Michael Prim commented on TEZ-3435:

[~hitesh] this sounds like the explanation. We have of course other nodes with 
broken disks, but in this case the disk was entirely broken and dead. Whereas 
in other cases the disks had only bad blocks and got mounted as read-only file 
systems until we replaced the disk. So if there was something in the filecache 
on these read-only devices it would explain why we did not hit the issue 

Thanks for you help, not sure how to proceed. Most likely not a TEZ UI issue, 
so you may want to close it.

> WebUIService thread tries to use blacklisted disk, dies, and kills AM
> ---------------------------------------------------------------------
>                 Key: TEZ-3435
>                 URL: https://issues.apache.org/jira/browse/TEZ-3435
>             Project: Apache Tez
>          Issue Type: Bug
>          Components: UI
>    Affects Versions: 0.8.4
>            Reporter: Michael Prim
>            Priority: Critical
> We recently hit an issue that certain TEZ jobs died when scheduled on a node 
> that had a broken disk. The disk was already marked as broken and excluded by 
> YARN node manager. Other applications worked fine on that node, only TEZ jobs 
> died.
> The error where ClassNotFound exceptions, of basic hadoop classes, which 
> should be available everywhere. After some investigation we found out that 
> the WebUIService thread, spawned by the AM tries to utilize that broken disk. 
> See stacktrace, disk3 was excluded by node manager.
> {code}
>  [WARN] [ServiceThread:org.apache.tez.dag.app.web.WebUIService] 
> |mortbay.log|: Failed to read file: 
> /volumes/disk3/yarn/nm/filecache/9017/hadoop-mapreduce-client-core-2.6.0.jar
> java.util.zip.ZipException: error in opening zip file
>       at java.util.zip.ZipFile.open(Native Method)
>       at java.util.zip.ZipFile.<init>(ZipFile.java:219)
>       at java.util.zip.ZipFile.<init>(ZipFile.java:149)
>       at java.util.jar.JarFile.<init>(JarFile.java:166)
>       at java.util.jar.JarFile.<init>(JarFile.java:130)
>       at 
> org.mortbay.jetty.webapp.TagLibConfiguration.configureWebApp(TagLibConfiguration.java:174)
>       at 
> org.mortbay.jetty.webapp.WebAppContext.startContext(WebAppContext.java:1279)
>       at 
> org.mortbay.jetty.handler.ContextHandler.doStart(ContextHandler.java:518)
>       at 
> org.mortbay.jetty.webapp.WebAppContext.doStart(WebAppContext.java:499)
>       at 
> org.mortbay.component.AbstractLifeCycle.start(AbstractLifeCycle.java:50)
>       at 
> org.mortbay.jetty.handler.HandlerCollection.doStart(HandlerCollection.java:152)
>       at 
> org.mortbay.jetty.handler.ContextHandlerCollection.doStart(ContextHandlerCollection.java:156)
>       at 
> org.mortbay.component.AbstractLifeCycle.start(AbstractLifeCycle.java:50)
>       at 
> org.mortbay.jetty.handler.HandlerWrapper.doStart(HandlerWrapper.java:130)
>       at org.mortbay.jetty.Server.doStart(Server.java:224)
>       at 
> org.mortbay.component.AbstractLifeCycle.start(AbstractLifeCycle.java:50)
>       at org.apache.hadoop.http.HttpServer2.start(HttpServer2.java:900)
>       at org.apache.hadoop.yarn.webapp.WebApps$Builder.start(WebApps.java:273)
>       at 
> org.apache.tez.dag.app.web.WebUIService.serviceStart(WebUIService.java:94)
>       at 
> org.apache.hadoop.service.AbstractService.start(AbstractService.java:193)
>       at 
> org.apache.tez.dag.app.DAGAppMaster$ServiceWithDependency.start(DAGAppMaster.java:1827)
>       at 
> org.apache.tez.dag.app.DAGAppMaster$ServiceThread.run(DAGAppMaster.java:1848)
> {code}
> Which did lead to the ClassNotFound exceptions and killing the AM. 
> Interesting enough the DAGAppMaster was aware of this broken disk and did 
> exclude it from the localDirs. It contains only the remaining disks of the 
> node.
> {code}
> [INFO] [main] |app.DAGAppMaster|: Creating DAGAppMaster for 
> applicationId=application_1472223062609_42648, attemptNum=1, 
> AMContainerId=container_1472223062609_42648_01_000001, jvmPid=2538, 
> userFromEnv=muhammad, cliSessionOption=true, 
> pwd=/volumes/disk1/yarn/nm/usercache/muhammad/appcache/application_1472223062609_42648/container_1472223062609_42648_01_000001,
> localDirs=/volumes/disk1/yarn/nm/usercache/muhammad/appcache/application_1472223062609_42648,/volumes/disk10/yarn/nm/usercache/muhammad/appcache/application_1472223062609_42648,/volumes/disk4/yarn/nm/usercache/muhammad/appcache/application_1472223062609_42648,/volumes/disk5/yarn/nm/usercache/muhammad/appcache/application_1472223062609_42648,/volumes/disk6/yarn/nm/usercache/muhammad/appcache/application_1472223062609_42648,/volumes/disk7/yarn/nm/usercache/muhammad/appcache/application_1472223062609_42648,/volumes/disk8/yarn/nm/usercache/muhammad/appcache/application_1472223062609_42648,/volumes/disk9/yarn/nm/usercache/muhammad/appcache/application_1472223062609_42648,
> logDirs=/var/log/hadoop-yarn/container/application_1472223062609_42648/container_1472223062609_42648_01_000001
> {code}
> Actually this is quite an issue as in a huge data center you always have some 
> broken disks and by chance your AM may scheduled on one of this nodes.
> Summary: From my point of view it looks like as if the WebUIService thread 
> does somehow not properly take into account the local directories that are 
> excluded by the node manager.

This message was sent by Atlassian JIRA

Reply via email to