Simon Beck
d645b44b67
|
||
|---|---|---|
| .vscode | ||
| config | ||
| controllers | ||
| hack | ||
| k8s | ||
| kopia | ||
| logger | ||
| .dockerignore | ||
| .gitignore | ||
| Dockerfile | ||
| Makefile | ||
| PROJECT | ||
| README.md | ||
| appLogger.go | ||
| command_kopia.go | ||
| command_kopia_backup.go | ||
| command_kopia_maintenance.go | ||
| command_operator.go | ||
| command_operator_backup.go | ||
| go.mod | ||
| go.sum | ||
| main.go | ||
| operator.go | ||
README.md
Kopia-k8s
This is a POC how a new, very simple backup operator for K8s could look like. Its main purposes is to make backups for my homelab and to serve as a POC for a K8up v3.
How does it work?
Kopia-k8s has two operation modes:
- Operator: This mode is intended to run within or against a k8s cluster. It doesn't trigger Kopia directly, but rather spawn jobs on K8s that run Kopia. It also runs the defined pre-backup scripts beforehand.
- Kopia: This mode runs actual Kopia commands. They are intended to run within a container.
When running kopia-k8s operator backup on either a workstation with a kubeconfig, or within a cluster as a job, it will first list all the running pods on the cluster. Then it filters out those that have a pre-backup annotation.
The command found in the pre-backup command annotation will be run sequentially and block until finished. They are executed by using exec on the pods. If one of the pre-backup commands fail with an exit code != 0 then the whole backup will be aborted and also exit with a non-zero code.
Once the pre-backup commands have finished, it will start to spawn the actual backup jobs within k8s. By default, it spawns 3 parallel jobs. Each of those jobs has pod-affinity rules, so that it's scheduled on the same host as the running pod. This ensures that the backup jobs can read the data from the same RWO PVCs. If Kopia encounters a critical error, it will exit with a non-zero exit code, thus failing the entire job. So Kopia-k8s jobs can be monitored by simply monitoring for failed jobs on the cluster.
To-dos
Some to-dos:
- Currently two parallel
kopia-k8s operator backupinstances could clash, because the spawned jobs don't have randomized names - Maybe add some metrics, for example if there were some non-critical errors (e.g. a file could not be read)