pscls commented on a change in pull request #15140: URL: https://github.com/apache/flink/pull/15140#discussion_r596962037
########## File path: flink-connectors/flink-connector-rabbitmq2/src/main/java/org/apache/flink/connector/rabbitmq2/sink/writer/specialized/RabbitMQSinkWriterExactlyOnce.java ########## @@ -0,0 +1,139 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.flink.connector.rabbitmq2.sink.writer.specialized; + +import org.apache.flink.api.common.serialization.SerializationSchema; +import org.apache.flink.api.connector.sink.SinkWriter; +import org.apache.flink.connector.rabbitmq2.common.RabbitMQConnectionConfig; +import org.apache.flink.connector.rabbitmq2.sink.RabbitMQSink; +import org.apache.flink.connector.rabbitmq2.sink.common.RabbitMQSinkMessageWrapper; +import org.apache.flink.connector.rabbitmq2.sink.common.RabbitMQSinkPublishOptions; +import org.apache.flink.connector.rabbitmq2.sink.common.SerializableReturnListener; +import org.apache.flink.connector.rabbitmq2.sink.state.RabbitMQSinkWriterState; +import org.apache.flink.connector.rabbitmq2.sink.writer.RabbitMQSinkWriterBase; + +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.io.IOException; +import java.util.ArrayList; +import java.util.Collections; +import java.util.List; + +/** + * A {@link SinkWriter} implementation for {@link RabbitMQSink} that provides exactly-once delivery + * guarantees, which means incoming stream elements will be delivered to RabbitMQ exactly once. For + * this, checkpointing needs to be enabled. + * + * <p>Exactly-once consistency is implemented using a transactional RabbitMQ channel. All incoming + * stream elements are buffered in the state of this writer until the next checkpoint is triggered. + * All buffered {@code messages} are then send to RabbitMQ in a single transaction. When successful, + * all messages committed get removed from the state. If the transaction is aborted, all messages + * are put back into the state and send on the next checkpoint. + * + * <p>The transactional channel is heavyweight and will decrease throughput. If the system is under + * heavy load, consecutive checkpoints can be delayed if commits take longer than the checkpointing + * interval specified. Only use exactly-once if necessary (no duplicated messages in RabbitMQ + * allowed), otherwise consider using at-least-once. + * + * @param <T> Type of the elements in this sink + */ +public class RabbitMQSinkWriterExactlyOnce<T> extends RabbitMQSinkWriterBase<T> { + private static final Logger LOG = LoggerFactory.getLogger(RabbitMQSinkWriterExactlyOnce.class); + + // All messages that arrived and could not be committed this far. + private List<RabbitMQSinkMessageWrapper<T>> messages; + + /** + * Create a new RabbitMQSinkWriterExactlyOnce. + * + * @param connectionConfig configuration parameters used to connect to RabbitMQ + * @param queueName name of the queue to publish to + * @param serializationSchema serialization schema to turn elements into byte representation + * @param publishOptions optionally used to compute routing/exchange for messages + * @param returnListener return listener + */ + public RabbitMQSinkWriterExactlyOnce( + RabbitMQConnectionConfig connectionConfig, + String queueName, + SerializationSchema<T> serializationSchema, + RabbitMQSinkPublishOptions<T> publishOptions, + SerializableReturnListener returnListener) { + super(connectionConfig, queueName, serializationSchema, publishOptions, returnListener); + messages = Collections.synchronizedList(new ArrayList<>()); + } + + /** + * On recover the messages are set to the outstanding messages from the states. + * + * @param states a list of states to recover the reader with + */ + @Override + public void recoverFromStates(List<RabbitMQSinkWriterState<T>> states) { + List<RabbitMQSinkMessageWrapper<T>> messages = + Collections.synchronizedList(new ArrayList<>()); + for (RabbitMQSinkWriterState<T> state : states) { + messages.addAll(state.getOutstandingMessages()); + } + this.messages = messages; + } + + @Override + protected void configureChannel() throws IOException { + // puts channel in commit mode + getRmqChannel().txSelect(); + } + + @Override + public void write(T element, Context context) { + messages.add( + new RabbitMQSinkMessageWrapper<>( + element, getSerializationSchema().serialize(element))); + } + + @Override + public List<RabbitMQSinkWriterState<T>> snapshotState() { + commitMessages(); + return Collections.singletonList(new RabbitMQSinkWriterState<>(messages)); + } + + private void commitMessages() { + List<RabbitMQSinkMessageWrapper<T>> messagesToSend = messages.subList(0, messages.size()); + try { + for (RabbitMQSinkMessageWrapper<T> msg : messagesToSend) { + getRmqSinkConnection().send(msg); + } + getRmqChannel().txCommit(); + LOG.info("Successfully committed {} messages.", messagesToSend.size()); + messagesToSend.clear(); + } catch (IOException e) { + LOG.error( + "Error during commit of {} messages. Rollback Messages. Error: {}", + messagesToSend.size(), + e.getMessage()); + try { + getRmqChannel().txRollback(); Review comment: We added a Runtime exception, so as soon as a transactional commit is not working we will throw it immediately. I guess the is the same discussion as the maxRetry counter we had earlier for sending an at-least-once message. You probably do not want to fail immediately but rather at some point in the future. Like you said this could cause a heavily increased checkpoint size. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
