debian-mirror-gitlab/doc/administration/restart_gitlab.md
2020-01-01 13:55:28 +05:30

5.2 KiB

How to restart GitLab

Depending on how you installed GitLab, there are different methods to restart its service(s).

If you want the TL;DR versions, jump to:

Omnibus installations

If you have used the Omnibus packages to install GitLab, then you should already have gitlab-ctl in your PATH.

gitlab-ctl interacts with the Omnibus packages and can be used to restart the GitLab Rails application (Unicorn) as well as the other components, like:

  • GitLab Workhorse
  • Sidekiq
  • PostgreSQL (if you are using the bundled one)
  • NGINX (if you are using the bundled one)
  • Redis (if you are using the bundled one)
  • Mailroom
  • Logrotate

Omnibus GitLab restart

There may be times in the documentation where you will be asked to restart GitLab. In that case, you need to run the following command:

sudo gitlab-ctl restart

The output should be similar to this:

ok: run: gitlab-workhorse: (pid 11291) 1s
ok: run: logrotate: (pid 11299) 0s
ok: run: mailroom: (pid 11306) 0s
ok: run: nginx: (pid 11309) 0s
ok: run: postgresql: (pid 11316) 1s
ok: run: redis: (pid 11325) 0s
ok: run: sidekiq: (pid 11331) 1s
ok: run: unicorn: (pid 11338) 0s

To restart a component separately, you can append its service name to the restart command. For example, to restart only NGINX you would run:

sudo gitlab-ctl restart nginx

To check the status of GitLab services, run:

sudo gitlab-ctl status

Notice that all services say ok: run.

Sometimes, components time out (look for timeout in the logs) during the restart and sometimes they get stuck. In that case, you can use gitlab-ctl kill <service> to send the SIGKILL signal to the service, for example sidekiq. After that, a restart should perform fine.

As a last resort, you can try to reconfigure GitLab instead.

Omnibus GitLab reconfigure

There may be times in the documentation where you will be asked to reconfigure GitLab. Remember that this method applies only for the Omnibus packages.

Reconfigure Omnibus GitLab with:

sudo gitlab-ctl reconfigure

Reconfiguring GitLab should occur in the event that something in its configuration (/etc/gitlab/gitlab.rb) has changed.

When you run this command, Chef, the underlying configuration management application that powers Omnibus GitLab, will make sure that all directories, permissions, services, etc., are in place and in the same shape that they were initially shipped.

It will also restart GitLab components where needed, if any of their configuration files have changed.

If you manually edit any files in /var/opt/gitlab that are managed by Chef, running reconfigure will revert the changes AND restart the services that depend on those files.

Installations from source

If you have followed the official installation guide to install GitLab from source, run the following command to restart GitLab:

sudo service gitlab restart

The output should be similar to this:

Shutting down GitLab Unicorn
Shutting down GitLab Sidekiq
Shutting down GitLab Workhorse
Shutting down GitLab MailRoom
...
GitLab is not running.
Starting GitLab Unicorn
Starting GitLab Sidekiq
Starting GitLab Workhorse
Starting GitLab MailRoom
...
The GitLab Unicorn web server with pid 28059 is running.
The GitLab Sidekiq job dispatcher with pid 28176 is running.
The GitLab Workhorse with pid 28122 is running.
The GitLab MailRoom email processor with pid 28114 is running.
GitLab and all its components are up and running.

This should restart Unicorn, Sidekiq, GitLab Workhorse and Mailroom (if enabled). The init service file that does all the magic can be found on your server in /etc/init.d/gitlab.


If you are using other init systems, like systemd, you can check the GitLab Recipes repository for some unofficial services. These are not officially supported so use them at your own risk.

Helm chart installations

There is no single command to restart the entire GitLab application installed via the cloud native Helm Chart. Usually, it should be enough to restart a specific component separately (gitaly, unicorn, workhorse, gitlab-shell, etc.) by deleting all the pods related to it:

kubectl delete pods -l release=<helm release name>,app=<component name>

The release name can be obtained from the output of the helm list command.