Parallel code scanning with CodeQL

documentation Improvements or additions to documentation

If you have a large repository containing various independent projects (a "monorepo"), the time taken to scan your code with CodeQL can be significantly reduced by splitting the scanning work into various parallel jobs which will individually analyze only a subset of the files in the repository.

This repository contains an example of a GitHub Actions workflow which does precisely that. The strategy implemented here works however only for the interpreted languages supported by CodeQL (e.g. Python and JavaScript). As a first step in this workflow, all non-hidden subdirectories under the repository's root directory are detected; parallel code scanning jobs are then dynamically generated for each subdirectory containing a software project. The subdirectories project-1, project-2 and project-3 here represent three independent software projects inside the repository, each one of which will be scanned in a dedicated job (i.e., three jobs will be generated in total). Adding a new software project to this repository (e.g. project-4) requires no changes to the workflow file as a dedicated code scanning job will be automatically generated for it when the workflow is executed.

If the workflow is triggered by a pull request the list of sub-directories that will be scanned will be limited to the subdirectories that contain changes. The changes are based on a git diff between the base and head repositories specified in the pull request.

This strategy is possible because GitHub Actions workflows accept JSON input to define a job matrix, and the JSON contents can be generated during the workflow's execution. In other words, the job matrix can be defined dynamically.

NOTE: The approach presented here must be taken with care as accidentally splitting a software component in this manner may reduce CodeQL's ability to recognize certain types of vulnerabilities in that component. For instance, it may not be able to entirely map how data flows inside the component and therefore miss possible attacks against it. Please make sure you understand the general capabilities of CodeQL before doing this.

Answers to common questions

1. Every code scanning job checks out the repository in parallel. If a change is made to the repository during that time (e.g. a subdirectory is added or removed, or a file in a pre-existing subdirectory is modified), you essentially have a race condition which is not being properly handled.

This situation will not occur because the actions/checkout action only fetches a single commit by default, for the ref/SHA which triggered the workflow. This implies that the same code snapshot will be checked out in all jobs triggered inside the Code scanning workflow.

If the amount of data fetched in each checkout step is large, you may get a performance improvement by generating an artifact containing the code in the very first job which is executed in the workflow and then consuming that artifact in all downstream jobs. The actions/upload-artifact and actions/download-artifact actions will help you accomplish this.