I noticed the following issues with the capacity scheduler doc: ./hadoop-mapreduce-project/hadoop-yarn/hadoop-yarn-site/src/site/apt/CapacityScheduler.apt.vm
- In overview section, 3rd paragraph, sentence "There is an added benefit that an organization can access any excess capacity no being used by others". No should be not.
- in overview section, 4th paragraph. dispropotionate misspelled
- in features section, under multitenancy - monopolizing is misspelled.
- in features section, under operability - it doesn't say if you can delete queues at runtime? I see there is a note at the end but perhaps that can be added into the other sections to since its easy to miss that Note at the very end.
- in features section - hierarchy and Hierarchical mispelled.
- under configuration section the class to turn on to use capacity scheduler should be: org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler
- section on setting up queues, 4th sentence - hierarchy misspelled as heirarcy and heirarchy.
- I think specifying how a user has to specify the queue when running a job/app would be useful information. Especially with the new hierarchical queues. Does the user have to specify the entire path like a.b.c or can they just specify c.
- under "Running and Pending Application Limits" section, property "yarn.scheduler.capacity.maximum-applications", they are referred to them as jobs, I believe that should be applications.
- misspelled concurrently as concurently in same section of maximum-applications.
- I think it should specify the defaults (if any) for the config vars. Also what format are they specified in - int, float,etc?
- might be nice to say it doesn't support preemption.
- under admin options yarn.scheduler.capacity.<queue-path>.state - queues misspelled as queueus
- under changing queue configuration it should have "yarn" in front of the "rmadmin -refreshQueues". Similarly a few lines down at "$YARN_HOME/bin/rmadmin -refreshQueues"