Gadgit

Mirrors a Git repo to which web applications can make efficient queries for information that the GitHub API does not provide.

Demo

Serves pytorch repo

Features

Event-driven repo maintenance

Keeps repo clone up-to-date by re-fetching upon GitHub push and pull_request events.

Efficient information retrieval

• Bulk git commit metadata retrieval
• Queries
  • is-ancestor queries
  • merge base queries
  • Determine Pull Request with a given head commit
  • Determine head commit of a given Pull Request

Diagnostics

• Logs all received GitHub events
• Logs all fetch operations

Repo hosting

• uses a "bare" repo clone
• rate-limits new fetches to 1 per minute
• protects against simultaneous fetch or clone operations

Deployment

Intended for hosting on Elastic Beanstalk.

Use this command to set the GitHub webhook secret:

eb setenv GITHUB_WEBHOOK_SECRET=your_secret

To deploy the application, run:

./deploy.sh

Webhook

Set up a GitHub webhook to post to http://gadgit.pytorch.org/github-webhook-event upon pushes. This will trigger a re-fetch of the relevant refs.

Testing

Local

Run:

eb-flask/application.py

Exercise an endpoint with curl:

curl --data '["0c7537c40939f7682c179813a4b7a50020f08152", "7ed9a3ec4895f0f501cf435fec88ff974d93f3da"]' http://localhost:5000/commit-metadata

Troubleshooting

Sometimes the repo becomes corrupted, with commands returning error messages like:

fatal: Not a git repository: '/tmp/repo/pytorch.git'

This can be a symptom of a missing HEAD file.

To fix this, create the file /tmp/repo/pytorch.git/HEAD with the content:

ref: refs/heads/master