Thank you, all of you. explode() is helpful: df.selectExpr("explode(bizs) as e").select("e.*").show()
2016-10-19 lk_spark 发件人:Hyukjin Kwon <gurwls...@gmail.com> 发送时间:2016-10-19 13:16 主题:Re: how to extract arraytype data to file 收件人:"Divya Gehlot"<divya.htco...@gmail.com> 抄送:"lk_spark"<lk_sp...@163.com>,"user.spark"<user@spark.apache.org> This reminds me of https://github.com/databricks/spark-xml/issues/141#issuecomment-234835577 Maybe using explode() would be helpful. Thanks! 2016-10-19 14:05 GMT+09:00 Divya Gehlot <divya.htco...@gmail.com>: http://stackoverflow.com/questions/33864389/how-can-i-create-a-spark-dataframe-from-a-nested-array-of-struct-element Hope this helps Thanks, Divya On 19 October 2016 at 11:35, lk_spark <lk_sp...@163.com> wrote: hi,all: I want to read a json file and search it by sql . the data struct should be : bid: string (nullable = true) code: string (nullable = true) and the json file data should be like : {bid":"MzI4MTI5MzcyNw==","code":"罗甸网警"} {"bid":"MzI3MzQ5Nzc2Nw==","code":"西早君"} but in fact my json file data is : {"bizs":[ {bid":"MzI4MTI5MzcyNw==","code":"罗甸网警"},{"bid":"MzI3MzQ5Nzc2Nw==","code":"西早君"}]} {"bizs":[ {bid":"MzI4MTI5Mzcy00==","code":"罗甸网警"},{"bid":"MzI3MzQ5Nzc201==","code":"西早君"}]} I load it by spark ,data schema shows like this : root |-- bizs: array (nullable = true) | |-- element: struct (containsNull = true) | | |-- bid: string (nullable = true) | | |-- code: string (nullable = true) I can select columns by : df.select("bizs.id","bizs.name") but the colume values is in array type: +--------------------+--------------------+ | id| code| +--------------------+--------------------+ |[4938200, 4938201...|[罗甸网警, 室内设计师杨焰红, ...| |[4938300, 4938301...|[SDCS十全九美, 旅梦长大, ...| |[4938400, 4938401...|[日重重工液压行走回转, 氧老家,...| |[4938500, 4938501...|[PABXSLZ, 陈少燕, 笑蜜...| |[4938600, 4938601...|[税海微云, 西域美农云家店, 福...| +--------------------+--------------------+ what I want is I can read colum in normal row type. how I can do it ? 2016-10-19 lk_spark