Document a new migrations testing technique

spec/migrations/README.md

+# Testing migrations
+
+In order to reliably test a migration, we need to test it against a database
+schema that this migration has been written for. In order to achieve that we
+have some _migration helpers_ and RSpec test tag, called `:migration`.
+
+If you want to write a test for a migration consider adding `:migration` tag to
+the test signature, like `describe SomeMigrationClass, :migration`.
+
+## How does it work?
+
+Adding a `:migration` tag to a test signature injects a few before / after
+hooks to the test.
+
+The most important change is that adding a `:migration` tag adds a `before`
+hook that will revert all migrations to the point that a migration under test
+is not yet migrated.
+
+In other words, our custom RSpec hooks will find a previous migration, and
+migrate the database **down** to the previous migration version.
+
+With this approach you can test a migration against a database schema that this
+migration has been written for.
+
+Use `migrate!` helper to run the migration that is under test.
+
+The `after` hook will migrate the database **up** and reinstitutes the latest
+schema version, so that the process does not affect subsequent specs and
+ensures proper isolation.
+
+## Available helpers
+
+Use `table` helper to create a temporary `ActiveRecord::Base` derived model
+for a table.
+
+Use `migrate!` helper to run the migration that is under test. It will not only
+run migration, but will also bump the schema version in the `schema_migrations`
+table. It is necessary because in the `after` hook we trigger the rest of
+the migrations, and we need to know where to start.
+
+See `spec/support/migrations_helpers.rb` for all the available helpers.
+
+## An example
+
+```ruby
+require 'spec_helper'
+
+# Load a migration class.
+
+require Rails.root.join('db', 'post_migrate', '20170526185842_migrate_pipeline_stages.rb')
+
+describe MigratePipelineStages, :migration do
+
+  # Create test data - pipeline and CI/CD jobs.
+
+  let(:jobs) { table(:ci_builds) }
+  let(:stages) { table(:ci_stages) }
+  let(:pipelines) { table(:ci_pipelines) }
+  let(:projects) { table(:projects) }
+
+  before do
+    projects.create!(id: 123, name: 'gitlab1', path: 'gitlab1')
+    pipelines.create!(id: 1, project_id: 123, ref: 'master', sha: 'adf43c3a')
+    jobs.create!(id: 1, commit_id: 1, project_id: 123, stage_idx: 2, stage: 'build')
+    jobs.create!(id: 2, commit_id: 1, project_id: 123, stage_idx: 1, stage: 'test')
+  end
+
+  # Test the migration.
+
+  it 'correctly migrates pipeline stages' do
+    expect(stages.count).to be_zero
+
+    migrate!
+
+    expect(stages.count).to eq 2
+    expect(stages.all.pluck(:name)).to match_array %w[test build]
+  end
+end
+```
+
+## Best practices
+
+1. Use only one test example per migration unless there is a good reason to
+use more.
+1. Note that this type of tests do not run within the transaction, we use
+a truncation database cleanup strategy. Do not depend on transaction being
+present.