[SPARK-15519] Shuffle Service fails to start if first yarn.nodemanager.local-dirs is bad

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 1.6.1
• Component/s: Shuffle, Spark Core, YARN

Environment

Ubuntu 14.04 LTS, MapR 5.1, hadoop-2.7.0

• Assignee: Unassigned
• Reporter: Matthew Sharp
• Created: 25/May/16 00:08
• Updated: 17/May/20 18:13
• Resolved: 25/May/16 12:16

Description

yarn.nodemanager.local-dirs is set to /mnt/data0,/mnt/data1,/mnt/data2,/mnt/data3,/mnt/data4

/mnt/data0 was not mounted due to a disk failure, so it was an empty directory which users were not allowed to write to.

Starting up the node manager, we get this in the logs:

2016-05-24 15:41:56,456 INFO org.apache.spark.network.yarn.YarnShuffleService: Initializing YARN shuffle service for Spark
2016-05-24 15:41:56,456 INFO org.apache.hadoop.yarn.server.nodemanager.containermanager.AuxServices: Adding auxiliary service spark_shuffle, "spark_shuffle"
2016-05-24 15:41:56,609 ERROR org.apache.spark.network.shuffle.ExternalShuffleBlockResolver: error opening leveldb file /mnt/data0/registeredExecutors.ldb. Creating new file, will not be able to recover state for existing applications
org.fusesource.leveldbjni.internal.NativeDB$DBException: IO error: /mnt/data0/registeredExecutors.ldb/LOCK: Permission denied
at org.fusesource.leveldbjni.internal.NativeDB.checkStatus(NativeDB.java:200)
at org.fusesource.leveldbjni.internal.NativeDB.open(NativeDB.java:218)
at org.fusesource.leveldbjni.JniDBFactory.open(JniDBFactory.java:168)
at org.apache.spark.network.shuffle.ExternalShuffleBlockResolver.<init>(ExternalShuffleBlockResolver.java:100)
at org.apache.spark.network.shuffle.ExternalShuffleBlockResolver.<init>(ExternalShuffleBlockResolver.java:81)
at org.apache.spark.network.shuffle.ExternalShuffleBlockHandler.<init>(ExternalShuffleBlockHandler.java:56)
at org.apache.spark.network.yarn.YarnShuffleService.serviceInit(YarnShuffleService.java:128)
at org.apache.hadoop.service.AbstractService.init(AbstractService.java:163)
at org.apache.hadoop.yarn.server.nodemanager.containermanager.AuxServices.serviceInit(AuxServices.java:157)
at org.apache.hadoop.service.AbstractService.init(AbstractService.java:163)
at org.apache.hadoop.service.CompositeService.serviceInit(CompositeService.java:107)
at org.apache.hadoop.yarn.server.nodemanager.containermanager.ContainerManagerImpl.serviceInit(ContainerManagerImpl.java:250)
at org.apache.hadoop.service.AbstractService.init(AbstractService.java:163)
at org.apache.hadoop.service.CompositeService.serviceInit(CompositeService.java:107)
at org.apache.hadoop.yarn.server.nodemanager.NodeManager.serviceInit(NodeManager.java:256)
at org.apache.hadoop.service.AbstractService.init(AbstractService.java:163)
at org.apache.hadoop.yarn.server.nodemanager.NodeManager.initAndStartNodeManager(NodeManager.java:476)
at org.apache.hadoop.yarn.server.nodemanager.NodeManager.main(NodeManager.java:524)
2016-05-24 15:41:56,611 WARN org.apache.spark.network.shuffle.ExternalShuffleBlockResolver: error deleting /mnt/data0/registeredExecutors.ldb
2016-05-24 15:41:56,611 ERROR org.apache.spark.network.yarn.YarnShuffleService: Failed to initialize external shuffle service
java.io.IOException: Unable to create state store
at org.apache.spark.network.shuffle.ExternalShuffleBlockResolver.<init>(ExternalShuffleBlockResolver.java:129)
at org.apache.spark.network.shuffle.ExternalShuffleBlockResolver.<init>(ExternalShuffleBlockResolver.java:81)
at org.apache.spark.network.shuffle.ExternalShuffleBlockHandler.<init>(ExternalShuffleBlockHandler.java:56)
at org.apache.spark.network.yarn.YarnShuffleService.serviceInit(YarnShuffleService.java:128)
at org.apache.hadoop.service.AbstractService.init(AbstractService.java:163)
at org.apache.hadoop.yarn.server.nodemanager.containermanager.AuxServices.serviceInit(AuxServices.java:157)
at org.apache.hadoop.service.AbstractService.init(AbstractService.java:163)
at org.apache.hadoop.service.CompositeService.serviceInit(CompositeService.java:107)
at org.apache.hadoop.yarn.server.nodemanager.containermanager.ContainerManagerImpl.serviceInit(ContainerManagerImpl.java:250)
at org.apache.hadoop.service.AbstractService.init(AbstractService.java:163)
at org.apache.hadoop.service.CompositeService.serviceInit(CompositeService.java:107)
at org.apache.hadoop.yarn.server.nodemanager.NodeManager.serviceInit(NodeManager.java:256)
at org.apache.hadoop.service.AbstractService.init(AbstractService.java:163)
at org.apache.hadoop.yarn.server.nodemanager.NodeManager.initAndStartNodeManager(NodeManager.java:476)
at org.apache.hadoop.yarn.server.nodemanager.NodeManager.main(NodeManager.java:524)
Caused by: org.fusesource.leveldbjni.internal.NativeDB$DBException: IO error: /mnt/data0/registeredExecutors.ldb/LOCK: Permission denied
at org.fusesource.leveldbjni.internal.NativeDB.checkStatus(NativeDB.java:200)
at org.fusesource.leveldbjni.internal.NativeDB.open(NativeDB.java:218)
at org.fusesource.leveldbjni.JniDBFactory.open(JniDBFactory.java:168)
at org.apache.spark.network.shuffle.ExternalShuffleBlockResolver.<init>(ExternalShuffleBlockResolver.java:127)
... 14 more
2016-05-24 15:41:56,723 INFO org.apache.spark.network.yarn.YarnShuffleService: Started YARN shuffle service for Spark on port 7337. Authentication is not enabled. Registered executor file is /mnt/data0/registeredExecutors.ldb

Later on, when jobs run on that node, we get many of this message:

2016-05-24 15:39:57,171 ERROR org.apache.spark.network.TransportContext: Error while initializing Netty pipeline
java.lang.NullPointerException
at org.apache.spark.network.server.TransportRequestHandler.<init>(TransportRequestHandler.java:77)
at org.apache.spark.network.TransportContext.createChannelHandler(TransportContext.java:159)
at org.apache.spark.network.TransportContext.initializePipeline(TransportContext.java:135)
at org.apache.spark.network.server.TransportServer$1.initChannel(TransportServer.java:123)
at org.apache.spark.network.server.TransportServer$1.initChannel(TransportServer.java:116)
at io.netty.channel.ChannelInitializer.channelRegistered(ChannelInitializer.java:69)
at io.netty.channel.AbstractChannelHandlerContext.invokeChannelRegistered(AbstractChannelHandlerContext.java:158)
at io.netty.channel.AbstractChannelHandlerContext.fireChannelRegistered(AbstractChannelHandlerContext.java:144)
at io.netty.channel.DefaultChannelPipeline.fireChannelRegistered(DefaultChannelPipeline.java:732)
at io.netty.channel.AbstractChannel$AbstractUnsafe.register0(AbstractChannel.java:442)
at io.netty.channel.AbstractChannel$AbstractUnsafe.access$100(AbstractChannel.java:374)
at io.netty.channel.AbstractChannel$AbstractUnsafe$1.run(AbstractChannel.java:418)
at io.netty.util.concurrent.SingleThreadEventExecutor.runAllTasks(SingleThreadEventExecutor.java:380)
at io.netty.channel.nio.NioEventLoop.run(NioEventLoop.java:357)
at io.netty.util.concurrent.SingleThreadEventExecutor$2.run(SingleThreadEventExecutor.java:116)
at java.lang.Thread.run(Thread.java:745)

We were not seeing this issue with the 1.3.1 shuffle service.

We would expect that the shuffle service would fail over to the next local-dir if the first one fails.

Issue Links

• duplicates: SPARK-14963 YarnShuffleService should use YARN getRecoveryPath() for leveldb location