The demo Video can be found on Youtube here
In the first half of the video, I talked about the design considerations and some best practices and worked through the reconcile logic. The later half is the actual demo. (sorry if it is a bit lengthy)
A clusterscan operator that can execute configurable jobs and cronjobs.
You can use this link to see the changes added after the kubebuilder init:
- Adding a figuration management file or system logic to manage the variables and job-specific config.
- Implement the logic for filtering for
Targetand checking for concurrent operation whenAllowConcurrencyspec is set to False. - Adding a lastOp field to the
Statusto track of the last operation that the underlining job performed. - Set ownership for the jobs created by the cronjob to the same operator. With testing, it is found that the cron children's job will not inherit this ownership and the cluster can miss reconciliation for them. (The workaround is to have a periodic reconciliation but this sort of hinders the event-based approach we had here)
- Introduce more scheduling options beyond standard cron expressions(like dynamic scheduling based on cluster events or metrics thresholds that I am thinking of)
- Ensure that all resources created by the operator are properly cleaned up when the operator itself is deleted.
- Add unit tests to ensure code and branch coverages.
Some doc I found really helpful in this journey learning k8s, minikube, kubebuilder, and k8s operators:
https://book.kubebuilder.io/getting-started
https://minikube.sigs.k8s.io/docs/start/
https://kubernetes.io/docs/concepts/workloads/controllers/cron-jobs/
https://github.com/robfig/cron
https://github.com/kubernetes/api
https://kubernetes.io/docs/reference/using-api/api-concepts/