primarily two reasons: 1. There is no recurring communication channel between NM and AM.
2. Application must act on single source of truth, which is RM. Sharad On Sun, Nov 27, 2011 at 9:01 PM, Praveen Sripati <[email protected]>wrote: > Hi, > > According to > > > http://hadoop.apache.org/common/docs/r0.23.0/hadoop-yarn/hadoop-yarn-site/WritingYarnApplications.html > > >Completed containers: Once an ApplicationMaster triggers a launch an > allocated container, it will receive an update from the ResourceManager > when the container completes. The ApplicationMaster can look into the > status of the completed container and take appropriate actions such as > re-trying a particular sub-task in case of a failure. > > My understanding is NM sends the completed status of the containers to the > RM and the RM in turn sends the status to AM. Why can't the NM send the > status of the completed containers to the AM directly? The NM will anyway > know which AM has requested for launching the containers. > > Thanks, > Praveen >
