基于sparksql调用shell脚本执行SQL

[Author]: kwu

基于sparksql调用shell脚本执行SQL，sparksql提供了类似hive中的 -e  , -f ,-i的选项

1、定时调用脚本

 #!/bin/sh
 # upload logs to hdfs

 yesterday=`date --date='1 days ago' +%Y%m%d`

 /opt/modules/spark/bin/spark-sql -i /opt/bin/spark_opt/init.sql --master spark://10.130.2.20:7077 --executor-memory 6g --total-executor-cores 45 --conf spark.ui.port=4075   -e "\
 insert overwrite table st.stock_realtime_analysis PARTITION (DTYPE='01' )
   select t1.stockId as stockId,
          t1.url as url,
          t1.clickcnt as clickcnt,
          0,
          round((t1.clickcnt / (case when t2.clickcntyesday is null then   0 else t2.clickcntyesday end) - 1) * 100, 2) as LPcnt,
          '01' as type,
          t1.analysis_date as analysis_date,
          t1.analysis_time as analysis_time
     from (select stock_code stockId,
                  concat('http://stockdata.stock.hexun.com/', stock_code,'.shtml') url,
                  count(1) clickcnt,
                  substr(from_unixtime(unix_timestamp(),'yyyy-MM-dd HH:mm:ss'),1,10) analysis_date,
                  substr(from_unixtime(unix_timestamp(),'yyyy-MM-dd HH:mm:ss'),12,8) analysis_time
             from dms.tracklog_5min
            where stock_type = 'STOCK'
              and day =
                  substr(from_unixtime(unix_timestamp(), 'yyyyMMdd'), 1, 8)
            group by stock_code
            order by clickcnt desc limit 20) t1
     left join (select stock_code stockId, count(1) clickcntyesday
                  from dms.tracklog_5min a
                 where stock_type = 'STOCK'
                   and substr(datetime, 1, 10) = date_sub(from_unixtime(unix_timestamp(),'yyyy-MM-dd HH:mm:ss'),1)
                   and substr(datetime, 12, 5) <substr(from_unixtime(unix_timestamp(),'yyyy-MM-dd HH:mm:ss'), 12, 5)
                   and day = '${yesterday}'
                 group by stock_code) t2
       on t1.stockId = t2.stockId;
   "\

 sqoop export  --connect jdbc:mysql://10.130.2.245:3306/charts   --username guojinlian  --password Abcd1234  --table stock_realtime_analysis  --fields-terminated-by '\001' --columns "stockid,url,clickcnt,splycnt,lpcnt,type" --export-dir /dw/st/stock_realtime_analysis/dtype=01;

init.sql内容为加载udf:

 add jar /opt/bin/UDF/hive-udf.jar;
 create temporary function udtf_stockidxfund as 'com.hexun.hive.udf.stock.UDTFStockIdxFund';
 create temporary function udf_getbfhourstime as 'com.hexun.hive.udf.time.UDFGetBfHoursTime';
 create temporary function udf_getbfhourstime2 as 'com.hexun.hive.udf.time.UDFGetBfHoursTime2';
 create temporary function udf_stockidxfund as 'com.hexun.hive.udf.stock.UDFStockIdxFund';
 create temporary function udf_md5 as 'com.hexun.hive.udf.common.HashMD5UDF';
 create temporary function udf_murhash as 'com.hexun.hive.udf.common.HashMurUDF';
 create temporary function udf_url as 'com.hexun.hive.udf.url.UDFUrl';
 create temporary function url_host as 'com.hexun.hive.udf.url.UDFHost';
 create temporary function udf_ip as 'com.hexun.hive.udf.url.UDFIP';
 create temporary function udf_site as 'com.hexun.hive.udf.url.UDFSite';
 create temporary function udf_UrlDecode as 'com.hexun.hive.udf.url.UDFUrlDecode';
 create temporary function udtf_url as 'com.hexun.hive.udf.url.UDTFUrl';
 create temporary function udf_ua as 'com.hexun.hive.udf.useragent.UDFUA';
 create temporary function udf_ssh as 'com.hexun.hive.udf.useragent.UDFSSH';
 create temporary function udtf_ua as 'com.hexun.hive.udf.useragent.UDTFUA';
 create temporary function udf_kw as 'com.hexun.hive.udf.url.UDFKW';
 create temporary function udf_chdecode as 'com.hexun.hive.udf.url.UDFChDecode';

设置ui的端口

 --conf spark.ui.port=4075

默认为4040，会与其他正在跑的任务冲突，这里修改为4075

设定任务使用的内存与CPU资源

 --executor-memory 6g --total-executor-cores 45

原来的语句是用hive -e 执行的，修改为spark后速度大加快了。原来为15min，提升速度后为 45s.