Xeli commented on a change in pull request #6594: [FLINK-9311] [pubsub] Added PubSub source connector with support for checkpointing (ATLEAST_ONCE) URL: https://github.com/apache/flink/pull/6594#discussion_r300321983
########## File path: flink-connectors/flink-connector-gcp-pubsub/src/main/java/org/apache/flink/streaming/connectors/gcp/pubsub/PubSubSource.java ########## @@ -0,0 +1,292 @@ +/* + * 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.streaming.connectors.gcp.pubsub; + +import org.apache.flink.api.common.functions.RuntimeContext; +import org.apache.flink.api.common.serialization.DeserializationSchema; +import org.apache.flink.api.common.typeinfo.TypeInformation; +import org.apache.flink.api.java.typeutils.ResultTypeQueryable; +import org.apache.flink.configuration.Configuration; +import org.apache.flink.runtime.state.CheckpointListener; +import org.apache.flink.streaming.api.checkpoint.ListCheckpointed; +import org.apache.flink.streaming.api.functions.source.ParallelSourceFunction; +import org.apache.flink.streaming.api.functions.source.RichSourceFunction; +import org.apache.flink.streaming.api.operators.StreamingRuntimeContext; +import org.apache.flink.streaming.connectors.gcp.pubsub.common.AcknowledgeIdsForCheckpoint; +import org.apache.flink.streaming.connectors.gcp.pubsub.common.AcknowledgeOnCheckpoint; +import org.apache.flink.streaming.connectors.gcp.pubsub.common.Acknowledger; +import org.apache.flink.streaming.connectors.gcp.pubsub.common.PubSubDeserializationSchema; +import org.apache.flink.streaming.connectors.gcp.pubsub.common.PubSubSubscriber; +import org.apache.flink.streaming.connectors.gcp.pubsub.common.PubSubSubscriberFactory; +import org.apache.flink.util.Preconditions; + +import com.google.auth.Credentials; +import com.google.cloud.pubsub.v1.Subscriber; +import com.google.pubsub.v1.ProjectSubscriptionName; +import com.google.pubsub.v1.PubsubMessage; +import com.google.pubsub.v1.ReceivedMessage; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.io.IOException; +import java.io.Serializable; +import java.time.Duration; +import java.util.List; +import java.util.concurrent.CancellationException; + +import static com.google.cloud.pubsub.v1.SubscriptionAdminSettings.defaultCredentialsProviderBuilder; + +/** + * PubSub Source, this Source will consume PubSub messages from a subscription and Acknowledge them on the next checkpoint. + * This ensures every message will get acknowledged at least once. + */ +public class PubSubSource<OUT> extends RichSourceFunction<OUT> + implements ResultTypeQueryable<OUT>, ParallelSourceFunction<OUT>, CheckpointListener, ListCheckpointed<AcknowledgeIdsForCheckpoint<String>> { + public static final int NO_MAX_MESSAGES_TO_ACKNOWLEDGE_LIMIT = -1; + private static final Logger LOG = LoggerFactory.getLogger(PubSubSource.class); + protected final PubSubDeserializationSchema<OUT> deserializationSchema; + protected final PubSubSubscriberFactory pubSubSubscriberFactory; + protected final Credentials credentials; + protected final int maxMessagesToAcknowledge; + protected final AcknowledgeOnCheckpointFactory acknowledgeOnCheckpointFactory; + + protected transient AcknowledgeOnCheckpoint<String> acknowledgeOnCheckpoint; + protected transient PubSubSubscriber subscriber; + + protected transient volatile boolean isRunning; + + PubSubSource(PubSubDeserializationSchema<OUT> deserializationSchema, + PubSubSubscriberFactory pubSubSubscriberFactory, + Credentials credentials, + int maxMessagesToAcknowledge, + AcknowledgeOnCheckpointFactory acknowledgeOnCheckpointFactory) { + this.deserializationSchema = deserializationSchema; + this.pubSubSubscriberFactory = pubSubSubscriberFactory; + this.credentials = credentials; + this.maxMessagesToAcknowledge = maxMessagesToAcknowledge; + this.acknowledgeOnCheckpointFactory = acknowledgeOnCheckpointFactory; + } + + @Override + public void open(Configuration configuration) throws Exception { + super.open(configuration); + if (hasNoCheckpointingEnabled(getRuntimeContext())) { + throw new IllegalArgumentException("The PubSubSource REQUIRES Checkpointing to be enabled and " + + "the checkpointing frequency must be MUCH lower than the PubSub timeout for it to retry a message."); + } + + getRuntimeContext().getMetricGroup().gauge("PubSubMessagesProcessedNotAcked", this::getOutstandingMessagesToAck); + + createAndSetPubSubSubscriber(); + this.isRunning = true; + } + + private boolean hasNoCheckpointingEnabled(RuntimeContext runtimeContext) { + return !(runtimeContext instanceof StreamingRuntimeContext && ((StreamingRuntimeContext) runtimeContext).isCheckpointingEnabled()); + } + + @Override + public void run(SourceContext<OUT> sourceContext) throws Exception { + while (isRunning) { + try { + if (maxMessagesToAcknowledgeLimitReached()) { Review comment: Thank you for the clear explaination. Without this limit, we will run into OOM exceptions. Lowering the CP interval does not matter because it is always trying to checkpoint but they just take too long to complete. Increasing the checkpoint timeout would work, but if this is increased too far the source can still fill up the memory. I like the idea of throttling via backpressure but with a dataskew, I don't know how the 'non-backpressured' subtasks can be throttled implicitly and explicit throttling is basically what this limit does right? My fear is that if we remove this limit a job might get into a situation where it can never recover and end-users have no way of throttling the sources. So if we can replace this limit by another means of throttling let us do that, do you have a suggestion how we can throttle instead? Ps. Regarding concern 3, the maximum acknowledged messages won't grow at ~10 pull req/s but stop at the limit because of the `continue`, right? ---------------------------------------------------------------- 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: [email protected] With regards, Apache Git Services
