【详述】为了从2.5.20版本升级到3.3.7，重新搭建一套新集群，把生产数据迁到新集群后，发现BI查询很慢，sql相同返回时间不一样，取其一sql验证：2.5版本700多毫秒，3.3.7要1.5s返回，然后查询返回结果也不一样
【背景】生产环境版本升级
【业务影响】BI查询比较慢，查询结果不一样
【是否存算分离】否
【StarRocks版本】3.3.7
【集群规模】3fe（3follower+3be（fe与be混部）
【机器信息】128C/256G
【联系方式】社区群13-麦咪
【附件】
查询sql如下，查询时间慢是一个问题，还有就是一个结果返回不一样，好像是数据关联不上：
select statmonth,b.companycode,
case when premplanIFNULL(a.peroid,b.peroid)=0 then null else netprm_idate/(premplanIFNULL(a.peroid,b.peroid)) end as qB1C6B74ADEFB1D07E0530B6C010A845Dtimeachiverate_time
from (
select companycode companycode,sum(premplan*10000) premplan ,
(datediff(‘2024-12-31’,‘2024-01-01’) + 1)/(datediff(‘2024-12-31’,‘2024-01-01’)+1) as peroid
from pub_its_prmplan
where kindcode = ‘03’
and datadate >= ‘2024-01-01’ and datadate <= ‘2024-12-31’
and com_flag =‘companycode’
group by companycode ) b
left join (
select statmonth,
companycode ,
sum(netprm_idate) over (partition by companycode order by statmonth,companycode rows between unbounded PRECEDING and current row) netprm_idate,
(datediff(max(date_add(statmonth,1)) over (partition by companycode order by statmonth,companycode rows between unbounded PRECEDING and current row) , ‘2024-01-01’ ))/(datediff(‘2024-12-31’,‘2024-01-01’)+1) peroid
from (
select ‘2024-12-31’ as statmonth,
companycode as companycode,
sum(netprm_idate) netprm_idate
from app_ply_model_main_month
left join pub_its_company on app_ply_model_main_month.buiscomcode=pub_its_company.buiscomcode
left join pub_its_statdate on app_ply_model_main_month.incomedate=pub_its_statdate.statdate
where app_ply_model_main_month.datadate = ‘2024-12-31’
and app_ply_model_main_month.incomedate>=‘2024-01-01’
and app_ply_model_main_month.incomedate<=‘2024-12-31’
group by ‘2024-12-31’,companycode) a
order by 1) a
on a.companycode=b.companycode
2.5.20版本查询结果：
3.3.7版本查询结果：
