The error you're seeing suggests a data corruption issue, likely due to the absence of a 'hummock version checkpoint' from the object store in the /checkpoint directory. To resolve this, you should clean both the metadata stored in etcd and the data in the storage. If you're deploying a new cluster, ensure that you're not using an etcd from a previous cluster with an empty object store, as this can lead to inconsistencies. Additionally, if you're using MinIO, check for any 'DELETE' operations that might have removed the /checkpoint directory. It's also recommended to use S3 for better performance and consistency, and to ensure that etcd has a persistent volume attached to prevent data loss on restarts.