一、抓取批量耗时最久的SQL:
2、执行次数最多的TOP10SQL"
db2 "select substr(stmt_text,1,500) as stmt_text,NUM_EXECUTIONS,TOTAL_USR_CPU_TIME,TOTAL_EXEC_TIME,SORT_OVERFLOWS from sysibmadm.snapdyn_sql order by NUM_EXECUTIONS desc fetch first 10 rows only"|grep -iv selected
3、平均执行时间最长的TOP10SQL"
db2 "select substr(stmt_text,1,500) as stmt_text,average_execution_time_s,num_executions from sysibmadm.top_dynamic_sql order by average_execution_time_s desc fetch first 10 rows only"|grep -iv selected
4、排序最多的TOP10SQL"
db2 "select substr(stmt_text,1,500) as stmt_text,stmt_sorts,SORT_OVERFLOWS,TOTAL_USR_CPU_TIME,NUM_EXECUTIONS from sysibmadm.snapdyn_sql order by stmt_sorts desc fetch first 10 rows only"|grep -iv selected
5、通过以上SQL获得最耗时的SQL信息如下:
update tbhisfeyehz t set client_num=( select count(distinct a.in_client_no) from vsharedetail a, tbbranch b, tbtainfo c where a.ta_code=c.ta_code and a.open_branch=b.branch_no and substr(b.internal_branch,1,length(Rtrim(t.internal_branch))) = t.internal_branch and a.client_type=t.client_type and a.prd_code=t.prd_code and a.ta_code=t.ta_code and c.prd_type=t.prd_type and a.tot_vol > 0.0001 and a.ta_code=?) where exists(select 1 from vsharedetail a, tbbranch b, tbtainfo c where a.ta_code=c.ta_code and a.open_branch=b.branch_no and substr(b.internal_branch,1,length(Rtrim(t.internal_branch))) = t.internal_branch and a.client_type=t.client_type and a.prd_code=t.prd_code and a.ta_code=t.ta_code and c.prd_type=t.prd_type and a.tot_vol > 0.0001 and a.ta_code=?) and internal_branch in(select internal_branch from tbbranch where branch_level<>?) and sum_flag='0clear
三、为了在测试环境复现该执行计划我们需要从生产导出该SQL用到的所有基表的统计信息还原到测试环境下去。
1、导出统计信息:
db2look -d ifm30 -u ifm40 -m -t tbhisfeyehz -o tbhisfeyehz.sql
db2look -d ifm30 -u ifm40 -m -t tbsharedetail1 -o tbsharedetail1.sql
db2look -d ifm30 -u ifm40 -m -t tbsharedetail2 -o tbsharedetail2.sql
db2look -d ifm30 -u ifm40 -m -t tbsharedetail3 -o tbsharedetail3.sql
db2look -d ifm30 -u ifm40 -m -t tbsharedetail4 -o tbsharedetail4.sql
db2look -d ifm30 -u ifm40 -m -t tbsharedetail5 -o tbsharedetail5.sql
db2look -d ifm30 -u ifm40 -m -t tsys_branch -o tsys_branch.sql
db2look -d ifm30 -u ifm40 -m -t tbtainfo -o tbtainfo.sql
2、生效到测试环境:
db2 -tvf tbsharedetail1.sql
db2 -tvf tbsharedetail2.sql
db2 -tvf tbsharedetail3.sql
db2 -tvf tbsharedetail4.sql
db2 -tvf tbsharedetail5.sql
db2 -tvf tsys_branch.sql
db2 -tvf tbtainfo.sql
db2 -tvf tbhisfeyehz.sql
四、分析SQL执行计划,根据执行计划分析关键执行成本之处的SQL计划并做出调整计划
db2 connect to ifm30
db2 set current explain mode explain
db2 -tvf optim.sql
db2 set current explain mode no
db2exfmt -d ifm30 -l -o
五、使用优化器顾问程序给出的建议进行优化
db2advis -d ifm30 -i optim3.sql -m i -l -1 -t 0 -o product.txt
六、优化相关基表
1、根据优化引擎建议新建唯一索引在对应基表之上
-- index[1], 0.013MB
CREATE INDEX "IFM40 "."IDX1501210621430" ON "IFM40 "."TBSHAREDETAIL4"
("TOT_VOL" ASC) ALLOW REVERSE SCANS COLLECT SAMPLED DETAILED STATISTICS;
COMMIT WORK ;
-- index[2], 0.013MB
CREATE INDEX "IFM40 "."IDX1501210623270" ON "IFM40 "."TBSHAREDETAIL3"
("TOT_VOL" ASC) ALLOW REVERSE SCANS COLLECT SAMPLED DETAILED STATISTICS;
COMMIT WORK ;
-- index[3], 0.013MB
CREATE INDEX "IFM40 "."IDX1501210625230" ON "IFM40 "."TBSHAREDETAIL2"
("TOT_VOL" ASC) ALLOW REVERSE SCANS COLLECT SAMPLED DETAILED STATISTICS;
COMMIT WORK ;
-- index[4], 0.013MB
CREATE INDEX "IFM40 "."IDX1501210627190" ON "IFM40 "."TBSHAREDETAIL1"
("TOT_VOL" ASC) ALLOW REVERSE SCANS COLLECT SAMPLED DETAILED STATISTICS;
COMMIT WORK ;
-- index[5], 0.013MB
CREATE INDEX "IFM40 "."IDX1501210629150" ON "IFM40 "."TBSHAREDETAIL5"
("TOT_VOL" ASC) ALLOW REVERSE SCANS COLLECT SAMPLED DETAILED STATISTICS;
COMMIT WORK ;
-- index[6], 0.013MB
CREATE UNIQUE INDEX "IFM40 "."IDX1501210631080"
ON "IFM40 "."TBTAINFO" ("TA_CODE" ASC) INCLUDE ("PRD_TYPE")
ALLOW REVERSE SCANS COLLECT SAMPLED DETAILED STATISTICS;
COMMIT WORK ;
-- index[7], 0.013MB
CREATE INDEX "IFM40 "."IDX1501210632520" ON "IFM40 "."TBHISFEYEHZ"
("SUM_DATE" ASC, "SUM_FLAG" ASC, "PRD_TYPE" ASC, "TA_CODE"
ASC, "CLIENT_NUM" ASC, "INTERNAL_BRANCH" ASC, "CLIENT_TYPE"
ASC, "PRD_CODE" ASC) ALLOW REVERSE SCANS COLLECT SAMPLED DETAILED STATISTICS;
COMMIT WORK ;
-- index[8], 0.036MB
CREATE INDEX "IFM40 "."IDX1501210632500" ON "IFM40 "."TSYS_BRANCH"
("BRANCH_PATH" ASC, "BRANCH_LEVEL" ASC) ALLOW REVERSE
SCANS COLLECT SAMPLED DETAILED STATISTICS;
COMMIT WORK ;
-- index[9], 0.024MB
CREATE UNIQUE INDEX "IFM40 "."IDX1501210631010"
ON "IFM40 "."TSYS_BRANCH" ("BRANCH_CODE" ASC) INCLUDE
("BRANCH_PATH") ALLOW REVERSE SCANS COLLECT SAMPLED DETAILED STATISTICS;
COMMIT WORK ;
2、批量统计信息更新
使用如下shell脚本批量优化
cat tblist|while read a
db2 connect to ifm30 >>/dev/null
do
if [ $a <> ' ' ];
then
db2 "runstats on table ifm40.${a} with distribution and detailed indexes all"
else
exit
fi
done
3、业务验证,通过重新批量测试,由原来的30min降低到2min优化效果明显
阅读(12788) | 评论(1) | 转发(0) |