[
https://issues.apache.org/jira/browse/HADOOP-5272?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Amar Kamat updated HADOOP-5272:
-------------------------------
Attachment: HADOOP-5272-v1.0.patch
Attaching a patch that does 2 things
# Maintain the first attempt seen by the TIP. This provides 2 functions
## Any random attempt should not classify as the first attempt (possible in
trunk)
## Only the first seen/processed attempt is marked as first and hence across
restart, joining tasks (attemptid > 0) can also act as the first attempt.
# In JIP.addRunningTaskToTIP(), add the attempt to the tip before anything
else. This solves 2 problems
## Helps in computing (1)
## speculation counts are now computed in JIP.addRunningTaskToTIP() assuming
that the attempt is already added.
HADOOP-4191 should test this patch.
> JobTracker does not log TIP start information after restart
> -----------------------------------------------------------
>
> Key: HADOOP-5272
> URL: https://issues.apache.org/jira/browse/HADOOP-5272
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Reporter: Amar Kamat
> Priority: Critical
> Attachments: HADOOP-5272-v1.0.patch
>
>
> In case of JobTracker restart, attempt_0123456789_0001_m_000000_0 might not
> be the first attempt to get scheduled as the attempt id offset changes after
> every restart. For example,
> upon first restart the new attempt id will be
> attempt_0123456789_0001_m_000000_1000. Hence TIP start line never gets logged
> to the job history after the restart, TaskInProgress.isFirstAttempt() always
> returns *false*.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
