[ 
https://issues.apache.org/jira/browse/DRILL-4785?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15383276#comment-15383276
 ] 

Dechang Gu commented on DRILL-4785:
-----------------------------------

Reproduced the issue with TPCH data and query. 

Commit id 10afc708600ea9f4cb0e7c2cd981b5b1001fea0d:   the query takes 1.864 secs

drillbit.log:
2016-07-18 15:31:18,480 ucs-node5.perf.lab 
[2872a749-2a71-a4ad-ae64-5539c681ec34:foreman] INFO  
o.a.drill.exec.work.foreman.Foreman - Query text for query id 
2872a749-2a71-a4ad-ae64-5539c681ec34: SELECT * FROM
(
select
sum(l_extendedprice * l_discount) as revenue
from
lineitem
where
l_shipdate >= date '1997-01-01'
and l_shipdate < date '1997-01-01' + interval '1' year
and
l_discount between 0.03 - 0.01 and 0.03 + 0.01
and l_quantity < 24
)
LZT
LIMIT 0
2016-07-18 15:31:19,141 ucs-node5.perf.lab 
[2872a749-2a71-a4ad-ae64-5539c681ec34:foreman] INFO  
o.a.d.exec.store.parquet.Metadata - Took 431 ms to read metadata from cache file
2016-07-18 15:31:19,296 ucs-node5.perf.lab 
[2872a749-2a71-a4ad-ae64-5539c681ec34:foreman] INFO  
o.a.d.exec.store.parquet.Metadata - Took 153 ms to read metadata from cache file
2016-07-18 15:31:19,968 ucs-node5.perf.lab 
[2872a749-2a71-a4ad-ae64-5539c681ec34:frag:0:0] INFO  
o.a.d.e.w.fragment.FragmentExecutor - 2872a749-2a71-a4ad-ae64-5539c681ec34:0:0: 
State change requested AWAITING_ALLOCATION --> RUNNING
2016-07-18 15:31:19,969 ucs-node5.perf.lab 
[2872a749-2a71-a4ad-ae64-5539c681ec34:frag:0:0] INFO  
o.a.d.e.w.f.FragmentStatusReporter - 2872a749-2a71-a4ad-ae64-5539c681ec34:0:0: 
State to report: RUNNING

commit id  9f4fff800d128878094ae70b454201f79976135d:       the query takes      
          0.286 secs

drillbit.log:
2016-07-18 15:42:59,618 ucs-node5.perf.lab 
[2872a48c-1221-bef1-32f0-4445a69ef580:foreman] INFO  
o.a.drill.exec.work.foreman.Foreman - Query text for query id 
2872a48c-1221-bef1-32f0-4445a69ef580: SELECT * FROM
(
select
sum(l_extendedprice * l_discount) as revenue
from
lineitem
where
l_shipdate >= date '1997-01-01'
and l_shipdate < date '1997-01-01' + interval '1' year
and
l_discount between 0.03 - 0.01 and 0.03 + 0.01
and l_quantity < 24
)
LZT
LIMIT 0
2016-07-18 15:42:59,841 ucs-node5.perf.lab 
[2872a48c-1221-bef1-32f0-4445a69ef580:frag:0:0] INFO  
o.a.d.e.w.fragment.FragmentExecutor - 2872a48c-1221-bef1-32f0-4445a69ef580:0:0: 
State change requested AWAITING_ALLOCATION --> RUNNING
2016-07-18 15:42:59,841 ucs-node5.perf.lab 
[2872a48c-1221-bef1-32f0-4445a69ef580:frag:0:0] INFO  
o.a.d.e.w.f.FragmentStatusReporter - 2872a48c-1221-bef1-32f0-4445a69ef580:0:0: 
State to report: RUNNING
2016-07-18 15:42:59,852 ucs-node5.perf.lab 
[2872a48c-1221-bef1-32f0-4445a69ef580:frag:0:0] INFO  
o.a.d.e.w.fragment.FragmentExecutor - 2872a48c-1221-bef1-32f0-4445a69ef580:0:0: 
State change requested RUNNING --> FINISHED
2016-07-18 15:42:59,852 ucs-node5.perf.lab 
[2872a48c-1221-bef1-32f0-4445a69ef580:frag:0:0] INFO  
o.a.d.e.w.f.FragmentStatusReporter - 2872a48c-1221-bef1-32f0-4445a69ef580:0:0: 
State to report: FINISHED


Note that   "read metadata from cache file "  is where the extra time spent.

                                                                                
                    


> Limit 0 queries regressed in Drill 1.7.0 
> -----------------------------------------
>
>                 Key: DRILL-4785
>                 URL: https://issues.apache.org/jira/browse/DRILL-4785
>             Project: Apache Drill
>          Issue Type: Bug
>          Components: Functions - Drill
>    Affects Versions: 1.7.0
>         Environment: Redhat EL6
>            Reporter: Dechang Gu
>            Assignee: Venki Korukanti
>             Fix For: 1.8.0
>
>
> We noticed a bunch of limit 0 queries regressed quite a bit: +2500ms, while 
> the same queries took  ~400ms in Apache Drill 1.6.0. 5-6X regression. Further 
> investigation indicates that most likely the root cause of the regression is 
> in the commit: 
> vkorukanti committed with vkorukanti DRILL-4446: Support mandatory work 
> assignment to endpoint requirement…
> commit id:  10afc708600ea9f4cb0e7c2cd981b5b1001fea0d
> With drill build on this commit, query takes 3095ms
> and in the drillbit.log:
> 2016-07-15 17:27:55,048 ucs-node2.perf.lab 
> [28768074-4ed6-a70a-2e6a-add3201ab801:foreman] INFO  
> o.a.drill.exec.work.foreman.Foreman - Query text for query id 
> 28768074-4ed6-a70a-2e6a-add3201ab801: SELECT * FROM (SELECT 
> CAST(EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) AS 
> INTEGER) AS `mn_business_date_ok`,AVG((CASE WHEN ((CAST(EXTRACT(YEAR FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER) = 2014) AND 
> (CAST((EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) - 1) / 3 
> + 1 AS INTEGER) <= 4)) THEN `rfm_sales`.`pos_netsales` ELSE NULL END)) AS 
> `avg_Calculation_CIDBACJBCCCBHDGB_ok`,SUM((CASE WHEN ((CAST(EXTRACT(YEAR FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER) = 2014) AND 
> (CAST((EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) - 1) / 3 
> + 1 AS INTEGER) <= 4)) THEN `rfm_sales`.`pos_netsales` ELSE NULL END)) AS 
> `sum_Calculation_CIDBACJBCCCBHDGB_ok`,SUM((CASE WHEN ((CAST(EXTRACT(YEAR FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER) = 2014) AND 
> (CAST((EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) - 1) / 3 
> + 1 AS INTEGER) <= 4)) THEN 1 ELSE NULL END)) AS 
> `sum_Calculation_CJEBBAEBBFADBDFJ_ok`,SUM((CASE WHEN ((CAST(EXTRACT(YEAR FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER) = 2014) AND 
> (CAST((EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) - 1) / 3 
> + 1 AS INTEGER) <= 4)) THEN (`rfm_sales`.`pos_comps` + 
> `rfm_sales`.`pos_promos`) ELSE NULL END)) AS `sum_Net_Sales__YTD___copy__ok` 
> FROM `dfs.xxx`.`views/rfm_sales` `rfm_sales` GROUP BY CAST(EXTRACT(MONTH FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER)) T LIMIT 0
> 2016-07-15 17:27:55,664 ucs-node2.perf.lab 
> [28768074-4ed6-a70a-2e6a-add3201ab801:foreman] INFO  
> o.a.d.exec.store.parquet.Metadata - Took 208 ms to read metadata from cache 
> file
> 2016-07-15 17:27:56,783 ucs-node2.perf.lab 
> [28768074-4ed6-a70a-2e6a-add3201ab801:foreman] INFO  
> o.a.d.exec.store.parquet.Metadata - Took 129 ms to read metadata from cache 
> file
> 2016-07-15 17:27:57,960 ucs-node2.perf.lab 
> [28768074-4ed6-a70a-2e6a-add3201ab801:frag:0:0] INFO  
> o.a.d.e.w.fragment.FragmentExecutor - 
> 28768074-4ed6-a70a-2e6a-add3201ab801:0:0: State change requested 
> AWAITING_ALLOCATION --> RUNNING
> 2016-07-15 17:27:57,961 ucs-node2.perf.lab 
> [28768074-4ed6-a70a-2e6a-add3201ab801:frag:0:0] INFO  
> o.a.d.e.w.f.FragmentStatusReporter - 
> 28768074-4ed6-a70a-2e6a-add3201ab801:0:0: State to report: RUNNING
> 2016-07-15 17:27:57,989 ucs-node2.perf.lab 
> [28768074-4ed6-a70a-2e6a-add3201ab801:frag:0:0] INFO  
> o.a.d.e.w.fragment.FragmentExecutor - 
> 28768074-4ed6-a70a-2e6a-add3201ab801:0:0: State change requested RUNNING --> 
> FINISHED
> 2016-07-15 17:27:57,989 ucs-node2.perf.lab 
> [28768074-4ed6-a70a-2e6a-add3201ab801:frag:0:0] INFO  
> o.a.d.e.w.f.FragmentStatusReporter - 
> 28768074-4ed6-a70a-2e6a-add3201ab801:0:0: State to report: FINISHED
> while running the same query on the parent commit (commit id 
> 9f4fff800d128878094ae70b454201f79976135d), it only takes  492ms.
> and in the drillbit.log:
> 2016-07-15 17:19:27,309 ucs-node7.perf.lab 
> [2876826f-ee19-9466-0c0c-869f47c409f8:foreman] INFO  
> o.a.drill.exec.work.foreman.Foreman - Query text for query id 
> 2876826f-ee19-9466-0c0c-869f47c409f8: SELECT * FROM (SELECT 
> CAST(EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) AS 
> INTEGER) AS `mn_business_date_ok`,AVG((CASE WHEN ((CAST(EXTRACT(YEAR FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER) = 2014) AND 
> (CAST((EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) - 1) / 3 
> + 1 AS INTEGER) <= 4)) THEN `rfm_sales`.`pos_netsales` ELSE NULL END)) AS 
> `avg_Calculation_CIDBACJBCCCBHDGB_ok`,SUM((CASE WHEN ((CAST(EXTRACT(YEAR FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER) = 2014) AND 
> (CAST((EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) - 1) / 3 
> + 1 AS INTEGER) <= 4)) THEN `rfm_sales`.`pos_netsales` ELSE NULL END)) AS 
> `sum_Calculation_CIDBACJBCCCBHDGB_ok`,SUM((CASE WHEN ((CAST(EXTRACT(YEAR FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER) = 2014) AND 
> (CAST((EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) - 1) / 3 
> + 1 AS INTEGER) <= 4)) THEN 1 ELSE NULL END)) AS 
> `sum_Calculation_CJEBBAEBBFADBDFJ_ok`,SUM((CASE WHEN ((CAST(EXTRACT(YEAR FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER) = 2014) AND 
> (CAST((EXTRACT(MONTH FROM CAST(`rfm_sales`.`business_date` AS DATE)) - 1) / 3 
> + 1 AS INTEGER) <= 4)) THEN (`rfm_sales`.`pos_comps` + 
> `rfm_sales`.`pos_promos`) ELSE NULL END)) AS `sum_Net_Sales__YTD___copy__ok` 
> FROM `dfs.xxx`.`views/rfm_sales` `rfm_sales` GROUP BY CAST(EXTRACT(MONTH FROM 
> CAST(`rfm_sales`.`business_date` AS DATE)) AS INTEGER)) T LIMIT 0
> 2016-07-15 17:19:27,619 ucs-node7.perf.lab 
> [2876826f-ee19-9466-0c0c-869f47c409f8:frag:0:0] INFO  
> o.a.d.e.w.fragment.FragmentExecutor - 
> 2876826f-ee19-9466-0c0c-869f47c409f8:0:0: State change requested 
> AWAITING_ALLOCATION --> RUNNING
> 2016-07-15 17:19:27,619 ucs-node7.perf.lab 
> [2876826f-ee19-9466-0c0c-869f47c409f8:frag:0:0] INFO  
> o.a.d.e.w.f.FragmentStatusReporter - 
> 2876826f-ee19-9466-0c0c-869f47c409f8:0:0: State to report: RUNNING
> 2016-07-15 17:19:27,643 ucs-node7.perf.lab 
> [2876826f-ee19-9466-0c0c-869f47c409f8:frag:0:0] INFO  
> o.a.d.e.w.fragment.FragmentExecutor - 
> 2876826f-ee19-9466-0c0c-869f47c409f8:0:0: State change requested RUNNING --> 
> FINISHED
> 2016-07-15 17:19:27,644 ucs-node7.perf.lab 
> [2876826f-ee19-9466-0c0c-869f47c409f8:frag:0:0] INFO  
> o.a.d.e.w.f.FragmentStatusReporter - 
> 2876826f-ee19-9466-0c0c-869f47c409f8:0:0: State to report: FINISHED



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to