Please dive in the second part of a blog series based on a project delivered for one of our clients. If you miss the first part, please check it here.
Table of Contents
- Problem description
- General description of the solution
- Problem 1: Limited job output size in GitLab
- Problem 2: Limited duration of jobs running on shared runners
PART II
- Problem 7: Passing on artifacts between CI/CD jobs
- Problem 8: Starting docker build manually
- Problem 9: We cannot rely on the error code returned by Puppet
- Summary
Problem 3: Building a container image in the job
In several of the CI/CD jobs in our project, we create container images and upload them to GitLab's Registry. Building container images in GitLab's CI/CD requires infrastructure preparation, or we have to use a tool other than Docker. If we want to build images using Docker, then we must give the Docker client access to the Docker daemon socket, which is not recommended for security reasons. If we want to use shared GitLab.com runners, access to the daemon socket is not possible.
We can solve the problem in several ways:
- Using an executor in a runner that does not work in a container and which has access to a Docker socket (e.g. shell executor). In this case, we can't use GKE or shared GitLab.com runners. We need to install, configure and maintain such a runner ourselves.
- Using Kubernetes or Docker executor, but with Docker socket mounted inside containers. The solution requires an independent runner, prevents the use of shared runners and is bad from a security perspective.
- Using the
buildah
tool to build the container image inside the container and the skopeo
program to upload the image to the GitLab Registry. Both programs are part of the https://github.com/containers project and do not require a privileged daemon norroot
access (unlike Docker).
- Use
kaniko
. Kaniko is a tool dedicated to building container images in Kubernetes and in containers. Like buildah
, it doesn't require special permissions. An example of its use is described in the official GitLab documentation.
Our team has already had experience with Kaniko in other projects, so we made use of it in this project.
The definition of a job using kaniko in the .gitlab-ci.yml
looks like this:
build-base-image:
stage: prepare
image:
name: gcr.io/kaniko-project/executor:debug
entrypoint: [""]
script:
- echo
"{\"auths\":{\"$CI_REGISTRY\":{\"username\":\"$CI_REGISTRY_USER\",\"password\":\"$CI_REGISTRY_PASSWORD\"}}}" > /kaniko/.docker/config.json
- /kaniko/executor --context $CI_PROJECT_DIR --dockerfile
$CI_PROJECT_DIR/Dockerfile --destination
$CI_REGISTRY_IMAGE:$CI_COMMIT_TAG
As you can guess, after studying the above code fragment, Kaniko builds a container image and immediately uploads it to the indicated image registry.
Problem 4: The GitLab Registry token expires too quickly
GitLab has an integrated container image registry (Registry). A one-time password is created for each job, which allows you to use Registry without having to manually create a dedicated account. GitLab passes the login and temporary password to the Runner, which then sets them in the environment variables of the job process.
This is the end result of a successful job output that creates a container image using Kaniko and uploads it to the Registry built into GitLab:
In our project, some jobs take a very long time, longer than the lifetime of a one-time Registry password. Limiting the duration of such a one-time password is necessary from the security point of view.
Uploading the image to the Registry fails and ends with the following error:
We solved this problem in a traditional way:
- We created a dedicated technical account in GitLab.
- We gave the new account project privileges.
- We logged in to GitLab on this account and created the personal access token. This token doesn't have an expiration date.
- In the project, in the CI/CD settings, we created two new variables containing the login and password (token) of the technical account respectively
- We modified the job so that Kaniko used a dedicated technical account to perform operations in the Registry.
After these modifications, we no longer had problems uploading container images to the Registry, even if the CI/CD job took 3 hours.
Problem 5: In the paid GitLab.com plan we have a limit on the shared runners used time
If you use paid GitLab-as-a-service plans, you can use a certain number of minutes for the CI pipeline, i.e. shared Runners. Each of the paid plans has a different limit. Using shared Runners is very convenient because we don't have to worry about maintaining our own Runners and thus save time and money. The limit may be sufficient in some projects, but in our project we quickly reached the monthly limit.
You will see this type of message when you use the available time of shared Runners in a given month:
We solved this problem by setting up a dedicated Kubernetes cluster for Runners. We allow GitLab to decide which Runner to use (shared or our own). Thanks to this, the load is distributed to both types of Runners, we reduce expenses and shorten the time of pipeline execution.
Instructions for using Kubernetes cluster as the platform for Runners are described in the Problem 2 section.
Problem 6: User's names with national characters in GitLab
If the runner or container we use in our CI/CD has locales that do not support UTF-8, and the user making commits to the repository that has characters that are not ASCII in the name or surname, then the CI/CD job may end with the following error:
FAILURE: Build failed with an exception.
* What went wrong:
Could not set the value of environment variable 'GITLAB_USER_NAME':
could not convert string to current locale
There is even an issue with this in this GitLab project.
A workaround is also provided there. You can change the value of the GITLAB_USER_NAME
variable so that it doesn't contain non-ASCII characters. It can be assigned, for example, to the value of the variable containing the user's login (assuming that the login consists only of ASCII characters).
To the before_script
section in .gitlab-ci.yml
, add:
# Workaround for "Could not set the value of environment variable
'GITLAB_USER_NAME': could not convert string to current locale" problem.
# https:
- export GITLAB_USER_NAME=$(echo $GITLAB_USER_LOGIN)
Next part of the post
Follow our profile on Linkedin and stay up to date for the next part!