[ https://issues.apache.org/jira/browse/YARN-2820?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14323788#comment-14323788
]
Tsuyoshi OZAWA commented on YARN-2820:
--------------------------------------
[~xgong] [~zxu] Oh, I overlooked the point. Good point, Xuan. My first suggestion is to use
[DFS-level retry|https://issues.apache.org/jira/browse/YARN-1778?focusedCommentId=14319725&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14319725],
but if we support generic filesystems which are not related to HDFS, it looks better to implement
RMStateStore-level retry as [~zxu] suggested firstly.
> Improve FileSystemRMStateStore to customize hdfs client retries for locateFollowingBlock
and completeFile for better error recovery.
> ------------------------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-2820
> URL: https://issues.apache.org/jira/browse/YARN-2820
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: resourcemanager
> Affects Versions: 2.5.0
> Reporter: zhihai xu
> Assignee: zhihai xu
> Attachments: YARN-2820.000.patch, YARN-2820.001.patch
>
>
> Improve FileSystemRMStateStore to customize hdfs client retries for locateFollowingBlock
and completeFile for better error recovery.
> When we use FileSystemRMStateStore as yarn.resourcemanager.store.class, We saw the following
IOexception cause the RM shutdown.
> {code}
> 2014-10-29 23:49:12,202 INFO org.apache.hadoop.yarn.server.resourcemanager.recovery.FileSystemRMStateStore:
> Updating info for attempt: appattempt_1409135750325_109118_000001 at:
> /tmp/hadoop-yarn/yarn/system/rmstore/FSRMStateRoot/RMAppRoot/application_1409135750325_109118/
> appattempt_1409135750325_109118_000001
> 2014-10-29 23:49:19,495 INFO org.apache.hadoop.hdfs.DFSClient: Could not complete
> /tmp/hadoop-yarn/yarn/system/rmstore/FSRMStateRoot/RMAppRoot/application_1409135750325_109118/
> appattempt_1409135750325_109118_000001.new.tmp retrying...
> 2014-10-29 23:49:23,757 INFO org.apache.hadoop.hdfs.DFSClient: Could not complete
> /tmp/hadoop-yarn/yarn/system/rmstore/FSRMStateRoot/RMAppRoot/application_1409135750325_109118/
> appattempt_1409135750325_109118_000001.new.tmp retrying...
> 2014-10-29 23:49:31,120 INFO org.apache.hadoop.hdfs.DFSClient: Could not complete
> /tmp/hadoop-yarn/yarn/system/rmstore/FSRMStateRoot/RMAppRoot/application_1409135750325_109118/
> appattempt_1409135750325_109118_000001.new.tmp retrying...
> 2014-10-29 23:49:46,283 INFO org.apache.hadoop.yarn.server.resourcemanager.recovery.FileSystemRMStateStore:
> Error updating info for attempt: appattempt_1409135750325_109118_000001
> java.io.IOException: Unable to close file because the last block does not have enough
number of replicas.
> 2014-10-29 23:49:46,284 ERROR org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore:
> Error storing/updating appAttempt: appattempt_1409135750325_109118_000001
> 2014-10-29 23:49:46,916 FATAL org.apache.hadoop.yarn.server.resourcemanager.ResourceManager:
> Received a org.apache.hadoop.yarn.server.resourcemanager.RMFatalEvent of type STATE_STORE_OP_FAILED.
Cause:
> java.io.IOException: Unable to close file because the last block does not have enough
number of replicas.
> at org.apache.hadoop.hdfs.DFSOutputStream.completeFile(DFSOutputStream.java:2132)
> at org.apache.hadoop.hdfs.DFSOutputStream.close(DFSOutputStream.java:2100)
> at org.apache.hadoop.fs.FSDataOutputStream$PositionCache.close(FSDataOutputStream.java:70)
> at org.apache.hadoop.fs.FSDataOutputStream.close(FSDataOutputStream.java:103)
> at org.apache.hadoop.yarn.server.resourcemanager.recovery.FileSystemRMStateStore.writeFile(FileSystemRMStateStore.java:522)
> at org.apache.hadoop.yarn.server.resourcemanager.recovery.FileSystemRMStateStore.updateFile(FileSystemRMStateStore.java:534)
> at org.apache.hadoop.yarn.server.resourcemanager.recovery.FileSystemRMStateStore.updateApplicationAttemptStateInternal(FileSystemRMStateStore.java:389)
> at org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore.handleStoreEvent(RMStateStore.java:675)
> at org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$ForwardingEventHandler.handle(RMStateStore.java:766)
> at org.apache.hadoop.yarn.server.resourcemanager.recovery.RMStateStore$ForwardingEventHandler.handle(RMStateStore.java:761)
> at org.apache.hadoop.yarn.event.AsyncDispatcher.dispatch(AsyncDispatcher.java:173)
> at org.apache.hadoop.yarn.event.AsyncDispatcher$1.run(AsyncDispatcher.java:106)
> at java.lang.Thread.run(Thread.java:744)
> {code}
> It will be better to Improve FileSystemRMStateStore to configure "dfs.client.block.write.locateFollowingBlock.retries"
to a bigger value for better error recovery.
> The default value for "dfs.client.block.write.locateFollowingBlock.retries" is 5.
> {code}
> public static final int DFS_CLIENT_BLOCK_WRITE_LOCATEFOLLOWINGBLOCK_RETRIES_DEFAULT
= 5;
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
|