err := r.client.Get(context.TODO(), req.NamespacedName, res) will not get config cluster, since it is still cluster based

[houshengbo]

I am not sure if this is the bug with the client of controller-runtime, or it is a race condition in operator.

The config get function at
https://github.com/tektoncd/operator/blob/master/pkg/controller/config/config_controller.go#L166
will return NOT FOUND error, even if the CR was created with confirmation.

We need to remove ignoreRequest, otherwise we cannot do future action if pipeline deployment is removed as in #30.

We need to use err := r.client.Get(context.TODO(), types.NamespacedName{Name: req.Name}, res) to get the config cluster.

[vincent-pli]

@houshengbo
Correct, that's a bug of controller-runtime, exactly I mentioned in another issue:
kubernetes-sigs/controller-runtime#274

I want to explained more about the issue:
In our case, we watch two kinds of resource:

• Config: cluster scope
• Deployment: namespace scope and owner by Config

Notice, we watch Config by EnqueueRequestForObject and
watch Deployment by EnqueueRequestForOwner
that's means when any update on Config, Reconcile will receive request about cluster scope Config
and when Deployment update, Reconcile will receive request about namespace scope Config, that's not correct behavior, Reconcile should receive cluster scope Config, that's what exactly the bug talk about! the bug has already be fixed in Controller-runtime v0.2.0.*.

But your fix could work now, thanks