During GitLab hosts reimages, the partman config fails sometimes leaving the host in a grub recovery shell.

GitLab uses a special partman config because it uses a non-standard disk setup. Two 1TB SSDs are used for / and two 2TB SSDs are used for /srv/gitlab-backup. When rebooting or reimaging, the device names and raid names change between sda and sdd and md0 and md1. Sometimes the raid is created on two non-identical disks (with on 1TB disk and one 2TB disk).

This lead to failed or repeated reimages (after some tries the raids are configured properly).

Also the manual step described in https://wikitech.wikimedia.org/wiki/GitLab#Bootstrap_a_new_GitLab_instance is unreliable because of the changing devices names.

The partman config should be fixed or we should just use a standard partman config with 2x2TB SSDs to store all of the data.