lanyuanxiaoyao/hudi
1
0
Fork 0
Code Activity

[HUDI-3356][HUDI-3203] HoodieData for metadata index records; BloomFilter construction from index based on the type param (#4848)

Browse Source 
Rework of #4761 
This diff introduces following changes:

- Write stats are converted to metadata index records during the commit. Making them use the HoodieData type so that the record generation scales up with needs. 
- Metadata index init support for bloom filter and column stats partitions.
- When building the BloomFilter from the index records, using the type param stored in the payload instead of hardcoded type.
- Delta writes can change column ranges and the column stats index need to be properly updated with new ranges to be consistent with the table dataset. This fix add column stats index update support for the delta writes.

Co-authored-by: Manoj Govindassamy <manoj.govindassamy@gmail.com>
This commit is contained in:
Sagar Sumit
2022-03-08 21:09:04 +05:30
committed by GitHub
parent ed26c5265c
commit 575bc63468
24 changed files with 1051 additions and 533 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
hudi-client/hudi-client-common/src/main/java/org/apache/hudi/config/HoodieWriteConfig.java
View File

@@ -1492,6 +1492,10 @@ public class HoodieWriteConfig extends HoodieConfig {
return isMetadataTableEnabled() && getMetadataConfig().isMetadataColumnStatsIndexForAllColumnsEnabled();
}
public int getColumnStatsIndexParallelism() {
return metadataConfig.getColumnStatsIndexParallelism();
}
public int getBloomIndexKeysPerBucket() {
return getInt(HoodieIndexConfig.BLOOM_INDEX_KEYS_PER_BUCKET);
}

23
hudi-client/hudi-client-common/src/main/java/org/apache/hudi/io/HoodieAppendHandle.java
View File

@@ -24,6 +24,7 @@ import org.apache.hudi.common.engine.TaskContextSupplier;
import org.apache.hudi.common.fs.FSUtils;
import org.apache.hudi.common.model.BaseFile;
import org.apache.hudi.common.model.FileSlice;
import org.apache.hudi.common.model.HoodieColumnRangeMetadata;
import org.apache.hudi.common.model.HoodieDeltaWriteStat;
import org.apache.hudi.common.model.HoodieKey;
import org.apache.hudi.common.model.HoodieLogFile;
@@ -71,6 +72,9 @@ import java.util.Properties;
import java.util.concurrent.atomic.AtomicLong;
import java.util.stream.Collectors;
import static org.apache.hudi.metadata.HoodieTableMetadataUtil.accumulateColumnRanges;
import static org.apache.hudi.metadata.HoodieTableMetadataUtil.aggregateColumnStats;
/**
* IO Operation to append data onto an existing file.
*/
@@ -320,7 +324,7 @@ public class HoodieAppendHandle<T extends HoodieRecordPayload, I, K, O> extends
statuses.add(this.writeStatus);
}
private void processAppendResult(AppendResult result) {
private void processAppendResult(AppendResult result, List<IndexedRecord> recordList) {
HoodieDeltaWriteStat stat = (HoodieDeltaWriteStat) this.writeStatus.getStat();
if (stat.getPath() == null) {
@@ -339,6 +343,19 @@ public class HoodieAppendHandle<T extends HoodieRecordPayload, I, K, O> extends
updateWriteStatus(stat, result);
}
if (config.isMetadataIndexColumnStatsForAllColumnsEnabled()) {
Map<String, HoodieColumnRangeMetadata<Comparable>> columnRangeMap = stat.getRecordsStats().isPresent()
? stat.getRecordsStats().get().getStats() : new HashMap<>();
final String filePath = stat.getPath();
// initialize map of column name to map of stats name to stats value
Map<String, Map<String, Object>> columnToStats = new HashMap<>();
writeSchemaWithMetaFields.getFields().forEach(field -> columnToStats.putIfAbsent(field.name(), new HashMap<>()));
// collect stats for columns at once per record and keep iterating through every record to eventually find col stats for all fields.
recordList.forEach(record -> aggregateColumnStats(record, writeSchemaWithMetaFields, columnToStats, config.isConsistentLogicalTimestampEnabled()));
writeSchemaWithMetaFields.getFields().forEach(field -> accumulateColumnRanges(field, filePath, columnRangeMap, columnToStats));
stat.setRecordsStats(new HoodieDeltaWriteStat.RecordsStats<>(columnRangeMap));
}
resetWriteCounts();
assert stat.getRuntimeStats() != null;
LOG.info(String.format("AppendHandle for partitionPath %s filePath %s, took %d ms.", partitionPath,
@@ -376,7 +393,7 @@ public class HoodieAppendHandle<T extends HoodieRecordPayload, I, K, O> extends
if (blocks.size() > 0) {
AppendResult appendResult = writer.appendBlocks(blocks);
processAppendResult(appendResult);
processAppendResult(appendResult, recordList);
recordList.clear();
keysToDelete.clear();
}
@@ -419,7 +436,7 @@ public class HoodieAppendHandle<T extends HoodieRecordPayload, I, K, O> extends
// update final size, once for all log files
// TODO we can actually deduce file size purely from AppendResult (based on offset and size
// of the appended block)
for (WriteStatus status: statuses) {
for (WriteStatus status : statuses) {
long logFileSize = FSUtils.getFileSize(fs, new Path(config.getBasePath(), status.getStat().getPath()));
status.getStat().setFileSizeInBytes(logFileSize);
}

38
hudi-client/hudi-client-common/src/main/java/org/apache/hudi/io/HoodieKeyLookupHandle.java
View File

@@ -19,14 +19,9 @@
package org.apache.hudi.io;
import org.apache.hudi.common.bloom.BloomFilter;
import org.apache.hudi.common.bloom.BloomFilterTypeCode;
import org.apache.hudi.common.bloom.HoodieDynamicBoundedBloomFilter;
import org.apache.hudi.common.fs.FSUtils;
import org.apache.hudi.common.model.HoodieBaseFile;
import org.apache.hudi.common.model.HoodieRecordPayload;
import org.apache.hudi.common.util.HoodieTimer;
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.exception.HoodieIndexException;
@@ -39,8 +34,6 @@ import org.apache.log4j.LogManager;
import org.apache.log4j.Logger;
import java.io.IOException;
import java.nio.ByteBuffer;
import java.nio.charset.StandardCharsets;
import java.util.ArrayList;
import java.util.List;
@@ -53,27 +46,13 @@ public class HoodieKeyLookupHandle<T extends HoodieRecordPayload, I, K, O> exten
private final BloomFilter bloomFilter;
private final List<String> candidateRecordKeys;
private final boolean useMetadataTableIndex;
private Option<String> fileName = Option.empty();
private long totalKeysChecked;
public HoodieKeyLookupHandle(HoodieWriteConfig config, HoodieTable<T, I, K, O> hoodieTable,
Pair<String, String> partitionPathFileIDPair) {
this(config, hoodieTable, partitionPathFileIDPair, Option.empty(), false);
}
public HoodieKeyLookupHandle(HoodieWriteConfig config, HoodieTable<T, I, K, O> hoodieTable,
Pair<String, String> partitionPathFileIDPair, Option<String> fileName,
boolean useMetadataTableIndex) {
super(config, hoodieTable, partitionPathFileIDPair);
this.candidateRecordKeys = new ArrayList<>();
this.totalKeysChecked = 0;
if (fileName.isPresent()) {
ValidationUtils.checkArgument(FSUtils.getFileId(fileName.get()).equals(getFileId()),
"File name '" + fileName.get() + "' doesn't match this lookup handle fileid '" + getFileId() + "'");
this.fileName = fileName;
}
this.useMetadataTableIndex = useMetadataTableIndex;
this.bloomFilter = getBloomFilter();
}
@@ -81,25 +60,16 @@ public class HoodieKeyLookupHandle<T extends HoodieRecordPayload, I, K, O> exten
BloomFilter bloomFilter = null;
HoodieTimer timer = new HoodieTimer().startTimer();
try {
if (this.useMetadataTableIndex) {
ValidationUtils.checkArgument(this.fileName.isPresent(),
"File name not available to fetch bloom filter from the metadata table index.");
Option<ByteBuffer> bloomFilterByteBuffer =
hoodieTable.getMetadataTable().getBloomFilter(partitionPathFileIDPair.getLeft(), fileName.get());
if (!bloomFilterByteBuffer.isPresent()) {
throw new HoodieIndexException("BloomFilter missing for " + partitionPathFileIDPair.getRight());
}
bloomFilter =
new HoodieDynamicBoundedBloomFilter(StandardCharsets.UTF_8.decode(bloomFilterByteBuffer.get()).toString(),
BloomFilterTypeCode.DYNAMIC_V0);
if (config.isMetadataBloomFilterIndexEnabled()) {
bloomFilter = hoodieTable.getMetadataTable().getBloomFilter(partitionPathFileIDPair.getLeft(), partitionPathFileIDPair.getRight())
.orElseThrow(() -> new HoodieIndexException("BloomFilter missing for " + partitionPathFileIDPair.getRight()));
} else {
try (HoodieFileReader reader = createNewFileReader()) {
bloomFilter = reader.readBloomFilter();
}
}
} catch (IOException e) {
throw new HoodieIndexException(String.format("Error reading bloom filter from %s/%s - %s",
getPartitionPathFileIDPair().getLeft(), this.fileName, e));
throw new HoodieIndexException(String.format("Error reading bloom filter from %s", getPartitionPathFileIDPair()), e);
}
LOG.info(String.format("Read bloom filter from %s in %d ms", partitionPathFileIDPair, timer.endTimer()));
return bloomFilter;

163
hudi-client/hudi-client-common/src/main/java/org/apache/hudi/metadata/HoodieBackedTableMetadataWriter.java
View File

@@ -55,6 +55,7 @@ import org.apache.hudi.common.table.view.HoodieTableFileSystemView;
import org.apache.hudi.common.util.HoodieTimer;
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.HoodieCompactionConfig;
import org.apache.hudi.config.HoodieWriteConfig;
import org.apache.hudi.config.metrics.HoodieMetricsConfig;
@@ -81,11 +82,11 @@ import java.util.LinkedList;
import java.util.List;
import java.util.Map;
import java.util.Properties;
import java.util.concurrent.atomic.AtomicLong;
import java.util.stream.Collectors;
import static org.apache.hudi.common.table.HoodieTableConfig.ARCHIVELOG_FOLDER;
import static org.apache.hudi.metadata.HoodieTableMetadata.METADATA_TABLE_NAME_SUFFIX;
import static org.apache.hudi.metadata.HoodieTableMetadata.NON_PARTITIONED_NAME;
import static org.apache.hudi.metadata.HoodieTableMetadata.SOLO_COMMIT_TIMESTAMP;
/**
@@ -121,7 +122,7 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
* @param hadoopConf - Hadoop configuration to use for the metadata writer
* @param writeConfig - Writer config
* @param engineContext - Engine context
* @param actionMetadata - Optional action metadata to help decide bootstrap operations
* @param actionMetadata - Optional action metadata to help decide initialize operations
* @param <T> - Action metadata types extending Avro generated SpecificRecordBase
* @param inflightInstantTimestamp - Timestamp of any instant in progress
*/
@@ -203,7 +204,7 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
* @param metadataConfig - Table config
* @param metaClient - Meta client for the metadata table
* @param fsView - Metadata table filesystem view to use
* @param isBootstrapCompleted - Is metadata table bootstrap completed
* @param isBootstrapCompleted - Is metadata table initializing completed
*/
private void enablePartition(final MetadataPartitionType partitionType, final HoodieMetadataConfig metadataConfig,
final Option<HoodieTableMetaClient> metaClient, Option<HoodieTableFileSystemView> fsView, boolean isBootstrapCompleted) {
@@ -319,13 +320,13 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
/**
* Initialize the metadata table if it does not exist.
*
* If the metadata table does not exist, then file and partition listing is used to bootstrap the table.
* <p>
* If the metadata table does not exist, then file and partition listing is used to initialize the table.
*
* @param engineContext
* @param actionMetadata Action metadata types extending Avro generated SpecificRecordBase
* @param actionMetadata Action metadata types extending Avro generated SpecificRecordBase
* @param inflightInstantTimestamp Timestamp of an instant in progress on the dataset. This instant is ignored
* while deciding to bootstrap the metadata table.
* while deciding to initialize the metadata table.
*/
protected abstract <T extends SpecificRecordBase> void initialize(HoodieEngineContext engineContext,
Option<T> actionMetadata,
@@ -345,27 +346,25 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
}
/**
* Bootstrap the metadata table if needed.
* Initialize the metadata table if needed.
*
* @param engineContext - Engine context
* @param dataMetaClient - Meta client for the data table
* @param actionMetadata - Optional action metadata
* @param <T> - Action metadata types extending Avro generated SpecificRecordBase
* @param inflightInstantTimestamp - Timestamp of an instant in progress on the dataset. This instant is ignored
* @param dataMetaClient - meta client for the data table
* @param actionMetadata - optional action metadata
* @param inflightInstantTimestamp - timestamp of an instant in progress on the dataset
* @param <T> - action metadata types extending Avro generated SpecificRecordBase
* @throws IOException
*/
protected <T extends SpecificRecordBase> void bootstrapIfNeeded(HoodieEngineContext engineContext,
HoodieTableMetaClient dataMetaClient,
Option<T> actionMetadata,
Option<String> inflightInstantTimestamp) throws IOException {
protected <T extends SpecificRecordBase> void initializeIfNeeded(HoodieTableMetaClient dataMetaClient,
Option<T> actionMetadata,
Option<String> inflightInstantTimestamp) throws IOException {
HoodieTimer timer = new HoodieTimer().startTimer();
boolean exists = dataMetaClient.getFs().exists(new Path(metadataWriteConfig.getBasePath(),
HoodieTableMetaClient.METAFOLDER_NAME));
boolean rebootstrap = false;
boolean reInitialize = false;
// If the un-synced instants have been archived, then
// the metadata table will need to be bootstrapped again.
// the metadata table will need to be initialized again.
if (exists) {
HoodieTableMetaClient metadataMetaClient = HoodieTableMetaClient.builder().setConf(hadoopConf.get())
.setBasePath(metadataWriteConfig.getBasePath()).build();
@@ -378,39 +377,39 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
final Option<HoodieInstant> latestMetadataInstant =
metadataMetaClient.getActiveTimeline().filterCompletedInstants().lastInstant();
rebootstrap = isBootstrapNeeded(latestMetadataInstant, actionMetadata);
reInitialize = isBootstrapNeeded(latestMetadataInstant, actionMetadata);
}
if (rebootstrap) {
if (reInitialize) {
metrics.ifPresent(m -> m.updateMetrics(HoodieMetadataMetrics.REBOOTSTRAP_STR, 1));
LOG.info("Deleting Metadata Table directory so that it can be re-bootstrapped");
LOG.info("Deleting Metadata Table directory so that it can be re-initialized");
dataMetaClient.getFs().delete(new Path(metadataWriteConfig.getBasePath()), true);
exists = false;
}
if (!exists) {
// Initialize for the first time by listing partitions and files directly from the file system
if (bootstrapFromFilesystem(engineContext, dataMetaClient, inflightInstantTimestamp)) {
if (initializeFromFilesystem(dataMetaClient, inflightInstantTimestamp)) {
metrics.ifPresent(m -> m.updateMetrics(HoodieMetadataMetrics.INITIALIZE_STR, timer.endTimer()));
}
}
}
/**
* Whether bootstrap operation needed for this metadata table.
* Whether initialize operation needed for this metadata table.
* <p>
* Rollback of the first commit would look like un-synced instants in the metadata table.
* Action metadata is needed to verify the instant time and avoid erroneous bootstrapping.
* Action metadata is needed to verify the instant time and avoid erroneous initializing.
* <p>
* TODO: Revisit this logic and validate that filtering for all
* commits timeline is the right thing to do
*
* @return True if the bootstrap is not needed, False otherwise
* @return True if the initialize is not needed, False otherwise
*/
private <T extends SpecificRecordBase> boolean isBootstrapNeeded(Option<HoodieInstant> latestMetadataInstant,
Option<T> actionMetadata) {
if (!latestMetadataInstant.isPresent()) {
LOG.warn("Metadata Table will need to be re-bootstrapped as no instants were found");
LOG.warn("Metadata Table will need to be re-initialized as no instants were found");
return true;
}
@@ -423,7 +422,7 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
if (dataMetaClient.getActiveTimeline().getAllCommitsTimeline().isBeforeTimelineStarts(
latestMetadataInstant.get().getTimestamp())
&& !isCommitRevertedByInFlightAction(actionMetadata, latestMetadataInstantTimestamp)) {
LOG.error("Metadata Table will need to be re-bootstrapped as un-synced instants have been archived."
LOG.error("Metadata Table will need to be re-initialized as un-synced instants have been archived."
+ " latestMetadataInstant=" + latestMetadataInstant.get().getTimestamp()
+ ", latestDataInstant=" + dataMetaClient.getActiveTimeline().firstInstant().get().getTimestamp());
return true;
@@ -455,9 +454,7 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
case HoodieTimeline.ROLLBACK_ACTION:
List<HoodieInstantInfo> rollbackedInstants =
((HoodieRollbackMetadata) actionMetadata.get()).getInstantsRollback();
affectedInstantTimestamps = rollbackedInstants.stream().map(instant -> {
return instant.getCommitTime().toString();
}).collect(Collectors.toList());
affectedInstantTimestamps = rollbackedInstants.stream().map(HoodieInstantInfo::getCommitTime).collect(Collectors.toList());
if (affectedInstantTimestamps.contains(latestMetadataInstantTimestamp)) {
return true;
@@ -466,9 +463,7 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
case HoodieTimeline.RESTORE_ACTION:
List<HoodieInstantInfo> restoredInstants =
((HoodieRestoreMetadata) actionMetadata.get()).getRestoreInstantInfo();
affectedInstantTimestamps = restoredInstants.stream().map(instant -> {
return instant.getCommitTime().toString();
}).collect(Collectors.toList());
affectedInstantTimestamps = restoredInstants.stream().map(HoodieInstantInfo::getCommitTime).collect(Collectors.toList());
if (affectedInstantTimestamps.contains(latestMetadataInstantTimestamp)) {
return true;
@@ -484,14 +479,14 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
/**
* Initialize the Metadata Table by listing files and partitions from the file system.
*
* @param dataMetaClient {@code HoodieTableMetaClient} for the dataset.
* @param inflightInstantTimestamp
* @param dataMetaClient - {@code HoodieTableMetaClient} for the dataset.
* @param inflightInstantTimestamp - Current action instant responsible for this initialization
*/
private boolean bootstrapFromFilesystem(HoodieEngineContext engineContext, HoodieTableMetaClient dataMetaClient,
Option<String> inflightInstantTimestamp) throws IOException {
private boolean initializeFromFilesystem(HoodieTableMetaClient dataMetaClient,
Option<String> inflightInstantTimestamp) throws IOException {
ValidationUtils.checkState(enabled, "Metadata table cannot be initialized as it is not enabled");
// We can only bootstrap if there are no pending operations on the dataset
// We can only initialize if there are no pending operations on the dataset
List<HoodieInstant> pendingDataInstant = dataMetaClient.getActiveTimeline()
.getInstants().filter(i -> !i.isCompleted())
.filter(i -> !inflightInstantTimestamp.isPresent() || !i.getTimestamp().equals(inflightInstantTimestamp.get()))
@@ -499,7 +494,7 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
if (!pendingDataInstant.isEmpty()) {
metrics.ifPresent(m -> m.updateMetrics(HoodieMetadataMetrics.BOOTSTRAP_ERR_STR, 1));
LOG.warn("Cannot bootstrap metadata table as operation(s) are in progress on the dataset: "
LOG.warn("Cannot initialize metadata table as operation(s) are in progress on the dataset: "
+ Arrays.toString(pendingDataInstant.toArray()));
return false;
}
@@ -514,15 +509,10 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
initTableMetadata();
initializeEnabledFileGroups(dataMetaClient, createInstantTime);
// List all partitions in the basePath of the containing dataset
LOG.info("Initializing metadata table by using file listings in " + dataWriteConfig.getBasePath());
engineContext.setJobStatus(this.getClass().getSimpleName(), "Bootstrap: initializing metadata table by listing files and partitions");
List<DirectoryInfo> dirInfoList = listAllPartitions(dataMetaClient);
// During bootstrap, the list of files to be committed can be huge. So creating a HoodieCommitMetadata out of these
// large number of files and calling the existing update(HoodieCommitMetadata) function does not scale well.
// Hence, we have a special commit just for the bootstrap scenario.
bootstrapCommit(dirInfoList, createInstantTime);
// During cold startup, the list of files to be committed can be huge. So creating a HoodieCommitMetadata out
// of these large number of files and calling the existing update(HoodieCommitMetadata) function does not scale
// well. Hence, we have a special commit just for the initialization scenario.
initialCommit(createInstantTime);
return true;
}
@@ -651,6 +641,14 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
}
}
private MetadataRecordsGenerationParams getRecordsGenerationParams() {
return new MetadataRecordsGenerationParams(
dataMetaClient, enabledPartitionTypes, dataWriteConfig.getBloomFilterType(),
dataWriteConfig.getBloomIndexParallelism(),
dataWriteConfig.isMetadataIndexColumnStatsForAllColumnsEnabled(),
dataWriteConfig.getColumnStatsIndexParallelism());
}
/**
* Interface to assist in converting commit metadata to List of HoodieRecords to be written to metadata table.
* Updates of different commit metadata uses the same method to convert to HoodieRecords and hence.
@@ -681,8 +679,8 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
*/
@Override
public void update(HoodieCommitMetadata commitMetadata, String instantTime, boolean isTableServiceAction) {
processAndCommit(instantTime, () -> HoodieTableMetadataUtil.convertMetadataToRecords(engineContext, enabledPartitionTypes,
commitMetadata, dataMetaClient, dataWriteConfig.isMetadataIndexColumnStatsForAllColumnsEnabled(), instantTime), !isTableServiceAction);
processAndCommit(instantTime, () -> HoodieTableMetadataUtil.convertMetadataToRecords(
engineContext, commitMetadata, instantTime, getRecordsGenerationParams()), !isTableServiceAction);
}
/**
@@ -693,8 +691,8 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
*/
@Override
public void update(HoodieCleanMetadata cleanMetadata, String instantTime) {
processAndCommit(instantTime, () -> HoodieTableMetadataUtil.convertMetadataToRecords(engineContext, enabledPartitionTypes,
cleanMetadata, dataMetaClient, instantTime), false);
processAndCommit(instantTime, () -> HoodieTableMetadataUtil.convertMetadataToRecords(engineContext,
cleanMetadata, getRecordsGenerationParams(), instantTime), false);
}
/**
@@ -706,7 +704,7 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
@Override
public void update(HoodieRestoreMetadata restoreMetadata, String instantTime) {
processAndCommit(instantTime, () -> HoodieTableMetadataUtil.convertMetadataToRecords(engineContext,
enabledPartitionTypes, metadataMetaClient.getActiveTimeline(), restoreMetadata, dataMetaClient, instantTime,
metadataMetaClient.getActiveTimeline(), restoreMetadata, getRecordsGenerationParams(), instantTime,
metadata.getSyncedInstantTime()), false);
}
@@ -732,8 +730,8 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
}
Map<MetadataPartitionType, HoodieData<HoodieRecord>> records =
HoodieTableMetadataUtil.convertMetadataToRecords(engineContext, enabledPartitionTypes,
metadataMetaClient.getActiveTimeline(), rollbackMetadata, dataMetaClient, instantTime,
HoodieTableMetadataUtil.convertMetadataToRecords(engineContext, metadataMetaClient.getActiveTimeline(),
rollbackMetadata, getRecordsGenerationParams(), instantTime,
metadata.getSyncedInstantTime(), wasSynced);
commit(instantTime, records, false);
}
@@ -845,20 +843,29 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
}
/**
* This is invoked to bootstrap metadata table for a dataset. Bootstrap Commit has special handling mechanism due to its scale compared to
* This is invoked to initialize metadata table for a dataset. Bootstrap Commit has special handling mechanism due to its scale compared to
* other regular commits.
*
*/
protected void bootstrapCommit(List<DirectoryInfo> partitionInfoList, String createInstantTime) {
List<String> partitions = partitionInfoList.stream().map(p ->
p.getRelativePath().isEmpty() ? NON_PARTITIONED_NAME : p.getRelativePath()).collect(Collectors.toList());
final int totalFiles = partitionInfoList.stream().mapToInt(p -> p.getTotalFiles()).sum();
private void initialCommit(String createInstantTime) {
// List all partitions in the basePath of the containing dataset
LOG.info("Initializing metadata table by using file listings in " + dataWriteConfig.getBasePath());
engineContext.setJobStatus(this.getClass().getSimpleName(), "Initializing metadata table by listing files and partitions");
List<DirectoryInfo> partitionInfoList = listAllPartitions(dataMetaClient);
List<String> partitions = new ArrayList<>();
AtomicLong totalFiles = new AtomicLong(0);
Map<String, Map<String, Long>> partitionToFilesMap = partitionInfoList.stream().map(p -> {
final String partitionName = HoodieTableMetadataUtil.getPartition(p.getRelativePath());
partitions.add(partitionName);
totalFiles.addAndGet(p.getTotalFiles());
return Pair.of(partitionName, p.getFileNameToSizeMap());
}).collect(Collectors.toMap(Pair::getKey, Pair::getValue));
final Map<MetadataPartitionType, HoodieData<HoodieRecord>> partitionToRecordsMap = new HashMap<>();
// Record which saves the list of all partitions
HoodieRecord allPartitionRecord = HoodieMetadataPayload.createPartitionListRecord(partitions);
if (partitions.isEmpty()) {
// in case of bootstrapping of a fresh table, there won't be any partitions, but we need to make a boostrap commit
// in case of initializing of a fresh table, there won't be any partitions, but we need to make a boostrap commit
final HoodieData<HoodieRecord> allPartitionRecordsRDD = engineContext.parallelize(
Collections.singletonList(allPartitionRecord), 1);
partitionToRecordsMap.put(MetadataPartitionType.FILES, allPartitionRecordsRDD);
@@ -866,7 +873,7 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
return;
}
HoodieData<HoodieRecord> partitionRecords = engineContext.parallelize(Arrays.asList(allPartitionRecord), 1);
HoodieData<HoodieRecord> filesPartitionRecords = engineContext.parallelize(Arrays.asList(allPartitionRecord), 1);
if (!partitionInfoList.isEmpty()) {
HoodieData<HoodieRecord> fileListRecords = engineContext.parallelize(partitionInfoList, partitionInfoList.size()).map(partitionInfo -> {
Map<String, Long> fileNameToSizeMap = partitionInfo.getFileNameToSizeMap();
@@ -878,29 +885,41 @@ public abstract class HoodieBackedTableMetadataWriter implements HoodieTableMeta
// Record which saves files within a partition
return HoodieMetadataPayload.createPartitionFilesRecord(
partitionInfo.getRelativePath().isEmpty() ? NON_PARTITIONED_NAME : partitionInfo.getRelativePath(), Option.of(validFileNameToSizeMap), Option.empty());
HoodieTableMetadataUtil.getPartition(partitionInfo.getRelativePath()), Option.of(validFileNameToSizeMap), Option.empty());
});
partitionRecords = partitionRecords.union(fileListRecords);
filesPartitionRecords = filesPartitionRecords.union(fileListRecords);
}
ValidationUtils.checkState(filesPartitionRecords.count() == (partitions.size() + 1));
partitionToRecordsMap.put(MetadataPartitionType.FILES, filesPartitionRecords);
if (enabledPartitionTypes.contains(MetadataPartitionType.BLOOM_FILTERS)) {
final HoodieData<HoodieRecord> recordsRDD = HoodieTableMetadataUtil.convertFilesToBloomFilterRecords(
engineContext, Collections.emptyMap(), partitionToFilesMap, getRecordsGenerationParams(), createInstantTime);
partitionToRecordsMap.put(MetadataPartitionType.BLOOM_FILTERS, recordsRDD);
}
if (enabledPartitionTypes.contains(MetadataPartitionType.COLUMN_STATS)) {
final HoodieData<HoodieRecord> recordsRDD = HoodieTableMetadataUtil.convertFilesToColumnStatsRecords(
engineContext, Collections.emptyMap(), partitionToFilesMap, getRecordsGenerationParams());
partitionToRecordsMap.put(MetadataPartitionType.COLUMN_STATS, recordsRDD);
}
LOG.info("Committing " + partitions.size() + " partitions and " + totalFiles + " files to metadata");
ValidationUtils.checkState(partitionRecords.count() == (partitions.size() + 1));
partitionToRecordsMap.put(MetadataPartitionType.FILES, partitionRecords);
commit(createInstantTime, partitionToRecordsMap, false);
}
/**
* A class which represents a directory and the files and directories inside it.
*
* <p>
* A {@code PartitionFileInfo} object saves the name of the partition and various properties requires of each file
* required for bootstrapping the metadata table. Saving limited properties reduces the total memory footprint when
* a very large number of files are present in the dataset being bootstrapped.
* required for initializing the metadata table. Saving limited properties reduces the total memory footprint when
* a very large number of files are present in the dataset being initialized.
*/
static class DirectoryInfo implements Serializable {
// Relative path of the directory (relative to the base directory)
private final String relativePath;
// Map of filenames within this partition to their respective sizes
private HashMap<String, Long> filenameToSizeMap;
private final HashMap<String, Long> filenameToSizeMap;
// List of directories within this partition
private final List<Path> subDirectories = new ArrayList<>();
// Is this a hoodie partition