-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/29765/
-----------------------------------------------------------
Review request for Ambari and Myroslav Papirkovskyy.
Bugs: AMBARI-9064
https://issues.apache.org/jira/browse/AMBARI-9064
Repository: ambari
Description
-------
we need to change the nm retry settings to resolve the bug. The proposed new
settings in yarn-site.xml are:
yarn.client.nodemanager-connect.max-wait-ms = 60,000
yarn.client.nodemanager-connect.retry-interval-ms = 10,000
If the underlying rpc retry takes 1 sec * 50 times to finish (50 secs), this
will make each round of retry to be 10 + 50 secs = 1 minute. From the code,
there will be 60,000 / 10,000 = 6 times of retries. So, in total, if the node
manager accidentally crashed and never goes up, we well retry 1 minute * 6
times = 6 minutes. The whole calculation is based on the rpc settings (50
times, 1 sec sleep time),
Diffs
-----
ambari-server/src/main/resources/stacks/HDP/2.2.GlusterFS/services/YARN/configuration/yarn-site.xml
2eb1bcc
ambari-server/src/main/resources/stacks/HDP/2.2/services/YARN/configuration/yarn-site.xml
847711a
ambari-server/src/main/resources/stacks/HDPWIN/2.2/services/YARN/configuration/yarn-site.xml
847711a
Diff: https://reviews.apache.org/r/29765/diff/
Testing
-------
[INFO] ------------------------------------------------------------------------
[INFO] Reactor Summary:
[INFO]
[INFO] Ambari Views ...................................... SUCCESS [4.585s]
[INFO] Ambari Server ..................................... SUCCESS [33.352s]
[INFO] Ambari Agent ...................................... SUCCESS [15.611s]
[INFO] ------------------------------------------------------------------------
[INFO] BUILD SUCCESS
[INFO] ------------------------------------------------------------------------
[INFO] Total time: 56.009s
[INFO] Finished at: Fri Jan 09 18:43:37 FET 2015
[INFO] Final Memory: 38M/381M
[INFO] ------------------------------------------------------------------------
Thanks,
Dmitro Lisnichenko