}/src/bzimage
rm -r ${EXTRACT_PATH}
rm -f ${DEB_PATH}

In this example the post-checkout hook downloads a deb image, extracts the contents of it and places it in linux-loader/src/bzimage. The unit tests will use the relative path src/bzimage which does not depend on the image being downloaded.

Integration Tests

In addition to the one-liner tests defined in the Builkite Pipeline, the rust-vmm-ci also has more complex tests defined in integration_tests.

Test Profiles

The integration tests support two test profiles:

• devel: this is the recommended profile for running the integration tests on a local development machine.
• ci (default option): this is the profile used when running the integration tests as part of the the Continuous Integration (CI).

The test profiles are applicable to tests that run using pytest. Currently only the coverage test follows this model as all the other integration tests are run using the Buildkite pipeline.

The difference between is declaring tests as passed or failed:

• with the devel profile the coverage test passes if the current coverage is equal or higher than the upstream coverage value. In case the current coverage is higher, the coverage file is updated to the new coverage value.
• with the ci profile the coverage test passes only if the current coverage is equal to the upstream coverage value.

Further details about the coverage test can be found in the Adaptive Coverage section.

Adaptive Coverage

The line coverage is saved in tests/coverage. To update the coverage before submitting a PR, run the coverage test:

docker run --device=/dev/kvm \
           -it \
           --security-opt seccomp=unconfined \
           --volume $(pwd)/kvm-ioctls:/kvm-ioctls \
           rustvmm/dev:v5
cd kvm-ioctls/
pytest --profile=devel rust-vmm-ci/integration_tests/test_coverage.py

If the PR coverage is higher than the upstream coverage, the coverage file needs to be manually added to the commit before submitting the PR:

git add tests/coverage

Failing to do so will generate a fail on the CI pipeline when publishing the PR.

NOTE: The coverage file is only updated in the devel test profile. In the ci profile the coverage test will fail if the current coverage is higher than the coverage reported in tests/coverage