BPF CI scripts

The Jenkinsfile is crafted to work with a kernel tree. A rough overview of all the steps:

1. Clone or fetch this repository from GitHub
2. Copy the Vagrantfile and VM related steps
3. Run stages
   • Install latest LLVM snapshot in newly booted kernel
   • Run Cilium tests (skipped)
   • Install bpftool
   • Run BPF selftest with LLVM versions 3.8.1 - 5.0.0 and snapshot
4. Get artifacts and cleanup
5. Send notifications via email

This currently only works with x86_64 but support for ARM64 is being worked on.

New Jenkins instance

Setting up a Jenkins instance to use this setup is pretty straightforward. The suggested plugins from the Jenkins community will cover most of the plugins but to avoid issues check that the ones listed in Jenkins plugins are installed. The regular tools like Git CLI, Virtualbox, and Vagrant are also required. Any of the missing plugins can cause weird error messages on Jenkins, so check that first in case of errors.

When Jenkins is up and running with all the dependencies. Copy the contents of Jenkinsfile and add it to /configfiles with the exact same name. For screenshots on how do this see vaimr/pipeline-multibranch-defaults-plugin

Valid SMTP credentials must be setup for email notifications to work. Otherwise it could cause a build failure.

Adding a new tree

The following steps assume you have a Jenkins instance up with all of the dependencies. Also please keep in mind that fetching a kernel tree might take 10+ min., so let it timeout and retry if necessary.

To add a tree first create a new Multibranch Pipeline with defaults job by visiting https://jenkins.cilium.io/view/all/newJob

From the General tab under Branch Sources use Add source to select Git, there you can enter the project repository (git URL) and if necessary add a filter. Make sure the Build Configuration section has the mode set to by default Jenkinsfile. It's also recommended to pick a Orphaned Item Strategy, f. ex. only keep 1-3 builds.

If the new tree is very similar to a existing job, the copy job feature could be used. Then you would just have to modify the relevant fields instead.

Why use a filter? In the linux-stable tree case a lot of branches are outdated and not relevant so a expression like this ^(master|linux-3\.18|linux-4\.4|linux-4\.9|linux-4\.14).* is enough to ignore the old ones.

Maintenance

This setup currently relies on Vagrant, the cilium/packer-ubuntu-16.10 image and LLVM repositories. Things are working most of the time on x86_64 but there are some known flakes.

• One of the stages installs the latest LLVM snapshot and at times there are network issues with apt. The cause for this has not been determined.
• The VM base image has not caused any issues but it's worth noting that any changes there is a separate process and requires getting changes merged in that repository and there might be a delay for VM image changes to take affect.
• If the disk space gets full the tests could start failing or issues with SCM triggered builds not being started.

Another thing that might be confusing is the directory layout. Since the kernel does not have a Jenkinsfile one is added globally under the /configfiles. Whenever this file is modified the changes have to be approved manually in /scriptApproval. The Jenkinsfile is not edited often so it's not a big issue. Since part of the build stage is fetching recent changes from this repository, the contents from scripts/ is copied over. To avoid conflicts with the kernel the scripts directory a prefix is used workspace/scripts. The paths in the Jenkinsfile using the scripts from the VM need to have the right prefix.

Jenkins plugins

• Pipeline: Multibranch with defaults
• Email-ext plugin
• Pipeline

Vagrant plugins

• vagrant-reload
• vagrant-scp

Credits

These files are based on other ones, mostly copied from the below ones

• https://github.com/cilium/cilium/blob/master/Jenkinsfile
• https://github.com/tgraf/net-next-vagrant