We need a repeatable process that can take current database state and the current system state, use those to create a plan of action, then apply that plan to the current cluster.
The snapshot apply process in the control-server binary is that system.
- Take a point in time snapshot of everything. (Prepare)
- Feed that snapshot into functional code that generate desired system states. (Generate)
- Apply any changes need to go from the current state to the desired state on all systems. (Apply)
- Record the status for all desired state pieces (Report)
- Broadcast the result of the overall attempt (Broadcast)
Prepare
- We need a summary of everything in the database and everything in the current system state. For that we use the system state summarizer.
- Then we need to create the target snapshots for the different systems. (KubeSnapshot and KeyCloakSnapshot)
Generation
- For each target snapshot use the summarized system state with functional modules to generate the target system specifications.
- Store each kube or keycloak target resource in the database (ResourcePath for kube)
Apply
- Remove any target system configuration or resource that already match with what’s there. For kubernetes this is done via sha hmac
KubeExt.Hashing - Update any matching resources are successfully applied
- Push each of the kube resource targets to kubernetes via
KubeExt.ApplyResource - Push each of the keycloak resource targets to keycloak.
- Trigger any post apply operations needed for keycloak
Report
- Record the per resource target results
- Compute an overall result
Broadcast
- Send the latest result via Phoenix pub sub and
EventCenter