I am trying and failing to install GitLab (EE) onto an on-premises bare-metal server PC running Ubuntu Server 20.04. The server is intended to later have internet-facing access, but for now it needs to be internal LAN only, accessed by IP or hostname if possible. The PC is currently set up with a static internal IP via netplan. We do not have a domain name reserved for this server or an SSL certificate at the moment.
I have a working Git server set up on the machine with successful access via SSH and via HTTP (Apache2, then switched to Nginx). My goal is to set up GitLab on the machine running in parallel with the existing Git server.
After getting Nginx running correctly yesterday, I attempted to install GitLab through Omnibus, using the procedure described at the official GitLab install page for Ubuntu, first the curl adding the repository, then running EXTERNAL_URL="http://10.7.2.xxx" apt install gitlab-ee (internal IP, but censored just in case). The process finished without errors, but when attempting to connect to the server with that URL, I only got a 403 from Nginx. I suspect this was due to Nginx already running previously for plain Git on /srv/www/git. I attempted to remove it with apt remove gitlab-ee, followed by deleting the the remaining directories at /etc/gitlab, /var/opt/gitlab, /opt/gitlab, and /var/log/gitlab. At the time I did not know that was the wrong way to uninstall GitLab.
This morning I attempted a reinstall. First I stopped and disabling Nginx with systemctl, backed up my config by moving it out of /etc/nginx/enabled-sites into my home directory, then uninstalling it with apt remove nginx. After a reboot and confirming neither it nor Apache2 were running, I attempted to reinstall GitLab with apt install gitlab-ee again. It finished unexpectedly quickly, as it appears its resources were still present on the system. Connecting from a browser failed completely, no connection at all.
I attempted a reinstall again, this time uninstalling GitLab fully with the following commands:
# gitlab-ctl uninstall
# gitlab-ctl cleanse
# gitlab-ctl remove-accounts
# dpkg -P gitlab-ee
Removing accounts hung for 10min with no output so I had to ^C it. After removing it with dpkg I then removed the same four directories as before, though only /opt/gitlab and /var/log/gitlab remained.
After a supposedly clean uninstall, I rebooted and attempted what I thought to be a clean reinstall. It hung on * ruby_block[wait for logrotate service socket] action run endlessly, for over an hour. Research pointed to opening up another terminal and running the following commands while omnibus was hung:
# nohup /opt/gitlab/embedded/bin/runsvdir-start &
# gitlab-ctl reconfigure
This didn't work, gitlab-ctl reconfigure hung for 10 mins before I gave up, and Omnibus on the other terminal didn't continue. It wouldn't even respond to ^C. I had to force a reboot from another session, and attempted a "clean" uninstall, reboot, and reinstall again.
Uninstall this time did not have any problems with remove-accounts (Pastebin log, username, hostname, IP censored). However, once again it hung at logrotate, for over an hour again, and again not responding to ^C. Pastebin output, Pastebin log from /var/log/gitlab/reconfigure/1636492507.log.
What has not solved my problem:
systemctl start logrotate.service (immediately stops/finishes, awaiting the timer), logrotate.timer (was already running). No response from Omnibus when starting either.- Running
runsvdir-start and gitlab-ctl reconfigure (see above)
- Running
gitlab-ctl tail (throws tail: cannot follow '-' by name)
multi-user.target is loaded and active, according to systemctl -t target.
All the other results I can find are related to specific aspects of GitLab failing, rather than the initial install. I'm stumped and I can't find anything else relevant. What do I do?
