Add GitLab-specific concerns to code review guideenhance-backend-review-guide

1 files changed, 47 insertions, 0 deletions

diff --git a/doc/development/code_review.md b/doc/development/code_review.md
index 4ed89146072..5f732ee7ad6 100644
--- a/doc/development/code_review.md
+++ b/doc/development/code_review.md
@@ -133,6 +133,53 @@ reviewee.
   tomorrow. When you are not able to find the right balance, ask other people
   about their opinion.
 
+### GitLab-specific concerns
+
+GitLab is used in a lot of places. Many users use
+our [Omnibus packages](https://about.gitlab.com/installation/), but some use
+the [Docker images](https://docs.gitlab.com/omnibus/docker/), some are
+[installed from source](https://docs.gitlab.com/ce/install/installation.html),
+and there are other installation methods available. GitLab.com itself is a large
+Enterprise Edition instance. This has some implications:
+
+1. **Query changes** should be tested to ensure that they don't result in worse
+   performance at the scale of GitLab.com:
+  1. Generating large quantities of data locally can help.
+  2. Asking for query plans from GitLab.com is the most reliable way to validate
+     these.
+2. **Database migrations** must be:
+  1. Reversible.
+  2. Performant at the scale of GitLab.com - ask a maintainer to test the
+     migration on the staging environment if you aren't sure.
+  3. Categorised correctly:
+     - Regular migrations run before the new code is running on the instance.
+     - [Post-deployment migrations](post_deployment_migrations.md) run _after_
+       the new code is deployed, when the instance is configured to do that.
+     - [Background migrations](background_migrations.md) run in Sidekiq, and
+       should only be done for migrations that would take an extreme amount of
+       time at GitLab.com scale.
+3. **Sidekiq workers** cannot change in a backwards-incompatible way:
+  1. Sidekiq queues are not drained before a deploy happens, so there will be
+     workers in the queue from the previous version of GitLab.
+  2. If you need to change a method signature, try to do so across two releases,
+     and accept both the old and new arguments in the first of those.
+  3. Similarly, if you need to remove a worker, stop it from being scheduled in
+     one release, then remove it in the next. This will allow existing jobs to
+     execute.
+  4. Don't forget, not every instance will upgrade to every version, so try to
+     be liberal in accepting the old format if it is cheap to do so.
+4. **Cached values** may persist across releases. If you are changing the type a
+   cached value returns (say, from a string or nil to an array), change the
+   cache key at the same time.
+5. **Settings** should be added as a
+   [last resort](https://about.gitlab.com/handbook/product/#convention-over-configuration).
+   If you're adding a new setting in `gitlab.yml`:
+  1. Try to avoid that, and add to `ApplicationSetting` instead.
+  2. Ensure that it is also
+     [added to Omnibus](https://docs.gitlab.com/omnibus/settings/gitlab.yml.html#adding-a-new-setting-to-gitlab-yml).
+6. **Filesystem access** can be slow, so try to avoid
+   [shared files](shared_files.md) when an alternative solution is available.
+
 ### Credits
 
 Largely based on the [thoughtbot code review guide].