Thread: Query performance issue
Hi,
Below question got in one discussion.Appreciate any guidance on this.
Below is a query which is running for ~40 seconds. As it's a query which is executed from UI , we were expecting it to finish in <~5 seconds. It has a "IN" and a "NOT IN" subquery , from the execution path it seems the total response time is mainly, to be sum of the "IN" and the "NOT IN" subquery section. My thought was that both "IN" and "NOT IN" should be executed/evaluated in parallel but not in serial fashion.
In the execution path below , the line number marked in bold are the top lines for the IN and NOT IN subquery evaluation and they are showing "Actual time" as Approx ~9 seconds and ~8 seconds and they seems to be summed up and the top lines showing it to be ~19 seconds. Then onwards it keeps on increasing with other "nested loop" joins.
Note:- This query is running on a MYSQL 8.0 database. So I'm wondering if there is any mysql list similar to Oracle list , in which i can share this issue?
Added the query in below path:-
https://gist.github.com/databasetech0073/95bce00c3a6bd4ae8d195401e0383185
SELECT ......
FROM R_CON_ESTS RC_STS,
R_CON rc,
D_LKP D_LKP_STS,
D_LKP D_LKP_FRQ,
(select RCE.DRV_DT, RCE.AID, RCE.R_CON_ID, RCE.R_CON_VER_NB
from R_CON_E RCE
where RCE.MTNE_ID in (SELECT MI1.MTNE_ID
FROM M_INF mi1 WHERE MI1.AID = :AID
UNION
SELECT rg.RG_MF_SK_ID
from RG_M_F_INF rg where rg.AS_ID =:AID
UNION
SELECT fti.FT_SRK_ID
from M_FT fti where fti.AS_ID= :AID
)
and (RCE.DRV_DT, RCE.AID, RCE.R_CON_ID, RCE.R_CON_ver_nb) NOT IN
(SELECT RCE_NS.DRV_DT, RCE_NS.AID, RCE_NS.R_CON_ID, RCE_NS.R_CON_VER_NB
FROM R_CON_E RCE_NS
WHERE RCE_NS.MTNE_ID NOT IN (select MI2.MTNE_ID
from M_INF MI2 where MI2.AID = :AID
UNION
SELECT RG2.RG_MF_SK_ID
from RG_M_F_INF RG2 where RG2.AS_ID =:AID
UNION
SELECT FTI1.FT_SRK_ID
from M_FT FTI1 where FTI1.AS_ID= :AID
))
) b
where RC_STS.RR_FRQ_NB = D_LKP_FRQ.D_LKP_NB
and RC_STS.R_CON_ESTS_NB = D_LKP_STS.D_LKP_NB
and RC_STS.R_CON_ID = rc.R_CON_ID
and RC_STS.R_CON_VER_NB = rc.R_CON_VER_NB
and RC_STS.AID = rc.AID
and RC_STS.AID = b.AID
and RC_STS.R_CON_ID = b.R_CON_ID
and RC_STS.R_CON_VER_NB = b.R_CON_VER_NB
order by 3,4,2;
-> Sort: RC_STS.R_CON_ID, RC_STS.R_CON_VER_NB, RC_STS.R_EX_RID (actual time=44392.655..44644.844 rows=745483 loops=1)
-> Stream results (cost=311479029610.37 rows=860847650219) (actual time=8957.556..42133.969 rows=745483 loops=1)
-> Nested loop inner join (cost=311479029610.37 rows=860847650219) (actual time=8957.548..40891.903 rows=745483 loops=1)
-> Nested loop inner join (cost=225393084569.25 rows=860847650219) (actual time=8957.541..40597.741 rows=745483 loops=1)
-> Nested loop inner join (cost=139307139528.12 rows=860847650219) (actual time=8957.530..40092.267 rows=745483 loops=1)
-> Nested loop antijoin (cost=53221194487.00 rows=532199430400) (actual time=8957.477..29529.382 rows=671352 loops=1)
-> Nested loop inner join (cost=886687.00 rows=729520) (actual time=0.123..19714.306 rows=692583 loops=1)
-> Filter: <in_optimizer>(RCE.MTNE_ID,<exists>(select #3)) (cost=84215.00 rows=729520) (actual time=0.085..9045.124 rows=692583 loops=1)
-> Covering index scan on RCE using R_58 (cost=84215.00 rows=729520) (actual time=0.055..534.110 rows=742706 loops=1)
-> Select #3 (subquery in condition; dependent)
-> Limit: 1 row(s) (cost=4.41..4.41 rows=1) (actual time=0.010..0.010 rows=1 loops=742706)
-> Table scan on <union temporary> (cost=4.41..5.70 rows=2) (actual time=0.010..0.010 rows=1 loops=742706)
-> Union materialize with deduplication (cost=3.18..3.18 rows=2) (actual time=0.010..0.010 rows=1 loops=742706)
-> Limit table size: 1 unique row(s)
-> Limit: 1 row(s) (cost=1.13 rows=1) (actual time=0.006..0.006 rows=1 loops=742706)
-> Covering index lookup on mi1 using M_INF_AID_index (AID='XXXXXXXXXXXXXXXXXXX', MTNE_ID=<cache>(RCE.MTNE_ID)) (cost=1.13 rows=1) (actual time=0.006..0.006 rows=1 loops=742706)
-> Limit table size: 1 unique row(s)
-> Limit: 1 row(s) (cost=1.10 rows=1) (actual time=0.003..0.003 rows=1 loops=132294)
-> Single-row covering index lookup on rg using PRIMARY (RG_MF_SK_ID=<cache>(RCE.MTNE_ID), AS_ID='XXXXXXXXXXXXXXXXXXX') (cost=1.10 rows=1) (actual time=0.003..0.003 rows=1 loops=132294)
-> Limit table size: 1 unique row(s)
-> Limit: 1 row(s) (cost=0.74 rows=0.05) (actual time=0.003..0.003 rows=0 loops=50123)
-> Filter: (fti.AS_ID = 'XXXXXXXXXXXXXXXXXXX') (cost=0.74 rows=0.05) (actual time=0.003..0.003 rows=0 loops=50123)
-> Covering index lookup on fti using AK_MFTI (FT_SRK_ID=<cache>(RCE.MTNE_ID)) (cost=0.74 rows=2) (actual time=0.003..0.003 rows=0 loops=50123)
-> Index lookup on rc using R_26 (AID=RCE.AID, R_CON_ID=RCE.R_CON_ID, R_CON_VER_NB=RCE.R_CON_VER_NB) (cost=1.00 rows=1) (actual time=0.014..0.015 rows=1 loops=692583)
-> Single-row index lookup on <subquery7> using <auto_distinct_key> (DRV_DT=RCE.DRV_DT, AID=RCE.AID, R_CON_ID=RCE.R_CON_ID, R_CON_VER_NB=RCE.R_CON_VER_NB) (cost=157167.31..157167.31 rows=1) (actual time=0.014..0.014 rows=0 loops=692583)
-> Materialize with deduplication (cost=157167.00..157167.00 rows=729520) (actual time=8957.347..8957.347 rows=25843 loops=1)
-> Filter: ((RCE_NS.DRV_DT is not null) and (RCE_NS.AID is not null) and (RCE_NS.R_CON_ID is not null) and (RCE_NS.R_CON_VER_NB is not null)) (cost=84215.00 rows=729520) (actual time=1737.420..8871.505 rows=50123 loops=1)
-> Filter: <in_optimizer>(RCE_NS.MTNE_ID,<exists>(select #8) is false) (cost=84215.00 rows=729520) (actual time=1737.417..8860.489 rows=50123 loops=1)
-> Covering index scan on RCE_NS using R_58 (cost=84215.00 rows=729520) (actual time=0.039..531.571 rows=742706 loops=1)
-> Select #8 (subquery in condition; dependent)
-> Limit: 1 row(s) (cost=4.41..4.41 rows=1) (actual time=0.010..0.010 rows=1 loops=742706)
-> Table scan on <union temporary> (cost=4.41..5.70 rows=2) (actual time=0.010..0.010 rows=1 loops=742706)
-> Union materialize with deduplication (cost=3.18..3.18 rows=2) (actual time=0.010..0.010 rows=1 loops=742706)
-> Limit table size: 1 unique row(s)
-> Limit: 1 row(s) (cost=1.13 rows=1) (actual time=0.007..0.007 rows=1 loops=742706)
-> Covering index lookup on MI2 using M_INF_AID_index (AID='XXXXXXXXXXXXXXXXXXX', MTNE_ID=<cache>(RCE_NS.MTNE_ID)) (cost=1.13 rows=1) (actual time=0.006..0.006 rows=1 loops=742706)
-> Limit table size: 1 unique row(s)
-> Limit: 1 row(s) (cost=1.10 rows=1) (actual time=0.004..0.004 rows=1 loops=132294)
-> Single-row covering index lookup on RG2 using PRIMARY (RG_MF_SK_ID=<cache>(RCE_NS.MTNE_ID), AS_ID='XXXXXXXXXXXXXXXXXXX') (cost=1.10 rows=1) (actual time=0.003..0.003 rows=1 loops=132294)
-> Limit table size: 1 unique row(s)
-> Limit: 1 row(s) (cost=0.74 rows=0.05) (actual time=0.003..0.003 rows=0 loops=50123)
-> Filter: (FTI1.AS_ID = 'XXXXXXXXXXXXXXXXXXX') (cost=0.74 rows=0.05) (actual time=0.003..0.003 rows=0 loops=50123)
-> Covering index lookup on FTI1 using AK_MFTI (FT_SRK_ID=<cache>(RCE_NS.MTNE_ID)) (cost=0.74 rows=2) (actual time=0.003..0.003 rows=0 loops=50123)
-> Index lookup on RC_STS using RCE_STS (AID=RCE.AID, R_CON_ID=RCE.R_CON_ID, R_CON_VER_NB=RCE.R_CON_VER_NB) (cost=1.62 rows=2) (actual time=0.013..0.016 rows=1 loops=671352)
-> Single-row index lookup on D_LKP_STS using PRIMARY (D_LKP_NB=RC_STS.R_CON_ESTS_NB) (cost=1.00 rows=1) (actual time=0.000..0.000 rows=1 loops=745483)
-> Single-row index lookup on D_LKP_FRQ using PRIMARY (D_LKP_NB=RC_STS.RR_FRQ_NB) (cost=1.00 rows=1) (actual time=0.000..0.000 rows=1 loops=745483)
On 10/16/24 10:50 AM, yudhi s wrote: > Hi, > > Below question got in one discussion.Appreciate any guidance on this. > > Below is a query which is running for ~40 seconds. As it's a query which > is executed from UI , we were expecting it to finish in <~5 seconds. It > has a "IN" and a "NOT IN" subquery , from the execution path it seems > the total response time is mainly, to be sum of the "IN" and the "NOT > IN" subquery section. My thought was that both "IN" and "NOT IN" should > be executed/evaluated in parallel but not in serial fashion. > > In the execution path below , the line number marked in *bold* are the > top lines for the IN and NOT IN subquery evaluation and they are showing > "Actual time" as Approx ~9 seconds and ~8 seconds and they seems to be > summed up and the top lines showing it to be ~19 seconds. Then onwards > it keeps on increasing with other "nested loop" joins. > > *Note*:- This query is running on a MYSQL 8.0 database. So I'm wondering > if there is any mysql list similar to Oracle list , in which i can share > this issue? > This is the Postgres list. As to your question maybe one of the forums?: https://forums.mysql.com/ In particular Performance: https://forums.mysql.com/list.php?24 -- Adrian Klaver adrian.klaver@aklaver.com
On Thu, Oct 17, 2024 at 3:06 AM Peter J. Holzer <hjp-pgsql@hjp.at> wrote:
On 2024-10-16 23:20:36 +0530, yudhi s wrote:
> Below is a query which is running for ~40 seconds.
[...]
> In the execution path below , the line number marked in bold are the top lines
> for the IN and NOT IN subquery evaluation and they are showing "Actual time" as
> Approx ~9 seconds and ~8 seconds and they seems to be summed up and the top
> lines showing it to be ~19 seconds. Then onwards it keeps on increasing with
> other "nested loop" joins.
>
> Note:- This query is running on a MYSQL 8.0 database. So I'm wondering if there
> is any mysql list similar to Oracle list , in which i can share this issue?
The execution plan looks like a postgresql execution plan, not a mysql
execution plan. Did you run this query on postgresql? That may be
interesting for comparison purposese, but ultimately it is useless: You
won't get mysql to work like postgresql, and any tips to speed up this
query on postgresql (which is all you can expect on a postgresql mailing
list) probably won't work on mysql.
Agreed. Postgres and mysql may have differences in how the optimizer is interpreting the stats and coming up with the execution oath. However, I was looking if the query can be written efficiently by tweaking the current logic. It's actually spending the majority of the time doing the "IN" and "NOT IN" evaluation and in that it's using the same exact subquery for the "UNION ALL" . And the overall execution time is summation of the IN and NOT IN clause evaluation. So I was thinking of a better way of writing the same logically.
On Thu, Oct 17, 2024 at 3:06 AM Peter J. Holzer <hjp-pgsql@hjp.at> wrote:
The execution plan looks like a postgresql execution plan, not a mysql
execution plan. Did you run this query on postgresql? That may be
interesting for comparison purposese, but ultimately it is useless: You
won't get mysql to work like postgresql, and any tips to speed up this
query on postgresql (which is all you can expect on a postgresql mailing
list) probably won't work on mysql.
Tried running the same in postgres and below is the plan from bothe postgres and mysql. Can you please guide me to understand , if anything else can be done to make it better?
On 10/21/24 23:31, yudhi s wrote:
On Thu, Oct 17, 2024 at 3:06 AM Peter J. Holzer <hjp-pgsql@hjp.at> wrote:
The execution plan looks like a postgresql execution plan, not a mysql
execution plan. Did you run this query on postgresql? That may be
interesting for comparison purposese, but ultimately it is useless: You
won't get mysql to work like postgresql, and any tips to speed up this
query on postgresql (which is all you can expect on a postgresql mailing
list) probably won't work on mysql.
Tried running the same in postgres and below is the plan from bothe postgres and mysql. Can you please guide me to understand , if anything else can be done to make it better?
Here's your postgres query as analyzed and made more readable by the most excellent explain.depesz.com.
https://explain.depesz.com/s/VyeM#html
Under the hints tab are suggestions to bump your work_mem to avoid writing sorts out to disk.
Jeff
To be frank, there is so much wrong with this query that it is hard to know where to start. But a few top items:
* Make sure all of the tables involved have been analyzed. You might want to bump default_statistics_target up and see if that helps.
* As mentioned already, increase work_mem, as you have things spilling to disk (e.g. external merge Disk: 36280kB)
* Don't use the "FROM table1, table2, table3" syntax but use "FROM table1 JOIN table2 ON (...) JOIN table3 ON (...)
* Try not to use subselects. Things like WHERE x IN (SELECT ...) are expensive and hard to optimize.
* You have useless GROUP BY clauses in there. Remove to simplify the query
* There is no LIMIT. Does the client really need all 135,214 rows?
Cheers,
Greg
* Make sure all of the tables involved have been analyzed. You might want to bump default_statistics_target up and see if that helps.
* As mentioned already, increase work_mem, as you have things spilling to disk (e.g. external merge Disk: 36280kB)
* Don't use the "FROM table1, table2, table3" syntax but use "FROM table1 JOIN table2 ON (...) JOIN table3 ON (...)
* Try not to use subselects. Things like WHERE x IN (SELECT ...) are expensive and hard to optimize.
* You have useless GROUP BY clauses in there. Remove to simplify the query
* There is no LIMIT. Does the client really need all 135,214 rows?
Cheers,
Greg
On Tue, Oct 22, 2024 at 3:02 PM Greg Sabino Mullane <htamfids@gmail.com> wrote:
[snip]
* Don't use the "FROM table1, table2, table3" syntax but use "FROM table1 JOIN table2 ON (...) JOIN table3 ON (...)
Death to <Redacted>, and butter sauce.
Don't boil me, I'm still alive.
<Redacted> crustacean!
On Tue, Oct 22, 2024 at 1:01 PM Ron Johnson <ronljohnsonjr@gmail.com> wrote:
On Tue, Oct 22, 2024 at 3:02 PM Greg Sabino Mullane <htamfids@gmail.com> wrote:[snip]* Don't use the "FROM table1, table2, table3" syntax but use "FROM table1 JOIN table2 ON (...) JOIN table3 ON (...)Why?
Readability is improved when done correctly; and you need to know it for outer joins anyway, so consistency.
David J.
On Wed, Oct 23, 2024 at 12:32 AM Greg Sabino Mullane <htamfids@gmail.com> wrote:
To be frank, there is so much wrong with this query that it is hard to know where to start. But a few top items:
* Make sure all of the tables involved have been analyzed. You might want to bump default_statistics_target up and see if that helps.
* As mentioned already, increase work_mem, as you have things spilling to disk (e.g. external merge Disk: 36280kB)
* Don't use the "FROM table1, table2, table3" syntax but use "FROM table1 JOIN table2 ON (...) JOIN table3 ON (...)
* Try not to use subselects. Things like WHERE x IN (SELECT ...) are expensive and hard to optimize.
* You have useless GROUP BY clauses in there. Remove to simplify the query
* There is no LIMIT. Does the client really need all 135,214 rows?
https://gist.github.com/databasetech0073/746353a9e76d5e29b2fc6abdc80cdef8
Again , not able to clearly understand the third point you said below. Can you please clarify a bit more. Do you mean we should write it as exists /not exists rather IN and NOT IN and that will improve the performance? I hope the third point doesn't matter much as we have all equijoin used here. Correct me if I'm wrong.
Again , not able to clearly understand the third point you said below. Can you please clarify a bit more. Do you mean we should write it as exists /not exists rather IN and NOT IN and that will improve the performance? I hope the third point doesn't matter much as we have all equijoin used here. Correct me if I'm wrong.
"Try not to use subselects. Things like WHERE x IN (SELECT ...) are expensive and hard to optimize."
Additionally in the plan which mysql makes and showing the highest response time, is it suffering because of differences of the speed of the underlying IO/storage or is it just because of the optimization features which are available in postgres and not there in mysql ? Trying to understand if it can be identified from the execution plan itself.
On Tue, 2024-10-22 at 13:06 -0700, David G. Johnston wrote: > On Tue, Oct 22, 2024 at 1:01 PM Ron Johnson <ronljohnsonjr@gmail.com> wrote: > > On Tue, Oct 22, 2024 at 3:02 PM Greg Sabino Mullane <htamfids@gmail.com> wrote: > > > * Don't use the "FROM table1, table2, table3" syntax but use "FROM table1 JOIN table2 ON (...) JOIN table3 ON (...) > > > > Why? > > Readability is improved when done correctly; and you need to know it for outer joins anyway, so consistency. Also, you cannot accidentally forget a join condition. Yours, Laurenz Albe
On Wed, Oct 23, 2024 at 2:06 AM yudhi s <learnerdatabase99@gmail.com> wrote:
On Wed, Oct 23, 2024 at 12:32 AM Greg Sabino Mullane <htamfids@gmail.com> wrote:To be frank, there is so much wrong with this query that it is hard to know where to start. But a few top items:
* Make sure all of the tables involved have been analyzed. You might want to bump default_statistics_target up and see if that helps.
* As mentioned already, increase work_mem, as you have things spilling to disk (e.g. external merge Disk: 36280kB)
* Don't use the "FROM table1, table2, table3" syntax but use "FROM table1 JOIN table2 ON (...) JOIN table3 ON (...)
* Try not to use subselects. Things like WHERE x IN (SELECT ...) are expensive and hard to optimize.
* You have useless GROUP BY clauses in there. Remove to simplify the query
* There is no LIMIT. Does the client really need all 135,214 rows?I tried running the query by removing both the "group by" from the inner subqueries (I think the initial thought was that they will give distinct records to the outer query and will thus help), and added limit 500 at the last and also set the work_mem to 2GB for that session before running the query. But seeing the response increased to ~5 seconds (from ~3.1 seconds earlier). Below I have updated the execution plan for the same at the last section.https://gist.github.com/databasetech0073/746353a9e76d5e29b2fc6abdc80cdef8
Again , not able to clearly understand the third point you said below. Can you please clarify a bit more. Do you mean we should write it as exists /not exists rather IN and NOT IN and that will improve the performance? I hope the third point doesn't matter much as we have all equijoin used here. Correct me if I'm wrong."Try not to use subselects. Things like WHERE x IN (SELECT ...) are expensive and hard to optimize."Additionally in the plan which mysql makes and showing the highest response time, is it suffering because of differences of the speed of the underlying IO/storage or is it just because of the optimization features which are available in postgres and not there in mysql ? Trying to understand if it can be identified from the execution plan itself.
OR its any optimization feature which helps postgres to give it an edge in performance here over mysql?
In mysql plan:-
-> Index lookup on EX_STS using EX_STS_INDEX (AID=b3.AID, RC_ID=b3.RC_ID, RC_VNB=b3.RC_VNB) (cost=0.43 rows=2) (actual time=0.014..0.021 rows=2 loops=70904)
VS
In postgres plan:-
-> Index Scan using EX_STS_INDEX on RCE_STS EX_STS (cost=0.42..0.82 rows=1 width=424) (actual time=0.006..0.007 rows=2 loops=70904)
**************
In mysql plan
-> Covering index lookup on mns using M_INF_AID_index (AID='XXXXXXXXXXXXXXXXXXX') (cost=9187.54 rows=72748) (actual time=0.058..19.637 rows=35980 loops=1)
-> Filter: (RNS.ASID = 'XXXXXXXXXXXXXXXXXXX') (cost=43.50 rows=42) (actual time=0.042..0.218 rows=97 loops=1)
VS
In postgres plan
-> Bitmap Index Scan on M_INF_AID_index (cost=0.00..406.98 rows=36074 width=0) (actual time=0.790..0.790 rows=35980 loops=1)
In mysql plan:-
-> Index lookup on EX_STS using EX_STS_INDEX (AID=b3.AID, RC_ID=b3.RC_ID, RC_VNB=b3.RC_VNB) (cost=0.43 rows=2) (actual time=0.014..0.021 rows=2 loops=70904)
VS
In postgres plan:-
-> Index Scan using EX_STS_INDEX on RCE_STS EX_STS (cost=0.42..0.82 rows=1 width=424) (actual time=0.006..0.007 rows=2 loops=70904)
**************
In mysql plan
-> Covering index lookup on mns using M_INF_AID_index (AID='XXXXXXXXXXXXXXXXXXX') (cost=9187.54 rows=72748) (actual time=0.058..19.637 rows=35980 loops=1)
-> Filter: (RNS.ASID = 'XXXXXXXXXXXXXXXXXXX') (cost=43.50 rows=42) (actual time=0.042..0.218 rows=97 loops=1)
VS
In postgres plan
-> Bitmap Index Scan on M_INF_AID_index (cost=0.00..406.98 rows=36074 width=0) (actual time=0.790..0.790 rows=35980 loops=1)
Index Cond: ((AID)::text = 'XXXXXXXXXXXXXXXXXXX'::text)
Additionally in the plan which mysql makes and showing the highest response time, is it suffering because of differences of the speed of the underlying IO/storage or is it just because of the optimization features which are available in postgres and not there in mysql ? Trying to understand if it can be identified from the execution plan itself.
I think trying to compare postgres and mysql plans against each other is not a very useful endeavor. There are fundamental design decisions between the two. Focus on making your Postgres query the best it can be, full stop. Optimize your queries, make sure the database is analyzed, and tweak some configs as needed.
Also, you cannot accidentally forget a join condition.
Yes, this is the primary reason. Cartesian joins hurt.
Again , not able to clearly understand the third point you said below. Can you please clarify a bit more. Do you mean we should write it as exists /not exists rather IN and NOT IN and that will improve the performance?
It gives Postgres more options on how to do things, so yes, it can be better.
Cheers,
Greg