[
https://issues.apache.org/jira/browse/HDFS-11208?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15788385#comment-15788385
]
Chris Nauroth commented on HDFS-11208:
--------------------------------------
This is somewhat similar to a deadlock we encountered on
{{S3AFileSystem#close}}, reported in HADOOP-13599. In that case, we were able
to resolve the problem by removing {{synchronized}} from
{{S3AFileSystem#close}}. This one is going to be trickier.
> Deadlock in WebHDFS on shutdown
> -------------------------------
>
> Key: HDFS-11208
> URL: https://issues.apache.org/jira/browse/HDFS-11208
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: webhdfs
> Affects Versions: 2.8.0, 2.7.3, 2.6.5, 3.0.0-alpha1
> Reporter: Erik Krogen
> Assignee: Erik Krogen
> Attachments: HDFS-11208-test-deadlock.patch
>
>
> Currently on the client side if the {{DelegationTokenRenewer}} attempts to
> renew a WebHdfs delegation token while the client system is shutting down
> (i.e. {{FileSystem.Cache.ClientFinalizer}} is running) a deadlock may occur.
> This happens because {{ClientFinalizer}} calls
> {{FileSystem.Cache.closeAll()}} which first takes a lock on the
> {{FileSystem.Cache}} object and then locks each file system in the cache as
> it iterates over them. {{DelegationTokenRenewer}} takes a lock on a
> filesystem object while it is renewing that filesystem's token, but within
> {{TokenAspect.TokenManager.renew()}} (used for renewal of WebHdfs tokens)
> {{FileSystem.get}} is called, which in turn takes a lock on the FileSystem
> cache object, potentially causing deadlock if {{ClientFinalizer}} is
> currently running.
> See below for example deadlock output:
> {code}
> Found one Java-level deadlock:
> =============================
> "Thread-8572":
> waiting to lock monitor 0x00007eff401f9878 (object 0x000000051ec3f930, a
> dali.hdfs.web.WebHdfsFileSystem),
> which is held by "FileSystem-DelegationTokenRenewer"
> "FileSystem-DelegationTokenRenewer":
> waiting to lock monitor 0x00007f005c08f5c8 (object 0x000000050389c8b8, a
> dali.fs.FileSystem$Cache),
> which is held by "Thread-8572"
> Java stack information for the threads listed above:
> ===================================================
> "Thread-8572":
> at dali.hdfs.web.WebHdfsFileSystem.close(WebHdfsFileSystem.java:864)
> - waiting to lock <0x000000051ec3f930> (a
> dali.hdfs.web.WebHdfsFileSystem)
> at dali.fs.FilterFileSystem.close(FilterFileSystem.java:449)
> at dali.fs.FileSystem$Cache.closeAll(FileSystem.java:2407)
> - locked <0x000000050389c8b8> (a dali.fs.FileSystem$Cache)
> at dali.fs.FileSystem$Cache$ClientFinalizer.run(FileSystem.java:2424)
> - locked <0x000000050389c8d0> (a
> dali.fs.FileSystem$Cache$ClientFinalizer)
> at dali.util.ShutdownHookManager$1.run(ShutdownHookManager.java:54)
> "FileSystem-DelegationTokenRenewer":
> at dali.fs.FileSystem$Cache.getInternal(FileSystem.java:2343)
> - waiting to lock <0x000000050389c8b8> (a dali.fs.FileSystem$Cache)
> at dali.fs.FileSystem$Cache.get(FileSystem.java:2332)
> at dali.fs.FileSystem.get(FileSystem.java:369)
> at
> dali.hdfs.web.TokenAspect$TokenManager.getInstance(TokenAspect.java:92)
> at dali.hdfs.web.TokenAspect$TokenManager.renew(TokenAspect.java:72)
> at dali.security.token.Token.renew(Token.java:373)
> at
>
> dali.fs.DelegationTokenRenewer$RenewAction.renew(DelegationTokenRenewer.java:127)
> - locked <0x000000051ec3f930> (a dali.hdfs.web.WebHdfsFileSystem)
> at
>
> dali.fs.DelegationTokenRenewer$RenewAction.access$300(DelegationTokenRenewer.java:57)
> at dali.fs.DelegationTokenRenewer.run(DelegationTokenRenewer.java:258)
> Found 1 deadlock.
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
