[SPARK-19645] structured streaming job restart bug

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Duplicate
• Affects Version/s: 2.1.0
• Component/s: Structured Streaming
• Assignee: Unassigned
• Reporter: Feng Gui
• Created: 17/Feb/17 06:21
• Updated: 28/Feb/17 18:54
• Resolved: 28/Feb/17 18:54

Description

We are trying to use Structured Streaming in product, however currently there exists a bug refer to the process of streaming job restart.
The following is the concrete error message:

Caused by: java.lang.IllegalStateException: Error committing version 2 into HDFSStateStore[id = (op=0, part=136), dir = /tmp/Pipeline_112346-continueagg-bxaxs/state/0/136]
at org.apache.spark.sql.execution.streaming.state.HDFSBackedStateStoreProvider$HDFSBackedStateStore.commit(HDFSBackedStateStoreProvider.scala:162)
at org.apache.spark.sql.execution.streaming.StateStoreSaveExec$$anonfun$doExecute$3.apply(StatefulAggregate.scala:173)
at org.apache.spark.sql.execution.streaming.StateStoreSaveExec$$anonfun$doExecute$3.apply(StatefulAggregate.scala:123)
at org.apache.spark.sql.execution.streaming.state.StateStoreRDD.compute(StateStoreRDD.scala:64)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:87)
at org.apache.spark.scheduler.Task.run(Task.scala:99)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:282)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745)
Caused by: java.io.IOException: Failed to rename /tmp/Pipeline_112346-continueagg-bxaxs/state/0/136/temp--5345709896617324284 to /tmp/Pipeline_112346-continueagg-bxaxs/state/0/136/2.delta
at org.apache.spark.sql.execution.streaming.state.HDFSBackedStateStoreProvider.org$apache$spark$sql$execution$streaming$state$HDFSBackedStateStoreProvider$$commitUpdates(HDFSBackedStateStoreProvider.scala:259)
at org.apache.spark.sql.execution.streaming.state.HDFSBackedStateStoreProvider$HDFSBackedStateStore.commit(HDFSBackedStateStoreProvider.scala:156)
... 14 more

The bug can be easily reproduced, just modify val metadataRoot = "hdfs://localhost:8020/tmp/checkpoint" in StreamTest and then run the test sort after aggregate in complete mode in StreamingAggregationSuite. The main reason is that when restart streaming job spark will recompute WAL offsets and generate the same hdfs delta file(latest delta file generated before restart and named "currentBatchId.delta") . In my opinion, this is a bug. If you guy consider that this is a bug also, I can fix it.

Issue Links

• duplicates: SPARK-19677 HDFSBackedStateStoreProvider fails to overwrite existing file