Collector instance not created if name + namespace exceeds 63 due to how the labels are built

[mimatache]

When applying a CR where the name + namespace length is over 63 characters the operator fails to create the instances dues to a failure when setting the labels:

{"level":"error","ts":1638457753.904282,"logger":"controllers.OpenTelemetryCollector","msg":"failed to reconcile config maps","error":"failed to reconcile the expected configmaps: failed to create: ConfigMap \"xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx-collector\" is invalid: metadata.labels: Invalid value: \"nsxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx.xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx\": must be no more than 63 characters","stacktrace":"github.com/open-telemetry/opentelemetry-operator/controllers.(*OpenTelemetryCollectorReconciler).Reconcile\n\t/workspace/controllers/opentelemetrycollector_controller.go:147\nsigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller).reconcileHandler\n\t/go/pkg/mod/sigs.k8s.io/[email protected]/pkg/internal/controller/controller.go:298\nsigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller).processNextWorkItem\n\t/go/pkg/mod/sigs.k8s.io/[email protected]/pkg/internal/controller/controller.go:253\nsigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller).Start.func2.2\n\t/go/pkg/mod/sigs.k8s.io/[email protected]/pkg/internal/controller/controller.go:214"}

As the Kubernetes standards state, a label can be no more than 63 characters. The same goes for instance names.

Since the names of items (CR & namespace) are allowed to be up to 63 characters, having the label set automatically to name + namespace seems like an issue with the operator.

As solutions I have 2 propositions:

1. Split that label into 2 labels: 1 for name and 1 for namespace
2. If the label size exceeds 63 characters create some kind of hash that is limited to 63 characters and replace the label with that.

[mmatache]

I would like to fix this issue. Are there any preferences on the approach? Are the labels used in the operator anywhere for lookups?
I am partial to the second approach since it would remain backwards compatible to anybody that is currently using the operator.

[jpkrohling]

To be clear, this is about the label app.kubernetes.io/instance, but we would have the same problem elsewhere if the name is too big. We have a routine for the Jaeger Operator to deal with this, perhaps we could have the same here:

https://github.com/jaegertracing/jaeger-operator/blob/91e3b69ee5c8761bbda9d3cf431400a73fc1112a/pkg/service/query.go#L69

[mmatache]

@jpkrohling thank you for the information!
Is it ok to use the same functions Jaeger is using?
If yes, should we make a copy of the functions, or add a dependency to Jaeger and use them directly?
If no, is there any specific functionality you are looking for?

[jpkrohling]

Yes, I think it's OK to use the same, but copy them. There's no need to add a dependency just because of that.

[mmatache]

My thoughts exactly, but since this is not my project, I still wanted to confirm with you.

[mmatache]

@jpkrohling I have found an issue in the jaeger trimming code which was not covered by any tests.
I have created #641 to handle this case. You should probably also look into whether this should be fixed in the Jaeger code as well.

[jpkrohling]

Would you be open to creating an issue there as well? Bonus points if you can send in a PR, which should be similar to this one here.

[mmatache]

@jpkrohling One hand scratches another 😄 : jaegertracing/jaeger-operator#1678