Summary
The user is reviewing documentation on enabling gang scheduling with the Yunikorn scheduler and has questions about the uniqueness of the task-group-name among pods in the same Flyte execution. They are unsure if this name needs to be unique across all pods or just within the execution. The user notes a broken link in the Flyte documentation and emphasizes the importance of the applicationId
label for identifying pods in the same application. They ask how to set the applicationId label for Flyte pods, which currently have a unique execution-id label. The user suggests using the pod template argument of @task to add a label that generates a UUID at registration but is concerned about running the same workflow multiple times. They conclude that all worker pods in the same PyTorchJob need a shared unique value for their applicationId. Additionally, the user seeks guidance on setting labels and annotations in template.metadata using flytectl and provides an expected tf yaml example for a TFJob configuration, questioning if the applicationId is required for the TFJob or if the one under the template is sufficient.