sqoop资源优化记录

场景：

利用sqoop 导入千万级大概1K8W条Oracle数据到hive多分区表中

集群资源：132G，96cores

队列highway资源：

yarn.scheduler.capacity.root.highway.capacity=40

yarn.scheduler.capacity.root.highway.maximum-capacity=70

yarn.scheduler.capacity.root.highway.minimum-user-limit-percent=80

yarn.scheduler.capacity.root.highway.state=RUNNING

yarn.scheduler.capacity.root.highway.user-limit-factor=2

分区字段：

原本sqoop脚本：

复制代码

sqoop import --D mapred.job.queue.name=highway \
--connect "jdbc:oracle:thin:@//localhost:61521/LZY2" \
--username LZSHARE \
--password '123456' \
--query "SELECT 
    TO_CHAR(GCRQ, 'YYYY') AS gcrq_year,
    TO_CHAR(GCRQ, 'MM') AS gcrq_month,
    TO_CHAR(GCRQ, 'DD') AS gcrq_day,
    YEAR,
    TO_CHAR(GCRQ, 'YYYY-MM-DD HH24:MI:SS') AS GCRQ,
    GCZBS,
    HOUR,
    MINUTE,
    ......
    DELETE_BY,
    TO_CHAR(DELETE_TIME, 'YYYY-MM-DD HH24:MI:SS') AS DELETE_TIME,
    CREATE_BY,
    TO_CHAR(CREATE_TIME, 'YYYY-MM-DD HH24:MI:SS') AS CREATE_TIME,
    UPDATE_BY,
    TO_CHAR(UPDATE_TIME, 'YYYY-MM-DD HH24:MI:SS') AS UPDATE_TIME,
    TO_CHAR(INSERT_TIME, 'YYYY-MM-DD HH24:MI:SS') AS INSERT_TIME
	FROM LZJHGX.dat_dcsj_time
WHERE TO_CHAR(GCRQ , 'YYYY-MM-DD') < TO_CHAR(SYSDATE, 'YYYY-MM-DD') AND \$CONDITIONS" \
--split-by MINUTE \
--hcatalog-database dw \
--hcatalog-table ods_pre_dat_dcsj_time \
--hcatalog-storage-stanza 'stored as orc' \
--num-mappers 5

问题1：Error: Java heap space Out of Memory

解决思路：分析splitby字段，这是作为splitby字段MINUTE的情况：

如果按照上述划分，如果5个mapper，平均一个mapper处理4.5百万数据。明显不合理，另选一个splitby字段 （由于没有id和自增键），情况如下：

范围是1~288,每个分组6W多条数据。

第二，增加mapper个数，设定每个mapper所使用的个数

-D mapreduce.map.memory.mb=4096 \
-D mapreduce.map.java.opts=-Xmx3072m \

--num-mappers 20

其实还有一个，尽量避免复杂查询。

bash 复制代码

sqoop import -D mapred.job.queue.name=highway \
-D mapreduce.map.memory.mb=4096 \
-D mapreduce.map.java.opts=-Xmx3072m \
--connect "jdbc:oracle:thin:@//localhost:61521/LZY2" \
--username LZSHARE \
--password '123456' \
--query "SELECT 
    TO_CHAR(GCRQ, 'YYYY') AS gcrq_year,
    TO_CHAR(GCRQ, 'MM') AS gcrq_month,
    TO_CHAR(GCRQ, 'DD') AS gcrq_day,
    YEAR,
    TO_CHAR(GCRQ, 'YYYY-MM-DD HH24:MI:SS') AS GCRQ,
    GCZBS,
    .......
    ERR_CODE,
    ERR_DESC,
    DELETE_BY,
    TO_CHAR(DELETE_TIME, 'YYYY-MM-DD HH24:MI:SS') AS DELETE_TIME,
    CREATE_BY,
    TO_CHAR(CREATE_TIME, 'YYYY-MM-DD HH24:MI:SS') AS CREATE_TIME,
    UPDATE_BY,
    TO_CHAR(UPDATE_TIME, 'YYYY-MM-DD HH24:MI:SS') AS UPDATE_TIME,
    TO_CHAR(INSERT_TIME, 'YYYY-MM-DD HH24:MI:SS') AS INSERT_TIME
	FROM LZJHGX.dat_dcsj_time
WHERE TO_CHAR(GCRQ , 'YYYY-MM-DD') < TO_CHAR(SYSDATE, 'YYYY-MM-DD') AND \$CONDITIONS" \
--split-by sjxh \
--hcatalog-database dw \
--hcatalog-table ods_pre_dat_dcsj_time \
--hcatalog-storage-stanza 'stored as orc' \
--num-mappers 20

最后再次运行：耗时4分钟左右

成功导入：