Deploying with Git

Last Updated: 02 February 2015

cli deploy git

Git is a powerful decentralized revision control system, and is the means for deploying apps to Heroku. You don’t need to be proficient with Git to use it for deploying code to Heroku, but you may find it valuable to learn the basics.

Tracking your app in git

Before you can push an app to Heroku, you’ll need to initialize a local Git repo and commit your files to it. For example, if you have an app in a directory, myapp, then create a new repo for it:

$ cd myapp
$ git init
Initialized empty Git repository in .git/
$ git add .
$ git commit -m "my first commit"
Created initial commit 5df2d09: my first commit
 44 files changed, 8393 insertions(+), 0 deletions(-)
 create mode 100644 README
 create mode 100644 Procfile
 create mode 100644 app/controllers/source_file
...

This is a local repository, now residing inside the .git directory. Nothing has been sent anywhere yet; you’ll need to create a remote and do a push to deploy your code to Heroku.

Creating a Heroku remote

Git remotes are references to remote repositories. You can have any number of these, but for now we’ll focus on just the remote to Heroku. The heroku create command creates a new application on Heroku – along with a git remote that must be used to receive your application source.

$ heroku create
Creating falling-wind-1624... done, stack is cedar-14
http://falling-wind-1624.herokuapp.com/ | https://git.heroku.com/falling-wind-1624.git
Git remote heroku added

By default, Heroku configures HTTP as the Git transport. Heroku Toolbelt will automatically place credentials in the .netrc file on heroku login. The Git client uses cURL when interacting with HTTP remotes, and cURL will use the credentials from the .netrc file. See the Authentication section and the CLI authentication article for details.

You can verify the remote in your git configuration as well:

$ git remote -v
heroku  https://git.heroku.com/falling-wind-1624.git (fetch)
heroku  https://git.heroku.com/falling-wind-1624.git (push)

You can also take an existing Git repo and add a remote using the git URL provided when you created your app. You may need to do this to associate a Git repo with an existing application. The heroku git:remote command will add this remote for you based on your applications git url.

$ heroku git:remote -a falling-wind-1624
Git remote heroku added.

The remote is named heroku in this example, but you can name the remote anything you want by passing -r other_remote_name. You may find it easier to follow the examples if you stick to using the heroku remote rather than using one with a different name.

There is one special remote name: origin, which is the default for pushes. Using origin as the remote name will allow you to type just git push instead of git push heroku, but we recommend using an explicitly named remote.

Deploying code

Your Heroku app starts with a blank repository – it has no branches and no code. So the first time you deploy, you’ll need to specify a remote branch to push to. You can do your first push:

$ git push heroku master
Initializing repository, done.
updating 'refs/heads/master'
...

This will push your code to the heroku remote, created earlier. Use this whenever you want to deploy the latest code committed in Git to Heroku.

During the start of your first build, Initializing repository will be displayed while your app’s repository is created on Heroku. On subsequent builds, Fetching repository will be displayed while your app’s repository is fetched and prepared to accept your push.

Branches pushed to Heroku other than master will be ignored by this command. If you’re working out of another branch locally, you can either merge to master before pushing, or specify that you want to push your local branch to a remote master. To push a branch other than master, use this syntax:

$ git push heroku yourbranch:master

Canceling a deployment

When deploying code using git push via SSH Git transport, you can interrupt the build process on Heroku by pressing Ctrl + C on your machine. As long as this is not done as the build process completes (generally around the “Launching…” message), and you’ve taken latency into account, then it will halt the build process and no new release will be generated. Build cancelation via HTTP Git transport is currently not supported.

HTTP Git authentication

The Heroku HTTP Git endpoint only accepts API-key based HTTP Basic authentication. A username is not required and any value passed for username is ignored.

If, for any reason, you authenticate to the Git service with incorrect credentials, you’ll get this error:

remote: ! WARNING:
remote: ! Do not authenticate with username and password using git.
remote: ! Run `heroku login` to update your credentials, then retry the git command.
remote: ! See documentation for details: https://devcenter.heroku.com/articles/http-git#authentication

When you do heroku login, Toolbelt will write an entry for git.heroku.com into your .netrc file (or its Windows equivalent). Since the Git client uses cURL when interacting with HTTP Git remotes, correct authentication will now happen transparently.

If you’re using other Git clients, such as EGit or Tower, configure them to use an empty string for username (or any string you like – it’s ignored) and your account API key for password. The API key is available from Toolbelt and in Dashboard.

SSH Git transport

The default Git transport configured by Heroku toolbelt is HTTP, but SSH transport is also supported. SSH and HTTP transport can be used interchangeably by the same user and by multiple users collaborating on the same app. To have Heroku Toolbelt configure SSH transport, you can pass a --ssh-git flag to the heroku create, heroku git:remote and heroku git:clone commands.

$ heroku create --ssh-git

To use SSH Git transport, you have to register your SSH key with Heroku. See the Managing SSH Keys article for details.

If you want to always use SSH Git with Heroku on a particular machine, you can add the following global config:

$ git config --global url.ssh://git@heroku.com/.insteadOf https://git.heroku.com/

HTTP URLs will still be written to .git folders but Git will rewrite, on the fly, all Heroku HTTP Git URLs to use SSH.

To remove this rewrite setting, run:

$ git config --global --remove-section url.ssh://git@heroku.com/

Multiple remotes and environments

The same techniques used to deploy to production can be used to deploy a development branch of your application to a staging application on Heroku, as described in Managing Multiple Environments for an App.

Build cache

Buildpacks can optionally cache content for re-use between builds. A typical use-case for the buildpack is to speed up builds by caching dependencies so that they don’t have to be re-fetched on every build. This greatly speeds up builds.

If you suspect that a build-problem is related to this caching, you can use the heroku-repo plugin to clear the cache.

Repo size

While there is not a hard limit on your repository size, very large repositories (over 600MB) are not recommended; they may cause timeouts and slow pushes overall. Running heroku apps:info will show you your repository size. The app build cache is stored inside the app repository, so don’t be surprised if the repository is larger remotely than locally.

Common causes of large repositories are binary files checked into the repo (git is notoriously bad at handling binaries) or constantly-changing development logs. Removing files committed by accident can be done with git filter-branch, though after running it you will have to push with the --force option, which is something that requires coordination among your team.

Other limits

To protect the Git service, Heroku imposes certain limits on Git repo use and content size.

Users are limited to a rolling window of 75 Git requests per hour, per user, per app. Once this limit is reached, Git requests are denied until request levels drop below the limit for a few minutes, with the error message:

 !  Too many requests for this Git repo. Please try again later.

If you reach this limit, ensure there are not automated processes or scripts polling the Git repo.

In addition, the uncompressed size of a checkout of HEAD from the repo, combined with the size of restored submodules, cannot exceed 1 GB.

Using subversion or other revision control systems

What if you’re already using Subversion or another revision control system to track your source code? Although we believe that Git is one of the best choices available for revision control, you don’t need to stop using your current revision control system. Git can be purely a deployment mechanism, existing side-by-side with your other tool.

For example, if you are using Subversion, initialize your Git repo as described above. Then, add a .gitignore file to tell Git to ignore your Subversion directories.

$ git init
$ echo .svn > .gitignore
$ git add .
$ git commit -m "using git for heroku deployment"

Now tell Subversion to ignore Git:

$ svn propset svn:ignore .git .
property 'svn:ignore' set on '.'
$ svn commit -m "ignoring git folder (git is used for heroku deployment)"

Each time you wish to deploy to Heroku:

$ git add -A
$ git commit -m "commit for deploy to heroku"
...

$ git push -f heroku

Additional resources

• Git on Rails shows common conventions for using Git to track Rails apps.
• Git cheat sheets for web and print consumption.
• Git - SVN Crash Course
• Pro Git and The Git Community Book are great books covering all of Git.