Build controller panics with nil pointer dereference on every build in 2.10.0

[Diesel93]

Camel K version: 2.10.0 (Helm chart)

Kubernetes: EKS 1.31, eu-west-2

After upgrading from 2.8.0 to 2.10.0 the operator panics on every build. The build pod never spawns — the reconcile loop crashes and retries indefinitely.

Operator logs:

{"level":"error","msg":"Observed a panic","controller":"build-controller","Build":{"name":"kit-d7njpnevtj8s73903ip0","namespace":"camel-k"},"panic":"runtime error: invalid memory address or nil pointer dereference"}
{"level":"error","msg":"Reconciler error","error":"panic: runtime error: invalid memory address or nil pointer dereference [recovered]"}

Always the same sequence — build queues fine, then immediately panics:

INFO | Invoking action build
INFO | State transition
INFO | Build queue duration 357ms
INFO | Invoking action build
ERROR | Observed a panic

Stack trace:

github.com/apache/camel-k/v2/pkg/apis/camel/v1.(*Build).GetObjectKind
k8s.io/client-go/tools/reference.GetReference
k8s.io/client-go/tools/events.(*recorderImpl).eventf
github.com/apache/camel-k/v2/pkg/event/manager.go:210
github.com/apache/camel-k/v2/pkg/controller/build/build_controller.go:196

Root cause:

PR #6492 bumped controller-runtime to 0.23.1 and migrated from record.EventRecorder to events.EventRecorder. The new recorder calls reference.GetReference(scheme, obj) which calls obj.GetObjectKind(). Since client.Get() strips TypeMeta from returned objects (known controller-runtime behaviour — kubernetes-sigs/controller-runtime#1735), GetObjectKind() ends up dereferencing a nil pointer.

The old record.EventRecorder didn't go through GetReference so this wasn't an issue in 2.9.x.

Reproducing:

Install 2.10.0 via Helm and trigger any build. Panic happens on the first state transition event after the build queues — doesn't matter what registry or build strategy.

Works fine on 2.9.1, broken on 2.10.0.

Fix:

Populate TypeMeta on the Build object before it reaches the event recorder, e.g. in build_controller.go:

instance.TypeMeta = metav1.TypeMeta{
    Kind:       "Build",
    APIVersion: "camel.apache.org/v1",
}

[squakez]

Thanks for reporting. I wonder if this is only happening with Helm installation procedure. We have e2e tests for pod strategy with Kustomization installation procedure though, and it is working fine apparently. Can you verify if you're able to reproduce with Kustomization as well? and surely a PR is welcome if you have time to contribute.

[Diesel93]

Dug into this a bit more — it's not Helm specific. The panic is a Go typed-nil-interface issue in NotifyError. When an action returns (nil, err), the typed nil *v1.Build passes the newResource != nil check because the interface has type information even though the pointer is nil. This then panics in recorder.Eventf → GetReference → GetObjectKind().

Tested a fix on a live cluster — PR up at #6589.