Uh oh!
There was an error while loading. Please reload this page.
[fix](scanner) Fix incorrect _max_thread_num in scanner context when many queries are running. - #41273
Merged
Merged
Conversation
doris-robot
commented
Sep 25, 2024
Thank you for your contribution to Apache Doris. Since 2024-03-18, the Document has been moved to doris-website. |
RoanHeNaN
commented
Sep 25, 2024
ContributorAuthor
run buildall |
RoanHeNaN
commented
Sep 25, 2024
ContributorAuthor
almost same with #40569. |
doris-robot
commented
Sep 25, 2024
TeamCity be ut coverage result: |
RoanHeNaN
commented
Sep 27, 2024
ContributorAuthor
run buildall |
RoanHeNaNforce-pushed
the
refactor-scanner
branch
from
September 27, 2024 02:42
0925b96 to
c27cf18CompareRoanHeNaN
commented
Sep 27, 2024
ContributorAuthor
run buildall |
doris-robot
commented
Sep 27, 2024
TPC-H: Total hot run time: 41063 ms |
doris-robot
commented
Sep 27, 2024
TeamCity be ut coverage result: |
doris-robot
commented
Sep 27, 2024
TPC-DS: Total hot run time: 191933 ms |
doris-robot
commented
Sep 27, 2024
ClickBench: Total hot run time: 32.74 s |
RoanHeNaN
commented
Sep 27, 2024
ContributorAuthor
run buildall |
doris-robot
commented
Sep 27, 2024
TPC-H: Total hot run time: 40778 ms |
doris-robot
commented
Sep 27, 2024
TeamCity be ut coverage result: |
doris-robot
commented
Sep 27, 2024
TPC-DS: Total hot run time: 190996 ms |
doris-robot
commented
Sep 27, 2024
ClickBench: Total hot run time: 32.26 s |
Contributor
PR approved by at least one committer and no changes requested. |
Contributor
PR approved by anyone and no changes requested. |
yiguolei
reviewed
Oct 8, 2024
Uh oh!
There was an error while loading. Please reload this page.
RoanHeNaNforce-pushed
the
refactor-scanner
branch
from
October 8, 2024 09:29
c843245 to
2ab35e1CompareUh oh!
There was an error while loading. Please reload this page.
RoanHeNaN
commented
Oct 9, 2024
ContributorAuthor
run buildall |
2 similar comments
RoanHeNaN
commented
Oct 9, 2024
ContributorAuthor
run buildall |
RoanHeNaN
commented
Oct 9, 2024
ContributorAuthor
run buildall |
RoanHeNaNforce-pushed
the
refactor-scanner
branch
from
October 10, 2024 15:45
f5855b7 to
b67039dCompareRoanHeNaN
commented
Oct 10, 2024
ContributorAuthor
run buildall |
doris-robot
commented
Oct 10, 2024
TeamCity be ut coverage result: |
RoanHeNaNforce-pushed
the
refactor-scanner
branch
from
October 11, 2024 02:57
b67039d to
74c20dcCompareUh oh!
There was an error while loading. Please reload this page.
RoanHeNaN
commented
Oct 11, 2024
ContributorAuthor
run buildall |
doris-robot
commented
Oct 11, 2024
TeamCity be ut coverage result: |
Contributor
PR approved by at least one committer and no changes requested. |
yiguolei
approved these changes
Oct 11, 2024
RoanHeNaN added a commit
to RoanHeNaN/doris
that referenced
this pull request
Oct 17, 2024
…many queries are running. (apache#41273) 1. Minor refactor for scanner constructor, calculation of _max_thread_num is moved to init method 2. The expected value of _max_thread_num is changed. There is no need to submit too many scan task to scan scheduler, since thread num is limited. 3. Calculation of _max_bytes_in_queue is changed. _max_bytes_in_queue for each scan instance is limited to 100MB by default. ``` mysql [tpch]>select count(*) from supplier; -------------- select count(*) from supplier -------------- +----------+ | count(*) | +----------+ | 1000000 | +----------+ 1 row in set (0.04 sec) mysql [tpch]>select count(*) from revenue0; -------------- select count(*) from revenue0 -------------- +----------+ | count(*) | +----------+ | 1000000 | +----------+ 1 row in set (0.19 sec) ``` To illustrate the effect, we need to create much scanners, so ``` set global experimental_parallel_scan_min_rows_per_scanner=29715 ``` default value is `2097152`, we can make scanner num almost equal to `experimental_parallel_scan_max_scanners_count` which is 48. Lets use mysqlslap to do concurrent test. Current master: ```text [hezhiqiang@VM-10-8-centos be_1]$ mysqlslap -hxxxx -uroot -Pyyyy --create-schema=tpch -c 20 -i 5 -q "select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey;" Benchmark Average number of seconds to run all queries: 12.480 seconds Minimum number of seconds to run all queries: 12.159 seconds Maximum number of seconds to run all queries: 12.843 seconds Number of clients running queries: 20 Average number of queries per client: 1 [hezhiqiang@VM-10-8-centos be_1]$ mysqlslap -hyyyy -uroot -Pyyyy --create-schema=tpch -c 25 -i 5 -q "select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey;" mysqlslap: Cannot run query select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey; ERROR : errCode = 2, detailMessage = (10.16.10.8)[TOO_MANY_TASKS]Failed to submit scanner to scanner pool reason:Thread pool Scan_normal is at capacity (192/192 tasks running, 102400/102400 tasks queued)|type:0 ``` After this pr ``` [hezhiqiang@VM-10-8-centos lib]$ mysqlslap -hxxx -uroot -Pxxx --create-schema=tpch -c 50 -i 5 -q "select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey;" Benchmark Average number of seconds to run all queries: 31.520 seconds Minimum number of seconds to run all queries: 30.164 seconds Maximum number of seconds to run all queries: 34.131 seconds Number of clients running queries: 50 Average number of queries per client: 1 ``` The max concurrency increased from 25 to 50. Actually, for sequential query test, the performance does not decrease, `submit_many_scan_tasks_for_potential_performance_issue` can be remove in the future.
RoanHeNaN added a commit
to RoanHeNaN/doris
that referenced
this pull request
Oct 17, 2024
…many queries are running. (apache#41273) 1. Minor refactor for scanner constructor, calculation of _max_thread_num is moved to init method 2. The expected value of _max_thread_num is changed. There is no need to submit too many scan task to scan scheduler, since thread num is limited. 3. Calculation of _max_bytes_in_queue is changed. _max_bytes_in_queue for each scan instance is limited to 100MB by default. ``` mysql [tpch]>select count(*) from supplier; -------------- select count(*) from supplier -------------- +----------+ | count(*) | +----------+ | 1000000 | +----------+ 1 row in set (0.04 sec) mysql [tpch]>select count(*) from revenue0; -------------- select count(*) from revenue0 -------------- +----------+ | count(*) | +----------+ | 1000000 | +----------+ 1 row in set (0.19 sec) ``` To illustrate the effect, we need to create much scanners, so ``` set global experimental_parallel_scan_min_rows_per_scanner=29715 ``` default value is `2097152`, we can make scanner num almost equal to `experimental_parallel_scan_max_scanners_count` which is 48. Lets use mysqlslap to do concurrent test. Current master: ```text [hezhiqiang@VM-10-8-centos be_1]$ mysqlslap -hxxxx -uroot -Pyyyy --create-schema=tpch -c 20 -i 5 -q "select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey;" Benchmark Average number of seconds to run all queries: 12.480 seconds Minimum number of seconds to run all queries: 12.159 seconds Maximum number of seconds to run all queries: 12.843 seconds Number of clients running queries: 20 Average number of queries per client: 1 [hezhiqiang@VM-10-8-centos be_1]$ mysqlslap -hyyyy -uroot -Pyyyy --create-schema=tpch -c 25 -i 5 -q "select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey;" mysqlslap: Cannot run query select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey; ERROR : errCode = 2, detailMessage = (10.16.10.8)[TOO_MANY_TASKS]Failed to submit scanner to scanner pool reason:Thread pool Scan_normal is at capacity (192/192 tasks running, 102400/102400 tasks queued)|type:0 ``` After this pr ``` [hezhiqiang@VM-10-8-centos lib]$ mysqlslap -hxxx -uroot -Pxxx --create-schema=tpch -c 50 -i 5 -q "select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey;" Benchmark Average number of seconds to run all queries: 31.520 seconds Minimum number of seconds to run all queries: 30.164 seconds Maximum number of seconds to run all queries: 34.131 seconds Number of clients running queries: 50 Average number of queries per client: 1 ``` The max concurrency increased from 25 to 50. Actually, for sequential query test, the performance does not decrease, `submit_many_scan_tasks_for_potential_performance_issue` can be remove in the future.
HappenLee pushed a commit
to HappenLee/incubator-doris
that referenced
this pull request
Apr 24, 2026
…many queries are running. (apache#41273) 1. Minor refactor for scanner constructor, calculation of _max_thread_num is moved to init method 2. The expected value of _max_thread_num is changed. There is no need to submit too many scan task to scan scheduler, since thread num is limited. 3. Calculation of _max_bytes_in_queue is changed. _max_bytes_in_queue for each scan instance is limited to 100MB by default. ``` mysql [tpch]>select count(*) from supplier; -------------- select count(*) from supplier -------------- +----------+ | count(*) | +----------+ | 1000000 | +----------+ 1 row in set (0.04 sec) mysql [tpch]>select count(*) from revenue0; -------------- select count(*) from revenue0 -------------- +----------+ | count(*) | +----------+ | 1000000 | +----------+ 1 row in set (0.19 sec) ``` To illustrate the effect, we need to create much scanners, so ``` set global experimental_parallel_scan_min_rows_per_scanner=29715 ``` default value is `2097152`, we can make scanner num almost equal to `experimental_parallel_scan_max_scanners_count` which is 48. Lets use mysqlslap to do concurrent test. Current master: ```text [hezhiqiang@VM-10-8-centos be_1]$ mysqlslap -hxxxx -uroot -Pyyyy --create-schema=tpch -c 20 -i 5 -q "select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey;" Benchmark Average number of seconds to run all queries: 12.480 seconds Minimum number of seconds to run all queries: 12.159 seconds Maximum number of seconds to run all queries: 12.843 seconds Number of clients running queries: 20 Average number of queries per client: 1 [hezhiqiang@VM-10-8-centos be_1]$ mysqlslap -hyyyy -uroot -Pyyyy --create-schema=tpch -c 25 -i 5 -q "select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey;" mysqlslap: Cannot run query select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey; ERROR : errCode = 2, detailMessage = (10.16.10.8)[TOO_MANY_TASKS]Failed to submit scanner to scanner pool reason:Thread pool Scan_normal is at capacity (192/192 tasks running, 102400/102400 tasks queued)|type:0 ``` After this pr ``` [hezhiqiang@VM-10-8-centos lib]$ mysqlslap -hxxx -uroot -Pxxx --create-schema=tpch -c 50 -i 5 -q "select s_suppkey, s_name, s_address, s_phone, total_revenue from supplier, revenue0 where s_suppkey = supplier_no and total_revenue = ( select max(total_revenue) from revenue0 ) order by s_suppkey;" Benchmark Average number of seconds to run all queries: 31.520 seconds Minimum number of seconds to run all queries: 30.164 seconds Maximum number of seconds to run all queries: 34.131 seconds Number of clients running queries: 50 Average number of queries per client: 1 ``` The max concurrency increased from 25 to 50. Actually, for sequential query test, the performance does not decrease, `submit_many_scan_tasks_for_potential_performance_issue` can be remove in the future.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
To illustrate the effect, we need to create much scanners, so
default value is
2097152, we can make scanner num almost equal toexperimental_parallel_scan_max_scanners_countwhich is 48.Lets use mysqlslap to do concurrent test.
Current master:
After this pr
The max concurrency increased from 25 to 50.
Actually, for sequential query test, the performance does not decrease,
submit_many_scan_tasks_for_potential_performance_issuecan be remove in the future.