lanyuanxiaoyao/hudi
1
0
Fork 0
Code Activity

[HUDI-2501] Add HoodieData abstraction and refactor compaction actions in hudi-client module (#3741)

Browse Source
This commit is contained in:
Y Ethan Guo
2021-10-22 12:58:51 -07:00
committed by GitHub
parent 1e285dc399
commit 5ed35bff83
41 changed files with 1084 additions and 1019 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
hudi-client/hudi-flink-client/src/main/java/org/apache/hudi/client/HoodieFlinkWriteClient.java
View File

@@ -19,6 +19,7 @@
package org.apache.hudi.client;
import org.apache.hudi.client.common.HoodieFlinkEngineContext;
import org.apache.hudi.common.data.HoodieList;
import org.apache.hudi.common.engine.HoodieEngineContext;
import org.apache.hudi.common.fs.FSUtils;
import org.apache.hudi.common.model.FileSlice;
@@ -58,7 +59,7 @@ import org.apache.hudi.table.HoodieFlinkTable;
import org.apache.hudi.table.HoodieTable;
import org.apache.hudi.table.HoodieTimelineArchiveLog;
import org.apache.hudi.table.action.HoodieWriteMetadata;
import org.apache.hudi.table.action.compact.FlinkCompactHelpers;
import org.apache.hudi.table.action.compact.CompactHelpers;
import org.apache.hudi.table.marker.WriteMarkersFactory;
import org.apache.hudi.table.upgrade.FlinkUpgradeDowngradeHelper;
import org.apache.hudi.table.upgrade.UpgradeDowngrade;
@@ -346,8 +347,8 @@ public class HoodieFlinkWriteClient<T extends HoodieRecordPayload> extends
List<WriteStatus> writeStatuses,
Option<Map<String, String>> extraMetadata) throws IOException {
HoodieFlinkTable<T> table = getHoodieTable();
HoodieCommitMetadata metadata = FlinkCompactHelpers.newInstance().createCompactionMetadata(
table, compactionInstantTime, writeStatuses, config.getSchema());
HoodieCommitMetadata metadata = CompactHelpers.getInstance().createCompactionMetadata(
table, compactionInstantTime, HoodieList.of(writeStatuses), config.getSchema());
extraMetadata.ifPresent(m -> m.forEach(metadata::addMetadata));
completeCompaction(metadata, writeStatuses, table, compactionInstantTime);
}
@@ -364,7 +365,7 @@ public class HoodieFlinkWriteClient<T extends HoodieRecordPayload> extends
// commit to data table after committing to metadata table.
finalizeWrite(table, compactionCommitTime, writeStats);
LOG.info("Committing Compaction {} finished with result {}.", compactionCommitTime, metadata);
FlinkCompactHelpers.newInstance().completeInflightCompaction(table, compactionCommitTime, metadata);
CompactHelpers.getInstance().completeInflightCompaction(table, compactionCommitTime, metadata);
if (compactionTimer != null) {
long durationInMs = metrics.getDurationInMs(compactionTimer.stop());
@@ -383,7 +384,8 @@ public class HoodieFlinkWriteClient<T extends HoodieRecordPayload> extends
protected List<WriteStatus> compact(String compactionInstantTime, boolean shouldComplete) {
// only used for metadata table, the compaction happens in single thread
try {
List<WriteStatus> writeStatuses = FlinkCompactHelpers.compact(compactionInstantTime, this);
List<WriteStatus> writeStatuses =
getHoodieTable().compact(context, compactionInstantTime).getWriteStatuses();
commitCompaction(compactionInstantTime, writeStatuses, Option.empty());
return writeStatuses;
} catch (IOException e) {

20
hudi-client/hudi-flink-client/src/main/java/org/apache/hudi/client/common/HoodieFlinkEngineContext.java
View File

@@ -20,6 +20,10 @@ package org.apache.hudi.client.common;
import org.apache.hudi.client.FlinkTaskContextSupplier;
import org.apache.hudi.common.config.SerializableConfiguration;
import org.apache.hudi.common.data.HoodieAccumulator;
import org.apache.hudi.common.data.HoodieAtomicLongAccumulator;
import org.apache.hudi.common.data.HoodieData;
import org.apache.hudi.common.data.HoodieList;
import org.apache.hudi.common.engine.EngineProperty;
import org.apache.hudi.common.engine.HoodieEngineContext;
import org.apache.hudi.common.engine.TaskContextSupplier;
@@ -32,6 +36,7 @@ import org.apache.hudi.common.util.Option;
import org.apache.flink.api.common.functions.RuntimeContext;
import java.util.Collections;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
@@ -72,6 +77,21 @@ public class HoodieFlinkEngineContext extends HoodieEngineContext {
this.runtimeContext = ((FlinkTaskContextSupplier) taskContextSupplier).getFlinkRuntimeContext();
}
@Override
public HoodieAccumulator newAccumulator() {
return HoodieAtomicLongAccumulator.create();
}
@Override
public <T> HoodieData<T> emptyHoodieData() {
return HoodieList.of(Collections.emptyList());
}
@Override
public <T> HoodieData<T> parallelize(List<T> data) {
return HoodieList.of(data);
}
public RuntimeContext getRuntimeContext() {
return this.runtimeContext;
}

21
hudi-client/hudi-flink-client/src/main/java/org/apache/hudi/table/HoodieFlinkCopyOnWriteTable.java
View File

@@ -74,7 +74,8 @@ import java.util.Map;
* <p>
* UPDATES - Produce a new version of the file, just replacing the updated records with new values
*/
public class HoodieFlinkCopyOnWriteTable<T extends HoodieRecordPayload> extends HoodieFlinkTable<T> {
public class HoodieFlinkCopyOnWriteTable<T extends HoodieRecordPayload>
extends HoodieFlinkTable<T> implements HoodieCompactionHandler<T> {
private static final Logger LOG = LoggerFactory.getLogger(HoodieFlinkCopyOnWriteTable.class);
@@ -265,7 +266,8 @@ public class HoodieFlinkCopyOnWriteTable<T extends HoodieRecordPayload> extends
}
@Override
public HoodieWriteMetadata<List<WriteStatus>> compact(HoodieEngineContext context, String compactionInstantTime) {
public HoodieWriteMetadata<List<WriteStatus>> compact(
HoodieEngineContext context, String compactionInstantTime) {
throw new HoodieNotSupportedException("Compaction is not supported on a CopyOnWrite table");
}
@@ -329,9 +331,10 @@ public class HoodieFlinkCopyOnWriteTable<T extends HoodieRecordPayload> extends
// -------------------------------------------------------------------------
// Used for compaction
// -------------------------------------------------------------------------
public Iterator<List<WriteStatus>> handleUpdate(String instantTime, String partitionPath, String fileId,
Map<String, HoodieRecord<T>> keyToNewRecords, HoodieBaseFile oldDataFile) throws IOException {
@Override
public Iterator<List<WriteStatus>> handleUpdate(
String instantTime, String partitionPath, String fileId,
Map<String, HoodieRecord<T>> keyToNewRecords, HoodieBaseFile oldDataFile) throws IOException {
// these are updates
HoodieMergeHandle upsertHandle = getUpdateHandle(instantTime, partitionPath, fileId, keyToNewRecords, oldDataFile);
return handleUpdateInternal(upsertHandle, instantTime, fileId);
@@ -366,9 +369,11 @@ public class HoodieFlinkCopyOnWriteTable<T extends HoodieRecordPayload> extends
}
}
public Iterator<List<WriteStatus>> handleInsert(String instantTime, String partitionPath, String fileId,
Map<String, HoodieRecord<? extends HoodieRecordPayload>> recordMap) {
HoodieCreateHandle<?,?,?,?> createHandle =
@Override
public Iterator<List<WriteStatus>> handleInsert(
String instantTime, String partitionPath, String fileId,
Map<String, HoodieRecord<? extends HoodieRecordPayload>> recordMap) {
HoodieCreateHandle<?, ?, ?, ?> createHandle =
new HoodieCreateHandle(config, instantTime, this, partitionPath, fileId, recordMap, taskContextSupplier);
createHandle.write();
return Collections.singletonList(createHandle.close()).iterator();

20
hudi-client/hudi-flink-client/src/main/java/org/apache/hudi/table/HoodieFlinkMergeOnReadTable.java
View File

@@ -30,14 +30,14 @@ import org.apache.hudi.common.table.timeline.HoodieInstant;
import org.apache.hudi.common.util.Option;
import org.apache.hudi.common.util.ValidationUtils;
import org.apache.hudi.config.HoodieWriteConfig;
import org.apache.hudi.exception.HoodieNotSupportedException;
import org.apache.hudi.io.FlinkAppendHandle;
import org.apache.hudi.io.HoodieWriteHandle;
import org.apache.hudi.table.action.HoodieWriteMetadata;
import org.apache.hudi.table.action.commit.delta.FlinkUpsertDeltaCommitActionExecutor;
import org.apache.hudi.table.action.commit.delta.FlinkUpsertPreppedDeltaCommitActionExecutor;
import org.apache.hudi.table.action.compact.BaseScheduleCompactionActionExecutor;
import org.apache.hudi.table.action.compact.FlinkScheduleCompactionActionExecutor;
import org.apache.hudi.table.action.compact.HoodieFlinkMergeOnReadTableCompactor;
import org.apache.hudi.table.action.compact.RunCompactionActionExecutor;
import org.apache.hudi.table.action.compact.ScheduleCompactionActionExecutor;
import org.apache.hudi.table.action.rollback.BaseRollbackPlanActionExecutor;
import org.apache.hudi.table.action.rollback.MergeOnReadRollbackActionExecutor;
@@ -97,15 +97,19 @@ public class HoodieFlinkMergeOnReadTable<T extends HoodieRecordPayload>
HoodieEngineContext context,
String instantTime,
Option<Map<String, String>> extraMetadata) {
BaseScheduleCompactionActionExecutor scheduleCompactionExecutor = new FlinkScheduleCompactionActionExecutor(
context, config, this, instantTime, extraMetadata);
ScheduleCompactionActionExecutor scheduleCompactionExecutor = new ScheduleCompactionActionExecutor(
context, config, this, instantTime, extraMetadata,
new HoodieFlinkMergeOnReadTableCompactor());
return scheduleCompactionExecutor.execute();
}
@Override
public HoodieWriteMetadata<List<WriteStatus>> compact(HoodieEngineContext context, String compactionInstantTime) {
throw new HoodieNotSupportedException("Compaction is supported as a separate pipeline, "
+ "should not invoke directly through HoodieFlinkMergeOnReadTable");
public HoodieWriteMetadata<List<WriteStatus>> compact(
HoodieEngineContext context, String compactionInstantTime) {
RunCompactionActionExecutor compactionExecutor = new RunCompactionActionExecutor(
context, config, this, compactionInstantTime, new HoodieFlinkMergeOnReadTableCompactor(),
new HoodieFlinkCopyOnWriteTable(config, context, getMetaClient()));
return convertMetadata(compactionExecutor.execute());
}
@Override

9
hudi-client/hudi-flink-client/src/main/java/org/apache/hudi/table/HoodieFlinkTable.java
View File

@@ -20,6 +20,7 @@ package org.apache.hudi.table;
import org.apache.hudi.client.WriteStatus;
import org.apache.hudi.client.common.HoodieFlinkEngineContext;
import org.apache.hudi.common.data.HoodieData;
import org.apache.hudi.common.engine.HoodieEngineContext;
import org.apache.hudi.common.model.HoodieKey;
import org.apache.hudi.common.model.HoodieRecord;
@@ -35,12 +36,15 @@ import org.apache.hudi.index.HoodieIndex;
import org.apache.hudi.metadata.FlinkHoodieBackedTableMetadataWriter;
import org.apache.hudi.metadata.HoodieTableMetadata;
import org.apache.hudi.metadata.HoodieTableMetadataWriter;
import org.apache.hudi.table.action.HoodieWriteMetadata;
import org.apache.hadoop.fs.Path;
import java.io.IOException;
import java.util.List;
import static org.apache.hudi.common.data.HoodieList.getList;
public abstract class HoodieFlinkTable<T extends HoodieRecordPayload>
extends HoodieTable<T, List<HoodieRecord<T>>, List<HoodieKey>, List<WriteStatus>>
implements ExplicitWriteHandleTable<T> {
@@ -87,6 +91,11 @@ public abstract class HoodieFlinkTable<T extends HoodieRecordPayload>
return hoodieFlinkTable;
}
public static HoodieWriteMetadata<List<WriteStatus>> convertMetadata(
HoodieWriteMetadata<HoodieData<WriteStatus>> metadata) {
return metadata.clone(getList(metadata.getWriteStatuses()));
}
@Override
protected HoodieIndex<T, List<HoodieRecord<T>>, List<HoodieKey>, List<WriteStatus>> getIndex(HoodieWriteConfig config, HoodieEngineContext context) {
return FlinkHoodieIndex.createIndex((HoodieFlinkEngineContext) context, config);

147
hudi-client/hudi-flink-client/src/main/java/org/apache/hudi/table/action/compact/FlinkCompactHelpers.java
View File

@@ -1,147 +0,0 @@
/*
* 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.hudi.table.action.compact;
import org.apache.hudi.avro.model.HoodieCompactionPlan;
import org.apache.hudi.client.HoodieFlinkWriteClient;
import org.apache.hudi.client.WriteStatus;
import org.apache.hudi.common.model.CompactionOperation;
import org.apache.hudi.common.model.HoodieCommitMetadata;
import org.apache.hudi.common.model.HoodieKey;
import org.apache.hudi.common.model.HoodieRecord;
import org.apache.hudi.common.model.HoodieRecordPayload;
import org.apache.hudi.common.model.HoodieWriteStat;
import org.apache.hudi.common.table.timeline.HoodieInstant;
import org.apache.hudi.common.table.timeline.HoodieTimeline;
import org.apache.hudi.common.table.timeline.TimelineMetadataUtils;
import org.apache.hudi.common.util.CompactionUtils;
import org.apache.hudi.table.HoodieFlinkCopyOnWriteTable;
import org.apache.hudi.table.HoodieFlinkTable;
import org.apache.hudi.table.HoodieTable;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.stream.Collectors;
import static java.util.stream.Collectors.toList;
/**
* A flink implementation of {@link AbstractCompactHelpers}.
*
* @param <T>
*/
public class FlinkCompactHelpers<T extends HoodieRecordPayload> extends
AbstractCompactHelpers<T, List<HoodieRecord<T>>, List<HoodieKey>, List<WriteStatus>> {
private static final Logger LOG = LoggerFactory.getLogger(FlinkCompactHelpers.class);
private FlinkCompactHelpers() {
}
private static class CompactHelperHolder {
private static final FlinkCompactHelpers FLINK_COMPACT_HELPERS = new FlinkCompactHelpers();
}
public static FlinkCompactHelpers newInstance() {
return CompactHelperHolder.FLINK_COMPACT_HELPERS;
}
@Override
public HoodieCommitMetadata createCompactionMetadata(HoodieTable<T, List<HoodieRecord<T>>, List<HoodieKey>, List<WriteStatus>> table,
String compactionInstantTime,
List<WriteStatus> writeStatuses,
String schema) throws IOException {
byte[] planBytes = table.getActiveTimeline().readCompactionPlanAsBytes(
HoodieTimeline.getCompactionRequestedInstant(compactionInstantTime)).get();
HoodieCompactionPlan compactionPlan = TimelineMetadataUtils.deserializeCompactionPlan(planBytes);
List<HoodieWriteStat> updateStatusMap = writeStatuses.stream().map(WriteStatus::getStat).collect(Collectors.toList());
org.apache.hudi.common.model.HoodieCommitMetadata metadata = new org.apache.hudi.common.model.HoodieCommitMetadata(true);
for (HoodieWriteStat stat : updateStatusMap) {
metadata.addWriteStat(stat.getPartitionPath(), stat);
}
metadata.addMetadata(org.apache.hudi.common.model.HoodieCommitMetadata.SCHEMA_KEY, schema);
if (compactionPlan.getExtraMetadata() != null) {
compactionPlan.getExtraMetadata().forEach(metadata::addMetadata);
}
return metadata;
}
@SuppressWarnings("unchecked, rawtypes")
public static List<WriteStatus> compact(
HoodieFlinkWriteClient writeClient,
String compactInstantTime,
CompactionOperation compactionOperation) throws IOException {
HoodieFlinkMergeOnReadTableCompactor compactor = new HoodieFlinkMergeOnReadTableCompactor();
return compactor.compact(
new HoodieFlinkCopyOnWriteTable<>(
writeClient.getConfig(),
writeClient.getEngineContext(),
writeClient.getHoodieTable().getMetaClient()),
writeClient.getHoodieTable().getMetaClient(),
writeClient.getConfig(),
compactionOperation,
compactInstantTime);
}
/**
* Called by the metadata table compactor code path.
*/
@SuppressWarnings("unchecked, rawtypes")
public static List<WriteStatus> compact(String compactionInstantTime, HoodieFlinkWriteClient writeClient) throws IOException {
HoodieFlinkTable table = writeClient.getHoodieTable();
HoodieTimeline pendingCompactionTimeline = table.getActiveTimeline().filterPendingCompactionTimeline();
HoodieInstant inflightInstant = HoodieTimeline.getCompactionInflightInstant(compactionInstantTime);
if (pendingCompactionTimeline.containsInstant(inflightInstant)) {
writeClient.rollbackInflightCompaction(inflightInstant, table);
table.getMetaClient().reloadActiveTimeline();
}
// generate compaction plan
// should support configurable commit metadata
HoodieCompactionPlan compactionPlan = CompactionUtils.getCompactionPlan(
table.getMetaClient(), compactionInstantTime);
if (compactionPlan == null || (compactionPlan.getOperations() == null)
|| (compactionPlan.getOperations().isEmpty())) {
// do nothing.
LOG.info("No compaction plan for instant " + compactionInstantTime);
return Collections.emptyList();
} else {
HoodieInstant instant = HoodieTimeline.getCompactionRequestedInstant(compactionInstantTime);
// Mark instant as compaction inflight
table.getActiveTimeline().transitionCompactionRequestedToInflight(instant);
table.getMetaClient().reloadActiveTimeline();
List<CompactionOperation> operations = compactionPlan.getOperations().stream()
.map(CompactionOperation::convertFromAvroRecordInstance).collect(toList());
LOG.info("Compacting " + operations + " files");
List<WriteStatus> writeStatusList = new ArrayList<>();
for (CompactionOperation operation : operations) {
List<WriteStatus> statuses = compact(writeClient, compactionInstantTime, operation);
writeStatusList.addAll(statuses);
}
return writeStatusList;
}
}
}

156
hudi-client/hudi-flink-client/src/main/java/org/apache/hudi/table/action/compact/FlinkScheduleCompactionActionExecutor.java
View File

@@ -1,156 +0,0 @@
/*
* 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.hudi.table.action.compact;
import org.apache.hudi.avro.model.HoodieCompactionPlan;
import org.apache.hudi.client.WriteStatus;
import org.apache.hudi.common.engine.HoodieEngineContext;
import org.apache.hudi.common.model.HoodieFileGroupId;
import org.apache.hudi.common.model.HoodieKey;
import org.apache.hudi.common.model.HoodieRecord;
import org.apache.hudi.common.model.HoodieRecordPayload;
import org.apache.hudi.common.table.timeline.HoodieActiveTimeline;
import org.apache.hudi.common.table.timeline.HoodieInstant;
import org.apache.hudi.common.table.timeline.HoodieTimeline;
import org.apache.hudi.common.table.view.SyncableFileSystemView;
import org.apache.hudi.common.util.Option;
import org.apache.hudi.common.util.collection.Pair;
import org.apache.hudi.config.HoodieWriteConfig;
import org.apache.hudi.exception.HoodieCompactionException;
import org.apache.hudi.table.HoodieTable;
import org.apache.log4j.LogManager;
import org.apache.log4j.Logger;
import java.io.IOException;
import java.text.ParseException;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.stream.Collectors;
@SuppressWarnings("checkstyle:LineLength")
public class FlinkScheduleCompactionActionExecutor<T extends HoodieRecordPayload> extends
BaseScheduleCompactionActionExecutor<T, List<HoodieRecord<T>>, List<HoodieKey>, List<WriteStatus>> {
private static final Logger LOG = LogManager.getLogger(FlinkScheduleCompactionActionExecutor.class);
private final Option<Map<String, String>> extraMetadata;
public FlinkScheduleCompactionActionExecutor(HoodieEngineContext context,
HoodieWriteConfig config,
HoodieTable<T, List<HoodieRecord<T>>, List<HoodieKey>, List<WriteStatus>> table,
String instantTime,
Option<Map<String, String>> extraMetadata) {
super(context, config, table, instantTime, extraMetadata);
this.extraMetadata = extraMetadata;
}
@Override
protected HoodieCompactionPlan scheduleCompaction() {
LOG.info("Checking if compaction needs to be run on " + config.getBasePath());
// judge if we need to compact according to num delta commits and time elapsed
boolean compactable = needCompact(config.getInlineCompactTriggerStrategy());
if (compactable) {
LOG.info("Generating compaction plan for merge on read table " + config.getBasePath());
HoodieFlinkMergeOnReadTableCompactor compactor = new HoodieFlinkMergeOnReadTableCompactor();
try {
SyncableFileSystemView fileSystemView = (SyncableFileSystemView) table.getSliceView();
Set<HoodieFileGroupId> fgInPendingCompactionAndClustering = fileSystemView.getPendingCompactionOperations()
.map(instantTimeOpPair -> instantTimeOpPair.getValue().getFileGroupId())
.collect(Collectors.toSet());
// exclude files in pending clustering from compaction.
fgInPendingCompactionAndClustering.addAll(fileSystemView.getFileGroupsInPendingClustering().map(Pair::getLeft).collect(Collectors.toSet()));
return compactor.generateCompactionPlan(context, table, config, instantTime, fgInPendingCompactionAndClustering);
} catch (IOException e) {
throw new HoodieCompactionException("Could not schedule compaction " + config.getBasePath(), e);
}
}
return new HoodieCompactionPlan();
}
public Pair<Integer, String> getLatestDeltaCommitInfo(CompactionTriggerStrategy compactionTriggerStrategy) {
Option<HoodieInstant> lastCompaction = table.getActiveTimeline().getCommitTimeline()
.filterCompletedInstants().lastInstant();
HoodieTimeline deltaCommits = table.getActiveTimeline().getDeltaCommitTimeline();
String latestInstantTs;
int deltaCommitsSinceLastCompaction = 0;
if (lastCompaction.isPresent()) {
latestInstantTs = lastCompaction.get().getTimestamp();
deltaCommitsSinceLastCompaction = deltaCommits.findInstantsAfter(latestInstantTs, Integer.MAX_VALUE).countInstants();
} else {
latestInstantTs = deltaCommits.firstInstant().get().getTimestamp();
deltaCommitsSinceLastCompaction = deltaCommits.findInstantsAfterOrEquals(latestInstantTs, Integer.MAX_VALUE).countInstants();
}
return Pair.of(deltaCommitsSinceLastCompaction, latestInstantTs);
}
public boolean needCompact(CompactionTriggerStrategy compactionTriggerStrategy) {
boolean compactable;
// get deltaCommitsSinceLastCompaction and lastCompactionTs
Pair<Integer, String> latestDeltaCommitInfo = getLatestDeltaCommitInfo(compactionTriggerStrategy);
int inlineCompactDeltaCommitMax = config.getInlineCompactDeltaCommitMax();
int inlineCompactDeltaSecondsMax = config.getInlineCompactDeltaSecondsMax();
switch (compactionTriggerStrategy) {
case NUM_COMMITS:
compactable = inlineCompactDeltaCommitMax <= latestDeltaCommitInfo.getLeft();
if (compactable) {
LOG.info(String.format("The delta commits >= %s, trigger compaction scheduler.", inlineCompactDeltaCommitMax));
}
break;
case TIME_ELAPSED:
compactable = inlineCompactDeltaSecondsMax <= parsedToSeconds(instantTime) - parsedToSeconds(latestDeltaCommitInfo.getRight());
if (compactable) {
LOG.info(String.format("The elapsed time >=%ss, trigger compaction scheduler.", inlineCompactDeltaSecondsMax));
}
break;
case NUM_OR_TIME:
compactable = inlineCompactDeltaCommitMax <= latestDeltaCommitInfo.getLeft()
|| inlineCompactDeltaSecondsMax <= parsedToSeconds(instantTime) - parsedToSeconds(latestDeltaCommitInfo.getRight());
if (compactable) {
LOG.info(String.format("The delta commits >= %s or elapsed_time >=%ss, trigger compaction scheduler.", inlineCompactDeltaCommitMax,
inlineCompactDeltaSecondsMax));
}
break;
case NUM_AND_TIME:
compactable = inlineCompactDeltaCommitMax <= latestDeltaCommitInfo.getLeft()
&& inlineCompactDeltaSecondsMax <= parsedToSeconds(instantTime) - parsedToSeconds(latestDeltaCommitInfo.getRight());
if (compactable) {
LOG.info(String.format("The delta commits >= %s and elapsed_time >=%ss, trigger compaction scheduler.", inlineCompactDeltaCommitMax,
inlineCompactDeltaSecondsMax));
}
break;
default:
throw new HoodieCompactionException("Unsupported compaction trigger strategy: " + config.getInlineCompactTriggerStrategy());
}
return compactable;
}
public Long parsedToSeconds(String time) {
long timestamp;
try {
timestamp = HoodieActiveTimeline.COMMIT_FORMATTER.parse(time).getTime() / 1000;
} catch (ParseException e) {
throw new HoodieCompactionException(e.getMessage(), e);
}
return timestamp;
}
}

204
hudi-client/hudi-flink-client/src/main/java/org/apache/hudi/table/action/compact/HoodieFlinkMergeOnReadTableCompactor.java
View File

@@ -18,54 +18,17 @@
package org.apache.hudi.table.action.compact;
import org.apache.hudi.avro.HoodieAvroUtils;
import org.apache.hudi.avro.model.HoodieCompactionOperation;
import org.apache.hudi.avro.model.HoodieCompactionPlan;
import org.apache.hudi.client.FlinkTaskContextSupplier;
import org.apache.hudi.client.WriteStatus;
import org.apache.hudi.common.engine.HoodieEngineContext;
import org.apache.hudi.common.fs.FSUtils;
import org.apache.hudi.common.model.CompactionOperation;
import org.apache.hudi.common.model.HoodieBaseFile;
import org.apache.hudi.common.model.HoodieFileGroupId;
import org.apache.hudi.common.data.HoodieData;
import org.apache.hudi.common.model.HoodieKey;
import org.apache.hudi.common.model.HoodieLogFile;
import org.apache.hudi.common.model.HoodieRecord;
import org.apache.hudi.common.model.HoodieRecordPayload;
import org.apache.hudi.common.model.HoodieTableType;
import org.apache.hudi.common.model.HoodieWriteStat.RuntimeStats;
import org.apache.hudi.common.table.HoodieTableMetaClient;
import org.apache.hudi.common.table.log.HoodieMergedLogRecordScanner;
import org.apache.hudi.common.table.timeline.HoodieInstant;
import org.apache.hudi.common.table.timeline.HoodieTimeline;
import org.apache.hudi.common.table.view.TableFileSystemView.SliceView;
import org.apache.hudi.common.util.CollectionUtils;
import org.apache.hudi.common.util.CompactionUtils;
import org.apache.hudi.common.util.Option;
import org.apache.hudi.common.util.ValidationUtils;
import org.apache.hudi.common.util.collection.Pair;
import org.apache.hudi.config.HoodieWriteConfig;
import org.apache.hudi.io.IOUtils;
import org.apache.hudi.table.HoodieFlinkCopyOnWriteTable;
import org.apache.hudi.table.HoodieTable;
import org.apache.hudi.table.action.compact.strategy.CompactionStrategy;
import org.apache.avro.Schema;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.log4j.LogManager;
import org.apache.log4j.Logger;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Iterator;
import java.util.List;
import java.util.Set;
import java.util.concurrent.atomic.AtomicLong;
import java.util.stream.Collectors;
import java.util.stream.StreamSupport;
import static java.util.stream.Collectors.toList;
/**
* Compacts a hoodie table with merge on read storage. Computes all possible compactions,
@@ -75,164 +38,21 @@ import static java.util.stream.Collectors.toList;
* <p>Note: the compaction logic is invoked through the flink pipeline.
*/
@SuppressWarnings("checkstyle:LineLength")
public class HoodieFlinkMergeOnReadTableCompactor<T extends HoodieRecordPayload> implements HoodieCompactor<T, List<HoodieRecord<T>>, List<HoodieKey>, List<WriteStatus>> {
private static final Logger LOG = LogManager.getLogger(HoodieFlinkMergeOnReadTableCompactor.class);
// Accumulator to keep track of total log files for a table
private AtomicLong totalLogFiles;
// Accumulator to keep track of total log file slices for a table
private AtomicLong totalFileSlices;
public class HoodieFlinkMergeOnReadTableCompactor<T extends HoodieRecordPayload>
extends HoodieCompactor<T, List<HoodieRecord<T>>, List<HoodieKey>, List<WriteStatus>> {
@Override
public List<WriteStatus> compact(HoodieEngineContext context, HoodieCompactionPlan compactionPlan,
HoodieTable hoodieTable, HoodieWriteConfig config, String compactionInstantTime) throws IOException {
throw new UnsupportedOperationException("HoodieFlinkMergeOnReadTableCompactor does not support compact directly, "
+ "the function works as a separate pipeline");
}
public List<WriteStatus> compact(HoodieFlinkCopyOnWriteTable hoodieCopyOnWriteTable,
HoodieTableMetaClient metaClient,
HoodieWriteConfig config,
CompactionOperation operation,
String instantTime) throws IOException {
FileSystem fs = metaClient.getFs();
Schema readerSchema = HoodieAvroUtils.addMetadataFields(new Schema.Parser().parse(config.getSchema()), config.allowOperationMetadataField());
LOG.info("Compacting base " + operation.getDataFileName() + " with delta files " + operation.getDeltaFileNames()
+ " for commit " + instantTime);
// TODO - FIX THIS
// Reads the entire avro file. Always only specific blocks should be read from the avro file
// (failure recover).
// Load all the delta commits since the last compaction commit and get all the blocks to be
// loaded and load it using CompositeAvroLogReader
// Since a DeltaCommit is not defined yet, reading all the records. revisit this soon.
String maxInstantTime = metaClient
.getActiveTimeline().getTimelineOfActions(CollectionUtils.createSet(HoodieTimeline.COMMIT_ACTION,
HoodieTimeline.ROLLBACK_ACTION, HoodieTimeline.DELTA_COMMIT_ACTION))
.filterCompletedInstants().lastInstant().get().getTimestamp();
// TODO(danny): make it configurable
long maxMemoryPerCompaction = IOUtils.getMaxMemoryPerCompaction(new FlinkTaskContextSupplier(null), config);
LOG.info("MaxMemoryPerCompaction => " + maxMemoryPerCompaction);
List<String> logFiles = operation.getDeltaFileNames().stream().map(
p -> new Path(FSUtils.getPartitionPath(metaClient.getBasePath(), operation.getPartitionPath()), p).toString())
.collect(toList());
HoodieMergedLogRecordScanner scanner = HoodieMergedLogRecordScanner.newBuilder()
.withFileSystem(fs)
.withBasePath(metaClient.getBasePath())
.withLogFilePaths(logFiles)
.withReaderSchema(readerSchema)
.withLatestInstantTime(maxInstantTime)
.withMaxMemorySizeInBytes(maxMemoryPerCompaction)
.withReadBlocksLazily(config.getCompactionLazyBlockReadEnabled())
.withReverseReader(config.getCompactionReverseLogReadEnabled())
.withBufferSize(config.getMaxDFSStreamBufferSize())
.withSpillableMapBasePath(config.getSpillableMapBasePath())
.withDiskMapType(config.getCommonConfig().getSpillableDiskMapType())
.withBitCaskDiskMapCompressionEnabled(config.getCommonConfig().isBitCaskDiskMapCompressionEnabled())
.build();
if (!scanner.iterator().hasNext()) {
return new ArrayList<>();
public void preCompact(
HoodieTable table, HoodieTimeline pendingCompactionTimeline, String compactionInstantTime) {
HoodieInstant inflightInstant = HoodieTimeline.getCompactionInflightInstant(compactionInstantTime);
if (pendingCompactionTimeline.containsInstant(inflightInstant)) {
table.rollbackInflightCompaction(inflightInstant);
table.getMetaClient().reloadActiveTimeline();
}
Option<HoodieBaseFile> oldDataFileOpt =
operation.getBaseFile(metaClient.getBasePath(), operation.getPartitionPath());
// Compacting is very similar to applying updates to existing file
Iterator<List<WriteStatus>> result;
// If the dataFile is present, perform updates else perform inserts into a new base file.
if (oldDataFileOpt.isPresent()) {
result = hoodieCopyOnWriteTable.handleUpdate(instantTime, operation.getPartitionPath(),
operation.getFileId(), scanner.getRecords(),
oldDataFileOpt.get());
} else {
result = hoodieCopyOnWriteTable.handleInsert(instantTime, operation.getPartitionPath(), operation.getFileId(),
scanner.getRecords());
}
Iterable<List<WriteStatus>> resultIterable = () -> result;
return StreamSupport.stream(resultIterable.spliterator(), false).flatMap(Collection::stream).peek(s -> {
s.getStat().setTotalUpdatedRecordsCompacted(scanner.getNumMergedRecordsInLog());
s.getStat().setTotalLogFilesCompacted(scanner.getTotalLogFiles());
s.getStat().setTotalLogRecords(scanner.getTotalLogRecords());
s.getStat().setPartitionPath(operation.getPartitionPath());
s.getStat()
.setTotalLogSizeCompacted(operation.getMetrics().get(CompactionStrategy.TOTAL_LOG_FILE_SIZE).longValue());
s.getStat().setTotalLogBlocks(scanner.getTotalLogBlocks());
s.getStat().setTotalCorruptLogBlock(scanner.getTotalCorruptBlocks());
s.getStat().setTotalRollbackBlocks(scanner.getTotalRollbacks());
RuntimeStats runtimeStats = new RuntimeStats();
runtimeStats.setTotalScanTime(scanner.getTotalTimeTakenToReadAndMergeBlocks());
s.getStat().setRuntimeStats(runtimeStats);
scanner.close();
}).collect(toList());
}
@Override
public HoodieCompactionPlan generateCompactionPlan(HoodieEngineContext context,
HoodieTable<T, List<HoodieRecord<T>>, List<HoodieKey>, List<WriteStatus>> hoodieTable,
HoodieWriteConfig config, String compactionCommitTime,
Set<HoodieFileGroupId> fgIdsInPendingCompactionAndClustering)
throws IOException {
totalLogFiles = new AtomicLong(0);
totalFileSlices = new AtomicLong(0);
ValidationUtils.checkArgument(hoodieTable.getMetaClient().getTableType() == HoodieTableType.MERGE_ON_READ,
"Can only compact table of type " + HoodieTableType.MERGE_ON_READ + " and not "
+ hoodieTable.getMetaClient().getTableType().name());
// TODO : check if maxMemory is not greater than JVM or flink.executor memory
// TODO - rollback any compactions in flight
HoodieTableMetaClient metaClient = hoodieTable.getMetaClient();
LOG.info("Compacting " + metaClient.getBasePath() + " with commit " + compactionCommitTime);
List<String> partitionPaths = FSUtils.getAllPartitionPaths(context, config.getMetadataConfig(), metaClient.getBasePath());
// filter the partition paths if needed to reduce list status
partitionPaths = config.getCompactionStrategy().filterPartitionPaths(config, partitionPaths);
if (partitionPaths.isEmpty()) {
// In case no partitions could be picked, return no compaction plan
return null;
}
SliceView fileSystemView = hoodieTable.getSliceView();
LOG.info("Compaction looking for files to compact in " + partitionPaths + " partitions");
context.setJobStatus(this.getClass().getSimpleName(), "Looking for files to compact");
List<HoodieCompactionOperation> operations = context.flatMap(partitionPaths, partitionPath -> fileSystemView
.getLatestFileSlices(partitionPath)
.filter(slice -> !fgIdsInPendingCompactionAndClustering.contains(slice.getFileGroupId()))
.map(s -> {
List<HoodieLogFile> logFiles =
s.getLogFiles().sorted(HoodieLogFile.getLogFileComparator()).collect(Collectors.toList());
totalLogFiles.addAndGet(logFiles.size());
totalFileSlices.addAndGet(1L);
// Avro generated classes are not inheriting Serializable. Using CompactionOperation POJO
// for flink Map operations and collecting them finally in Avro generated classes for storing
// into meta files.
Option<HoodieBaseFile> dataFile = s.getBaseFile();
return new CompactionOperation(dataFile, partitionPath, logFiles,
config.getCompactionStrategy().captureMetrics(config, s));
})
.filter(c -> !c.getDeltaFileNames().isEmpty()), partitionPaths.size()).stream().map(CompactionUtils::buildHoodieCompactionOperation).collect(toList());
LOG.info("Total of " + operations.size() + " compactions are retrieved");
LOG.info("Total number of latest files slices " + totalFileSlices.get());
LOG.info("Total number of log files " + totalLogFiles.get());
LOG.info("Total number of file slices " + totalFileSlices.get());
// Filter the compactions with the passed in filter. This lets us choose most effective
// compactions only
HoodieCompactionPlan compactionPlan = config.getCompactionStrategy().generateCompactionPlan(config, operations,
CompactionUtils.getAllPendingCompactionPlans(metaClient).stream().map(Pair::getValue).collect(toList()));
ValidationUtils.checkArgument(
compactionPlan.getOperations().stream().noneMatch(
op -> fgIdsInPendingCompactionAndClustering.contains(new HoodieFileGroupId(op.getPartitionPath(), op.getFileId()))),
"Bad Compaction Plan. FileId MUST NOT have multiple pending compactions. "
+ "Please fix your strategy implementation. FileIdsWithPendingCompactions :" + fgIdsInPendingCompactionAndClustering
+ ", Selected workload :" + compactionPlan);
if (compactionPlan.getOperations().isEmpty()) {
LOG.warn("After filtering, Nothing to compact for " + metaClient.getBasePath());
}
return compactionPlan;
public void maybePersist(HoodieData<WriteStatus> writeStatus, HoodieWriteConfig config) {
// No OP
}
}