[Hadoop-common-dev] [jira] Created: (HADOOP-5367) After some jobs have finished, Reducer will run new job's reduce tasks sequentially and not in parallel (mapred.JobTracker: Serious problem. While updating status, cannot find taskid...)

[ https://issues.apache.org/jira/browse/HADOOP-5367?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Thibaut updated HADOOP-5367:
----------------------------

Component/s: mapred

[ https://issues.apache.org/jira/browse/HADOOP-5367?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Thibaut updated HADOOP-5367:
----------------------------

Description:
Hi,

After I while, my cluster will only run the reduce tasks sequentially (each reducer running on the same node), the other nodes stay empty. The map phase however will run the jobs on all the nodes, also after such a "long" reduce phase has completed. But the reduce phase will then be again executed sequentially. This happens in my cluster after about 160 successfully completed jobs. (Some jobs have reducer set to 0!).
As possible solution I have to restart the mapreduce service.

I didn't notice this behaviour in version 0.19.0. I can't use version 0.19.0 because of the multipleoutput bug when setting reducers to 0.

Anoter site node which might be related. I also tried running the jobs with speculative execution set to on. My cluster would always hold back one reducer and only run it (in multiple instances) after the first of the other 6 reducers had finished, instead of launching all of them at the same time.


Below is a short extract from related logfile. It's full of these kind of entries.

09/02/28 12:48:07 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0051_r_000006_1
09/02/28 12:48:08 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0041_r_000002_1
09/02/28 12:48:08 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0083_r_000006_1
09/02/28 12:48:08 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0041_r_000005_1
09/02/28 12:48:10 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0105_r_000006_1
09/02/28 12:48:10 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0102_r_000006_1
09/02/28 12:48:12 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0051_r_000006_1
09/02/28 12:48:13 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0041_r_000002_1
09/02/28 12:48:13 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0083_r_000006_1
09/02/28 12:48:13 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0041_r_000005_1


was:
Hi,

After I while, my cluster will only run the reduce tasks sequentially (each reducer running on the same node), the other nodes stay empty. The map phase however will run the jobs on all the nodes. This happens in my cluster after about 160 successfully completed jobs. (Some jobs have reducer set to 0!).
As possible solution I have to restart the mapreduce service.

I didn't notice this behaviour in version 0.19.0. I can't use version 0.19.0 because of the multipleoutput bug when setting reducers to 0.

Anoter site node which might be related. I also tried running the jobs with speculative execution set to on. My cluster would always hold back one reducer and only run it (in multiple instances) after the first of the other 6 reducers had finished, instead of launching all of them at the same time.


Below is a short extract from related logfile. It's full of these kind of entries.

09/02/28 12:48:07 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0051_r_000006_1
09/02/28 12:48:08 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0041_r_000002_1
09/02/28 12:48:08 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0083_r_000006_1
09/02/28 12:48:08 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0041_r_000005_1
09/02/28 12:48:10 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0105_r_000006_1
09/02/28 12:48:10 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0102_r_000006_1
09/02/28 12:48:12 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0051_r_000006_1
09/02/28 12:48:13 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0041_r_000002_1
09/02/28 12:48:13 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0083_r_000006_1
09/02/28 12:48:13 INFO mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200902271700_0041_r_000005_1

[ https://issues.apache.org/jira/browse/HADOOP-5367?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12677883#action_12677883 ]

Amareshwari Sriramadasu commented on HADOOP-5367:
-------------------------------------------------

This could be mostly because of HADOOP-5269 and HADOOP-5235. There are committed to branch 0.19.2. Can you please try out 0.19.2 by checking out branch 0.19 ?

[ https://issues.apache.org/jira/browse/HADOOP-5367?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12682698#action_12682698 ]

schubert zhang commented on HADOOP-5367:
----------------------------------------

I also meet this such issue.

After a long time running of MapReduce (about 200 jobs have completed). The MapReduce job is huaguped forever.
(1) The JobTracker always logs:
2009-03-17 16:29:39,997 INFO org.apache.hadoop.mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200903171247_0387_m_000015_1

(2) The Job cannot complete and stopp at 79% forever.

(3) All TaskTrackers may hungup at the sametime, since the logs of each TaskTracer stop at that time.


And nefore the hangup. I can also find odds and ends such logs of JobTracker, such as.
2009-03-17 16:29:21,767 INFO org.apache.hadoop.mapred.JobTracker: Serious problem. While updating status, cannot find taskid attempt_200903171247_0387_m_000015_1


And another experience is:
One time, I found the task slot cannot reach the capability, maybe some slot is also hungup.

[ https://issues.apache.org/jira/browse/HADOOP-5367?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12689809#action_12689809 ]

schubert zhang commented on HADOOP-5367:
----------------------------------------

I am using branch-0.19, it seems fine.

[ https://issues.apache.org/jira/browse/HADOOP-5367?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Amareshwari Sriramadasu resolved HADOOP-5367.
---------------------------------------------

Resolution: Fixed

Got fixed in 0.19.2