Description
@platform-delivery-tooling-support folks, I will describe the steps to reproduce an outstanding issue with the operator in air-gapped environments.
In short, if the Controller Manager can’t reach the W&B endpoint, the entire stack will fail to be deployed.
In other words, this is a blocker.
- Deploy the operator using the helm chart
helm upgrade --install -n wandb operator wandb/operator
-
cut the access of cluster nodes from the internet (this step was used only to repro)
-
apply the CRD
kubectl apply -f wandb.yaml
At this point, nothing will happen. Not a single container will be created.
- check the Controller Manager logs
{"level":"dpanic","ts":"2024-08-07T13:33:40Z","msg":"non-string key argument passed to logging, ignoring all later arguments","controller":"weightsandbiases","controllerGroup":"apps.wandb.com","controllerKind":"WeightsAndBiases","WeightsAndBiases":{"name":"wandb","namespace":"default"},"namespace":"default","name":"wandb","reconcileID":"08ea8f72-2387-4b9b-b445-0afc2b2caa58","invalid key":"Secret \"wandb-latest-cached-release\" not found","stacktrace":"github.com/wandb/operator/controllers.(*WeightsAndBiasesReconciler).Reconcile\n\t/workspace/controllers/weightsandbiases_controller.go:135\nsigs.k8s.io/controller-runtime/pkg/internal/controller.(Controller).Reconcile\n\t/go/pkg/mod/sigs.k8s.io/controller-runtime@v0.14.1/pkg/internal/controller/controller.go:122\nsigs.k8s.io/controller-runtime/pkg/internal/controller.(Controller).reconcileHandler\n\t/go/pkg/mod/sigs.k8s.io/controller-runtime@v0.14.1/pkg/internal/controller/controller.go:323\nsigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller).processNextWorkItem\n\t/go/pkg/mod/sigs.k8s.io/controller-runtime@v0.14.1/pkg/internal/controller/controller.go:274\nsigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller).Start.func2.2\n\t/go/pkg/mod/sigs.k8s.io/controller-runtime@v0.14.1/pkg/internal/controller/controller.go:235"}
{"level":"error","ts":"2024-08-07T13:33:40Z","msg":"No cached release found for deployer spec","controller":"weightsandbiases","controllerGroup":"apps.wandb.com","controllerKind":"WeightsAndBiases","WeightsAndBiases":{"name":"wandb","namespace":"default"},"namespace":"default","name":"wandb","reconcileID":"08ea8f72-2387-4b9b-b445-0afc2b2caa58","error":"Secret \"wandb-latest-cached-release\" not...
Issue created in Slack from a [message](https://weightsandbiases.slack.com/archives/C06CKRTPKDF/p1723039059602709?thread_ts=1723039059.602709&cid=C06CKRTPKDF).
Description
@platform-delivery-tooling-support folks, I will describe the steps to reproduce an outstanding issue with the operator in air-gapped environments.
In short, if the Controller Manager can’t reach the W&B endpoint, the entire stack will fail to be deployed.
In other words, this is a blocker.
helm upgrade --install -n wandb operator wandb/operatorcut the access of cluster nodes from the internet (this step was used only to repro)
apply the CRD
kubectl apply -f wandb.yamlAt this point, nothing will happen. Not a single container will be created.