Document Multi-node DB migration procedure for Container Registry
Context
From a large self-managed user:
Multi-node DB migration procedure: We're using a multi-node setup based on omnibus-gitlab for our frontend and registry nodes, following the zero downtime upgrade guide for new versions. Since the registry DB migrations have to be run manually, we do so during the update of the primary deployment node. This means that all migrations are applied while the secondary nodes still use an older version of the registry. Is this the intended flow or should the initial DB migration be run with --skip-post-deployment
(see https://gitlab.com/gitlab-org/container-registry/-/blob/master/docs/database-migrations.md?ref_type=heads#apply-up-migrations) and only after all nodes have been upgraded, applying the DB migration again without the flag?
Problem
We should clarify the documentation on how to run container registry DB migration on multi-node setups. All normal migrations need to be ran before any registry nodes can come online. In charts, the registry instances wait until migrations are complete before they attempt to come online. For additional context this is the MR for charts accomplishing this: Registry: Add wait-for-deps and basic db-migrate (gitlab-org/build/CNG!603 - merged)