chewy 0.8.4 → 7.3.4

data/lib/chewy/index/import/bulk_builder.rb

@@ -0,0 +1,311 @@
+module Chewy
+  class Index
+    module Import
+      # This class purpose is to build ES client-acceptable bulk
+      # request body from the passed objects for index and deletion.
+      # It handles parent-child relationships as well by fetching
+      # existing documents from ES and database, taking their join field values and
+      # using it in the bulk body.
+      # If fields are passed - it creates partial update entries except for
+      # the cases when the type has parent and parent_id has been changed.
+      class BulkBuilder
+        # @param index [Chewy::Index] desired index
+        # @param to_index [Array<Object>] objects to index
+        # @param delete [Array<Object>] objects or ids to delete
+        # @param fields [Array<Symbol, String>] and array of fields for documents update
+        def initialize(index, to_index: [], delete: [], fields: [])
+          @index = index
+          @to_index = to_index
+          @delete = delete
+          @fields = fields.map!(&:to_sym)
+        end
+
+        # Returns ES API-ready bulk requiest body.
+        # @see https://github.com/elastic/elasticsearch-ruby/blob/master/elasticsearch-api/lib/elasticsearch/api/actions/bulk.rb
+        # @return [Array<Hash>] bulk body
+        def bulk_body
+          populate_cache
+
+          @bulk_body ||= @to_index.flat_map(&method(:index_entry)).concat(
+            @delete.flat_map(&method(:delete_entry))
+          ).uniq
+        end
+
+        # The only purpose of this method is to cache document ids for
+        # all the passed object for index to avoid ids recalculation.
+        #
+        # @return [Hash[String => Object]] an ids-objects index hash
+        def index_objects_by_id
+          @index_objects_by_id ||= index_object_ids.invert.stringify_keys!
+        end
+
+      private
+
+        def crutches_for_index
+          @crutches_for_index ||= Chewy::Index::Crutch::Crutches.new @index, @to_index
+        end
+
+        def index_entry(object)
+          entry = {}
+          entry[:_id] = index_object_ids[object] if index_object_ids[object]
+          entry[:routing] = routing(object) if join_field?
+
+          parent = cache(entry[:_id])
+          data = data_for(object) if parent.present?
+          if parent.present? && parent_changed?(data, parent)
+            reindex_entries(object, data) + reindex_descendants(object)
+          elsif @fields.present?
+            return [] unless entry[:_id]
+
+            entry[:data] = {doc: data_for(object, fields: @fields)}
+            [{update: entry}]
+          else
+            entry[:data] = data || data_for(object)
+            [{index: entry}]
+          end
+        end
+
+        def reindex_entries(object, data, root: object)
+          entry = {}
+          entry[:_id] = index_object_ids[object] || entry_id(object)
+          entry[:data] = data
+          entry[:routing] = routing(root) || routing(object) if join_field?
+          delete = delete_single_entry(object, root: root).first
+          index = {index: entry}
+          [delete, index]
+        end
+
+        def reindex_descendants(root)
+          descendants = load_descendants(root)
+          crutches = Chewy::Index::Crutch::Crutches.new @index, [root, *descendants]
+          descendants.flat_map do |object|
+            reindex_entries(
+              object,
+              data_for(object, crutches: crutches),
+              root: root
+            )
+          end
+        end
+
+        def delete_entry(object)
+          delete_single_entry(object)
+        end
+
+        def delete_single_entry(object, root: object)
+          entry = {}
+          entry[:_id] = entry_id(object)
+          entry[:_id] ||= object.as_json
+
+          return [] if entry[:_id].blank?
+
+          if join_field?
+            cached_parent = cache(entry[:_id])
+            entry_parent_id =
+              if cached_parent
+                cached_parent[:parent_id]
+              else
+                find_parent_id(object)
+              end
+
+            entry[:routing] = existing_routing(root.try(:id)) || existing_routing(object.id)
+            entry[:parent] = entry_parent_id if entry_parent_id
+          end
+
+          [{delete: entry}]
+        end
+
+        def load_descendants(root)
+          root_type = join_field_type(root)
+          return [] unless root_type
+
+          descendant_ids = []
+          grouped_parents = {root_type => [root.id]}
+          # iteratively fetch all the descendants (with grouped_parents as a queue for next iteration)
+          until grouped_parents.empty?
+            children_data = grouped_parents.flat_map do |parent_type, parent_ids|
+              @index.query(
+                has_parent: {
+                  parent_type: parent_type,
+                  # ignore_unmapped to avoid error for the leaves of the tree
+                  # (types without children)
+                  ignore_unmapped: true,
+                  query: {ids: {values: parent_ids}}
+                }
+              ).pluck(:_id, join_field).map { |id, join| [join['name'], id] }
+            end
+            descendant_ids |= children_data.map(&:last)
+
+            grouped_parents = {}
+            children_data.each do |name, id|
+              next unless name
+
+              grouped_parents[name] ||= []
+              grouped_parents[name] << id
+            end
+          end
+          # query the primary database to load the descentants' records
+          @index.adapter.load(descendant_ids, _index: @index.base_name, raw_import: @index._default_import_options[:raw_import])
+        end
+
+        def populate_cache
+          @cache = load_cache
+        end
+
+        def cache(id)
+          @cache[id.to_s]
+        end
+
+        def load_cache
+          return {} unless join_field?
+
+          @index
+            .filter(ids: {values: ids_for_cache})
+            .order('_doc')
+            .pluck(:_id, :_routing, join_field)
+            .map do |id, routing, join|
+              [
+                id,
+                {routing: routing, parent_id: join['parent']}
+              ]
+            end.to_h
+        end
+
+        def existing_routing(id)
+          # All objects needed here should be cached in #load_cache,
+          # if not, we return nil. In some cases we don't have existing routing cached,
+          # e.g. for loaded descendants
+          return unless cache(id)
+
+          cache(id)[:routing]
+        end
+
+        # Two types of ids:
+        # * of parents of the objects to be indexed
+        # * of objects to be deleted
+        def ids_for_cache
+          ids = @to_index.flat_map do |object|
+            [find_parent_id(object), object.id] if object.respond_to?(:id)
+          end
+          ids.concat(@delete.map do |object|
+            object.id if object.respond_to?(:id)
+          end)
+          ids.uniq.compact
+        end
+
+        def routing(object)
+          # filter out non-model objects, early return on object==nil
+          return unless object.respond_to?(:id)
+
+          parent_id = find_parent_id(object)
+          if parent_id
+            routing(index_objects_by_id[parent_id.to_s]) || existing_routing(parent_id)
+          else
+            object.id.to_s
+          end
+        end
+
+        def find_parent_id(object)
+          return unless object.respond_to?(:id)
+
+          join = data_for(object, fields: [join_field.to_sym])[join_field]
+          join['parent'] if join
+        end
+
+        def join_field
+          return @join_field if defined?(@join_field)
+
+          @join_field = find_join_field
+        end
+
+        def find_join_field
+          type_settings = @index.mappings_hash[:mappings]
+          return unless type_settings
+
+          properties = type_settings[:properties]
+          join_fields = properties.find { |_, options| options[:type] == :join }
+          return unless join_fields
+
+          join_fields.first.to_s
+        end
+
+        def join_field_type(object)
+          return unless join_field?
+
+          raw_object =
+            if @index._default_import_options[:raw_import]
+              @index._default_import_options[:raw_import].call(object.attributes)
+            else
+              object
+            end
+
+          join_field_value = data_for(
+            raw_object,
+            fields: [join_field.to_sym], # build only the field that is needed
+            crutches: Chewy::Index::Crutch::Crutches.new(@index, [raw_object])
+          )[join_field]
+
+          case join_field_value
+          when String
+            join_field_value
+          when Hash
+            join_field_value['name']
+          end
+        end
+
+        def join_field?
+          join_field && !join_field.empty?
+        end
+
+        def data_for(object, fields: [], crutches: crutches_for_index)
+          @index.compose(object, crutches, fields: fields)
+        end
+
+        def parent_changed?(data, old_parent)
+          return false unless old_parent
+          return false unless join_field?
+          return false unless @fields.include?(join_field.to_sym)
+          return false unless data.key?(join_field)
+
+          # The join field value can be a hash, e.g.:
+          # {"name": "child", "parent": "123"} for a child
+          # {"name": "parent"} for a parent
+          # but it can also be a string: (e.g. "parent") for a parent:
+          # https://www.elastic.co/guide/en/elasticsearch/reference/current/parent-join.html#parent-join
+          new_join_field_value = data[join_field]
+          if new_join_field_value.is_a? Hash
+            # If we have a hash in the join field,
+            # we're taking the `parent` field that holds the parent id.
+            new_parent_id = new_join_field_value['parent']
+            new_parent_id != old_parent[:parent_id]
+          else
+            # If there is a non-hash value (String or nil), it means that the join field is changed
+            # and the current object is no longer a child.
+            true
+          end
+        end
+
+        def entry_id(object)
+          if type_root.id
+            type_root.compose_id(object)
+          else
+            id = object.id if object.respond_to?(:id)
+            id ||= object[:id] || object['id'] if object.is_a?(Hash)
+            id = id.to_s if defined?(BSON) && id.is_a?(BSON::ObjectId)
+            id
+          end
+        end
+
+        def index_object_ids
+          @index_object_ids ||= @to_index.each_with_object({}) do |object, result|
+            id = entry_id(object)
+            result[object] = id if id.present?
+          end
+        end
+
+        def type_root
+          @type_root ||= @index.root
+        end
+      end
+    end
+  end
+end

data/lib/chewy/index/import/bulk_request.rb

@@ -0,0 +1,77 @@
+module Chewy
+  class Index
+    module Import
+      # Adds additional features to elasticsearch-api bulk method:
+      # * supports Chewy index suffix if necessary;
+      # * supports bulk_size, devides the passed body in chunks
+      #   and peforms a separate request for each chunk;
+      # * returns only errored document entries from the response
+      #   if any present.
+      #
+      # @see https://github.com/elastic/elasticsearch-ruby/blob/master/elasticsearch-api/lib/elasticsearch/api/actions/bulk.rb
+      class BulkRequest
+        # @param index [Chewy::Index] an index for the request
+        # @param suffix [String] an index name optional suffix
+        # @param bulk_size [Integer] bulk size in bytes
+        # @param bulk_options [Hash] options passed to the elasticsearch-api bulk method
+        def initialize(index, suffix: nil, bulk_size: nil, **bulk_options)
+          @index = index
+          @suffix = suffix
+          @bulk_size = bulk_size - 1.kilobyte if bulk_size # 1 kilobyte for request header and newlines
+          @bulk_options = bulk_options
+
+          raise ArgumentError, '`bulk_size` can\'t be less than 1 kilobyte' if @bulk_size && @bulk_size <= 0
+        end
+
+        # Performs a bulk request with the passed body, returns empty
+        # array if everything is fine and array filled with errored
+        # document entries if something went wrong.
+        #
+        # @param body [Array<Hash>] a standard bulk request body
+        # @return [Array<Hash>] an array of bulk errors
+        def perform(body)
+          return [] if body.blank?
+
+          request_bodies(body).each_with_object([]) do |request_body, results|
+            response = @index.client.bulk(**request_base.merge(body: request_body)) if request_body.present?
+
+            next unless response.try(:[], 'errors')
+
+            response_items = (response.try(:[], 'items') || [])
+              .select { |item| item.values.first['error'] }
+            results.concat(response_items)
+          end
+        end
+
+      private
+
+        def request_base
+          @request_base ||= {
+            index: @index.index_name(suffix: @suffix)
+          }.merge!(@bulk_options)
+        end
+
+        def request_bodies(body)
+          if @bulk_size
+            serializer = ::Elasticsearch::API.serializer
+            pieces = body.each_with_object(['']) do |piece, result|
+              operation, meta = piece.to_a.first
+              data = meta.delete(:data)
+              piece = serializer.dump(operation => meta)
+              piece << "\n" << serializer.dump(data) if data.present?
+
+              if result.last.bytesize + piece.bytesize > @bulk_size
+                result.push(piece)
+              else
+                result[-1].blank? ? (result[-1] = piece) : (result[-1] << "\n" << piece)
+              end
+            end
+            pieces.each { |piece| piece << "\n" }
+          else
+            [body]
+          end
+        end
+      end
+    end
+  end
+end

data/lib/chewy/index/import/journal_builder.rb

@@ -0,0 +1,44 @@
+module Chewy
+  class Index
+    module Import
+      class JournalBuilder
+        def initialize(index, to_index: [], delete: [])
+          @index = index
+          @to_index = to_index
+          @delete = delete
+        end
+
+        def bulk_body
+          Chewy::Index::Import::BulkBuilder.new(
+            Chewy::Stash::Journal,
+            to_index: [
+              entries(:index, @to_index),
+              entries(:delete, @delete)
+            ].compact
+          ).bulk_body.each do |item|
+            item.values.first.merge!(
+              _index: Chewy::Stash::Journal.index_name
+            )
+          end
+        end
+
+      private
+
+        def entries(action, objects)
+          return unless objects.present?
+
+          {
+            index_name: @index.derivable_name,
+            action: action,
+            references: identify(objects).map { |item| Base64.encode64(::Elasticsearch::API.serializer.dump(item)) },
+            created_at: Time.now.utc
+          }
+        end
+
+        def identify(objects)
+          @index.adapter.identify(objects)
+        end
+      end
+    end
+  end
+end

data/lib/chewy/index/import/routine.rb

@@ -0,0 +1,139 @@
+module Chewy
+  class Index
+    module Import
+      # This class performs the import routine for the options and objects given.
+      #
+      # 0. Create target and journal indexes if needed.
+      # 1. Iterate over all the passed objects in batches.
+      # 2. For each batch {#process} method is called:
+      #   * creates a bulk request body;
+      #   * appends journal entries for the current batch to the request body;
+      #   * prepends a leftovers bulk to the request body, which is calculated
+      #     basing on the previous iteration errors;
+      #   * performs the bulk request;
+      #   * composes new leftovers bulk for the next iteration basing on the response errors if `update_failover` is true;
+      #   * appends the rest of unfixable errors to the instance level errors array.
+      # 4. Perform the request for the last leftovers bulk if present using {#extract_leftovers}.
+      # 3. Return the result errors array.
+      #
+      # At the moment, it tries to restore only from the partial document update errors in cases
+      # when the document doesn't exist only if `update_failover` option is true. In order to
+      # restore, it indexes such an objects completely on the next iteration.
+      #
+      # @see Chewy::Index::Import::ClassMethods#import
+      class Routine
+        BULK_OPTIONS = %i[
+          suffix bulk_size
+          refresh timeout fields pipeline
+          consistency replication
+          wait_for_active_shards routing _source _source_exclude _source_include
+        ].freeze
+
+        DEFAULT_OPTIONS = {
+          refresh: true,
+          update_fields: [],
+          update_failover: true,
+          batch_size: Chewy::Index::Adapter::Base::BATCH_SIZE
+        }.freeze
+
+        attr_reader :options, :parallel_options, :errors, :stats, :leftovers
+
+        # Basically, processes passed options, extracting bulk request specific options.
+        # @param index [Chewy::Index] chewy index
+        # @param options [Hash] import options, see {Chewy::Index::Import::ClassMethods#import}
+        def initialize(index, **options)
+          @index = index
+          @options = options
+          @options.reverse_merge!(@index._default_import_options)
+          @options.reverse_merge!(journal: Chewy.configuration[:journal])
+          @options.reverse_merge!(DEFAULT_OPTIONS)
+          @bulk_options = @options.slice(*BULK_OPTIONS)
+          @parallel_options = @options.delete(:parallel)
+          if @parallel_options && !@parallel_options.is_a?(Hash)
+            @parallel_options = if @parallel_options.is_a?(Integer)
+              {in_processes: @parallel_options}
+            else
+              {}
+            end
+          end
+          @errors = []
+          @stats = {}
+          @leftovers = []
+        end
+
+        # Creates the journal index and the corresponding index if necessary.
+        # @return [Object] whatever
+        def create_indexes!
+          Chewy::Stash::Journal.create if @options[:journal] && !Chewy.configuration[:skip_journal_creation_on_import]
+          return if Chewy.configuration[:skip_index_creation_on_import]
+
+          @index.create!(**@bulk_options.slice(:suffix)) unless @index.exists?
+        end
+
+        # The main process method. Converts passed objects to the bulk request body,
+        # appends journal entries, performs this request and handles errors performing
+        # failover procedures if applicable.
+        #
+        # @param index [Array<Object>] any acceptable objects for indexing
+        # @param delete [Array<Object>] any acceptable objects for deleting
+        # @return [true, false] the result of the request, true if no errors
+        def process(index: [], delete: [])
+          bulk_builder = BulkBuilder.new(@index, to_index: index, delete: delete, fields: @options[:update_fields])
+          bulk_body = bulk_builder.bulk_body
+
+          if @options[:journal]
+            journal_builder = JournalBuilder.new(@index, to_index: index, delete: delete)
+            bulk_body.concat(journal_builder.bulk_body)
+          end
+
+          bulk_body.unshift(*flush_leftovers)
+
+          perform_bulk(bulk_body) do |response|
+            @leftovers = extract_leftovers(response, bulk_builder.index_objects_by_id)
+            @stats[:index] = @stats[:index].to_i + index.count if index.present?
+            @stats[:delete] = @stats[:delete].to_i + delete.count if delete.present?
+          end
+        end
+
+        # Performs a bulk request for the passed body.
+        #
+        # @param body [Array<Hash>] a standard bulk request body
+        # @return [true, false] the result of the request, true if no errors
+        def perform_bulk(body)
+          response = bulk.perform(body)
+          yield response if block_given?
+          Chewy.wait_for_status
+          @errors.concat(response)
+          response.blank?
+        end
+
+      private
+
+        def flush_leftovers
+          leftovers = @leftovers
+          @leftovers = []
+          leftovers
+        end
+
+        def extract_leftovers(errors, index_objects_by_id)
+          return [] unless @options[:update_fields].present? && @options[:update_failover] && errors.present?
+
+          failed_partial_updates = errors.select do |item|
+            item.keys.first == 'update' && item.values.first['error']['type'] == 'document_missing_exception'
+          end
+          failed_ids_hash = failed_partial_updates.index_by { |item| item.values.first['_id'].to_s }
+          failed_ids_for_reimport = failed_ids_hash.keys & index_objects_by_id.keys
+          errors_to_cleanup = failed_ids_hash.values_at(*failed_ids_for_reimport)
+          errors_to_cleanup.each { |error| errors.delete(error) }
+
+          failed_objects = index_objects_by_id.values_at(*failed_ids_for_reimport)
+          BulkBuilder.new(@index, to_index: failed_objects).bulk_body
+        end
+
+        def bulk
+          @bulk ||= BulkRequest.new(@index, **@bulk_options)
+        end
+      end
+    end
+  end
+end