Details
-
Improvement
-
Status: Closed
-
Major
-
Resolution: Duplicate
-
None
-
None
-
None
Description
We've been having problems where the task tracker is destabilizing HDFS by filling disks and then releasing them. Various tasks can also interfere with each other this way.
We should have a configurable max working space for a task (and allow the setting of lower per task limits). The task tracker should track use against this limit and terminate a job if it overruns it.
Attachments
Issue Links
- duplicates
-
HADOOP-27 MapRed tries to allocate tasks to nodes that have no available disk space
- Closed
- is related to
-
HADOOP-324 "IOException: No space left on device" is handled incorrectly
- Closed