OpenLooKeng性能测试方案

Openlookeng
OpenLooKeng

一、测试环境

1.1 硬件信息

操作系统 CPU 内存 磁盘
Centos 7 6核 32GB 1.5TB
Centos 7 6核 32GB 1.5TB
Centos 7 6核 32GB 1.5TB

1.2 软件信息

组件名称 版本
JDK 1.8.0_322
OpenLooKeng 1.5.0
Jmeter 5.4.3
presto-benchmark-driver 1.5.0
hive-testbench /
Nginx 1.20.2

二、测试方法

本次测试主要针对以下两个方面：

  1. 使用官方提供的presto-benchmark-driver-1.5.0-executable.jar对OpenLooKeng进行性能测试。
    1. Scan query
    2. Aggregation query
    3. Join query
  2. 使用JMeter对OpenLooKeng的查询引擎进行并发吞吐量测试，分别设置并发数为10、50、100进行。

注：本文以Hive数据源为例，OpenLooKeng使用Hive Connector进行连接。

三、测试数据

使用开源数据生成工具hive_testbench生成测试数据，生成数据总量为10GB，本次用到的测试数据大约2GB。

四、执行性能测试

4.1 SQL性能测试

4.1.1 Scan query

select
    ss_sold_time_sk,
    ss_item_sk,
    ss_customer_sk
from
    store_sales
where
    ss_cdemo_sk > 20000
limit
    1000;

4.1.2 Aggregation query

select
    dt.d_year,
    item.i_brand_id brand_id,
    item.i_brand brand,
    sum(ss_sales_price) sum_agg
from
    date_dim dt,
    store_sales,
    item
where
    dt.d_date_sk = store_sales.ss_sold_date_sk
    and store_sales.ss_item_sk = item.i_item_sk
    and item.i_manufact_id = 816
    and dt.d_moy = 11
group by
    dt.d_year,
    item.i_brand,
    item.i_brand_id
order by
    dt.d_year,
    sum_agg desc,
    brand_id
limit
    1000;

4.1.3 Join query

select cd_gender, cd_marital_status, cd_education_status, cd_purchase_estimate, cd_credit_rating, cd_dep_count, cd_dep_employed_count, cd_dep_college_count from customer c,customer_address ca,customer_demographics where c.c_current_addr_sk = ca.ca_address_sk and cd_demo_sk = c.c_current_cdemo_sk limit 1000;
select
    cd_gender,
    cd_marital_status,
    cd_education_status,
    cd_purchase_estimate,
    cd_credit_rating,
    cd_dep_count,
    cd_dep_employed_count,
    cd_dep_college_count
from
    customer c,
    customer_address ca,
    customer_demographics
where
    c.c_current_addr_sk = ca.ca_address_sk
    and cd_demo_sk = c.c_current_cdemo_sk
limit
    1000;

4.2 JMeter测试

Jdbc请求脚本统一用：

select
    cd_gender,
    cd_marital_status,
    cd_education_status,
    cd_purchase_estimate,
    cd_credit_rating,
    cd_dep_count,
    cd_dep_employed_count,
    cd_dep_college_count
from
    customer c,
    customer_address ca,
    customer_demographics
where
    c.c_current_addr_sk = ca.ca_address_sk
    and cd_demo_sk = c.c_current_cdemo_sk
limit
    1000

4.2.1 并发数10

结果：

4.2.1 并发数50

结果：

4.2.1 并发数100

结果：

五、结果总结

5.1 SQL性能测试

参数\组件 OpenLooKeng
Query scan Aggregation scan Join scan
第一次执行耗时 5.83 s 36.06 s 12.58 s
第二次执行耗时 5.54 s 32.01 s 11.23 s
CPU 50% + 70% + 50% +
内存 10% + 20% + 20% +

5.2 JMeter测试

参数\组件 OpenLooKeng
Samples 10 50 100
Average 14893 72955 47878
Min 14046 66442 37451
Max 15513 76075 54587
Throughput 0.62/sec 0.68/sec 1.8/sec
Received KB/sec 28.54 30.02 83.96

上表参数解释如下：

  • samples：请求线程数
  • average：平均响应时间（单位：毫秒）
  • min：最小相应时间（单位：毫秒）
  • max：最大响应时间（单位：毫秒）
  • throughput：吞吐量–默认情况下表示每秒完成的请求数
  • received KB/sec：每秒从服务端接收到的数据量

六、参考

【1】OpenLooKeng性能测试方案：https://dtwave.yuque.com/wm7zom/xxznbo/nlgz0f