源码见  https://cwiki.apache.org/confluence/display/Hive/LanguageManual+UDFhttps://cwiki.apache.org/confluence/display/Hive/LanguageManual+UDF

INSERT OVERWRITE DIRECTORY，当map 或者reduce 数量不一样多时  不会删除之前的文件，MR执行架构有变化时，

set hive.exec.compress.output=true;set mapred.output.compression.codec=org.apache.hadoop.io.compress.GzipCodec;SnappyCodec

org.apache.hadoop.hive.cli.CliDriver

describe formatted table_name partition (ptdate=20130909)

set mapred.job.name = my_job_name

PARTIAL1 标识map阶段,会调用iterate()和terminatePartial()PARTIAL2 标识combiner阶段，会调用merge()和terminatePartial()FINAL    标识reduce阶段调用merge()和terminate()COMPLETE 标识没有reduce阶段map，调用iterate()和terminate()

describe formatted table partition (ptdate='20170123');

hive 常用语句创建表=================================================create EXTERNAL table IF NOT EXISTS t1 (c1 string, c2 string, c3 string, c34 string)partitioned by (ptdate string)ROW FORMA

同一个reducer在HIVE UDAF中的buffer始终是一个对象，调用顺序是new==>init==>reset==>merge==>merge==>merge==>terminate    merge的个数，是有多少个mapper用了同一个key。一次循环处理一个key。reset==>merge==>merge==>merge==>terminatereset==>merge