Merge branch 'mk-add-old-attachments-to-uploads-table' into 'master'

Add old files to uploads table

See merge request gitlab-org/gitlab-ce!15270

4 files changed, 447 insertions, 2 deletions

diff --git a/lib/gitlab/background_migration/populate_untracked_uploads.rb b/lib/gitlab/background_migration/populate_untracked_uploads.rb
new file mode 100644
index 00000000000..81e95e5832d
--- /dev/null
+++ b/lib/gitlab/background_migration/populate_untracked_uploads.rb
@@ -0,0 +1,259 @@
+# frozen_string_literal: true
+
+module Gitlab
+  module BackgroundMigration
+    # This class processes a batch of rows in `untracked_files_for_uploads` by
+    # adding each file to the `uploads` table if it does not exist.
+    class PopulateUntrackedUploads # rubocop:disable Metrics/ClassLength
+      # This class is responsible for producing the attributes necessary to
+      # track an uploaded file in the `uploads` table.
+      class UntrackedFile < ActiveRecord::Base # rubocop:disable Metrics/ClassLength, Metrics/LineLength
+        self.table_name = 'untracked_files_for_uploads'
+
+        # Ends with /:random_hex/:filename
+        FILE_UPLOADER_PATH = %r{/\h+/[^/]+\z}
+        FULL_PATH_CAPTURE = %r{\A(.+)#{FILE_UPLOADER_PATH}}
+
+        # These regex patterns are tested against a relative path, relative to
+        # the upload directory.
+        # For convenience, if there exists a capture group in the pattern, then
+        # it indicates the model_id.
+        PATH_PATTERNS = [
+          {
+            pattern: %r{\A-/system/appearance/logo/(\d+)/},
+            uploader: 'AttachmentUploader',
+            model_type: 'Appearance'
+          },
+          {
+            pattern: %r{\A-/system/appearance/header_logo/(\d+)/},
+            uploader: 'AttachmentUploader',
+            model_type: 'Appearance'
+          },
+          {
+            pattern: %r{\A-/system/note/attachment/(\d+)/},
+            uploader: 'AttachmentUploader',
+            model_type: 'Note'
+          },
+          {
+            pattern: %r{\A-/system/user/avatar/(\d+)/},
+            uploader: 'AvatarUploader',
+            model_type: 'User'
+          },
+          {
+            pattern: %r{\A-/system/group/avatar/(\d+)/},
+            uploader: 'AvatarUploader',
+            model_type: 'Namespace'
+          },
+          {
+            pattern: %r{\A-/system/project/avatar/(\d+)/},
+            uploader: 'AvatarUploader',
+            model_type: 'Project'
+          },
+          {
+            pattern: FILE_UPLOADER_PATH,
+            uploader: 'FileUploader',
+            model_type: 'Project'
+          }
+        ].freeze
+
+        def to_h
+          @upload_hash ||= {
+            path: upload_path,
+            uploader: uploader,
+            model_type: model_type,
+            model_id: model_id,
+            size: file_size,
+            checksum: checksum
+          }
+        end
+
+        def upload_path
+          # UntrackedFile#path is absolute, but Upload#path depends on uploader
+          @upload_path ||=
+            if uploader == 'FileUploader'
+              # Path relative to project directory in uploads
+              matchd = path_relative_to_upload_dir.match(FILE_UPLOADER_PATH)
+              matchd[0].sub(%r{\A/}, '') # remove leading slash
+            else
+              path
+            end
+        end
+
+        def uploader
+          matching_pattern_map[:uploader]
+        end
+
+        def model_type
+          matching_pattern_map[:model_type]
+        end
+
+        def model_id
+          return @model_id if defined?(@model_id)
+
+          pattern = matching_pattern_map[:pattern]
+          matchd = path_relative_to_upload_dir.match(pattern)
+
+          # If something is captured (matchd[1] is not nil), it is a model_id
+          # Only the FileUploader pattern will not match an ID
+          @model_id = matchd[1] ? matchd[1].to_i : file_uploader_model_id
+        end
+
+        def file_size
+          File.size(absolute_path)
+        end
+
+        def checksum
+          Digest::SHA256.file(absolute_path).hexdigest
+        end
+
+        private
+
+        def matching_pattern_map
+          @matching_pattern_map ||= PATH_PATTERNS.find do |path_pattern_map|
+            path_relative_to_upload_dir.match(path_pattern_map[:pattern])
+          end
+
+          unless @matching_pattern_map
+            raise "Unknown upload path pattern \"#{path}\""
+          end
+
+          @matching_pattern_map
+        end
+
+        def file_uploader_model_id
+          matchd = path_relative_to_upload_dir.match(FULL_PATH_CAPTURE)
+          not_found_msg = <<~MSG
+            Could not capture project full_path from a FileUploader path:
+              "#{path_relative_to_upload_dir}"
+          MSG
+          raise not_found_msg unless matchd
+
+          full_path = matchd[1]
+          project = Project.find_by_full_path(full_path)
+          return nil unless project
+
+          project.id
+        end
+
+        # Not including a leading slash
+        def path_relative_to_upload_dir
+          upload_dir = Gitlab::BackgroundMigration::PrepareUntrackedUploads::RELATIVE_UPLOAD_DIR # rubocop:disable Metrics/LineLength
+          base = %r{\A#{Regexp.escape(upload_dir)}/}
+          @path_relative_to_upload_dir ||= path.sub(base, '')
+        end
+
+        def absolute_path
+          File.join(CarrierWave.root, path)
+        end
+      end
+
+      # This class is used to query the `uploads` table.
+      class Upload < ActiveRecord::Base
+        self.table_name = 'uploads'
+      end
+
+      def perform(start_id, end_id)
+        return unless migrate?
+
+        files = UntrackedFile.where(id: start_id..end_id)
+        processed_files = insert_uploads_if_needed(files)
+        processed_files.delete_all
+
+        drop_temp_table_if_finished
+      end
+
+      private
+
+      def migrate?
+        UntrackedFile.table_exists? && Upload.table_exists?
+      end
+
+      def insert_uploads_if_needed(files)
+        filtered_files, error_files = filter_error_files(files)
+        filtered_files = filter_existing_uploads(filtered_files)
+        filtered_files = filter_deleted_models(filtered_files)
+        insert(filtered_files)
+
+        processed_files = files.where.not(id: error_files.map(&:id))
+        processed_files
+      end
+
+      def filter_error_files(files)
+        files.partition do |file|
+          begin
+            file.to_h
+            true
+          rescue => e
+            msg = <<~MSG
+              Error parsing path "#{file.path}":
+                #{e.message}
+                #{e.backtrace.join("\n  ")}
+            MSG
+            Rails.logger.error(msg)
+            false
+          end
+        end
+      end
+
+      def filter_existing_uploads(files)
+        paths = files.map(&:upload_path)
+        existing_paths = Upload.where(path: paths).pluck(:path).to_set
+
+        files.reject do |file|
+          existing_paths.include?(file.upload_path)
+        end
+      end
+
+      # There are files on disk that are not in the uploads table because their
+      # model was deleted, and we don't delete the files on disk.
+      def filter_deleted_models(files)
+        ids = deleted_model_ids(files)
+
+        files.reject do |file|
+          ids[file.model_type].include?(file.model_id)
+        end
+      end
+
+      def deleted_model_ids(files)
+        ids = {
+          'Appearance' => [],
+          'Namespace' => [],
+          'Note' => [],
+          'Project' => [],
+          'User' => []
+        }
+
+        # group model IDs by model type
+        files.each do |file|
+          ids[file.model_type] << file.model_id
+        end
+
+        ids.each do |model_type, model_ids|
+          model_class = Object.const_get(model_type)
+          found_ids = model_class.where(id: model_ids.uniq).pluck(:id)
+          deleted_ids = ids[model_type] - found_ids
+          ids[model_type] = deleted_ids
+        end
+
+        ids
+      end
+
+      def insert(files)
+        rows = files.map do |file|
+          file.to_h.merge(created_at: 'NOW()')
+        end
+
+        Gitlab::Database.bulk_insert('uploads',
+                                     rows,
+                                     disable_quote: :created_at)
+      end
+
+      def drop_temp_table_if_finished
+        if UntrackedFile.all.empty?
+          UntrackedFile.connection.drop_table(:untracked_files_for_uploads,
+                                              if_exists: true)
+        end
+      end
+    end
+  end
+end
diff --git a/lib/gitlab/background_migration/prepare_untracked_uploads.rb b/lib/gitlab/background_migration/prepare_untracked_uploads.rb
new file mode 100644
index 00000000000..476c46341ae
--- /dev/null
+++ b/lib/gitlab/background_migration/prepare_untracked_uploads.rb
@@ -0,0 +1,163 @@
+# frozen_string_literal: true
+
+module Gitlab
+  module BackgroundMigration
+    # This class finds all non-hashed uploaded file paths and saves them to a
+    # `untracked_files_for_uploads` table.
+    class PrepareUntrackedUploads # rubocop:disable Metrics/ClassLength
+      # For bulk_queue_background_migration_jobs_by_range
+      include Database::MigrationHelpers
+
+      FIND_BATCH_SIZE = 500
+      RELATIVE_UPLOAD_DIR = "uploads".freeze
+      ABSOLUTE_UPLOAD_DIR = "#{CarrierWave.root}/#{RELATIVE_UPLOAD_DIR}".freeze
+      FOLLOW_UP_MIGRATION = 'PopulateUntrackedUploads'.freeze
+      START_WITH_CARRIERWAVE_ROOT_REGEX = %r{\A#{CarrierWave.root}/}
+      EXCLUDED_HASHED_UPLOADS_PATH = "#{ABSOLUTE_UPLOAD_DIR}/@hashed/*".freeze
+      EXCLUDED_TMP_UPLOADS_PATH = "#{ABSOLUTE_UPLOAD_DIR}/tmp/*".freeze
+
+      # This class is used to iterate over batches of
+      # `untracked_files_for_uploads` rows.
+      class UntrackedFile < ActiveRecord::Base
+        include EachBatch
+
+        self.table_name = 'untracked_files_for_uploads'
+      end
+
+      def perform
+        ensure_temporary_tracking_table_exists
+
+        # Since Postgres < 9.5 does not have ON CONFLICT DO NOTHING, and since
+        # doing inserts-if-not-exists without ON CONFLICT DO NOTHING would be
+        # slow, start with an empty table for Postgres < 9.5.
+        # That way we can do bulk inserts at ~30x the speed of individual
+        # inserts (~20 minutes worth of inserts at GitLab.com scale instead of
+        # ~10 hours).
+        # In all other cases, installations will get both bulk inserts and the
+        # ability for these jobs to retry without having to clear and reinsert.
+        clear_untracked_file_paths unless can_bulk_insert_and_ignore_duplicates?
+
+        store_untracked_file_paths
+
+        schedule_populate_untracked_uploads_jobs
+      end
+
+      private
+
+      def ensure_temporary_tracking_table_exists
+        table_name = :untracked_files_for_uploads
+        unless UntrackedFile.connection.table_exists?(table_name)
+          UntrackedFile.connection.create_table table_name do |t|
+            t.string :path, limit: 600, null: false
+            t.index :path, unique: true
+          end
+        end
+      end
+
+      def clear_untracked_file_paths
+        UntrackedFile.delete_all
+      end
+
+      def store_untracked_file_paths
+        return unless Dir.exist?(ABSOLUTE_UPLOAD_DIR)
+
+        each_file_batch(ABSOLUTE_UPLOAD_DIR, FIND_BATCH_SIZE) do |file_paths|
+          insert_file_paths(file_paths)
+        end
+      end
+
+      def each_file_batch(search_dir, batch_size, &block)
+        cmd = build_find_command(search_dir)
+
+        Open3.popen2(*cmd) do |stdin, stdout, status_thread|
+          yield_paths_in_batches(stdout, batch_size, &block)
+
+          raise "Find command failed" unless status_thread.value.success?
+        end
+      end
+
+      def yield_paths_in_batches(stdout, batch_size, &block)
+        paths = []
+
+        stdout.each_line("\0") do |line|
+          paths << line.chomp("\0").sub(START_WITH_CARRIERWAVE_ROOT_REGEX, '')
+
+          if paths.size >= batch_size
+            yield(paths)
+            paths = []
+          end
+        end
+
+        yield(paths)
+      end
+
+      def build_find_command(search_dir)
+        cmd = %W[find -L #{search_dir}
+                 -type f
+                 ! ( -path #{EXCLUDED_HASHED_UPLOADS_PATH} -prune )
+                 ! ( -path #{EXCLUDED_TMP_UPLOADS_PATH} -prune )
+                 -print0]
+
+        ionice = which_ionice
+        cmd = %W[#{ionice} -c Idle] + cmd if ionice
+
+        log_msg = "PrepareUntrackedUploads find command: \"#{cmd.join(' ')}\""
+        Rails.logger.info log_msg
+
+        cmd
+      end
+
+      def which_ionice
+        Gitlab::Utils.which('ionice')
+      rescue StandardError
+        # In this case, returning false is relatively safe,
+        # even though it isn't very nice
+        false
+      end
+
+      def insert_file_paths(file_paths)
+        sql = insert_sql(file_paths)
+
+        ActiveRecord::Base.connection.execute(sql)
+      end
+
+      def insert_sql(file_paths)
+        if postgresql_pre_9_5?
+          "INSERT INTO #{table_columns_and_values_for_insert(file_paths)};"
+        elsif postgresql?
+          "INSERT INTO #{table_columns_and_values_for_insert(file_paths)}"\
+            " ON CONFLICT DO NOTHING;"
+        else # MySQL
+          "INSERT IGNORE INTO"\
+            " #{table_columns_and_values_for_insert(file_paths)};"
+        end
+      end
+
+      def table_columns_and_values_for_insert(file_paths)
+        values = file_paths.map do |file_path|
+          ActiveRecord::Base.send(:sanitize_sql_array, ['(?)', file_path]) # rubocop:disable GitlabSecurity/PublicSend, Metrics/LineLength
+        end.join(', ')
+
+        "#{UntrackedFile.table_name} (path) VALUES #{values}"
+      end
+
+      def postgresql?
+        @postgresql ||= Gitlab::Database.postgresql?
+      end
+
+      def can_bulk_insert_and_ignore_duplicates?
+        !postgresql_pre_9_5?
+      end
+
+      def postgresql_pre_9_5?
+        @postgresql_pre_9_5 ||= postgresql? &&
+          Gitlab::Database.version.to_f < 9.5
+      end
+
+      def schedule_populate_untracked_uploads_jobs
+        bulk_queue_background_migration_jobs_by_range(
+          UntrackedFile, FOLLOW_UP_MIGRATION)
+      end
+    end
+  end
+end
diff --git a/lib/gitlab/database.rb b/lib/gitlab/database.rb
index 96922e1a62f..e51794fef99 100644
--- a/lib/gitlab/database.rb
+++ b/lib/gitlab/database.rb
@@ -120,15 +120,21 @@ module Gitlab
     #        values.
     # return_ids - When set to true the return value will be an Array of IDs of
     #              the inserted rows, this only works on PostgreSQL.
-    def self.bulk_insert(table, rows, return_ids: false)
+    # disable_quote - A key or an Array of keys to exclude from quoting (You
+    #                 become responsible for protection from SQL injection for
+    #                 these keys!)
+    def self.bulk_insert(table, rows, return_ids: false, disable_quote: [])
       return if rows.empty?
 
       keys = rows.first.keys
       columns = keys.map { |key| connection.quote_column_name(key) }
       return_ids = false if mysql?
 
+      disable_quote = Array(disable_quote).to_set
       tuples = rows.map do |row|
-        row.values_at(*keys).map { |value| connection.quote(value) }
+        keys.map do |k|
+          disable_quote.include?(k) ? row[k] : connection.quote(row[k])
+        end
       end
 
       sql = <<-EOF
diff --git a/lib/gitlab/utils.rb b/lib/gitlab/utils.rb
index abb3d3a02c3..b3baaf036d8 100644
--- a/lib/gitlab/utils.rb
+++ b/lib/gitlab/utils.rb
@@ -46,5 +46,22 @@ module Gitlab
     def random_string
       Random.rand(Float::MAX.to_i).to_s(36)
     end
+
+    # See: http://stackoverflow.com/questions/2108727/which-in-ruby-checking-if-program-exists-in-path-from-ruby
+    # Cross-platform way of finding an executable in the $PATH.
+    #
+    #   which('ruby') #=> /usr/bin/ruby
+    def which(cmd, env = ENV)
+      exts = env['PATHEXT'] ? env['PATHEXT'].split(';') : ['']
+
+      env['PATH'].split(File::PATH_SEPARATOR).each do |path|
+        exts.each do |ext|
+          exe = File.join(path, "#{cmd}#{ext}")
+          return exe if File.executable?(exe) && !File.directory?(exe)
+        end
+      end
+
+      nil
+    end
   end
 end