[HDFS-14027] DFSStripedOutputStream should implement both hsync methods

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 3.0.0
• Fix Version/s: 3.0.4, 3.1.2, 3.3.0, 3.2.1
• Component/s: erasure-coding
• Target Version/s: 3.0.4, 3.1.2, 3.3.0, 3.2.1
• Assignee: Xiao Chen
• Reporter: Xiao Chen
• Created: 24/Oct/18 22:30
• Updated: 30/Oct/18 03:22
• Resolved: 30/Oct/18 02:14

Description

In an internal spark investigation, it appears that when EventLoggingListener writes to an EC file, one may get exceptions reading, or get odd outputs. A sample exception is

hdfs dfs -cat /user/spark/applicationHistory/application_1540333573846_0003 | head -1
18/10/23 18:12:39 WARN impl.BlockReaderFactory: I/O error constructing remote block reader.
java.io.IOException: Got error, status=ERROR, status message opReadBlock BP-1488936467-HOST_IP-1540333392519:blk_-9223372036854774960_1085 received exception java.io.IOException:  Offset 0 and length 116161 don't match block BP-1488936467-HOST_IP-1540333392519:blk_-9223372036854774960_1085 ( blockLen 110296 ), for OP_READ_BLOCK, self=/HOST_IP:48610, remote=/HOST2_IP:20002, for file /user/spark/applicationHistory/application_1540333573846_0003, for pool BP-1488936467-HOST_IP-1540333392519 block -9223372036854774960_1085
	at org.apache.hadoop.hdfs.protocol.datatransfer.DataTransferProtoUtil.checkBlockOpStatus(DataTransferProtoUtil.java:134)
	at org.apache.hadoop.hdfs.protocol.datatransfer.DataTransferProtoUtil.checkBlockOpStatus(DataTransferProtoUtil.java:110)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderRemote.checkSuccess(BlockReaderRemote.java:440)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderRemote.newBlockReader(BlockReaderRemote.java:408)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderFactory.getRemoteBlockReader(BlockReaderFactory.java:848)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderFactory.getRemoteBlockReaderFromTcp(BlockReaderFactory.java:744)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderFactory.build(BlockReaderFactory.java:379)
	at org.apache.hadoop.hdfs.DFSInputStream.getBlockReader(DFSInputStream.java:644)
	at org.apache.hadoop.hdfs.DFSStripedInputStream.createBlockReader(DFSStripedInputStream.java:264)
	at org.apache.hadoop.hdfs.StripeReader.readChunk(StripeReader.java:299)
	at org.apache.hadoop.hdfs.StripeReader.readStripe(StripeReader.java:330)
	at org.apache.hadoop.hdfs.DFSStripedInputStream.readOneStripe(DFSStripedInputStream.java:326)
	at org.apache.hadoop.hdfs.DFSStripedInputStream.readWithStrategy(DFSStripedInputStream.java:419)
	at org.apache.hadoop.hdfs.DFSInputStream.read(DFSInputStream.java:829)
	at java.io.DataInputStream.read(DataInputStream.java:100)
	at org.apache.hadoop.io.IOUtils.copyBytes(IOUtils.java:92)
	at org.apache.hadoop.io.IOUtils.copyBytes(IOUtils.java:66)
	at org.apache.hadoop.io.IOUtils.copyBytes(IOUtils.java:127)
	at org.apache.hadoop.fs.shell.Display$Cat.printToStdout(Display.java:101)
	at org.apache.hadoop.fs.shell.Display$Cat.processPath(Display.java:96)
	at org.apache.hadoop.fs.shell.Command.processPathInternal(Command.java:367)
	at org.apache.hadoop.fs.shell.Command.processPaths(Command.java:331)
	at org.apache.hadoop.fs.shell.Command.processPathArgument(Command.java:304)
	at org.apache.hadoop.fs.shell.Command.processArgument(Command.java:286)
	at org.apache.hadoop.fs.shell.Command.processArguments(Command.java:270)
	at org.apache.hadoop.fs.shell.FsCommand.processRawArguments(FsCommand.java:119)
	at org.apache.hadoop.fs.shell.Command.run(Command.java:177)
	at org.apache.hadoop.fs.FsShell.run(FsShell.java:326)
	at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:76)
	at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:90)
	at org.apache.hadoop.fs.FsShell.main(FsShell.java:389)
18/10/23 18:12:39 WARN hdfs.DFSClient: Failed to connect to /HOST2_IP:20002 for blockBP-1488936467-HOST_IP-1540333392519:blk_-9223372036854774960_1085
java.io.IOException: Got error, status=ERROR, status message opReadBlock BP-1488936467-HOST_IP-1540333392519:blk_-9223372036854774960_1085 received exception java.io.IOException:  Offset 0 and length 116161 don't match block BP-1488936467-HOST_IP-1540333392519:blk_-9223372036854774960_1085 ( blockLen 110296 ), for OP_READ_BLOCK, self=/HOST_IP:48610, remote=/HOST2_IP:20002, for file /user/spark/applicationHistory/application_1540333573846_0003, for pool BP-1488936467-HOST_IP-1540333392519 block -9223372036854774960_1085
	at org.apache.hadoop.hdfs.protocol.datatransfer.DataTransferProtoUtil.checkBlockOpStatus(DataTransferProtoUtil.java:134)
	at org.apache.hadoop.hdfs.protocol.datatransfer.DataTransferProtoUtil.checkBlockOpStatus(DataTransferProtoUtil.java:110)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderRemote.checkSuccess(BlockReaderRemote.java:440)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderRemote.newBlockReader(BlockReaderRemote.java:408)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderFactory.getRemoteBlockReader(BlockReaderFactory.java:848)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderFactory.getRemoteBlockReaderFromTcp(BlockReaderFactory.java:744)
	at org.apache.hadoop.hdfs.client.impl.BlockReaderFactory.build(BlockReaderFactory.java:379)
	at org.apache.hadoop.hdfs.DFSInputStream.getBlockReader(DFSInputStream.java:644)
	at org.apache.hadoop.hdfs.DFSStripedInputStream.createBlockReader(DFSStripedInputStream.java:264)
	at org.apache.hadoop.hdfs.StripeReader.readChunk(StripeReader.java:299)
	at org.apache.hadoop.hdfs.StripeReader.readStripe(StripeReader.java:330)
	at org.apache.hadoop.hdfs.DFSStripedInputStream.readOneStripe(DFSStripedInputStream.java:326)
	at org.apache.hadoop.hdfs.DFSStripedInputStream.readWithStrategy(DFSStripedInputStream.java:419)
	at org.apache.hadoop.hdfs.DFSInputStream.read(DFSInputStream.java:829)
	at java.io.DataInputStream.read(DataInputStream.java:100)
	at org.apache.hadoop.io.IOUtils.copyBytes(IOUtils.java:92)
	at org.apache.hadoop.io.IOUtils.copyBytes(IOUtils.java:66)
	at org.apache.hadoop.io.IOUtils.copyBytes(IOUtils.java:127)
	at org.apache.hadoop.fs.shell.Display$Cat.printToStdout(Display.java:101)
	at org.apache.hadoop.fs.shell.Display$Cat.processPath(Display.java:96)
	at org.apache.hadoop.fs.shell.Command.processPathInternal(Command.java:367)
	at org.apache.hadoop.fs.shell.Command.processPaths(Command.java:331)
	at org.apache.hadoop.fs.shell.Command.processPathArgument(Command.java:304)
	at org.apache.hadoop.fs.shell.Command.processArgument(Command.java:286)
	at org.apache.hadoop.fs.shell.Command.processArguments(Command.java:270)
	at org.apache.hadoop.fs.shell.FsCommand.processRawArguments(FsCommand.java:119)
	at org.apache.hadoop.fs.shell.Command.run(Command.java:177)
	at org.apache.hadoop.fs.FsShell.run(FsShell.java:326)
	at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:76)
	at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:90)
	at org.apache.hadoop.fs.FsShell.main(FsShell.java:389)
18/10/23 18:12:39 WARN hdfs.DFSClient: [DatanodeInfoWithStorage[HOST2_IP:20002,DS-f5bc0566-eeb0-43aa-84b9-551a3a6d01a6,DISK]] are unavailable and all striping blocks on them are lost. IgnoredNodes = null
{"Event":"SparkListenerLogStart","Spark Version":"2.4.0-cdh6.x-SNAPSHOT"}

Also, there are clearly fsync logs in NN for the file.

Looking from code, the only way this can happen is through the hsync overload on DFSStripedOutputStream. We should make that consistent with the hsync without parameters. It seems this was simply missed from day0 implementation in HDFS-7889.

Credit to irashid for investigating this issue from Spark

Issue Links

• relates to: HADOOP-13327 Add OutputStream + Syncable to the Filesystem Specification
• relates to: SPARK-25855 Don't use Erasure Coding for event log files