Unexpected Ratio Between S3 Upload Size and Ephemeral Storage Usage on Kubernetes

[talevy18]

I’m using Goofys to mount an S3 bucket in a k8s pod. I’ve noticed that when my pod has a certain ephemeral storage limit, I’m able to write significantly more data to S3 than that limit (e.g., 1GB ephemeral limit allows ~10GB written, 100GB allows ~234GB, etc.). I understand that S3 multi-part uploads and buffering can impact local disk usage, but I’d like more clarity on how Goofys manages these uploads, why these ratios can vary so drastically, and how to better estimate (or configure) ephemeral usage when I need to upload large data sets (e.g., hundreds of gigabytes). Any guidance or best practices on managing ephemeral storage usage in this scenario would be greatly appreciated.