GitLab 11 specific changes

TLS v1.1 Deprecation

Beginning with GitLab 12.0, TLS v1.1 will be disabled by default to improve security.

This mitigates numerous issues including, but not limited to, Heartbleed and makes GitLab compliant out of the box with the PCI DSS 3.1 standard. Learn more about why TLS v1.1 is being deprecated in our blog.

Clients supporting TLS v1.2

  • Git-Credential-Manager - support since 1.14.0
  • Git on Red Hat Enterprise Linux 6 - support since 6.8
  • Git on Red Hat Enterprise Linux 7 - support since 7.2
  • JGit / Java - support since JDK 7
  • Visual Studio - support since version 2017

Modify or add these entries to gitlab.rb and run gitlab-ctl reconfigure to disable TLS v1.1 immediately: 

nginx['ssl_protocols'] = "TLSv1.2"

Upgrade prerequisites

For successfully upgrading to GitLab 11.0, users need to satisfy following requirements:

  1. Users should be running latest version in the 10.x series. At the time of writing this documentation, it is GitLab 10.8.7.

  2. The configurations that were deprecated (list below) in the 10.x series have been now removed. Users needs to remove them from /etc/gitlab/gitlab.rb. Then run gitlab-ctl reconfigure to apply the configuration changes.

If either of the above requirements are not satisfied, upgrade process will abort without making changes to user’s existing installation. This is to ensure that users does not end up with a broken GitLab due to these unsupported configurations.

Removed configurations

The following configurations were deprecated in the 10.x series and have now been removed:

  1. Mattermost related configurations - Support for most of the Mattermost related configuration have been removed, except for the essential ones that are needed for GitLab-Mattermost integration. Check out the official documentation for details

  2. Legacy git_data_dir configuration, which was used to set location of where data was to be stored. It has been now replaced with git_data_dirs configuration. Check out the official documentation for details

  3. Old format of git_data_dirs configuration has been replaced with a new format, allowing much more fine grain control. Check out the official documentation for details

Changes introduced in minor versions

11.2

Rack Attack is disabled by default. To continue using Rack Attack, you must enable it manually.

11.3

  1. A security patch removed the ability to get files from a repository by passing a private_token URL parameter. Instead, a redirect to /users/sign_in is now served. Update any CI scripts, custom automations, etc. to use the repository files API.

11.4

  1. Version of bundled Redis has been upgraded to 3.2.12. This is a critical security update that fixes multiple vulnerabilities. After upgrading to 11.4, run gitlab-ctl restart redis to ensure the new version is loaded.

  2. The bundled version of Prometheus has been upgraded to 2.4.2 and fresh installations will use it by default. Version 2 of Prometheus uses a data format incompatible with version 1.

    For users looking for preserving the Prometheus version 1 data, a command line tool is provided to upgrade their Prometheus service and migrate data to the format supported by new Prometheus version. This tool can be invoked using the following command: 

    sudo gitlab-ctl prometheus-upgrade

    This tool will convert existing data to a format supported by the latest Prometheus version. Depending on the volume of data, this process can take hours. If users do not want to migrate the data, but start with a clean database, they can pass --skip-data-migration flag to the above command.

    notePrometheus service will be stopped during the migration process.

    To know about other supported options, pass --help flag to the above command.

    This tool will not be automatically invoked during package upgrades. Users will have to run it manually to migrate to latest version of Prometheus, and are advised to do it as soon as possible. Therefore, existing users who are upgrading to 11.4 will continue to use Prometheus 1.x until they manually migrate to the 2.x version.

    Support for Prometheus 1.x versions that were shipped with earlier versions of GitLab has been deprecated and will be removed completely in GitLab 12.0. Users still using those versions will be presented with a deprecation warning during reconfigure. With GitLab 12.0 Prometheus will be upgraded to 2.x automatically, Prometheus 1.0 data will not be migrated.

  3. A security patch removed the ability to get files from a repository by passing a private_token URL parameter. Instead, a redirect to /users/sign_in is now served. Update any CI scripts, custom automations, etc. to use the repository files API.

11.5

  1. A security patch removed the ability to get files from a repository by passing a private_token URL parameter. Instead, a redirect to /users/sign_in is now served. Update any CI scripts, custom automations, etc. to use the repository files API.

11.6

  1. Sidekiq probe of GitLab Monitor will be disabled by default if GitLab is configured in Redis for scaling. To manually enable it, users can set gitlab_monitor['probe_sidekiq'] = true in /etc/gitlab/gitlab.rb file. However, when manually enabling it in Redis HA mode, users are expected to point the probe to a Redis instance connected to the instance using the gitlab_rails['redis_*'] settings.

    A valid example configuration is: 

    gitlab_monitor['probe_sidekiq'] = true
gitlab_rails['redis_host'] = <IP of Redis master node>
gitlab_rails['redis_port'] = <Port where Redis runs in master node>
gitlab_rails['redis_password'] = <Password to connect to Redis master>
    noteIn the above configuration, when a failover happens after the master node fails, GitLab Monitor will still be probing the original master node, since it is specified in gitlab.rb. Users will have to manually update gitlab.rb to point it to the new master node.

  2. Ruby has been updated to 2.5.3. GitLab will be down during the upgrade until the Unicorn processes have been restarted. The restart is done automatically at the end of gitlab-ctl reconfigure, which is run by default on upgrade.

    noteThe application will throw 500 http errors until the Unicorn restart is completed.

11.8

  1. The runit cookbook was updated to be closer to the latest version of the upstream runit cookbook. No user changes are necessary for this release.