clickhouse 子查询_TPCDS用于Clickhouse和Doris性能测试
大致介紹
TPC-DS采用星型、雪花型等多維數(shù)據(jù)模式。它包含7張事實表，17張緯度表平均每張表含有18列。其工作負載包含99個SQL查詢，覆蓋SQL99和2003的核心部分以及OLAP。這個測試集包含對大數(shù)據(jù)集的統(tǒng)計、報表生成、聯(lián)機查詢、數(shù)據(jù)挖掘等復雜應用，測試用的數(shù)據(jù)和值是有傾斜的，與真實數(shù)據(jù)一致。可以說TPC-DS是與真實場景非常接近的一個測試集，也是難度較大的一個測試集。
Clickhouse是俄羅斯Yandex公司開源的一個非常快的數(shù)據(jù)管理系統(tǒng)，性能非常強悍。Apache Doris是百度開源的另一個基于 MPP 的交互式 SQL 數(shù)據(jù)倉庫，主要用于解決報表和多維分析，成熟穩(wěn)定。
下載編譯
下載官方網址
也可以通過Git下載
cd tpcds-kit/tools
make OS=LINUX
1
2
3
[tools]$ ll *.sql
-rw-r----- 1 prodadmin prodadmin 13875 Mar 12 03:44 tpcds_ri.sql
-rw-r----- 1 prodadmin prodadmin 22153 Mar 12 03:44 tpcds_source.sql
-rw-r----- 1 prodadmin prodadmin 30001 Mar 12 03:44 tpcds.sql
1
2
3
4
5
[tools]$ ll ds*
-rwxr-x--- 1 prodadmin prodadmin 455880 Apr 16 19:15 dsdgen
-rwxr-x--- 1 prodadmin prodadmin 292254 Apr 16 19:15 dsqgen
1
2
3
4
[tools]$ ls ../query_templates/
ansi.tpl query12.tpl query18.tpl query23.tpl query29.tpl query34.tpl query3.tpl query45.tpl query50.tpl query56.tpl query61.tpl query67.tpl query72.tpl query78.tpl query83.tpl query89.tpl query94.tpl query9.tpl
db2.tpl query13.tpl query19.tpl query24.tpl query2.tpl query35.tpl query40.tpl query46.tpl query51.tpl query57.tpl query62.tpl query68.tpl query73.tpl query79.tpl query84.tpl query8.tpl query95.tpl README
netezza.tpl query14.tpl query1.tpl query25.tpl query30.tpl query36.tpl query41.tpl query47.tpl query52.tpl query58.tpl query63.tpl query69.tpl query74.tpl query7.tpl query85.tpl query90.tpl query96.tpl sqlserver.tpl
oracle.tpl query15.tpl query20.tpl query26.tpl query31.tpl query37.tpl query42.tpl query48.tpl query53.tpl query59.tpl query64.tpl query6.tpl query75.tpl query80.tpl query86.tpl query91.tpl query97.tpl templates.lst
query10.tpl query16.tpl query21.tpl query27.tpl query32.tpl query38.tpl query43.tpl query49.tpl query54.tpl query5.tpl query65.tpl query70.tpl query76.tpl query81.tpl query87.tpl query92.tpl query98.tpl
query11.tpl query17.tpl query22.tpl query28.tpl query33.tpl query39.tpl query44.tpl query4.tpl query55.tpl query60.tpl query66.tpl query71.tpl query77.tpl query82.tpl query88.tpl query93.tpl query99.tpl
1
2
3
4
5
6
7
8
建表語句
請根據(jù)Hive、Doris、Clickhouse等組件特點，修改建表語句，請注意，列是否為空，列的順序等和后面步驟的導入數(shù)據(jù)密切相關，請勿輕易修改。
1?Clickhouse數(shù)據(jù)類型
2?Doris建表和數(shù)據(jù)類型
create table customer_address
create table customer_demographics
create table date_dim
create table warehouse
create table ship_mode
create table time_dim
create table reason
create table income_band
create table item
create table store
create table call_center
create table customer
create table web_site
create table store_returns
create table household_demographics
create table web_page
create table promotion
create table catalog_page
create table inventory
create table catalog_returns
create table web_returns
create table web_sales
create table catalog_sales
create table store_sales
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
數(shù)據(jù)生成
可以建一個腳本，來生成數(shù)據(jù)
1 數(shù)據(jù)分隔符是“|”，空值默認為空。Clickhouse支持的格式，Doris Load格式。
2 scale單位為G，指生成的數(shù)據(jù)量大小，paralle指分割多少個文件，child指第幾個文件
echo $1
mkdir ../../data_tsv/
nohup ./dsdgen -scale 100 -dir ../../data_tsv/ -paralle 10 -child $1 > child$1.log &
1
2
3
4
數(shù)據(jù)導入
Clickhouse數(shù)據(jù)導入，一個例子，我寫了一個convert.py的小腳本，處理分隔符、空值、列順序等問題
cat ./data_tsv/dbgen_version_$1_10.dat|python convert.py ck|clickhouse-client --query="INSERT INTO default.dbgen_version_dist FORMAT CSV"
touch ./data_tsv/dbgen_version_$1_10.done
fi
1
2
3
4
Doris數(shù)據(jù)導入
1
生成Query
cat build_sql.sh./dsqgen \
-DIRECTORY ../query_templates \
-INPUT ../query_templates/templates.lst \
-VERBOSE Y \
-QUALIFY Y \
-SCALE 100 \
-DIALECT sqlserver \
-OUTPUT_DIR ../../query/
1
2
3
4
5
6
7
8
9
Query改寫
比如如下的SQL，在Doris中是可以正確的執(zhí)行的，但是在Clickhouse中不行，CK中需要子查詢嵌套或者用global inner join來顯示指定broadcast的字表。
selectc_last_name,c_first_name,substr(s_city,1,30),ss_ticket_number,amt,profit
from
(select ss_ticket_number
,ss_customer_sk
,store.s_city
,sum(ss_coupon_amt) amt
,sum(ss_net_profit) profit
from store_sales,date_dim,store,household_demographics
where store_sales.ss_sold_date_sk = date_dim.d_date_sk
and store_sales.ss_store_sk = store.s_store_sk
and store_sales.ss_hdemo_sk = household_demographics.hd_demo_sk
and (household_demographics.hd_dep_count = 1 or household_demographics.hd_vehicle_count > -1)
and date_dim.d_dow = 1
and date_dim.d_year in (2000,2000+1,2000+2)
and store.s_number_employees between 200 and 295
group by ss_ticket_number,ss_customer_sk,ss_addr_sk,store.s_city) ms,customer
where ss_customer_sk = c_customer_sk
order by c_last_name,c_first_name,substr(s_city,1,30), profit
limit 10;
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
改寫后
select c_last_name, c_first_name, substr(tbl1.s_city,1,30), ss_ticket_number, amt, profitfrom customer_dist
global inner join (
select ss_ticket_number, ss_customer_sk, s_city, sum(ss_coupon_amt) as amt, sum(ss_net_profit) as profit
from store_sales_dist
global inner join ( select d_date_sk from date_dim_dist where date_dim_dist.d_dow = 1 and d_year in (2000,2000+1,2000+2) ) on ss_sold_date_sk = d_date_sk
global inner join ( select s_store_sk, s_city from store_dist where s_number_employees between 200 and 295 ) on ss_store_sk = s_store_sk
global inner join ( select hd_demo_sk from household_demographics_dist where hd_dep_count = 1 or hd_vehicle_count > -1 ) on ss_hdemo_sk = hd_demo_sk
group by ss_ticket_number, ss_customer_sk, ss_addr_sk, s_city
) tbl1 on ss_customer_sk = c_customer_sk
order by c_last_name, c_first_name, substr(s_city,1,30), profit
limit 10;
1
2
3
4
5
6
7
8
9
10
11
12
測試結論
1 導入數(shù)據(jù)Clickhouse快
2 數(shù)據(jù)壓縮率Clickhouse好
3 單表查詢Clickhouse快
4 Join查詢兩者各有優(yōu)劣，數(shù)據(jù)量小情況下Clickhouse好，數(shù)據(jù)量大Doris好
5 Doris對SQL支持情況要好
總結
以上是生活随笔為你收集整理的clickhouse 子查询_TPCDS用于Clickhouse和Doris性能测试的全部內容，希望文章能夠幫你解決所遇到的問題。
- 上一篇： python画卡通人物用什么_干啥啥不行
- 下一篇： 开运算和闭运算_OpenCV计算机视觉学