Hadoop - force task attempt to start on different node -
i submitted job cluster of 4 hosts, can see correctly spread among 4 nodes, 1 map task per node.
later on, 1 of node failed.
i stopped tasktracker on failed node, added id of node excludes file , updated list of nodes hadoop mradmin -refreshnodes. failed node disappeared list of available nodes on hadoop administration pages.
then started tasktracker again, updated nodes mradmin, , observed node appeared in job tracker list again.
during time of node being down, hadoop re-scheduled map task execution on node, started run 2 map jobs. i've got cluster unbalanced:
- 2 nodes running 1 task each,
- 1 node running 2 tasks
- and 1 node (the 1 restarted) running no tasks.
i killed job hadoop job -kill-task attempt_201308010141_0001_m_000000_1 , looks never starts again - can see 3 nodes running 1 task each, 1 node no tasks @ , 1 pending task in list.
am missing something? correct way of 'moving' task 1 node one?
jobs keep list of blacklisted tasktrackers (there global blacklist , per job one). think that's why new attempt don't start again @ end on restarted task tracker.
you can try commands :
hadoop job -unblacklist <jobid> <hostname> hadoop job -unblacklist-tracker <hostname> from http://doc.mapr.com/display/mapr/tasktracker+blacklisting
Comments
Post a Comment