Hue's simpler Job Designer app (not the Oozie designer app) that lets you submit simple 1-job MR or singular Pig/Hive actions, uses this very approach you're discussing. If its of any help, you can checkout its sources at https://github.com/cloudera/hue/tree/master/apps/jobsub (Python based).
Note that for a fully disconnected setup you must also placed all the code libraries (mapper/reducer/etc.) already on HDFS, otherwise users will have to upload them somehow to HDFS when they submit things (HDFS access can be REST-done too, without needing a Hadoop install - using HTTPFS or WebHDFS methods). On Wed, Sep 11, 2013 at 8:10 PM, Lorrie Tomek <[email protected]> wrote: > I would like to be able to submit individual map/reduce jobs from a java > app running at an arbitrary remote node. As (1) Oozie has REST support for > job submission API (and that appears not to exist elsewhere), (2) Oozie > supports many versions/variants of Hadoop, and (3) a single map/reduce job > can be represented as a trivial one action workflow ... Can the Oozie > client be installed/run on an arbitrary node? (or say does it need to run > on a Hadoop EdgeNode)? [If it needs to run on a Hadoop EdgeNode, would the > answer be different if I wrote own REST client using say Apache Wink > client?] > > Also, since I only really want to submit individual map/reduce jobs, not > complex workflows, is there any reason I should not use Oozie (because > workflow orchestration makes Oozie more complex than needed for my specific > task). > > LAT -- Harsh J
