Geo: A rake task to perform post-omnibus Geo update steps in Zero downtime deployments
Problem to solve
When updating Geo in a zero-downtime, a systems administrator has to perform several manual steps per node. This is error prone and we aim to reduce the number of manual steps required to update Geo.
Intended users
Further details
This feature is part of our strategy to improve the user experience of Geo by reducing manual interventions.
Proposal
When updating a Geo deployment using the Zero Downtime instructions a systems administrator needs to manually perform 8 steps (after omnibus-gitlab!3562 (merged) is merged, 4 + 1 steps).
We aim to reduce these steps for both the primary and secondary nodes to a single rake task e.g. gitlab-rake gitlab:geo:update
In omnibus-gitlab#4637 (closed) we've realized we cannot create a single task for this, but we need a before and after. This is the current proposal:
| Before | After |
|---|---|
SKIP_POST_DEPLOYMENT_MIGRATIONS=true sudo gitlab-ctl reconfigure |
|
sudo gitlab-ctl hup unicorn (optional) |
gitlab-rake gitlab:geo:update |
sudo gitlab-ctl hup sidekiq (optional) |
|
| update GitLab on other machines in cluster | update GitLab on other machines in cluster |
sudo gitlab-rake db:migrate |
gitlab-rake gitlab:geo:post-update |
sudo gitlab-rake gitlab:geo:check |
Documentation
N/A
Testing
What does success look like, and how can we measure that?
- The number of manual interventions is reduced by 50%.
- Geo update procedure is the same on primary and secondaries.
- A single rake task for Geo updates
What is the type of buyer?
- Premium
- Ultimate