On 02/07/11 03:22, Raja Nagendra Kumar wrote:
Hi, Is there any plans for Hadoop to support annodations specially for the api level configurations eliminations.. eg. conf.setMapperClass(MaxTemperatureMapper.class); conf.setCombinerClass(MaxTemperatureReducer.class); conf.setReducerClass(MaxTemperatureReducer.class); can be easily elimiated through proper class level annotations.
How? Would something be expected to run through the entire **/*.class to look for annotated classes and somehow determine which ones a specific job is using? A lot of our JARs contain all the classes for many actions, with the combination of map and reduce chosen from job to job. Reading in all the annotations hits startup time and would cause confusion.
What could be done would be a spring-style file (JSON, etc) that declares the Map, Reduce, Combiner, other conf options and builds the conf from it, but that could be a job submission interface in front of the classes. Something that would work well with the outstanding "provide a long-haul job submission API that is robust against Hadoop version changes" issue in hadoop-mapreduce.
