Document Rails deploy node / server pattern in scaled out deployments (aka toolbox / task runner role)
Summary
I'm aware of at least two customers that have a dedicated Rails deploy node in their scaled-out environment.
Its purpose is primarily to support upgrades, but it'd also be a good place to run the rails console and run backups. In an environment using PgBouncer, it could then be left configured to by-pass PgBouncer in support of both upgrades and backups (and could do so automatically using a load balancer VIP - see #6204)
This server would be analogous to the Helm toolbox pod.
Question: if Puma and Sidekiq are disabled on an Omnibus installation, does the Rails configuration get maintained?
Ticket reference for GitLab team members: https://gitlab.zendesk.com/agent/tickets/285264
Proposal
- Verify if Omnibus does maintain the Rails configuration.
- Confirm with distribution if this is a supported pattern that will be maintained.
- Document it.