GitLab → Jenkins multibranch poller

This repository provides a Docker image and example .gitlab-ci.yml job which aims to allow teams which have their testing in Jenkins but workflow in GitLab to get the two working together loosely. It will let these teams see a build pass/fail from GitLab merge requests and commits.

This expects that you have a Git multibranch pipeline running in Jenkins and want to have a GitLab CI pipeline that polls for the creation and completion of a build for a commit.

Use

See the .gitlab-ci.yml.example for what you need to add to your .gitlab-ci.yml. There are also secrets and environment variables you can set:

• JENKINS_USER (secret) - read-only user name
• JENKINS_PASSWORD (secret) - read-only user's password/API key
• JENKINS_HOST (optional) - defaults to https://jenkins.mintel.ad
• MULTIBRANCH_JOB - path that Jenkins job for the target repository, e.g. EVERESTUI_jobs/web
• PYTHONHTTPSVERIFY (optional) - defaults to 0 so no SSL verification occurs so self-signed certificates can be used
• JENKINS_NODEBUG (optional) - Don't output debug info about the job. Defaults to off to preserve existing behaviour.
• JENKINS_CONSOLE_OUTPUT (optional) - Display the Jenkins console output. Defaults to off to preserve existing behaviour.

The job then uses the provided variables (including ones GitLab injects) to find which particular job is for this branch (e.g. EVERESTUI_jobs/web/CFD-4563) and it then polls the job looking for the latest build for the current commit. Once that is found, it polls the build until it has a result where it will exit with 0 if successful, or 1 otherwise.

This has no timeouts, so relies on the GitLab project's timeout settings which will terminate the pipeline if it takes too long.

If the password is an API key for a read-only account, then the JENKINS_USER and JENKINS_PASSWORD may not really be secrets - if they aren't you can enter them as variables. This will let the poll job work for non-protected branches.

Possible future work

The image is hundreds of megabytes, it could be stripped down to a few megabytes by using go and distroless to reduce download+cache cost.