I have a few Flink 1.17 workloads that save checkpoints to S3. They are running on Kubernetes and managed by Flink Operator. I have as well decided to RETAIN_ON_CANCELLATION for safety.