Git LFS

Managing large files such as audio, video and graphics files has always been one of the shortcomings of Git. The general recommendation is to not have Git repositories larger than 1GB to preserve performance.

How it works

Git LFS client talks with the GitLab server over HTTPS. It uses HTTP Basic Authentication to authorize client requests. Once the request is authorized, Git LFS client receives instructions from where to fetch or where to push the large file.

GitLab server configuration

Documentation for GitLab instance administrators is under LFS administration doc.

Requirements

  • Git LFS is supported in GitLab starting with version 8.2
  • Git LFS must be enabled under project settings
  • Git LFS client version 1.0.1 and up

Known limitations

  • Git LFS v1 original API is not supported since it was deprecated early in LFS development
  • When SSH is set as a remote, Git LFS objects still go through HTTPS
  • Any Git LFS request will ask for HTTPS credentials to be provided so a good Git credentials store is recommended
  • Git LFS always assumes HTTPS so if you have GitLab server on HTTP you will have to add the URL to Git config manually (see troubleshooting)

Note: With 8.12 GitLab added LFS support to SSH. The Git LFS communication still goes over HTTP, but now the SSH client passes the correct credentials to the Git LFS client, so no action is required by the user.

Using Git LFS

Lets take a look at the workflow when you need to check large files into your Git repository with Git LFS. For example, if you want to upload a very large file and check it into your Git repository:

git clone git@gitlab.example.com:group/project.git
git lfs install                       # initialize the Git LFS project
git lfs track "*.iso"                 # select the file extensions that you want to treat as large files

Once a certain file extension is marked for tracking as a LFS object you can use Git as usual without having to redo the command to track a file with the same extension:

cp ~/tmp/debian.iso ./                # copy a large file into the current directory
git add .                             # add the large file to the project
git commit -am "Added Debian iso"     # commit the file meta data
git push origin master                # sync the git repo and large file to the GitLab server

Note: Make sure that .gitattributes is tracked by git. Otherwise Git LFS will not be working properly for people cloning the project.

 git add .gitattributes

Cloning the repository works the same as before. Git automatically detects the LFS-tracked files and clones them via HTTP. If you performed the git clone command with a SSH URL, you have to enter your GitLab credentials for HTTP authentication.

git clone git@gitlab.example.com:group/project.git

If you already cloned the repository and you want to get the latest LFS object that are on the remote repository, eg. from branch master:

git lfs fetch master

Troubleshooting

error: Repository or object not found

There are a couple of reasons why this error can occur:

  • You don't have permissions to access certain LFS object

Check if you have permissions to push to the project or fetch from the project.

  • Project is not allowed to access the LFS object

LFS object you are trying to push to the project or fetch from the project is not available to the project anymore. Probably the object was removed from the server.

  • Local git repository is using deprecated LFS API

Invalid status for <url> : 501

Git LFS will log the failures into a log file. To view this log file, while in project directory:

git lfs logs last

If the status error 501 is shown, it is because:

  • Git LFS is not enabled in project settings. Check your project settings and enable Git LFS.

  • Git LFS support is not enabled on the GitLab server. Check with your GitLab administrator why Git LFS is not enabled on the server. See LFS administration documentation for instructions on how to enable LFS support.

  • Git LFS client version is not supported by GitLab server. Check your Git LFS version with git lfs version. Check the Git config of the project for traces of deprecated API with git lfs -l. If batch = false is set in the config, remove the line and try to update your Git LFS client. Only version 1.0.1 and newer are supported.

getsockopt: connection refused

If you push a LFS object to a project and you receive an error similar to: Post <URL>/info/lfs/objects/batch: dial tcp IP: getsockopt: connection refused, the LFS client is trying to reach GitLab through HTTPS. However, your GitLab instance is being served on HTTP.

This behaviour is caused by Git LFS using HTTPS connections by default when a lfsurl is not set in the Git config.

To prevent this from happening, set the lfs url in project Git config:

git config --add lfs.url "http://gitlab.example.com/group/project.git/info/lfs"

Credentials are always required when pushing an object

Note: With 8.12 GitLab added LFS support to SSH. The Git LFS communication still goes over HTTP, but now the SSH client passes the correct credentials to the Git LFS client, so no action is required by the user.

Given that Git LFS uses HTTP Basic Authentication to authenticate the user pushing the LFS object on every push for every object, user HTTPS credentials are required.

By default, Git has support for remembering the credentials for each repository you use. This is described in Git credentials man pages.

For example, you can tell Git to remember the password for a period of time in which you expect to push the objects:

git config --global credential.helper 'cache --timeout=3600'

This will remember the credentials for an hour after which Git operations will require re-authentication.

If you are using OS X you can use osxkeychain to store and encrypt your credentials. For Windows, you can use wincred or Microsoft's Git Credential Manager for Windows.

More details about various methods of storing the user credentials can be found on Git Credential Storage documentation.