[
https://issues.apache.org/jira/browse/MAPREDUCE-4442?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13444729#comment-13444729
]
Rahul Jain commented on MAPREDUCE-4442:
---------------------------------------
Here is the workaround we've put; the time where job counters, state etc are
unavailable appears to depend upon the number of mappers/reducers for the job:
the more tasks there are, the more time it will take to get access to counters
restored.
We didn't have to worry about this issue any more after the workaround
described below was put in place.
{code}
RunningJob job = jobThatCompleted;
// Allow 20 seconds wait, and 0.005 seconds additional per mapper/reducer
long maxWaitInSecs = 20 + Math.round((numMaps + numReduces) * 0.005);
long currWait = 0;
long waitIncr = 1;
do {
try {
return job.getCounters();
} catch (IOException e) {
LOG.error("Could not access job counters due to exception;" +
e);
if (currWait < maxWaitInSecs) {
LOG.info("Waiting for " + waitIncr + " seconds before
another attempt; max remaining wait="
+ (maxWaitInSecs - currWait));
try {
Thread.sleep(waitIncr * 1000);
} catch (InterruptedException ie) {
LOG.warn("interrupted waitForJobCountersAccess() ", ie);
}
currWait += waitIncr;
waitIncr += 1;
continue;
} else {
LOG.error("Exceeded maximum wait for job counters; aborting
with received exception", e);
throw new RuntimeException(e);
}
}
} while (currWait < maxWaitInSecs);
{code}
> Accessing hadoop counters from a job is unreliable in yarn during AM process
> cleanup window
> --------------------------------------------------------------------------------------------
>
> Key: MAPREDUCE-4442
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-4442
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Affects Versions: 2.0.0-alpha
> Reporter: Rahul Jain
> Attachments: am_logs_counter_failure.html,
> rsrc_mgr_logs_counter_failed.txt
>
>
> We found this issue during our tests moving from MapReduceV1 to MapReduceV2.
> A few of our applications access job counters multiple times:
> a) After submission of job, while job is execution (works fine)
> b) Right after job complete notification is received (works fine)
> c) Few seconds after job complete notification (fails most of the time).
> The error snippet is as follows:
> {code}
> 2012-07-12 19:12:29,039 WARN [Client] Unexpected error reading responses on
> connection Thread[IPC Client (1252749669) connection to
> sjc1-ciq-ibm-grid07.carrieriq.com/10.202.50.187:47944 from hadoop,5,main]
> java.lang.NullPointerException
> at
> org.apache.hadoop.ipc.Client$Connection.receiveResponse(Client.java:852)
> at org.apache.hadoop.ipc.Client$Connection.run(Client.java:781)
> 2012-07-12 19:12:29,044 INFO [ClientServiceDelegate] Application state is
> completed. FinalApplicationStatus=SUCCEEDED. Redirecting to job history server
> 2012-07-12 19:12:29,132 INFO [ClientServiceDelegate] Application state is
> completed. FinalApplicationStatus=SUCCEEDED. Redirecting to job history server
> 2012-07-12 19:12:29,216 ERROR [UserGroupInformation]
> PriviledgedActionException as:hadoop (auth:SIMPLE) cause:java.io.IOException
> 2012-07-12 19:12:29,216 WARN [BaseOutputStageJob] getJobCounters: Unable to
> retrieve counters. null
> java.io.IOException
> at
> org.apache.hadoop.mapred.ClientServiceDelegate.invoke(ClientServiceDelegate.java:315)
> at
> org.apache.hadoop.mapred.ClientServiceDelegate.getJobCounters(ClientServiceDelegate.java:335)
> at
> org.apache.hadoop.mapred.YARNRunner.getJobCounters(YARNRunner.java:470)
> at org.apache.hadoop.mapreduce.Job$8.run(Job.java:719)
> at org.apache.hadoop.mapreduce.Job$8.run(Job.java:716)
> at java.security.AccessController.doPrivileged(Native Method)
> at javax.security.auth.Subject.doAs(Subject.java:396)
> at
> org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1232)
> at org.apache.hadoop.mapreduce.Job.getCounters(Job.java:716)
> at
> org.apache.hadoop.mapred.JobClient$NetworkedJob.getCounters(JobClient.java:396)
> {code}
> The connection to 10.202.50.187:47944 is actually the connection to AM;
> appears that we are connecting to AM to get the counters for the successful
> job and not yet to the history server.
>
> I'll attach the logs for AM and resource mgr separately, however no unusual
> activity is seen in those.
> This makes me suspect that we have a race condition in the code trying to
> access job counters when AM is finishing up and the job hasn't moved to
> history server yet.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
