oriz.in / knowledge
type: decision
status: active
timestamp: 2026-06-24
tags: [decision, backup, disaster-recovery, metadata, cloudflare-r2, backblaze-b2, huggingface, github-migration]

Alternative free-forever backup channels for GitHub code and metadata

Alternative free backup channels repositories and their metadata (issues, PRs, wikis, releases) using Cloudflare R2, Backblaze B2, Hugging Face Datasets (with caveats), and the native GitHub Migration API. Integrated into our overall disaster recovery options.

Alternative free-forever backup channels

Context

Our primary backup strategy is the 6-host active git mirror (using GitHub Actions to push code). However, git mirroring only copies the source code and git history. It does NOT backup repository metadata:

This decision documents alternative, free-forever, automated methods to back up both repositories and their associated metadata.

1. Native GitHub Migration API (Full Export)

GitHub offers a native migration API that generates a single, downloadable .tar.gz archive containing the entire repository structure and metadata.

Capabilities

Automated Flow (GH Actions + Target Storage)

  1. Trigger: A weekly GitHub Actions cron runs a script to start the migration: 
    # Start migration for repository or organization
curl -X POST -H "Authorization: token $GH_PAT" \
  -H "Accept: application/vnd.github+json" \
  "https://api.github.com/orgs/chirag127/migrations" \
  -d '{"repositories":["workspace"],"exclude_attachments":false}'
  2. Poll: The workflow sleeps/polls the status endpoint until the migration state becomes exported.
  3. Download: Downloads the archive using the returned download URL: 
    curl -H "Authorization: token $GH_PAT" \
  -o backup.tar.gz \
  "https://api.github.com/orgs/chirag127/migrations/$MIGRATION_ID/archive"
  4. Push to Storage: Stream/copy the archive to one of the free-forever storage layers below.

2. Cloudflare R2 (10 GB S3-Compatible Free Storage) — PREFERRED

Cloudflare R2 provides a free-tier object storage bucket with zero egress fees, making it the perfect remote archive repository for backup bundles.

Limits

Automation via AWS CLI

Since R2 is S3-compatible, the pre-installed aws CLI on GitHub runners can copy backups directly:

- name: Upload to Cloudflare R2
  run: |
    aws s3 cp backup.tar.gz s3://oriz-backups/github-backup-$(date +%F).tar.gz \
      --endpoint-url ${{ secrets.R2_ENDPOINT }}
  env:
    AWS_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
    AWS_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
    AWS_DEFAULT_REGION: auto

3. Backblaze B2 (10 GB Free Storage) — SECONDARY

Backblaze B2 is an alternative object storage option offering a free-forever tier without requiring a credit card.

Limits

Automation via Rclone or B2 CLI

You can use rclone (pre-installed on runner or run-time script) to sync backups:

rclone copy backup.tar.gz b2:oriz-backups-bucket/ -v

4. Hugging Face Datasets — AVOID for generic file backup

?? CAVEAT 2026-06-24: Hugging Face’s “free unlimited” hosting is intended for public ML datasets and model weights. Using HF Datasets as a generic file backup target for tarballs / metadata archives:

Use R2 or B2 instead for backup tarballs. Only use HF Datasets for things that are genuinely ML-related (e.g. the oriz-ai-providers-data repo, or training datasets if those ever exist).

5. Open-Source Metadata Backup Tools (CLI)

For custom metadata extraction (saving issues/PRs as readable JSON rather than binary tarballs), the following CLI tools can be executed inside GitHub Actions:

  1. github-backup (Python):
    • Command: github-backup --token $GH_PAT --output-directory ./backup --all --private-embed-key chirag127
    • Backs up repositories, wikis, issues, pull requests, milestones, labels, and releases.
  2. gitbackup (Go):
    • Excellent for multi-platform clones.
  3. rclone (Go):
    • For syncing local folders of downloaded metadata files to GDrive/OneDrive free tiers.

Recommendation

Default channel: Native GitHub Migration API ? Cloudflare R2 weekly. R2’s zero egress + 10 GB free + S3-compatible CLI makes it the cleanest path. Use B2 as a secondary mirror only if you’ve outgrown R2’s 10 GB.

Skip HF Datasets for this purpose — not designed for it, real ToS risk.

Cross-refs

Edit on GitHub · Back to index