Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
3.4.0
-
Reviewed
Description
Currently inactive nodes are removed from the Yarn local memory irrespective of which state they are in. This makes the node removal process not too much configurable
After this patch: https://issues.apache.org/jira/browse/YARN-10854
If autoscaling is enabled, lot many nodes go into DECOMMISSIONED state but still other states like LOST, SHUTDOWN are very less and systems might want them to be still visible on UI for better tracking
The proposal is to introduce a new config, which when set, will allow only selective node states to be removed after going into untracked state.
PR: https://github.com/apache/hadoop/pull/5681
Any thoughts/suggestions/feedbacks are welcome!
Attachments
Issue Links
- links to