Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
2.6.0
-
Hadoop 2.6.0-cdh5.5.1 Express edition.
-
Important
Description
There is a business impact MR job which runs every day @ 2.00 PM PST and data size is about 1 - 1.5 TB (depends on the business days) . Ideal elapsed time of this job : 4 hrs. But the multiple mappers of this job simultaneously failing with the following error so job will take some times 11 and even 13 hours also like that.
Steps to prevent this problem : 1, Migrated the environment to Yarn .2 increased the ulimit 3. Added extra nodes to the cluster. 4. Disks replacement taking place regularly 5. Monitoring the cluster and terminating other jobs which impacts this job.
Few of the values that we tried increasing without any benefit are
1. increased open files
2. increase dfs.datanode.handler.count
3. increase dfs.datanode.max.xcievers
4. increase dfs.datanode.max.transfer.threads
But no luck.
org.apache.hadoop.hdfs.DFSClient: Error Recovery for block BP-854530680-69.194.253.58-1430267558563:blk_4683766046_1108754130089 in pipeline DatanodeInfoWithStorage
[10.0.1.37:50010,DS-ed333d2e-839a-4029-a1c9-b6615c322ed2,DISK],
DatanodeInfoWithStorage[74.120.143.19:50010,DS-5d10576e-adc3-474f-bc9d-f0d6fb3ae4c3,DISK],
DatanodeInfoWithStorage[74.120.143.6:50010,DS-a5299d68-2858-46c3-8e37-d2559895f979,DISK]
bad datanode DatanodeInfoWithStorage[10.0.1.37:50010,DS-ed333d2e-839a-4029-a1c9-b6615c322ed2,DISK]
org.apache.hadoop.hdfs.DFSClient: Error Recovery for block BP-854530680-69.194.253.58-1430267558563:blk_4683766046_1108754130089 in pipeline DatanodeInfoWithStorage[74.120.143.19:50010,DS-5d10576e-adc3-474f-bc9d-f0d6fb3ae4c3,DISK], DatanodeInfoWithStorage[74.120.143.6:50010,DS-a5299d68-2858-46c3-8e37-d2559895f979,DISK]: bad datanode DatanodeInfoWithStorage[74.120.143.19:50010,DS-5d10576e-adc3-474f-bc9d-f0d6fb3ae4c3,DISK]
org.apache.hadoop.mapred.YarnChild: Exception running child : java.io.IOException: java.io.IOException: All datanodes DatanodeInfoWithStorage[74.120.143.6:50010,DS-a5299d68-2858-46c3-8e37-d2559895f979,DISK] are bad. Aborting... at