从代码上看，期望是能在单个交易日期内对股票代码分组计算，在组内还要按照CLOSE进行筛选。由于分布式表的分区规则与group by需求相同，理论上可以对单个分区内的数据进行股票分组聚合计算，可以使用mr方法进行自定义的map-reduce任务计算。
def myMap(table){
result_temp = select SecurityID,date(first(TradeTime)),aggrTopN(func=min, funcArgs=CLOSE, sortingCol=CLOSE, top=0.20, ascending=false) as res from table group by SecurityID
return result_temp
}
res = mr(sqlDS(<select * from loadTable('dfs://SHSZ_OHLCV_1MIN','SHSZ_OHLCV_1MIN')>), myMap)
result = table(1000000:0,`SecurityID`date`res,`SYMBOL`TIMESTAMP`DOUBLE)
for (i in res[0])
{
result.append!(i)
}
也可以在sql后方加上map关键字，表示强制分区执行：
result_temp = select SecurityID,TradeDate,aggrTopN(func=min, funcArgs=CLOSE, sortingCol=CLOSE, top=0.20, ascending=false) as res from loadTable('dfs://SHSZ_OHLCV_1MIN','SHSZ_OHLCV_1MIN') group by SecurityID,date(TradeTime) as TradeDate map