Registry: Consider backing up files on S3

[jpmckinney]

To:

• Mitigate risk of Hetzner hardware issues
• Offload traffic (leading to better performance)
• Simplify server upgrades (less data to copy)
• Separate the frontend from the backend (allowing the frontend to run on Linode, and progressing Registry: Read-only backup server #289)

[jpmckinney]

When making the S3 estimate I forgot to add outbound, which is 0.09/GB.

Based on https://robot.hetzner.com/traffic we had 8TB from Aug 2024-Jul 2025, so $720/y for data transfer only.

So, I think we'll continue to look at hosting the files ourselves for now.

[jpmckinney]

Re-opening to make this about storing exported files, as backup. This might be implemented in the deploy repo instead.

Currently running:

aws s3 sync /data/storage/exporter s3://ocp-registry-backup/exporter/ --storage-class STANDARD_IA

[jpmckinney]

FYI @RobHooper

[jpmckinney]

Actually, there are some other directories to backup, so I'll transfer the issue to the deploy repo.

Directories: https://ocdsdeploy.readthedocs.io/en/latest/deploy/servers/data-registry.html#filesystem

For the exporter files: We could alternatively only backup the files of active jobs, but this would require changes to application code, and might not save a whole lot of S3 transfer and storage costs.

Total jobs adds up to 772G. Back in 2021 (#289), I had estimated active jobs to require tens of GBs.