Add authenticated GitHub PAT for Forgejo mirror sync #269

Merged
eblume merged 2 commits from feature/mirror-github-pat into main 2026-02-25 20:20:24 -08:00
7 changed files with 234 additions and 2 deletions

View file

@ -52,6 +52,10 @@ NO_REPLY_ADDRESS = noreply.indri
ENABLE_OPENID_SIGNIN = false
ENABLE_OPENID_SIGNUP = false
[mirror]
DEFAULT_INTERVAL = 8h
MIN_INTERVAL = 10m
[cron.update_checker]
ENABLED = false

View file

@ -0,0 +1 @@
Add authenticated GitHub mirror sync with PAT rotation tooling (`mirror-update-pats`, `mirror-create` auth support, how-to doc).

View file

@ -0,0 +1,147 @@
---
title: Manage Forgejo Mirrors
modified: 2026-02-25
tags:
- how-to
- forgejo
- git
---
# Manage Forgejo Mirrors
How Forgejo upstream mirrors work, how to create new mirrors, and how to rotate the GitHub PAT used for authenticated sync.
## Overview
BlumeOps mirrors upstream repositories (mostly from GitHub) into the `mirrors/` organization on forge. These are **pull mirrors** — Forgejo periodically fetches from the upstream URL and updates the local copy. ArgoCD and other consumers then read from forge instead of hitting upstream directly.
### Why Authenticate
GitHub rate-limits unauthenticated git fetch/clone over HTTPS. As of May 2025, these limits were tightened significantly. All mirrors should use an authenticated `clone_addr` (via a GitHub fine-grained PAT) to avoid throttling.
The GitHub PAT is stored in 1Password:
| Property | Value |
|----------|-------|
| **Vault** | blumeops (`vg6xf6vvfmoh5hqjjhlhbeoaie`) |
| **Item** | Forgejo Secrets (`w3663ffnvkewbftncqxtcpeavy`) |
| **Field** | `github-mirror-pat` |
| **op ref** | `op://blumeops/w3663ffnvkewbftncqxtcpeavy/github-mirror-pat` |
### Sync Interval
Mirror sync frequency is controlled by two settings in `app.ini`:
| Setting | Section | Default | Purpose |
|---------|---------|---------|---------|
| `DEFAULT_INTERVAL` | `[mirror]` | `8h` | How often each mirror checks for upstream changes |
| `MIN_INTERVAL` | `[mirror]` | `10m` | Floor for per-repo interval overrides |
| `SCHEDULE` | `[cron.update_mirrors]` | `@every 10m` | How often the cron scans for due mirrors |
With 1030 mirrors at 8h intervals, expect ~14 fetches/hour — well within any rate limit when authenticated.
The explicit configuration lives in `ansible/roles/forgejo/templates/app.ini.j2`.
## Prerequisites
- Access to 1Password blumeops vault
- Forgejo admin account on forge.ops.eblu.me
- `op` CLI authenticated
- For new mirrors: `mise run mirror-create`
## Create a New Mirror
```fish
mise run mirror-create https://github.com/org/repo.git
```
Options:
- `--name <name>` — override the repo name on forge (default: derived from URL)
- `--description <text>` — set the repo description
- `--dry-run` — preview without creating
For GitHub upstreams, the script automatically includes the GitHub PAT from 1Password so the mirror authenticates from the start. Non-GitHub upstreams (Codeberg, etc.) are created without upstream auth.
## Update All Mirror PATs
To update the GitHub PAT on all existing mirrors at once:
```fish
mise run mirror-update-pats
```
This SSHs into indri and rewrites the git remote URL in each mirror's bare repository to embed `eblume:<PAT>@` in the upstream URL. It reads the PAT from 1Password and skips mirrors that already have the current PAT.
Use `--dry-run` to preview:
```fish
mise run mirror-update-pats --dry-run
```
### How It Works
Forgejo stores mirror credentials directly in the bare repo's git config on disk (not in the database). The `remote_address` in SQLite stays as the clean URL; the actual fetch URL in `<repo>.git/config` contains the embedded credentials:
```
# Unauthenticated
url = https://github.com/org/repo.git
# Authenticated
url = https://eblume:<pat>@github.com/org/repo.git
```
The Forgejo API has no endpoint for updating pull mirror credentials, so the script updates the git config directly via SSH.
## Rotate the GitHub PAT
The GitHub fine-grained PAT has a 30-day expiry. Set a recurring reminder (every 20 days) to rotate it before it expires.
### 1. Create a New PAT on GitHub
Go to [GitHub fine-grained token settings](https://github.com/settings/personal-access-tokens/new) and create a new token:
- **Name:** `forgejo-mirror-sync` (or similar, include the date for tracking)
- **Expiration:** 30 days
- **Repository access:** Public repositories (read-only)
- **Permissions:** None required — fine-grained PATs automatically include read-only access to all public repos
Copy the new PAT to your clipboard.
### 2. Update 1Password
With the new PAT on your clipboard:
```fish
op item edit w3663ffnvkewbftncqxtcpeavy github-mirror-pat=(pbpaste) --vault blumeops
```
Verify the update:
```fish
op read "op://blumeops/w3663ffnvkewbftncqxtcpeavy/github-mirror-pat" | head -c 12
# Should print the first 12 chars of the new PAT (github_pat_...)
```
### 3. Push the PAT to All Mirrors
```fish
mise run mirror-update-pats
```
### 4. Delete the Old PAT on GitHub
Return to [GitHub token settings](https://github.com/settings/tokens?type=beta) and delete the previous token.
### 5. Verify
Trigger a manual sync on one mirror to confirm the new PAT works:
1. Go to any mirror repo on forge (e.g., `mirrors/cloudnative-pg`)
2. Click the sync button (circular arrows icon) next to the mirror status
3. Confirm the sync completes without errors
## Related
- [[forgejo]] — Forgejo service reference
- [[upstream-fork-strategy]] — Stacked-branch forking for repos with local modifications
- [[gandi-operations]] — Similar PAT rotation workflow for Gandi DNS

View file

@ -26,6 +26,7 @@ Task-oriented instructions for common BlumeOps operations. These guides assume y
| [[gandi-operations]] | Manage DNS records and cycle the Gandi API token |
| [[use-pypi-proxy]] | Configure pip and publish packages to devpi |
| [[expose-service-publicly]] | Expose a service to the public internet via Fly.io + Tailscale |
| [[manage-forgejo-mirrors]] | Create mirrors, update PATs, and rotate GitHub credentials |
| [[update-documentation]] | Publish docs via build-blumeops workflow |
| [[update-tooling-dependencies]] | Monthly update cycle for pre-commit, Fly, mise, and workflow deps |

View file

@ -50,6 +50,7 @@ Run `mise tasks --sort name` for the live list with descriptions.
| `container-build-and-release` | Trigger container build workflows via Forgejo API |
| `container-version-check` | Validate version consistency across Dockerfiles, nix, and manifests |
| `mirror-create` | Create an upstream mirror in the `mirrors/` Forgejo org |
| `mirror-update-pats` | Update GitHub PAT on all mirror repos on indri |
## Git & Forge

View file

@ -9,6 +9,7 @@ set -euo pipefail
FORGE_API="https://forge.ops.eblu.me/api/v1"
ORG="mirrors"
OP_TOKEN_REF="op://blumeops/w3663ffnvkewbftncqxtcpeavy/api-token"
OP_GITHUB_PAT_REF="op://blumeops/w3663ffnvkewbftncqxtcpeavy/github-mirror-pat"
url="${usage_url:?}"
@ -41,9 +42,16 @@ if [[ "${usage_dry_run:-}" == "true" ]]; then
exit 0
fi
echo "Reading Forgejo API token from 1Password..."
echo "Reading secrets from 1Password..."
token="$(op read "$OP_TOKEN_REF")"
# For GitHub upstreams, include the PAT for authenticated sync
auth_token=""
if [[ "$service" == "github" ]]; then
auth_token="$(op read "$OP_GITHUB_PAT_REF")"
echo "Using GitHub PAT for authenticated mirror sync"
fi
payload=$(cat <<ENDJSON
{
"clone_addr": "$url",
@ -51,7 +59,8 @@ payload=$(cat <<ENDJSON
"repo_owner": "$ORG",
"mirror": true,
"service": "$service",
"description": "$description"
"description": "$description",
"auth_token": "$auth_token"
}
ENDJSON
)

69
mise-tasks/mirror-update-pats Executable file
View file

@ -0,0 +1,69 @@
#!/usr/bin/env bash
eblume marked this conversation as resolved

I'm pretty sure there's a card somewhere that lists mise tasks, update that please

I'm pretty sure there's a card somewhere that lists mise tasks, update that please
#MISE description="Update GitHub PAT on all mirror repos on indri"
#USAGE flag "--dry-run" help="Show what would be done without changing anything"
set -euo pipefail
OP_GITHUB_PAT_REF="op://blumeops/w3663ffnvkewbftncqxtcpeavy/github-mirror-pat"
REPO_BASE="/opt/homebrew/var/forgejo/data/forgejo-repositories"
DB_PATH="/opt/homebrew/var/forgejo/data/forgejo.db"
DRY_RUN="${usage_dry_run:-false}"
echo "Reading GitHub PAT from 1Password..."
pat="$(op read "$OP_GITHUB_PAT_REF")"
if [[ -z "$pat" ]]; then
echo "Error: GitHub PAT is empty"
exit 1
fi
echo "Querying mirrors on indri..."
# Get all GitHub mirrors: org_name, repo_name, upstream_url
mirrors=$(ssh indri "sqlite3 '$DB_PATH' \"
SELECT u.name, r.name, m.remote_address
FROM mirror m
JOIN repository r ON m.repo_id = r.id
JOIN [user] u ON r.owner_id = u.id
WHERE m.remote_address LIKE '%github.com%'
\"" 2>/dev/null)
if [[ -z "$mirrors" ]]; then
echo "No GitHub mirrors found."
exit 0
fi
updated=0
skipped=0
while IFS='|' read -r org repo upstream_url; do
bare_repo="${REPO_BASE}/${org}/${repo}.git"
# Build authenticated URL: https://eblume:<pat>@github.com/...
auth_url="${upstream_url/https:\/\/github.com/https:\/\/eblume:${pat}@github.com}"
if [[ "$DRY_RUN" == "true" ]]; then
echo "[dry-run] ${org}/${repo}: would set origin to authenticated URL"
((updated++))
continue
fi
# Check current remote URL (< /dev/null prevents ssh from consuming loop stdin)
current_url=$(ssh indri "git -C '${bare_repo}' config remote.origin.url" < /dev/null 2>/dev/null || echo "")
if [[ "$current_url" == "$auth_url" ]]; then
echo " ${org}/${repo}: already up to date"
((skipped++))
continue
fi
ssh indri "git -C '${bare_repo}' remote set-url origin '${auth_url}'" < /dev/null 2>/dev/null
echo " ${org}/${repo}: updated"
((updated++))
done <<< "$mirrors"
echo
echo "Done. Updated: ${updated}, Skipped: ${skipped}"
if [[ "$DRY_RUN" == "true" ]]; then
echo "(dry-run mode — no changes were made)"
fi