stevenzwu commented on a change in pull request #4329: URL: https://github.com/apache/iceberg/pull/4329#discussion_r835777538
########## File path: flink/v1.14/flink/src/main/java/org/apache/iceberg/flink/source/enumerator/ContinuousSplitPlannerImpl.java ########## @@ -0,0 +1,160 @@ +/* + * 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.iceberg.flink.source.enumerator; + +import java.util.Collections; +import java.util.List; +import java.util.Optional; +import java.util.concurrent.ExecutorService; +import org.apache.flink.annotation.Internal; +import org.apache.flink.annotation.VisibleForTesting; +import org.apache.iceberg.Snapshot; +import org.apache.iceberg.Table; +import org.apache.iceberg.flink.source.FlinkSplitPlanner; +import org.apache.iceberg.flink.source.ScanContext; +import org.apache.iceberg.flink.source.StreamingStartingStrategy; +import org.apache.iceberg.flink.source.split.IcebergSourceSplit; +import org.apache.iceberg.util.SnapshotUtil; +import org.apache.iceberg.util.ThreadPools; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +@Internal +public class ContinuousSplitPlannerImpl implements ContinuousSplitPlanner { + private static final Logger LOG = LoggerFactory.getLogger(ContinuousSplitPlannerImpl.class); + + private final Table table; + private final ScanContext scanContext; + private final ExecutorService workerPool; + + public ContinuousSplitPlannerImpl(Table table, ScanContext scanContext, String threadPoolName) { + this.table = table; + this.scanContext = scanContext; + this.workerPool = ThreadPools.newWorkerPool( + "iceberg-plan-worker-pool-" + threadPoolName, scanContext.planParallelism()); + } + + @Override + public ContinuousEnumerationResult planSplits(IcebergEnumeratorPosition lastPosition) { + table.refresh(); + if (lastPosition != null) { + return discoverIncrementalSplits(lastPosition); + } else { + return discoverInitialSplits(); + } + } + + /** + * Discover incremental changes between @{code lastPosition} and current table snapshot + */ + private ContinuousEnumerationResult discoverIncrementalSplits(IcebergEnumeratorPosition lastPosition) { + // incremental discovery mode + Snapshot currentSnapshot = table.currentSnapshot(); + if (currentSnapshot.snapshotId() == lastPosition.endSnapshotId()) { + LOG.info("Current table snapshot is already enumerated: {}", currentSnapshot.snapshotId()); + return new ContinuousEnumerationResult(Collections.emptyList(), lastPosition); + } else { + ScanContext incrementalScan = scanContext + .copyWithAppendsBetween(lastPosition.endSnapshotId(), currentSnapshot.snapshotId()); + LOG.info("Incremental scan: startSnapshotId = {}, endSnapshotId = {}", + incrementalScan.startSnapshotId(), incrementalScan.endSnapshotId()); + List<IcebergSourceSplit> splits = FlinkSplitPlanner.planIcebergSourceSplits(table, incrementalScan, workerPool); + IcebergEnumeratorPosition position = IcebergEnumeratorPosition.builder() + .startSnapshotId(lastPosition.endSnapshotId()) + .startSnapshotTimestampMs(lastPosition.endSnapshotTimestampMs()) + .endSnapshotId(currentSnapshot.snapshotId()) + .endSnapshotTimestampMs(currentSnapshot.timestampMillis()) + .build(); + LOG.info("Discovered {} splits from incremental scan: {}", splits.size(), position); + return new ContinuousEnumerationResult(splits, position); + } + } + + /** + * Discovery initial set of splits based on {@link StreamingStartingStrategy}. + * + * <li>{@link ContinuousEnumerationResult#splits()} should contain initial splits + * discovered from table scan for {@link StreamingStartingStrategy#TABLE_SCAN_THEN_INCREMENTAL}. + * For all other strategies, splits collection should be empty. + * <li>{@link ContinuousEnumerationResult#position()} points to the starting position + * for the next incremental split discovery with exclusive behavior. Meaning files committed + * by the snapshot from the position in {@code ContinuousEnumerationResult} won't be included + * in the next incremental scan. + */ + private ContinuousEnumerationResult discoverInitialSplits() { + Optional<Snapshot> startSnapshotOptional = getStartSnapshot(table, scanContext); + if (!startSnapshotOptional.isPresent()) { + // set IcebergEnumeratorPosition to null in this case + return new ContinuousEnumerationResult(Collections.emptyList(), null); + } Review comment: agree it is not just the first snapshot. I was using it just to simplify the discussion. it doesn't really matter how many snapshots are committed before the first discovery. semantics are the same. I was saying `earliest` or `latest` snapshot can be interpreted as the earliest or latest valid snapshot of the table. Let's say when the Flink job started, it is an empty table. initial scan return null. * case 1: there are 3 snapshots (S1, S2, S3) committed. Now Flink source ran initial scan again. For earliest strategy, S1 will be used as starting snapshot. For latest strategy, S3 will be used as starting snapshot. * case 2: there is 1 snapshot committed. Now both earliest and latest strategy will discover S1 as starting snapshot regarding -1 snapshot id, is it a reserved special id? I wouldn't find the usage for snapshot id. Also, `IncrementalDataTableScan` seems to validate the start snapshot id. This may not work. ``` Preconditions.checkArgument( snapshotIdsRange.contains(newFromSnapshotId), "from snapshot id %s not in existing snapshot ids range (%s, %s]", newFromSnapshotId, context().fromSnapshotId(), newToSnapshotId); ``` @yittg I definitely see your perspective, which is a valid interpretation of earliest and latest for empty table. In your interpretation, earliest and latest will behave the same (with some imaginary snapshot as starting point). all future snapshot should be discovered. To implement this model, we need to change `appendsBetween` implementation to support a special starting snapshot id, which is invalid and yet indicates a special starting point. To me, both interpretations are reasonable. I also think for the practical purpose, how to interpret earliest and latest strategy for an empty table is not that important for streaming job. Most of the streaming job's bootstrap point is not really important, as it is arbitrary anyway. After bootstrapped, checkpoint contains valid snapshot id. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
