Race condition at startup
In the changes that have been done for public images, we don't perform the hg update
for gitlab-rails
at startup anymore.
It is done instead directly in the Docker image, making the startup sequence much faster.
As a consequence, the gitlab-ctl restart
that the heptapod wrapper performs often occurs too early, when the service manager (runsv) is not ready yet. Upon install of a new image, the startup sequence is usually still slow enough that at least a service has a successful restart, so that the whole command does not fail. But on a simple restart of a running container (fastest case), all services fail and the container stops.
I do have a fix, need to validate it more thoroughly