First try without systemd
During my previous mission, we used AWS ECR to manage the docker images.
Amazon EC2 Container Registry (ECR) is a fully-managed Docker container registry that makes it easy for developers to store, manage, and deploy Docker container images.
To be able to store/retrieve docker images, you have to be authenticated. AWS CLI provides a command which takes your credentials and generate a token, authenticating the user for 12 hours.
Our continuous integration servers (ie. Jenkins nodes) needed to access to the ECR so we had to provide an automated mechanism.
The first idea was to use 2 crons:
- The first on @reboot, to ensure the machine will be authenticated upon reboot.
- The second every 11 hours.
This was not a good idea, sometimes it did not work and it was difficult to know why. (In fact, the @reboot only works when using root, which we try to avoid)
So we used systemd and:
- it just worked
- we can monitor the executions
Here is the unit located here: /etc/systemd/system/docker_ecr_login.service
Dont forget to
systemctl daemon-reload and
systemctl enable docker_ecr_login
Now we only use one cron, every 11 hours it launches
systemctl start docker_ecr_login
Now it’s easier to monitor past executions with:
The error status 217/USER means that the user ‘builder’ doesn’t exist on this system.
We are looking into systemd timers with something like
OnUnitActiveSec= we should be able to completely get rid of cron and go full systemd.