oriz.in / knowledge
type: runbook
status: active
timestamp: 2026-06-24
tags: [runbook, backup, backblaze-b2, metadata, github-migration-api, weekly-cron, restore-test]

Backup metadata to Backblaze B2 (weekly, single umbrella workflow)

Weekly backup of repo metadata (issues/PRs/wiki/releases) to Backblaze B2

Backup metadata to Backblaze B2 (weekly)

What this solves

The 9-host mirror cron (mirror-all.yml) preserves git history — code, branches, tags, commits — across GitLab/Codeberg/Bitbucket/GitFlic/Azure DevOps/AWS CodeCommit. It does NOT capture GitHub-only metadata: issues, PR conversations, wiki pages, release binaries, milestones, labels, projects. This runbook adds a second weekly cron that uses the GitHub Migration API to grab a .tar.gz per repo with all of that data + uploads to B2.

What this is NOT

Not a code backup — code is in the 9-host mirrors. Don’t conflate the two.

Prerequisites

Workflow file

Path: .github/workflows/backup-metadata-b2.yml in chirag127/workspace.

name: Backup metadata to B2 (weekly)
# Captures issues/PRs/wiki/releases via GitHub Migration API.
# Code is mirrored separately by mirror-all.yml.
# Workspace is PUBLIC so Actions minutes are free + unlimited.

on:
  schedule:
    - cron: '0 21 * * 6'   # Saturday 21:00 UTC = Sunday 02:30 IST
  workflow_dispatch:

permissions:
  contents: read

jobs:
  discover:
    runs-on: ubuntu-latest
    outputs:
      repos: ${{ steps.collect.outputs.repos }}
    steps:
      - name: Collect repos from chirag127 + chirag127
        id: collect
        env:
          GH_TOKEN: ${{ secrets.GH_ADMIN_PAT }}
        run: |
          ORG=$(gh repo list chirag127 --limit 500 --json nameWithOwner --jq '.[].nameWithOwner')
          USR=$(gh repo list chirag127 --limit 500 --json nameWithOwner --jq '.[].nameWithOwner')
          printf '%s\n%s\n' "$ORG" "$USR" | jq -Rn '[inputs | select(length > 0)]' > repos.json
          echo "repos=$(cat repos.json | jq -c '.')" >> "$GITHUB_OUTPUT"

  backup:
    needs: discover
    runs-on: ubuntu-latest
    strategy:
      matrix:
        repo: ${{ fromJson(needs.discover.outputs.repos) }}
      max-parallel: 4   # rate-limit-friendly
      fail-fast: false
    steps:
      - name: Start migration
        id: start
        env:
          GH_TOKEN: ${{ secrets.GH_ADMIN_PAT }}
          REPO: ${{ matrix.repo }}
        run: |
          OWNER="${REPO%%/*}"
          NAME="${REPO##*/}"
          # Org migration API path differs slightly for personal accounts:
          if [ "$OWNER" = "chirag127" ]; then
            ENDPOINT="user/migrations"
            BODY=$(jq -n --arg r "$NAME" '{repositories:[$r], exclude_attachments:false, lock_repositories:false}')
          else
            ENDPOINT="orgs/$OWNER/migrations"
            BODY=$(jq -n --arg r "$REPO" '{repositories:[$r], exclude_attachments:false, lock_repositories:false}')
          fi
          MIG=$(gh api "$ENDPOINT" -X POST --input - <<< "$BODY")
          MID=$(echo "$MIG" | jq -r .id)
          echo "migration_id=$MID" >> "$GITHUB_OUTPUT"
          echo "endpoint=$ENDPOINT" >> "$GITHUB_OUTPUT"

      - name: Poll until exported
        env:
          GH_TOKEN: ${{ secrets.GH_ADMIN_PAT }}
        run: |
          MID="${{ steps.start.outputs.migration_id }}"
          ENDPOINT="${{ steps.start.outputs.endpoint }}"
          for i in $(seq 1 60); do
            STATE=$(gh api "$ENDPOINT/$MID" --jq .state)
            echo "attempt $i: $STATE"
            [ "$STATE" = "exported" ] && exit 0
            [ "$STATE" = "failed" ] && { echo "::error::migration failed"; exit 1; }
            sleep 20
          done
          echo "::error::migration timeout after 20min"; exit 1

      - name: Download archive
        env:
          GH_TOKEN: ${{ secrets.GH_ADMIN_PAT }}
        run: |
          MID="${{ steps.start.outputs.migration_id }}"
          ENDPOINT="${{ steps.start.outputs.endpoint }}"
          gh api -H "Accept: application/vnd.github+json" \
            "$ENDPOINT/$MID/archive" \
            > backup.tar.gz

      - name: Upload to B2 (S3-compatible)
        env:
          AWS_ACCESS_KEY_ID: ${{ secrets.B2_APPLICATION_KEY_ID }}
          AWS_SECRET_ACCESS_KEY: ${{ secrets.B2_APPLICATION_KEY }}
          BUCKET: ${{ secrets.B2_BUCKET_NAME }}
          ENDPOINT: ${{ secrets.B2_ENDPOINT }}
          REPO: ${{ matrix.repo }}
        run: |
          DATE=$(date -u +%Y-%m-%d)
          SAFE_REPO="${REPO//\//__}"
          aws s3 cp backup.tar.gz \
            "s3://${BUCKET}/metadata/${SAFE_REPO}/${DATE}.tar.gz" \
            --endpoint-url "${ENDPOINT}"

      - name: Cleanup migration on GitHub
        if: always()
        env:
          GH_TOKEN: ${{ secrets.GH_ADMIN_PAT }}
        run: |
          MID="${{ steps.start.outputs.migration_id }}"
          ENDPOINT="${{ steps.start.outputs.endpoint }}"
          gh api -X DELETE "$ENDPOINT/$MID/archive" || true

Pruning (10 GB free-tier ceiling)

10 GB is generous but not infinite. A separate workflow keeps the bucket clean:

# .github/workflows/backup-prune-b2.yml — runs after every backup-metadata-b2 successful run
# Keeps: last 4 weekly snapshots + last 1st-of-month for 12 months + last 1st-of-year forever
# Implemented as a small `aws s3 ls | awk` script — see knowledge bundle for full template.

Restore test (monthly)

Separate workflow .github/workflows/backup-restore-test.yml:

name: Backup restore test (monthly)
on:
  schedule:
    - cron: '0 4 1 * *'   # 1st of each month, 04:00 UTC
  workflow_dispatch:

jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - name: Pick random repo + latest backup
        env:
          AWS_ACCESS_KEY_ID: ${{ secrets.B2_APPLICATION_KEY_ID }}
          AWS_SECRET_ACCESS_KEY: ${{ secrets.B2_APPLICATION_KEY }}
          BUCKET: ${{ secrets.B2_BUCKET_NAME }}
          ENDPOINT: ${{ secrets.B2_ENDPOINT }}
        run: |
          # List repos under metadata/, pick one at random
          REPOS=$(aws s3 ls "s3://${BUCKET}/metadata/" --endpoint-url "${ENDPOINT}" | awk '/PRE/{gsub("/","",$2); print $2}')
          PICK=$(echo "$REPOS" | shuf -n 1)
          # Latest snapshot
          LATEST=$(aws s3 ls "s3://${BUCKET}/metadata/${PICK}/" --endpoint-url "${ENDPOINT}" | sort | tail -1 | awk '{print $NF}')
          aws s3 cp "s3://${BUCKET}/metadata/${PICK}/${LATEST}" backup.tar.gz --endpoint-url "${ENDPOINT}"
          echo "PICK=$PICK" >> "$GITHUB_ENV"

      - name: Unpack + smoke-test
        run: |
          mkdir -p restored && tar -xzf backup.tar.gz -C restored
          # Validate: should contain at least repositories/ + (issues_*.json OR pull_requests_*.json)
          [ -d restored/repositories ] || { echo "::error::no repositories/ in tarball"; exit 1; }
          # Count records present
          ISSUES=$(find restored -name 'issues_*.json' | xargs -r jq '.|length' 2>/dev/null | awk '{s+=$1} END {print s+0}')
          PRS=$(find restored -name 'pull_requests_*.json' | xargs -r jq '.|length' 2>/dev/null | awk '{s+=$1} END {print s+0}')
          echo "::notice::Restored ${{ env.PICK }}: ${ISSUES} issues, ${PRS} PRs"

      - name: Post status on failure
        if: failure()
        env:
          GH_TOKEN: ${{ secrets.GH_ADMIN_PAT }}
        run: gh issue create -R chirag127/workspace -t "Backup restore test failed $(date -u +%F)" -b "See run ${{ github.run_id }}"

Why not one workflow with code + metadata?

Cross-refs

Edit on GitHub · Back to index