Introduces `FastDestroyAll` module

3 files changed, 85 insertions, 32 deletions

diff --git a/app/models/ci/build_trace_chunk.rb b/app/models/ci/build_trace_chunk.rb
index 5dc43c6e00f..3f4a901b626 100644
--- a/app/models/ci/build_trace_chunk.rb
+++ b/app/models/ci/build_trace_chunk.rb
@@ -1,10 +1,12 @@
 module Ci
   class BuildTraceChunk < ActiveRecord::Base
+    include FastDestroyAll
     extend Gitlab::Ci::Model
 
     belongs_to :build, class_name: "Ci::Build", foreign_key: :build_id
 
     default_value_for :data_store, :redis
+    fast_destroy_all_with :delete_all_redis_data, :redis_all_data_keys
 
     WriteError = Class.new(StandardError)
 
@@ -19,27 +21,23 @@ module Ci
       db: 2
     }
 
-    def self.delayed_cleanup_blk
-      ids = all.redis.pluck(:build_id, :chunk_index).map do |data|
-        "gitlab:ci:trace:#{data.first}:chunks:#{data.second}"
+    class << self
+      def redis_data_key(build_id, chunk_index)
+        "gitlab:ci:trace:#{build_id}:chunks:#{chunk_index}"
       end
 
-      puts "before cleanup: #{ids.count}"
-
-      Proc.new do
-        puts "after cleanup: #{ids.count}"
-        Gitlab::Redis::SharedState.with do |redis|
-          redis.del(ids)
-        end unless ids.empty?
-
-        true
+      def redis_all_data_keys
+        redis.pluck(:build_id, :chunk_index).map do |data|
+          redis_data_key(data.first, data.second)
+        end
       end
-    end
 
-    def self.fast_destroy_all
-      delayed_cleanup_blk.tap do |cleanup|
-        delete_all
-        cleanup.call
+      def delete_all_redis_data(redis_keys)
+        if redis_keys.any?
+          Gitlab::Redis::SharedState.with do |redis|
+            redis.del(redis_keys)
+          end
+        end
       end
     end
 
@@ -130,26 +128,22 @@ module Ci
 
     def redis_data
       Gitlab::Redis::SharedState.with do |redis|
-        redis.get(redis_data_key)
+        redis.get(self.class.redis_data_key(build_id, chunk_index))
       end
     end
 
     def redis_set_data(data)
       Gitlab::Redis::SharedState.with do |redis|
-        redis.set(redis_data_key, data, ex: CHUNK_REDIS_TTL)
+        redis.set(self.class.redis_data_key(build_id, chunk_index), data, ex: CHUNK_REDIS_TTL)
       end
     end
 
     def redis_delete_data
       Gitlab::Redis::SharedState.with do |redis|
-        redis.del(redis_data_key)
+        redis.del(self.class.redis_data_key(build_id, chunk_index))
       end
     end
 
-    def redis_data_key
-      "gitlab:ci:trace:#{build_id}:chunks:#{chunk_index}"
-    end
-
     def redis_lock_key
       "trace_write:#{build_id}:chunks:#{chunk_index}"
     end
diff --git a/app/models/concerns/fast_destroy_all.rb b/app/models/concerns/fast_destroy_all.rb
new file mode 100644
index 00000000000..28ab541dd13
--- /dev/null
+++ b/app/models/concerns/fast_destroy_all.rb
@@ -0,0 +1,61 @@
+##
+# This module is for replacing `dependent: :destroy` and `before_destroy` hooks.
+#
+# In general, `destroy_all` is inefficient because it calls each callback with `DELETE` queries i.e. O(n), whereas,
+# `delete_all` is efficient as it deletes all rows with a single `DELETE` query.
+#
+# It's better to use `delete_all` as much as possible, however, in the real cases, it's hard to adopt because  
+# in some cases, external data (in ObjectStorage/FileStorage/Redis) is assosiated with rows.
+#
+# This module introduces a protocol to support the adoption with easy way.
+# You can use in the following scnenes
+# - When calling `destroy_all` explicitly
+#   e.g. `job.trace_chunks.fast_destroy_all``
+# - When a parent record is deleted and all children are deleted subsequently (cascade delete)
+#   e.g. `before_destroy -> { run_after_commit(&build_trace_chunks.delete_all_external_data_proc) }``
+module FastDestroyAll
+  extend ActiveSupport::Concern
+
+  included do
+    class_attribute :_delete_method, :_delete_params_generator
+  end
+
+  class_methods do
+    ##
+    # This method is for registering :delete_method and :delete_params_generator
+    # You have to define the method if you want to use `FastDestroyAll`` module.
+    #
+    # e.g. fast_destroy_all_with :delete_all_redis_data, :redis_all_data_keys
+    def fast_destroy_all_with(delete_method, delete_params_generator)
+      self._delete_method = delete_method
+      self._delete_params_generator = delete_params_generator
+    end
+
+    ##
+    # This method generates a proc to delete external data.
+    # It's useful especially when you want to hook parent record's deletion event.
+    #
+    # e.g. before_destroy -> { run_after_commit(&build_trace_chunks.delete_all_external_data_proc) }
+    def delete_all_external_data_proc
+      params = send self._delete_params_generator
+      callee = self # Preserve the callee class, otherwise `proc` uses a different class
+
+      proc do
+        callee.send callee._delete_method, params
+
+        true
+      end
+    end
+
+    ##
+    # This method deletes all rows at first and delete all external data at second.
+    # Before deleting the rows, it generates a proc to delete external data.
+    # So it won't lose the track of deleting external data, even if it happened after rows had been deleted.
+    def fast_destroy_all
+      delete_all_external_data_proc.tap do |delete_all_external_data|
+        delete_all
+        delete_all_external_data.call
+      end
+    end
+  end
+end
diff --git a/app/models/project.rb b/app/models/project.rb
index 48e30f0fd36..fd46c7b1c70 100644
--- a/app/models/project.rb
+++ b/app/models/project.rb
@@ -78,6 +78,12 @@ class Project < ActiveRecord::Base
   after_update :update_forks_visibility_level
 
   before_destroy :remove_private_deploy_keys
+
+  # This has to be defined before `has_many :builds, depenedent: :destroy`,
+  # otherwise we will not delete any data, due to trace chunks
+  # going through :builds
+  before_destroy -> { run_after_commit(&build_trace_chunks.delete_all_external_data_proc) }
+
   after_destroy -> { run_after_commit { remove_pages } }
   after_destroy :remove_exports
 
@@ -214,14 +220,6 @@ class Project < ActiveRecord::Base
   has_many :commit_statuses
   has_many :pipelines, class_name: 'Ci::Pipeline', inverse_of: :project
 
-  # This has to be defined before `has_many :builds, depenedent: :destroy`,
-  # otherwise we will not delete any data, due to trace chunks
-  # going through :builds
-  before_destroy do
-    puts "destroying all chunks"
-    self.run_after_commit(&build_trace_chunks.delayed_cleanup_blk)
-  end
-
   # Ci::Build objects store data on the file system such as artifact files and
   # build traces. Currently there's no efficient way of removing this data in
   # bulk that doesn't involve loading the rows into memory. As a result we're