[MAPREDUCE-7003] Indefinite retries of getJobSummary() if a job summary file is corrupt

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: jobhistoryserver
• Assignee: Unassigned
• Reporter: Attila Sasvári
• Created: 06/Nov/17 09:37
• Updated: 06/Nov/17 09:37

Description

Having a corrupt job summary file in the /user/history/done_intermediate directory in HDFS, e.g. /user/history/done_intermediate/oozie/job_1111111111111_111111.summary before moving it to /user/history/done, results in indefinite retries of org.apache.hadoop.mapreduce.v2.hs.HistoryFileManager.getJobSummary(). JHS will log recurring exceptions like:

2017-11-03 01:01:01,124 WARN org.apache.hadoop.hdfs.BlockReaderFactory: I/O error constructing remote block reader.
java.io.IOException: Got error for OP_READ_BLOCK, status=ERROR, self=/ABC.DEF.GHI:JKLMN, remote=/ABC.DEF.GHI:JKLMN, for file /user/history/done_intermediate/admin/job_1111111111111_1111.summary, for pool XX-999999999-ABC.DEF.GHI-1111111111111 block 1111111111_22222
	at org.apache.hadoop.hdfs.RemoteBlockReader2.checkSuccess(RemoteBlockReader2.java:467)
	at org.apache.hadoop.hdfs.RemoteBlockReader2.newBlockReader(RemoteBlockReader2.java:432)
	at org.apache.hadoop.hdfs.BlockReaderFactory.getRemoteBlockReader(BlockReaderFactory.java:881)
	at org.apache.hadoop.hdfs.BlockReaderFactory.getRemoteBlockReaderFromTcp(BlockReaderFactory.java:759)
	at org.apache.hadoop.hdfs.BlockReaderFactory.build(BlockReaderFactory.java:376)
	at org.apache.hadoop.hdfs.DFSInputStream.blockSeekTo(DFSInputStream.java:652)
	at org.apache.hadoop.hdfs.DFSInputStream.readWithStrategy(DFSInputStream.java:879)
	at org.apache.hadoop.hdfs.DFSInputStream.read(DFSInputStream.java:932)
	at org.apache.hadoop.hdfs.DFSInputStream.read(DFSInputStream.java:732)
	at java.io.DataInputStream.readUnsignedShort(DataInputStream.java:337)
	at java.io.DataInputStream.readUTF(DataInputStream.java:589)
	at java.io.DataInputStream.readUTF(DataInputStream.java:564)
	at org.apache.hadoop.mapreduce.v2.hs.HistoryFileManager.getJobSummary(HistoryFileManager.java:1059)

(INFO and ERROR logs are omitted)

To reproduce it:

• start JHS in debug mode (use JVM parameter -agentlib:jdwp=transport=dt_socket,server=y,address=45555,suspend=n when starting it)
• attach debugger to the process and add a break point to stop in org.apache.hadoop.mapreduce.v2.hs.HistoryFileManager.getJobSummary()
• start a map reduce job and wait until breakpoint is hit
• delete or rename physical block on the datanode(s) for the job summary file (e.g. use hdfs fsck /user/history/done_intermediate/oozie/job_1111111111111_111111.summary -blocks -locations -files to get the block name; search for the block the on datanode(s) and remove/ rename it)
• detach debugger
• examine JHS log files