Troubleshooting Git

Sometimes things don’t work the way they should or as you might expect when you’re using Git. Here are some tips on troubleshooting and resolving issues with Git.

Broken pipe errors on git push

‘Broken pipe’ errors can occur when attempting to push to a remote repository. When pushing you usually see:

Write failed: Broken pipe
fatal: The remote end hung up unexpectedly

To fix this issue, here are some possible solutions.

Increase the POST buffer size in Git

If you’re using Git over HTTP instead of SSH, you can try increasing the POST buffer size in Git’s configuration.

Example of an error during a clone: fatal: pack has bad object at offset XXXXXXXXX: inflate returned -5

Open a terminal and enter:

git config http.postBuffer 52428800

The value is specified in bytes, so in the above case the buffer size has been set to 50MB. The default is 1MB.

Check your SSH configuration

If pushing over SSH, first check your SSH configuration as ‘Broken pipe’ errors can sometimes be caused by underlying issues with SSH (such as authentication). Make sure that SSH is correctly configured by following the instructions in the SSH troubleshooting documentation.

If you’re a GitLab administrator with server access, you can also prevent session timeouts by configuring SSH keep-alive on the client or the server.

noteConfiguring both the client and the server is unnecessary.

To configure SSH on the client side:

  • On UNIX, edit ~/.ssh/config (create the file if it doesn’t exist) and add or edit:

    Host your-gitlab-instance-url.com
      ServerAliveInterval 60
      ServerAliveCountMax 5
    
  • On Windows, if you are using PuTTY, go to your session properties, then navigate to “Connection” and under “Sending of null packets to keep session active”, set Seconds between keepalives (0 to turn off) to 60.

To configure SSH on the server side, edit /etc/ssh/sshd_config and add:

ClientAliveInterval 60
ClientAliveCountMax 5

Running a git repack

If ‘pack-objects’ type errors are also being displayed, you can try to run a git repack before attempting to push to the remote repository again:

git repack
git push

Upgrade your Git client

In case you’re running an older version of Git (< 2.9), consider upgrading to >= 2.9 (see Broken pipe when pushing to Git repository).

ssh_exchange_identification error

Users may experience the following error when attempting to push or pull using Git over SSH:

Please make sure you have the correct access rights
and the repository exists.
...
ssh_exchange_identification: read: Connection reset by peer
fatal: Could not read from remote repository.

or

ssh_exchange_identification: Connection closed by remote host
fatal: The remote end hung up unexpectedly

This error usually indicates that SSH daemon’s MaxStartups value is throttling SSH connections. This setting specifies the maximum number of concurrent, unauthenticated connections to the SSH daemon. This affects users with proper authentication credentials (SSH keys) because every connection is ‘unauthenticated’ in the beginning. The default value is 10.

Increase MaxStartups on the GitLab server by adding or modifying the value in /etc/ssh/sshd_config:

MaxStartups 100:30:200

100:30:200 means up to 100 SSH sessions are allowed without restriction, after which 30% of connections are dropped until reaching an absolute maximum of 200.

Once configured, restart the SSH daemon for the change to take effect.

# Debian/Ubuntu
sudo systemctl restart ssh

# CentOS/RHEL
sudo service sshd restart

Timeout during git push / git pull

If pulling/pushing from/to your repository ends up taking more than 50 seconds, a timeout is issued. It contains a log of the number of operations performed and their respective timings, like the example below:

remote: Running checks for branch: master
remote: Scanning for LFS objects... (153ms)
remote: Calculating new repository size... (cancelled after 729ms)

This could be used to further investigate what operation is performing poorly and provide GitLab with more information on how to improve the service.

git clone over HTTP fails with transfer closed with outstanding read data remaining error

Sometimes, when cloning old or large repositories, the following error is thrown:

error: RPC failed; curl 18 transfer closed with outstanding read data remaining
fatal: The remote end hung up unexpectedly
fatal: early EOF
fatal: index-pack failed

This is a common problem with Git itself, due to its inability to handle large files or large quantities of files. Git LFS was created to work around this problem; however, even it has limitations. It’s usually due to one of these reasons:

  • The number of files in the repository.
  • The number of revisions in the history.
  • The existence of large files in the repository.

The root causes vary, so multiple potential solutions exist, and you may need to apply more than one:

  • If this error occurs when cloning a large repository, you can decrease the cloning depth to a value of 1. For example:

    variables:
      GIT_DEPTH: 1
    
  • You can increase the http.postBuffer value in your local Git configuration from the default 1 MB value to a value greater than the repository size. For example, if git clone fails when cloning a 500 MB repository, you should set http.postBuffer to 524288000:

    # Set the http.postBuffer size, in bytes
    git config http.postBuffer 524288000
    
  • You can increase the http.postBuffer on the server side:

    1. Modify the GitLab instance’s gitlab.rb file:

      omnibus_gitconfig['system'] = {
        # Set the http.postBuffer size, in bytes
        "http" => ["postBuffer" => 524288000]
      }
      
    2. After applying this change, apply the configuration change:

      sudo gitlab-ctl reconfigure
      

For example, if a repository has a very long history and no large files, changing the depth should fix the problem. However, if a repository has very large files, even a depth of 1 may be too large, thus requiring the postBuffer change. If you increase your local postBuffer but the NGINX value on the backend is still too small, the error persists.

Modifying the server is not always an option, and introduces more potential risk. Attempt local changes first.