Details
Description
The default value of yarn.timeline-service.app-collector.linger-period.ms is 1 sec which is very low. In production cluster, we observe that most of the entity information is not published because of 1sec linger period.
I would suggest to increase the default value to either 10sec or 30sec or 60 sec. As far as I have seen, 10sec should be sufficient since node heartbeat period is 3 sec by default, all NM gets request to stop the application.
cc:/ haibo.chen vrushalic
Attachments
Attachments
Issue Links
- is related to
-
YARN-8130 Race condition when container events are published for KILLED applications
- Resolved