lanyuanxiaoyao/hudi
1
0
Fork 0
Code Activity

[HUDI-3029] Transaction manager: avoid deadlock when doing begin and end transactions (#4363)

Browse Source 
* [HUDI-3029] Transaction manager: avoid deadlock when doing begin and end transactions

 - Transaction manager has begin and end transactions as synchronized methods.
   Based on the lock provider implementaion, this can lead to deadlock
   situation when the underlying lock() calls are blocking or with a long timeout.

 - Fixing transaction manager begin and end transactions to not get to deadlock
   and to not assume anything on the lock provider implementation.
This commit is contained in:
Manoj Govindassamy
2021-12-18 06:43:17 -08:00
committed by GitHub
parent 47852446e8
commit d1d48ed494
11 changed files with 256 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/client/SparkRDDWriteClient.java
View File

@@ -300,8 +300,8 @@ public class SparkRDDWriteClient<T extends HoodieRecordPayload> extends
String compactionCommitTime) {
this.context.setJobStatus(this.getClass().getSimpleName(), "Collect compaction write status and commit compaction");
List<HoodieWriteStat> writeStats = writeStatuses.map(WriteStatus::getStat).collect();
final HoodieInstant compactionInstant = new HoodieInstant(HoodieInstant.State.INFLIGHT, HoodieTimeline.COMPACTION_ACTION, compactionCommitTime);
try {
HoodieInstant compactionInstant = new HoodieInstant(HoodieInstant.State.INFLIGHT, HoodieTimeline.COMPACTION_ACTION, compactionCommitTime);
this.txnManager.beginTransaction(Option.of(compactionInstant), Option.empty());
finalizeWrite(table, compactionCommitTime, writeStats);
// commit to data table after committing to metadata table.
@@ -309,7 +309,7 @@ public class SparkRDDWriteClient<T extends HoodieRecordPayload> extends
LOG.info("Committing Compaction " + compactionCommitTime + ". Finished with result " + metadata);
CompactHelpers.getInstance().completeInflightCompaction(table, compactionCommitTime, metadata);
} finally {
this.txnManager.endTransaction();
this.txnManager.endTransaction(Option.of(compactionInstant));
}
WriteMarkersFactory.get(config.getMarkersType(), table, compactionCommitTime)
.quietDeleteMarkerDir(context, config.getMarkersDeleteParallelism());
@@ -378,8 +378,8 @@ public class SparkRDDWriteClient<T extends HoodieRecordPayload> extends
throw new HoodieClusteringException("Clustering failed to write to files:"
+ writeStats.stream().filter(s -> s.getTotalWriteErrors() > 0L).map(s -> s.getFileId()).collect(Collectors.joining(",")));
}
final HoodieInstant clusteringInstant = new HoodieInstant(HoodieInstant.State.INFLIGHT, HoodieTimeline.REPLACE_COMMIT_ACTION, clusteringCommitTime);
try {
HoodieInstant clusteringInstant = new HoodieInstant(HoodieInstant.State.INFLIGHT, HoodieTimeline.REPLACE_COMMIT_ACTION, clusteringCommitTime);
this.txnManager.beginTransaction(Option.of(clusteringInstant), Option.empty());
finalizeWrite(table, clusteringCommitTime, writeStats);
writeTableMetadataForTableServices(table, metadata,clusteringInstant);
@@ -395,7 +395,7 @@ public class SparkRDDWriteClient<T extends HoodieRecordPayload> extends
} catch (Exception e) {
throw new HoodieClusteringException("unable to transition clustering inflight to complete: " + clusteringCommitTime, e);
} finally {
this.txnManager.endTransaction();
this.txnManager.endTransaction(Option.of(clusteringInstant));
}
WriteMarkersFactory.get(config.getMarkersType(), table, clusteringCommitTime)
.quietDeleteMarkerDir(context, config.getMarkersDeleteParallelism());