Class: Sequent::Core::Persistors::ReplayOptimizedPostgresPersistor

Inherits:
Object
  • Object
show all
Includes:
Persistor
Defined in:
lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb

Overview

The ReplayOptimizedPostgresPersistor is optimized for bulk loading records in a Postgres database.

Depending on the amount of records it uses CSV import, otherwise statements are batched using normal sql.

Rebuilding the view state (or projection) of an aggregate typically consists of an initial insert and then many updates and maybe a delete. With a normal Persistor (like ActiveRecordPersistor) each action is executed to the database. This persistor creates an in-memory store first and finally flushes the in-memory store to the database. This can significantly reduce the amount of queries to the database. E.g. 1 insert, 6 updates is only a single insert using this Persistor.

After lot of experimenting this turned out to be the fastest way to to bulk inserts in the database. You can tweak the amount of records in the CSV via insert_with_csv_size before it flushes to the database to gain (or loose) speed.

It is highly recommended to create indices on the in memory record_store to speed up the processing. By default all records are indexed by aggregate_id if they have such a property.

Example:

class InvoiceProjector < Sequent::Core::Projector
  on RecipientMovedEvent do |event|
    update_all_records(
      InvoiceRecord,
      { aggregate_id: event.aggregate_id, recipient_id: event.recipient.aggregate_id },
      { recipient_street: event.recipient.street },
    end
  end
end

In this case it is wise to create an index on InvoiceRecord on the aggregate_id and recipient_id like you would in the database.

Example:

ReplayOptimizedPostgresPersistor.new(
  50,
  {InvoiceRecord => [[:aggregate_id, :recipient_id]]}
)

Defined Under Namespace

Modules: InitStruct Classes: Index

Constant Summary collapse

CHUNK_SIZE = 
1024

Instance Attribute Summary collapse

Class Method Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(insert_with_csv_size = 50, indices = {}) ⇒ ReplayOptimizedPostgresPersistor

insert_with_csv_size number of records to insert in a single batch

indices Hash of indices to create in memory. Greatly speeds up the replaying.

Key corresponds to the name of the 'Record'
Values contains list of lists on which columns to index.
E.g. [[:first_index_column], [:another_index, :with_to_columns]]



170
171
172
173
174
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 170

def initialize(insert_with_csv_size = 50, indices = {})
  @insert_with_csv_size = insert_with_csv_size
  @record_store = Hash.new { |h, k| h[k] = Set.new }
  @record_index = Index.new(indices)
end

Instance Attribute Details

#insert_with_csv_size ⇒ Object

Returns the value of attribute insert_with_csv_size.



57
58
59
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 57

def insert_with_csv_size
  @insert_with_csv_size
end

#record_store ⇒ Object (readonly)

Returns the value of attribute record_store.



56
57
58
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 56

def record_store
  @record_store
end

Class Method Details

.struct_cache ⇒ Object 



59
60
61
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 59

def self.struct_cache
  @struct_cache ||= {}
end

Instance Method Details

#clear ⇒ Object 



359
360
361
362
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 359

def clear
  @record_store.clear
  @record_index.clear
end

#commit ⇒ Object 



313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 313

def commit
  @record_store.each do |clazz, records|
    @column_cache ||= {}
    @column_cache[clazz.name] ||= clazz.columns.reduce({}) do |hash, column|
      hash.merge({column.name => column})
    end
    if records.size > @insert_with_csv_size
      csv = CSV.new(StringIO.new)
      column_names = clazz.column_names.reject { |name| name == 'id' }
      records.each do |record|
        csv << column_names.map do |column_name|
          cast_value_to_column_type(clazz, column_name, record)
        end
      end

      conn = Sequent::ApplicationRecord.connection.raw_connection
      copy_data = StringIO.new(csv.string)
      conn.transaction do
        conn.copy_data("COPY #{clazz.table_name} (#{column_names.join(',')}) FROM STDIN WITH csv") do
          while (out = copy_data.read(CHUNK_SIZE))
            conn.put_copy_data(out)
          end
        end
      end
    else
      clazz.unscoped do
        inserts = []
        column_names = clazz.column_names.reject { |name| name == 'id' }
        prepared_values = (1..column_names.size).map { |i| "$#{i}" }.join(',')
        records.each do |record|
          values = column_names.map do |column_name|
            cast_value_to_column_type(clazz, column_name, record)
          end
          inserts << values
        end
        sql = %{insert into #{clazz.table_name} (#{column_names.join(',')}) values (#{prepared_values})}
        inserts.each do |insert|
          clazz.connection.raw_connection.async_exec(sql, insert)
        end
      end
    end
  end
ensure
  clear
end

#create_or_update_record(record_class, values, created_at = Time.now) {|record| ... } ⇒ Object

Yields:

  • (record) 


233
234
235
236
237
238
239
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 233

def create_or_update_record(record_class, values, created_at = Time.now)
  record = get_record(record_class, values)
  record ||= create_record(record_class, values.merge(created_at: created_at))
  yield record if block_given?
  @record_index.update(record_class, record)
  record
end

#create_record(record_class, values) {|record| ... } ⇒ Object

Yields:

  • (record) 


189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 189

def create_record(record_class, values)
  column_names = record_class.column_names
  values = record_class.column_defaults.with_indifferent_access.merge(values)
  values.merge!(updated_at: values[:created_at]) if column_names.include?('updated_at')
  struct_class_name = "#{record_class}Struct"
  if self.class.struct_cache.key?(struct_class_name)
    struct_class = self.class.struct_cache[struct_class_name]
  else
    # We create a struct on the fly.
    # Since the replay happens in memory we implement the ==, eql? and hash methods
    # to point to the same object. A record is the same if and only if they point to
    # the same object. These methods are necessary since we use Set instead of [].
    class_def = <<-EOD
      #{struct_class_name} = Struct.new(*#{column_names.map(&:to_sym)})
      class #{struct_class_name}
        include InitStruct
        def ==(other)
          self.equal?(other)
        end
        def hash
          self.object_id.hash
        end
      end
    EOD
    # rubocop:disable Security/Eval
    eval(class_def.to_s)
    # rubocop:enable Security/Eval
    struct_class = ReplayOptimizedPostgresPersistor.const_get(struct_class_name)
    self.class.struct_cache[struct_class_name] = struct_class
  end
  record = struct_class.new.set_values(values)

  yield record if block_given?
  @record_store[record_class] << record

  @record_index.add(record_class, record)

  record
end

#create_records(record_class, array_of_value_hashes) ⇒ Object 



229
230
231
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 229

def create_records(record_class, array_of_value_hashes)
  array_of_value_hashes.each { |values| create_record(record_class, values) }
end

#delete_all_records(record_class, where_clause) ⇒ Object 



255
256
257
258
259
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 255

def delete_all_records(record_class, where_clause)
  find_records(record_class, where_clause).each do |record|
    delete_record(record_class, record)
  end
end

#delete_record(record_class, record) ⇒ Object 



261
262
263
264
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 261

def delete_record(record_class, record)
  @record_store[record_class].delete(record)
  @record_index.remove(record_class, record)
end

#do_with_record(record_class, where_clause) {|record| ... } ⇒ Object

Yields:

  • (record) 


283
284
285
286
287
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 283

def do_with_record(record_class, where_clause)
  record = get_record!(record_class, where_clause)
  yield record
  @record_index.update(record_class, record)
end

#do_with_records(record_class, where_clause) ⇒ Object 



275
276
277
278
279
280
281
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 275

def do_with_records(record_class, where_clause)
  records = find_records(record_class, where_clause)
  records.each do |record|
    yield record
    @record_index.update(record_class, record)
  end
end

#find_records(record_class, where_clause) ⇒ Object 



289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 289

def find_records(record_class, where_clause)
  if @record_index.use_index?(record_class, where_clause)
    @record_index.find(record_class, where_clause)
  else
    @record_store[record_class].select do |record|
      where_clause.all? do |k, v|
        expected_value = v.is_a?(Symbol) ? v.to_s : v
        actual_value = record[k.to_sym]
        actual_value = actual_value.to_s if actual_value.is_a? Symbol
        if expected_value.is_a?(Array)
          expected_value.include?(actual_value)
        else
          actual_value == expected_value
        end
      end
    end
  end.dup
end

#get_record(record_class, where_clause) ⇒ Object 



250
251
252
253
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 250

def get_record(record_class, where_clause)
  results = find_records(record_class, where_clause)
  results.empty? ? nil : results.first
end

#get_record!(record_class, where_clause) ⇒ Object 



241
242
243
244
245
246
247
248
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 241

def get_record!(record_class, where_clause)
  record = get_record(record_class, where_clause)
  unless record
    fail("record #{record_class} not found for #{where_clause}, store: #{@record_store[record_class]}")
  end

  record
end

#last_record(record_class, where_clause) ⇒ Object 



308
309
310
311
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 308

def last_record(record_class, where_clause)
  results = find_records(record_class, where_clause)
  results.empty? ? nil : results.last
end

#update_all_records(record_class, where_clause, updates) ⇒ Object 



266
267
268
269
270
271
272
273
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 266

def update_all_records(record_class, where_clause, updates)
  find_records(record_class, where_clause).each do |record|
    updates.each_pair do |k, v|
      record[k.to_sym] = v
    end
    @record_index.update(record_class, record)
  end
end

#update_record(record_class, event, where_clause = {aggregate_id: event.aggregate_id}, options = {}) {|record| ... } ⇒ Object

Yields:

  • (record) 


176
177
178
179
180
181
182
183
184
185
186
187
 
# File 'lib/sequent/core/persistors/replay_optimized_postgres_persistor.rb', line 176

def update_record(record_class, event, where_clause = {aggregate_id: event.aggregate_id}, options = {})
  record = get_record!(record_class, where_clause)
  record.updated_at = event.created_at if record.respond_to?(:updated_at)
  yield record if block_given?
  @record_index.update(record_class, record)
  update_sequence_number = if options.key?(:update_sequence_number)
                             options[:update_sequence_number]
                           else
                             record.respond_to?(:sequence_number=)
                           end
  record.sequence_number = event.sequence_number if update_sequence_number
end