lanyuanxiaoyao/hudi
1
0
Fork 0
Code Activity

[HUDI-796] Add deduping logic for upserts case (#1558)

Browse Source
This commit is contained in:
Pratyaksh Sharma
2020-09-18 17:07:52 +05:30
committed by GitHub
parent bf65269f66
commit 73e5b4c7bb
5 changed files with 195 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
hudi-cli/src/main/java/org/apache/hudi/cli/commands/RepairsCommand.java
View File

@@ -18,6 +18,7 @@
package org.apache.hudi.cli.commands;
import org.apache.hudi.cli.DeDupeType;
import org.apache.hudi.cli.HoodieCLI;
import org.apache.hudi.cli.HoodiePrintHelper;
import org.apache.hudi.cli.HoodieTableHeaderFields;
@@ -77,8 +78,13 @@ public class RepairsCommand implements CommandMarker {
help = "Spark executor memory") final String sparkMemory,
@CliOption(key = {"dryrun"},
help = "Should we actually remove duplicates or just run and store result to repairedOutputPath",
unspecifiedDefaultValue = "true") final boolean dryRun)
unspecifiedDefaultValue = "true") final boolean dryRun,
@CliOption(key = {"dedupeType"}, help = "Valid values are - insert_type, update_type and upsert_type",
unspecifiedDefaultValue = "insert_type") final String dedupeType)
throws Exception {
if (!DeDupeType.values().contains(DeDupeType.withName(dedupeType))) {
throw new IllegalArgumentException("Please provide valid dedupe type!");
}
if (StringUtils.isNullOrEmpty(sparkPropertiesPath)) {
sparkPropertiesPath =
Utils.getDefaultPropertiesFile(JavaConverters.mapAsScalaMapConverter(System.getenv()).asScala());
@@ -87,7 +93,7 @@ public class RepairsCommand implements CommandMarker {
SparkLauncher sparkLauncher = SparkUtil.initLauncher(sparkPropertiesPath);
sparkLauncher.addAppArgs(SparkMain.SparkCommand.DEDUPLICATE.toString(), master, sparkMemory,
duplicatedPartitionPath, repairedOutputPath, HoodieCLI.getTableMetaClient().getBasePath(),
String.valueOf(dryRun));
String.valueOf(dryRun), dedupeType);
Process process = sparkLauncher.launch();
InputStreamConsumer.captureOutput(process);
int exitCode = process.waitFor();

11
hudi-cli/src/main/java/org/apache/hudi/cli/commands/SparkMain.java
View File

@@ -18,6 +18,7 @@
package org.apache.hudi.cli.commands;
import org.apache.hudi.cli.DeDupeType;
import org.apache.hudi.DataSourceWriteOptions;
import org.apache.hudi.cli.DedupeSparkJob;
import org.apache.hudi.cli.utils.SparkUtil;
@@ -87,8 +88,8 @@ public class SparkMain {
returnCode = rollback(jsc, args[3], args[4]);
break;
case DEDUPLICATE:
assert (args.length == 7);
returnCode = deduplicatePartitionPath(jsc, args[3], args[4], args[5], args[6]);
assert (args.length == 8);
returnCode = deduplicatePartitionPath(jsc, args[3], args[4], args[5], Boolean.parseBoolean(args[6]), args[7]);
break;
case ROLLBACK_TO_SAVEPOINT:
assert (args.length == 5);
@@ -304,10 +305,10 @@ public class SparkMain {
}
private static int deduplicatePartitionPath(JavaSparkContext jsc, String duplicatedPartitionPath,
String repairedOutputPath, String basePath, String dryRun) {
String repairedOutputPath, String basePath, boolean dryRun, String dedupeType) {
DedupeSparkJob job = new DedupeSparkJob(basePath, duplicatedPartitionPath, repairedOutputPath, new SQLContext(jsc),
FSUtils.getFs(basePath, jsc.hadoopConfiguration()));
job.fixDuplicates(Boolean.parseBoolean(dryRun));
FSUtils.getFs(basePath, jsc.hadoopConfiguration()), DeDupeType.withName(dedupeType));
job.fixDuplicates(dryRun);
return 0;
}

28
hudi-cli/src/main/scala/org/apache/hudi/cli/DeDupeType.scala Normal file
View File

@@ -0,0 +1,28 @@
/*
* 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.cli
object DeDupeType extends Enumeration {
type dedupeType = Value
val INSERT_TYPE = Value("insert_type")
val UPDATE_TYPE = Value("update_type")
val UPSERT_TYPE = Value("upsert_type")
}

120
hudi-cli/src/main/scala/org/apache/hudi/cli/DedupeSparkJob.scala
View File

@@ -26,11 +26,10 @@ import org.apache.hudi.common.table.HoodieTableMetaClient
import org.apache.hudi.common.table.view.HoodieTableFileSystemView
import org.apache.hudi.exception.HoodieException
import org.apache.log4j.Logger
import org.apache.spark.sql.{DataFrame, SQLContext}
import org.apache.spark.sql.{DataFrame, Row, SQLContext}
import scala.collection.JavaConversions._
import scala.collection.mutable._
import scala.collection.mutable.{Buffer, HashMap, HashSet, ListBuffer}
/**
* Spark job to de-duplicate data present in a partition path
@@ -39,8 +38,8 @@ class DedupeSparkJob(basePath: String,
duplicatedPartitionPath: String,
repairOutputPath: String,
sqlContext: SQLContext,
fs: FileSystem) {
fs: FileSystem,
dedupeType: DeDupeType.Value) {
val sparkHelper = new SparkHelper(sqlContext, fs)
val LOG = Logger.getLogger(this.getClass)
@@ -98,33 +97,91 @@ class DedupeSparkJob(basePath: String,
ON h.`_hoodie_record_key` = d.dupe_key
"""
val dupeMap = sqlContext.sql(dupeDataSql).collectAsList().groupBy(r => r.getString(0))
val fileToDeleteKeyMap = new HashMap[String, HashSet[String]]()
// Mark all files except the one with latest commits for deletion
dupeMap.foreach(rt => {
val (key, rows) = rt
var maxCommit = -1L
rows.foreach(r => {
val c = r(3).asInstanceOf[String].toLong
if (c > maxCommit)
maxCommit = c
})
rows.foreach(r => {
val c = r(3).asInstanceOf[String].toLong
if (c != maxCommit) {
val f = r(2).asInstanceOf[String].split("_")(0)
if (!fileToDeleteKeyMap.contains(f)) {
fileToDeleteKeyMap(f) = HashSet[String]()
}
fileToDeleteKeyMap(f).add(key)
}
})
})
fileToDeleteKeyMap
getDedupePlan(dupeMap)
}
private def getDedupePlan(dupeMap: Map[String, Buffer[Row]]): HashMap[String, HashSet[String]] = {
val fileToDeleteKeyMap = new HashMap[String, HashSet[String]]()
dupeMap.foreach(rt => {
val (key, rows) = rt
dedupeType match {
case DeDupeType.UPDATE_TYPE =>
/*
This corresponds to the case where all duplicates have been updated at least once.
Once updated, duplicates are bound to have same commit time unless forcefully modified.
*/
rows.init.foreach(r => {
val f = r(2).asInstanceOf[String].split("_")(0)
if (!fileToDeleteKeyMap.contains(f)) {
fileToDeleteKeyMap(f) = HashSet[String]()
}
fileToDeleteKeyMap(f).add(key)
})
case DeDupeType.INSERT_TYPE =>
/*
This corresponds to the case where duplicates got created due to INSERT and have never been updated.
*/
var maxCommit = -1L
rows.foreach(r => {
val c = r(3).asInstanceOf[String].toLong
if (c > maxCommit)
maxCommit = c
})
rows.foreach(r => {
val c = r(3).asInstanceOf[String].toLong
if (c != maxCommit) {
val f = r(2).asInstanceOf[String].split("_")(0)
if (!fileToDeleteKeyMap.contains(f)) {
fileToDeleteKeyMap(f) = HashSet[String]()
}
fileToDeleteKeyMap(f).add(key)
}
})
case DeDupeType.UPSERT_TYPE =>
/*
This corresponds to the case where duplicates got created as a result of inserts as well as updates,
i.e few duplicate records have been updated, while others were never updated.
*/
var maxCommit = -1L
rows.foreach(r => {
val c = r(3).asInstanceOf[String].toLong
if (c > maxCommit)
maxCommit = c
})
val rowsWithMaxCommit = new ListBuffer[Row]()
rows.foreach(r => {
val c = r(3).asInstanceOf[String].toLong
if (c != maxCommit) {
val f = r(2).asInstanceOf[String].split("_")(0)
if (!fileToDeleteKeyMap.contains(f)) {
fileToDeleteKeyMap(f) = HashSet[String]()
}
fileToDeleteKeyMap(f).add(key)
} else {
rowsWithMaxCommit += r
}
})
rowsWithMaxCommit.toList.init.foreach(r => {
val f = r(2).asInstanceOf[String].split("_")(0)
if (!fileToDeleteKeyMap.contains(f)) {
fileToDeleteKeyMap(f) = HashSet[String]()
}
fileToDeleteKeyMap(f).add(key)
})
case _ => throw new IllegalArgumentException("Please provide valid type for deduping!")
}
})
LOG.debug(s"fileToDeleteKeyMap size: ${fileToDeleteKeyMap.size}, map: $fileToDeleteKeyMap")
fileToDeleteKeyMap
}
def fixDuplicates(dryRun: Boolean = true) = {
val metadata = new HoodieTableMetaClient(fs.getConf, basePath)
@@ -152,7 +209,7 @@ class DedupeSparkJob(basePath: String,
val newFilePath = new Path(s"$repairOutputPath/${fileNameToPathMap(fileName).getName}")
LOG.info(" Skipping and writing new file for : " + fileName)
SparkHelpers.skipKeysAndWriteNewFile(instantTime, fs, badFilePath, newFilePath, dupeFixPlan(fileName))
fs.delete(badFilePath, false)
fs.delete(badFilePath, true)
}
// 3. Check that there are no duplicates anymore.
@@ -175,7 +232,6 @@ class DedupeSparkJob(basePath: String,
throw new HoodieException("Some records in source are not found in fixed files. Inspect output!!")
}
println("No duplicates found & counts are in check!!!! ")
// 4. Prepare to copy the fixed files back.
fileNameToPathMap.foreach { case (_, filePath) =>