Make unlocking more reliable

[jstudler]

In some cases, the Prefix and IpAddress reconcilers are stuck because a resource update prior to unlocking failed (e.g. due to concurrent edit in the kube-api). This could be e.g. the following line https://github.com/netbox-community/netbox-operator/blob/main/internal/controller/prefix_controller.go#L182

The next run of the reconciler will have nil for ll in https://github.com/netbox-community/netbox-operator/blob/main/internal/controller/prefix_controller.go#L203 and thus not try to unlock it. Example log output of the netbox-operator:

2024-09-12T11:48:16+02:00       ERROR   Reconciler error        {"controller": "prefix", "controllerGroup": "netbox.dev", "controllerKind": "Prefix", "Prefix": {"name":"example","namespace":"default"}, "namespace": "default", "name": "example", "reconcileID": "1c20d680-fc77-48fb-a949-730a20ebd83a", "error": "Operation cannot be fulfilled on prefixes.netbox.dev \"example\": the object has been modified; please apply your changes to the latest version and try again"}
sigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller[...]).reconcileHandler
        /Users/joel/go/pkg/mod/sigs.k8s.io/[email protected]/pkg/internal/controller/controller.go:316
sigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller[...]).processNextWorkItem
        /Users/joel/go/pkg/mod/sigs.k8s.io/[email protected]/pkg/internal/controller/controller.go:263
sigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller[...]).Start.func2.2
        /Users/joel/go/pkg/mod/sigs.k8s.io/[email protected]/pkg/internal/controller/controller.go:224
2024-09-12T11:48:16+02:00       INFO    prefix reconcile loop started   {"controller": "prefix", "controllerGroup": "netbox.dev", "controllerKind": "Prefix", "Prefix": {"name":"example","namespace":"default"}, "namespace": "default", "name": "example", "reconcileID": "f12a1389-4102-4507-b5b0-21873447dc16"}

After this, the entire reconciler will be blocked because the next object trying to aquire the lock for the same parentPrefix will time out (to the configured 60 seconds) and thus all operations on the same CRs will be blocked for these 60 seconds (each reconcile loop is single threaded).

This code change enhances the unlocking by moving the leaselocker Unlock() call up in the reconciler to the position where it actually makes sense (right after the Prefix or IP was updated in NetBox). We also move up the status update right after unlocking in order not to lose the reference to the object in NetBox.

[bruelea]

Looks good!
Small note about the update of the status fields: If the update of the status fields fails the reference to the object in NetBox won't be lost, it will be added the next time the reconcile function runs.