hlteoh37 commented on code in PR #49: URL: https://github.com/apache/flink-connector-aws/pull/49#discussion_r1196848796
########## flink-connector-aws-kinesis-streams/src/main/java/org/apache/flink/connector/kinesis/source/enumerator/KinesisStreamsSourceEnumerator.java: ########## @@ -0,0 +1,338 @@ +/* + * 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.kinesis.source.enumerator; + +import org.apache.flink.annotation.Internal; +import org.apache.flink.api.connector.source.ReaderInfo; +import org.apache.flink.api.connector.source.SourceEvent; +import org.apache.flink.api.connector.source.SplitEnumerator; +import org.apache.flink.api.connector.source.SplitEnumeratorContext; +import org.apache.flink.api.connector.source.SplitsAssignment; +import org.apache.flink.connector.kinesis.source.config.SourceConfigConstants.InitialPosition; +import org.apache.flink.connector.kinesis.source.enumerator.assigner.ShardAssignerFactory; +import org.apache.flink.connector.kinesis.source.exception.KinesisStreamsSourceException; +import org.apache.flink.connector.kinesis.source.model.CompletedShardsEvent; +import org.apache.flink.connector.kinesis.source.proxy.StreamProxy; +import org.apache.flink.connector.kinesis.source.split.KinesisShardSplit; +import org.apache.flink.connector.kinesis.source.split.StartingPosition; + +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; +import software.amazon.awssdk.services.kinesis.model.Shard; + +import javax.annotation.Nullable; + +import java.io.IOException; +import java.time.Instant; +import java.util.ArrayList; +import java.util.Collections; +import java.util.HashMap; +import java.util.HashSet; +import java.util.List; +import java.util.Map; +import java.util.Properties; +import java.util.Set; + +import static org.apache.flink.connector.kinesis.source.config.SourceConfigConstants.DEFAULT_SHARD_DISCOVERY_INTERVAL_MILLIS; +import static org.apache.flink.connector.kinesis.source.config.SourceConfigConstants.DEFAULT_STREAM_INITIAL_POSITION; +import static org.apache.flink.connector.kinesis.source.config.SourceConfigConstants.SHARD_DISCOVERY_INTERVAL_MILLIS; +import static org.apache.flink.connector.kinesis.source.config.SourceConfigConstants.STREAM_INITIAL_POSITION; +import static org.apache.flink.connector.kinesis.source.config.SourceConfigUtil.parseStreamTimestampStartingPosition; + +/** + * This class is used to discover and assign Kinesis splits to subtasks on the Flink cluster. This + * runs on the JobManager. + */ +@Internal +public class KinesisStreamsSourceEnumerator + implements SplitEnumerator<KinesisShardSplit, KinesisStreamsSourceEnumeratorState> { + + private static final Logger LOG = LoggerFactory.getLogger(KinesisStreamsSourceEnumerator.class); + + private final SplitEnumeratorContext<KinesisShardSplit> context; + private final String streamArn; + private final Properties consumerConfig; + private final StreamProxy streamProxy; + private final KinesisShardAssigner shardAssigner; + private final KinesisShardAssigner.Context shardAssignerContext; + + private final Map<Integer, Set<KinesisShardSplit>> splitAssignment = new HashMap<>(); + private final Set<String> assignedSplitIds = new HashSet<>(); + private final Set<KinesisShardSplit> unassignedSplits; + private final Set<String> completedSplitIds; + + private String lastSeenShardId; + + public KinesisStreamsSourceEnumerator( + SplitEnumeratorContext<KinesisShardSplit> context, + String streamArn, + Properties consumerConfig, + StreamProxy streamProxy, + KinesisStreamsSourceEnumeratorState state) { + this.context = context; + this.streamArn = streamArn; + this.consumerConfig = consumerConfig; + this.streamProxy = streamProxy; + this.shardAssigner = ShardAssignerFactory.uniformShardAssigner(); + this.shardAssignerContext = new ShardAssignerContext(splitAssignment, context); + if (state == null) { + this.completedSplitIds = new HashSet<>(); + this.lastSeenShardId = null; + this.unassignedSplits = new HashSet<>(); + } else { + this.completedSplitIds = state.getCompletedSplitIds(); + this.lastSeenShardId = state.getLastSeenShardId(); + this.unassignedSplits = state.getUnassignedSplits(); + } + } + + @Override + public void start() { + if (lastSeenShardId == null) { + context.callAsync(this::initialDiscoverSplits, this::assignSplits); + } + + final long shardDiscoveryInterval = + Long.parseLong( + consumerConfig.getProperty( + SHARD_DISCOVERY_INTERVAL_MILLIS, + String.valueOf(DEFAULT_SHARD_DISCOVERY_INTERVAL_MILLIS))); + context.callAsync( + this::periodicallyDiscoverSplits, + this::assignSplits, + shardDiscoveryInterval, + shardDiscoveryInterval); + } + + @Override + public void handleSplitRequest(int subtaskId, @Nullable String requesterHostname) { + // Do nothing, since we assign splits eagerly + } + + @Override + public void addSplitsBack(List<KinesisShardSplit> splits, int subtaskId) { + if (!splitAssignment.containsKey(subtaskId)) { + LOG.warn( + "Unable to add splits back for subtask {} since it is not assigned any splits. Splits: {}", + subtaskId, + splits); + return; + } + + for (KinesisShardSplit split : splits) { + splitAssignment.get(subtaskId).remove(split); + assignedSplitIds.remove(split.splitId()); + unassignedSplits.add(split); + } + } + + @Override + public void addReader(int subtaskId) { + splitAssignment.putIfAbsent(subtaskId, new HashSet<>()); + } + + @Override + public KinesisStreamsSourceEnumeratorState snapshotState(long checkpointId) throws Exception { + return new KinesisStreamsSourceEnumeratorState( + completedSplitIds, unassignedSplits, lastSeenShardId); + } + + @Override + public void close() throws IOException {} + + @Override + public void handleSourceEvent(int subtaskId, SourceEvent sourceEvent) { + if (sourceEvent instanceof CompletedShardsEvent) { + Set<String> newlyCompletedSplitIds = + ((CompletedShardsEvent) sourceEvent).getCompletedSplitIds(); + LOG.info( + "Received CompletedShardsEvent from subtask {}. Marking the following splits as complete: {}", + subtaskId, + newlyCompletedSplitIds); + completedSplitIds.addAll(newlyCompletedSplitIds); + } else { + SplitEnumerator.super.handleSourceEvent(subtaskId, sourceEvent); + } + } + + private List<KinesisShardSplit> initialDiscoverSplits() { + List<Shard> shards = streamProxy.listShards(streamArn, lastSeenShardId); + return mapToSplits(shards, false); + } + + /** + * This method is used to discover Kinesis splits the job can subscribe to. It can be run in + * parallel, is important to not mutate any shared state. + * + * @return list of discovered splits + */ + private List<KinesisShardSplit> periodicallyDiscoverSplits() { + List<Shard> shards = streamProxy.listShards(streamArn, lastSeenShardId); + // Any shard discovered after the initial startup should be read from the start, since they + // come from resharding + return mapToSplits(shards, true); + } + + private List<KinesisShardSplit> mapToSplits(List<Shard> shards, boolean shouldReadFromStart) { + InitialPosition initialPositionFromConfig = + shouldReadFromStart + ? InitialPosition.TRIM_HORIZON + : InitialPosition.valueOf( + consumerConfig + .getOrDefault( + STREAM_INITIAL_POSITION, + DEFAULT_STREAM_INITIAL_POSITION) + .toString()); + StartingPosition startingPosition; + switch (initialPositionFromConfig) { + case LATEST: + // If LATEST is requested, we still set the starting position to the time of + // startup. This way, the job starts reading from a deterministic timestamp + // (i.e. time of job submission), even if it enters a restart loop immediately + // after submission. + startingPosition = StartingPosition.fromTimestamp(Instant.now()); + break; + case AT_TIMESTAMP: + startingPosition = + StartingPosition.fromTimestamp( + parseStreamTimestampStartingPosition(consumerConfig).toInstant()); + break; + case TRIM_HORIZON: + default: + startingPosition = StartingPosition.fromStart(); + } + + List<KinesisShardSplit> splits = new ArrayList<>(); + for (Shard shard : shards) { + splits.add(new KinesisShardSplit(streamArn, shard.shardId(), startingPosition)); + } + + return splits; + } + + /** + * This method assigns a given set of Kinesis splits to the readers currently registered on the + * cluster. This assignment is done via a side-effect on the {@link SplitEnumeratorContext} + * object. + * + * @param discoveredSplits list of discovered splits + * @param t throwable thrown when discovering splits. Will be null if no throwable thrown. + */ + private void assignSplits(List<KinesisShardSplit> discoveredSplits, Throwable t) { + if (t != null) { + throw new KinesisStreamsSourceException("Failed to list shards.", t); + } + + if (context.registeredReaders().isEmpty()) { + LOG.info("No registered readers, skipping assignment of discovered splits."); + unassignedSplits.addAll(discoveredSplits); + return; + } + + Map<Integer, List<KinesisShardSplit>> newSplitAssignments = new HashMap<>(); + for (KinesisShardSplit split : unassignedSplits) { + assignSplitToSubtask(split, newSplitAssignments); + } + unassignedSplits.clear(); + for (KinesisShardSplit split : discoveredSplits) { + assignSplitToSubtask(split, newSplitAssignments); + } + + updateLastSeenShardId(discoveredSplits); + updateSplitAssignment(newSplitAssignments); + context.assignSplits(new SplitsAssignment<>(newSplitAssignments)); Review Comment: Great callout. So I have checked that should be effectively synchronous across components due to our RPC setup. 1. **On JM:** assignSplit() runs in the coordinatorThread. [code here](https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/source/coordinator/SourceCoordinatorContext.java#L323). (This is the same thread that handles checkpointCoordinator() that triggers checkpoint on SourceCoordinator). This means the order of calls will be maintained here. 2. In both cases, these make RPC calls to the TM. [assignSplit() calls `sendOperatorEventToTask()`](https://github.com/apache/flink/blob/2c25292272306a6eb094be40be0a0fa5fc3cf644/flink-runtime/src/main/java/org/apache/flink/runtime/taskexecutor/TaskExecutorGateway.java#L284) and `checkpointCoordinator()` [calls `triggerCheckpoint()`](https://github.com/apache/flink/blob/2c25292272306a6eb094be40be0a0fa5fc3cf644/flink-runtime/src/main/java/org/apache/flink/runtime/taskexecutor/TaskExecutorGateway.java#L141). 3. Since RPC calls are FIFO, we can be assured that the `assignSplit()` will be always processed before `checkpointCoordinator()` -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
