From a08c18eb8154ecc76dd1ad6f9576794cd769ca29 Mon Sep 17 00:00:00 2001
From: Pierre de La Morinerie <kemenaran@gmail.com>
Date: Thu, 25 Mar 2021 16:46:18 +0000
Subject: [PATCH] db: fix ActiveStorage migration to Rails 6.1

1. Executing an `update_all` on a lot of records is slow. Instead we
backfill the column in batches (without locking the table).

2. We can't mark the colum as not-null, because already running
instances still using the old code will not use the service_name
column yet.

  A later migration will backfill the remaining data and make the column
non-null.
---
 ..._to_active_storage_blobs.active_storage.rb | 23 ++++++++++++++-----
 db/schema.rb                                  |  2 +-
 2 files changed, 18 insertions(+), 7 deletions(-)

diff --git a/db/migrate/20210318090000_add_service_name_to_active_storage_blobs.active_storage.rb b/db/migrate/20210318090000_add_service_name_to_active_storage_blobs.active_storage.rb
index 365bafc7f..edf231ebb 100644
--- a/db/migrate/20210318090000_add_service_name_to_active_storage_blobs.active_storage.rb
+++ b/db/migrate/20210318090000_add_service_name_to_active_storage_blobs.active_storage.rb
@@ -1,16 +1,27 @@
 # This migration comes from active_storage (originally 20190112182829)
 class AddServiceNameToActiveStorageBlobs < ActiveRecord::Migration[6.0]
+  disable_ddl_transaction!
+
   def up
     unless column_exists?(:active_storage_blobs, :service_name)
       add_column :active_storage_blobs, :service_name, :string
+    end
 
-      if (configured_service = ActiveStorage::Blob.service.name)
-        # rubocop:disable DS/Unscoped
-        ActiveStorage::Blob.unscoped.update_all(service_name: configured_service)
-        # rubocop:enable DS/Unscoped
+    # rubocop:disable DS/Unscoped
+    blobs_without_service_name = ActiveStorage::Blob.unscoped.where(service_name: nil)
+    # rubocop:enable DS/Unscoped
+
+    if (configured_service = ActiveStorage::Blob.service.name && blobs_without_service_name.count > 0)
+      # Backfill the existing blobs with the service.
+      # NB: during a continuous deployments, some blobs may still be created
+      # with an empty service_name. A later migration will fix those later.
+
+      say_with_time('backfill ActiveStorage::Blob.service.name. This could take a while…') do
+        blobs_without_service_name.in_batches do |relation|
+          relation.update_all service_name: configured_service
+          sleep(0.01) # throttle
+        end
       end
-
-      change_column :active_storage_blobs, :service_name, :string, null: false
     end
   end
 
diff --git a/db/schema.rb b/db/schema.rb
index 750d0157b..b4833672b 100644
--- a/db/schema.rb
+++ b/db/schema.rb
@@ -44,7 +44,7 @@ ActiveRecord::Schema.define(version: 2021_03_18_090001) do
     t.bigint "byte_size", null: false
     t.string "checksum", null: false
     t.datetime "created_at", null: false
-    t.string "service_name", null: false
+    t.string "service_name"
     t.index ["key"], name: "index_active_storage_blobs_on_key", unique: true
   end