zhengchenyu commented on PR #235: URL: https://github.com/apache/tez/pull/235#issuecomment-1220522116
> thanks for this patch @zhengchenyu! > > can you include a unit test to TestTaskScheduler which confirms that a TaskScheduler returns Resource(0,0) even if the RM client returned null? > > I'm not familiar with yarn federation, but defaulting to Resource(0,0) makes sense in edge cases can you please clarify if this is specific to yarn federation or can happen without yarn federation too? (it has never been reported yet) why does it return null? does it reflect the state of a specific RM or the whole cluster of RMs? It happen only in yarn federation, will never happen without yarn federation. In fact, YARN-8933 have fix it. After apply YARN-8933, it will never happen in yarn federation. I don't know it is necessary to continue it. Because it is not a problem for latest hadoop version, but still a problem for some popular version (For example: hadoop-3.2.1). If you think it is necessary, I will add some unit test. If you think it is not necessary, I will close it. ### For why return null in yarn federation? It is another issue about yarn. Yarn router use some async thread to connect rm. When all down streaming resourcemanager timeout, yarn router may return null. But After YARN-8933, will return Resource(0,0). -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@tez.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
