Deploying with Git
Last updated 11 January 2017
Table of Contents
- Tracking your app in git
- Creating a Heroku remote
- Deploying code
- Build deploy ordering
- Detach from build process
- HTTP Git authentication
- SSH Git transport
- Multiple remotes and environments
- Build cache
- Repository size
- Other limits
- Using subversion or other revision control systems
- Additional resources
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
Heroku apps expect the app directory structure at the root of the repository. If your app is inside a subdirectory in your repository, it won’t run when pushed to Heroku.
Before you can push an app to Heroku, you’ll need to initialize a local Git repository and commit your files to it. For example, if you have an app in a directory, myapp, then create a new repository 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. The Heroku CLI 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 repository and add a remote using the git URL provided
when you created your app. You may need to do this to associate a Git repository 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
git push heroku, but we recommend using an explicitly named remote.
To switch from SSH Git to HTTP Git, run
heroku git:remote in the directory holding your local Git repository. The CLI will override the
heroku Git remote to use the HTTP protocol.
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
Applications that rely on git submodules are supported, in addition to many other dependency resolution strategies.
git lfs is not supported and may cause pushes to fail.
Build deploy ordering
If multiple parallel builds are started for an app (either by the same user performing multiple pushes, by app collaborators pushing concurrently or because builds are created concurrently with other mechanisms like Build API or GitHub Sync), then the last build to complete will generally be the one that ends up being deployed for the app, even if that build was started before other builds.
Take an example of two builds, A and B: Build A is started, runs slowly and completes in 2 minutes. 30 seconds after build A is started, build B is started and completes in 1 minute. Build B is deployed for the app when it completes. 30 seconds later, build A completes and is deployed for the app. The end result is that build A is deployed on the app, even though build B was started later.
Detach from build process
When deploying code using
git push, you can detach from the build process by pressing Ctrl + C. However, your build will continue to process in the background and will create a new release as soon as it finished.
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.
You cannot authenticate with the Heroku HTTP Git endpoint using your Heroku username (email) and password. Use an API key as described in this section
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/git#http-git-authentication
When you do
heroku login, the CLI 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 in the CLI and in Dashboard.
SSH Git transport
The default Git transport configured by the Heroku CLI 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 the Heroku CLI configure SSH transport, you can pass a
--ssh-git flag to the
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://firstname.lastname@example.org/.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://email@example.com/
SSH is not supported for SSO users. SSO users must use HTTPS as a git transport.
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.
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.
While there is not a hard limit on your repository size, very large
repositories (over 600 MB) 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
repository (Git is notoriously bad at handling binaries) or
constantly-changing development logs. Removing files committed by
accident can be done with
though after running it you will have to push with the
option, which is something that requires coordination among your team.
To protect the Git service, Heroku imposes certain limits on Git repository 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 repository.
In addition, the uncompressed size of a checkout of
HEAD from the repository, 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.
You can learn much more about
.gitignore in our article on the topic.
For example, if you are using Subversion, initialize your Git repository 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)"
-f (force flag) is recommended in order to avoid conflicts with other developers’ pushes. Since you are not using Git for your revision control, but as a transport only, using the force flag is a reasonable practice.
Each time you wish to deploy to Heroku:
$ git add -A $ git commit -m "commit for deploy to heroku" ... $ git push -f heroku