[GitHub] hadoop pull request:

[smarella <gi...@git.apache.org>]

Github user smarella commented on the pull request:

    https://github.com/apache/hadoop/commit/968425e9f7b850ff9c2ab8ca37a64c3fdbe77dbf#commitcomment-14835835
  
    In hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/RMAppManager.java:
    In hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/RMAppManager.java on line 397:
    YARN-3305 seem to break the Myriad builds (see below). As indicated in YARN-3305, https://issues.apache.org/jira/browse/YARN-3996 is trying to fix the problem. YARN-3996 is currently work-in-progress, but we need to back port some of it immediately to unblock Myriad. Once YARN-3996 is fully fixed, it needs to be fully back ported. I've requested Sarjeet to raise two bugs for this (one to provide an immediate fix and second to back port YARN-3996 once it's resolved).
    
    Currently, the latest Myriad builds are effected by this change. The problem is that, the AM container resources are normalized during app submission with ```scheduler.getminimumResourceCapability()```, which comes from yarn.scheduler.minimum-allocation-{mb,vcores,disks}. These values are set to "0" by Myriad. Hence, the AM container allocated will be of zero size and is killed by NM as soon as it is launched.
    
    The reason why Myriad needs yarn.scheduler.minimum-allocation-{mb,vcores,disks} to be 0 is for discussed in https://issues.apache.org/jira/browse/MYRIAD-139.


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---