openinx commented on a change in pull request #1185: URL: https://github.com/apache/iceberg/pull/1185#discussion_r479874623
########## File path: flink/src/main/java/org/apache/iceberg/flink/sink/IcebergFilesCommitter.java ########## @@ -0,0 +1,229 @@ +/* + * 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.sink; + +import java.util.Comparator; +import java.util.List; +import java.util.Map; +import java.util.NavigableMap; +import java.util.SortedMap; +import org.apache.flink.api.common.state.ListState; +import org.apache.flink.api.common.state.ListStateDescriptor; +import org.apache.flink.api.common.typeinfo.BasicTypeInfo; +import org.apache.flink.api.common.typeinfo.TypeInformation; +import org.apache.flink.api.java.typeutils.ListTypeInfo; +import org.apache.flink.runtime.state.StateInitializationContext; +import org.apache.flink.runtime.state.StateSnapshotContext; +import org.apache.flink.streaming.api.operators.AbstractStreamOperator; +import org.apache.flink.streaming.api.operators.BoundedOneInput; +import org.apache.flink.streaming.api.operators.OneInputStreamOperator; +import org.apache.flink.streaming.runtime.streamrecord.StreamRecord; +import org.apache.flink.table.runtime.typeutils.SortedMapTypeInfo; +import org.apache.hadoop.conf.Configuration; +import org.apache.iceberg.AppendFiles; +import org.apache.iceberg.DataFile; +import org.apache.iceberg.Snapshot; +import org.apache.iceberg.Table; +import org.apache.iceberg.flink.TableLoader; +import org.apache.iceberg.hadoop.SerializableConfiguration; +import org.apache.iceberg.relocated.com.google.common.base.Preconditions; +import org.apache.iceberg.relocated.com.google.common.collect.ImmutableList; +import org.apache.iceberg.relocated.com.google.common.collect.Lists; +import org.apache.iceberg.relocated.com.google.common.collect.Maps; +import org.apache.iceberg.types.Comparators; +import org.apache.iceberg.types.Types; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +class IcebergFilesCommitter extends AbstractStreamOperator<Void> + implements OneInputStreamOperator<DataFile, Void>, BoundedOneInput { + + private static final long serialVersionUID = 1L; + private static final long INITIAL_CHECKPOINT_ID = -1L; + + private static final Logger LOG = LoggerFactory.getLogger(IcebergFilesCommitter.class); + private static final String FLINK_JOB_ID = "flink.job-id"; + + // The max checkpoint id we've committed to iceberg table. As the flink's checkpoint is always increasing, so we could + // correctly commit all the data files whose checkpoint id is greater than the max committed one to iceberg table, for + // avoiding committing the same data files twice. This id will be attached to iceberg's meta when committing the + // iceberg transaction. + private static final String MAX_COMMITTED_CHECKPOINT_ID = "flink.max-committed-checkpoint-id"; + + // TableLoader to load iceberg table lazily. + private final TableLoader tableLoader; + private final SerializableConfiguration hadoopConf; + + // A sorted map to maintain the completed data files for each pending checkpointId (which have not been committed + // to iceberg table). We need a sorted map here because there's possible that few checkpoints snapshot failed, for + // example: the 1st checkpoint have 2 data files <1, <file0, file1>>, the 2st checkpoint have 1 data files + // <2, <file3>>. Snapshot for checkpoint#1 interrupted because of network/disk failure etc, while we don't expect + // any data loss in iceberg table. So we keep the finished files <1, <file0, file1>> in memory and retry to commit + // iceberg table when the next checkpoint happen. + private final NavigableMap<Long, List<DataFile>> dataFilesPerCheckpoint = Maps.newTreeMap(); + + // The data files cache for current checkpoint. Once the snapshot barrier received, it will be flushed to the + // 'dataFilesPerCheckpoint'. + private final List<DataFile> dataFilesOfCurrentCheckpoint = Lists.newArrayList(); + + // It will have an unique identifier for one job. + private transient String flinkJobId; + private transient Table table; + private transient long maxCommittedCheckpointId; + + // All pending checkpoints states for this function. + private static final ListStateDescriptor<SortedMap<Long, List<DataFile>>> STATE_DESCRIPTOR = buildStateDescriptor(); + private transient ListState<SortedMap<Long, List<DataFile>>> checkpointsState; + + IcebergFilesCommitter(TableLoader tableLoader, Configuration hadoopConf) { + this.tableLoader = tableLoader; + this.hadoopConf = new SerializableConfiguration(hadoopConf); + } + + @Override + public void initializeState(StateInitializationContext context) throws Exception { + super.initializeState(context); + this.flinkJobId = getContainingTask().getEnvironment().getJobID().toString(); + + // Open the table loader and load the table. + this.tableLoader.open(hadoopConf.get()); + this.table = tableLoader.loadTable(); + this.maxCommittedCheckpointId = INITIAL_CHECKPOINT_ID; + + this.checkpointsState = context.getOperatorStateStore().getListState(STATE_DESCRIPTOR); + if (context.isRestored()) { + this.maxCommittedCheckpointId = getMaxCommittedCheckpointId(table, flinkJobId); Review comment: It's a great point that we could handle, I think we can attach both the `flinkJobId` and `maxCommittedCheckPointId` to the checkpoint state. when in restoring path, we read the `flinkJobId` and `maxCommittedCheckpointId` from states, and compare the current flink job id with `flinkJobId` from state. If the job ids are matched, then it is surely be the case that restoring without redploying(case#1), otherwise it's the case you said (case#2). For case#1, the current code should be correct. For case#2, we should use the old `flinkJobId` from state to parse the `maxCommittedCheckpointId` in iceberg table, and use that checkpoint id to filter all the committed data files. Does that make sense ? ---------------------------------------------------------------- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
