Hey team I was looking at how Metaflow handles log persistence across different orchestrators, and I noticed the AWS Batch implementation has a known limitation with OOM log loss but relies on CloudWatch as a fallback. I wanted to discuss how Kubernetes handles the same scenario.