Elasticsearch性能测试报告

Elasticsearch的性能测试报告分为基于TPCDS的性能测试报告和基于业务数据的性能测试报告。

测试环境

操作系统: CentOS release 6.2 (Final) ​
虚拟机版本: Java HotSpot(TM) 64-Bit Server VM (build 25.60-b23, mixed mode) ​
测试机器: client01v.qss.zzzc.qihoo.net、clientadmin.dfs.shbt.qihoo.net

XSQL配置

基于TPCDS的性能测试报告

测试目标

本次测试涵盖性能测试与功能测试,主要包含两个目的:

  • 展示XSQL在针对单数据源、多数据源的SQL解析及执行能力

  • 找出XSQL在复杂SQL处理以及大规模数据集计算时存在的问题

测试数据集大小:1G/9710124

测试语句集

编号 SQL语句
1 select * from myes.xitong_xsql_test.taxis
2 select * from myes.xitong_xsql_test.taxis where do_location_id="249"
3 select * from myes.xitong_xsql_test.taxis where fare_amount < 3
4 select * from myes.xitong_xsql_test.taxis where dropoff_datetime is null
5 select * from myes.xitong_xsql_test.taxis where dropoff_datetime is not null
6 select * from myes.xitong_xsql_test.taxis where rate_code_id in ("1","2")
7 select vendor_id,total_amount,trip_distance from myes.xitong_xsql_test.taxis where payment_type = "1" and tolls_amount = 0
8 select * from myes.xitong_xsql_test.taxis order by total_amount desc
9 select * from myes.xitong_xsql_test.taxis order by vendor_id
10 select * from myes.xitong_xsql_test.taxis limit 20
11 select count(payment_type),sum(total_amount) from myes.xitong_xsql_test.taxis
12 select avg(population) from myes.xitong_xsql_test.geonames_all
13 select payment_type from myes.xitong_xsql_test.taxis group by payment_type
14 select count(payment_type),payment_type from myes.xitong_xsql_test.taxis group by payment_type
15 select count(payment_type),payment_type,total_amount from myes.xitong_xsql_test.taxis group by payment_type,total_amount
16 select sum(total_amount) from myes.xitong_xsql_test.taxis group by vendor_id
17 select sum(population) from myes.xitong_xsql_test.geonames_all group by country_code,timezone
18 select sum(total_amount),vendor_id,payment_type from myes.xitong_xsql_test.taxis group by vendor_id,payment_type
19 select max(population) from myes.xitong_xsql_test.geonames_all group by country_code
20 select min(population),country_code from myes.xitong_xsql_test.geonames_all group by country_code
21 select avg(population) from myes.xitong_xsql_test.geonames_all group by country_code
22 select count(population),country_code from myes.xitong_xsql_test.geonames_all group by country_code order by country_code limit 100
23 select avg(population),max(population),timezone,country_code from myes.xitong_xsql_test.geonames_all group by country_code,timezone order by max(population) limit 100
24 select count(population),timezone,country_code from myes.xitong_xsql_test.geonames_all group by country_code,timezone order by count(population) limit 100
25 select distinct dem,location from myes.xitong_xsql_test.geonames_all
26 select count(distinct passenger_count) from myes.xitong_xsql_test.taxis

测试结果(单位:秒)

执行引擎 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 26
XSQL [Pushdown] 0.211 0.163 0.173 0.223 0.224 0.249 0.262 0.271 0.195 0.227 0.288 0.158 0.135 0.134 0.198 0.138 0.226 0.161 0.351 0.137 0.132 0.205 0.736 0.559 0.487 0.137
ES 0.147 0.088 0.102 0.177 0.148 0.162 0.165 0.16 0.161 0.178 0.098 0.113 0.08 0.08 0.1 0.098 0.173 0.114 0.116 0.097 0.114 0.145 0.286 0.243 0.262 0.114

结论

基于业务数据的性能测试报告

测试目标

本次测试涵盖性能测试与功能测试,主要包含两个目的:

  • 展示XSQL在针对单数据源、多数据源的SQL解析及执行能力

  • 找出XSQL在复杂SQL处理以及大规模数据集计算时存在的问题

测试数据集大小

数据源 50 G/63178212 docs 100 G/123714760 docs 200 G/260567692 docs 500G/371503959 docs 1 T /466168986 docs
Elasticsearch logsget-user-qdas-newuser-channel-all.msdocker_frontmidres1 logsget-user-qdas-newuser-channel-all.huajiao_frontmidres1 logsget-user-qdas-newuser-channel-all.chromium_frontmidres1 logsget-user-pc-profile-v1.profile logsget-user-profile.profile

测试语句集

编号 SQL
1 SELECT m2, active_channel FROM msdocker_frontmidres1 WHERE m2 IS NOT NULL LIMIT 10
2 SELECT m2, active_channel FROM huajiao_frontmidres1 WHERE m2 IS NOT NULL LIMIT 10
3 SELECT m2, active_channel FROM chromium_frontmidres1 WHERE m2 IS NOT NULL LIMIT 10
4 SELECT city, ip FROM logsget_user_pc_profile_v1.profile WHERE city IS NOT NULL LIMIT 10
5 SELECT city, ip FROM logsget_user_profile.profile WHERE city IS NOT NULL LIMIT 10
编号 SQL
1 SELECT appkey, COUNT(m2) FROM msdocker_frontmidres1 WHERE m2 IS NOT NULL GROUP BY appkey LIMIT 10
2 SELECT appkey, COUNT(m2) FROM huajiao_frontmidres1 WHERE m2 IS NOT NULL GROUP BY appkey LIMIT 10
3 SELECT appkey, COUNT(m2) FROM chromium_frontmidres1 WHERE m2 IS NOT NULL GROUP BY appkey LIMIT 10
4 SELECT province, COUNT(ip), SUM(pro_daohang), AVG(pro_daohang) FROM logsget_user_pc_profile_v1.profile WHERE ip IS NOT NULL GROUP BY province LIMIT 10
5 SELECT pro_zhushou, COUNT(m2), SUM(pro_info_flow), AVG(pro_info_flow) FROM logsget_user_profile.profile WHERE m2 IS NOT NULL GROUP BY pro_zhushou LIMIT 10

测试结果(单位:秒)

50G/63178212 docs 100G/123714760 docs 200G/260567692 docs 500G/371503959 docs 1T/466168986 docs
Elasticsearch 0.399 s 0.558 s 0.949 s 1.080 s 0.381 s
XSQL Pushdown [Elasticsearch] 0.426 s 0.575 s 1.025 s 1.107 s 0.407 s
XSQL [Elasticsearch] 324.813 568.389 1045.91 1121.31 271.082

50G/63178212 docs 100G/123714760 docs 200G/260567692 docs 500G/371503959 docs 1T/466168986 docs
Elasticsearch 0.774 s 1.436 s 2.665 s 0.096 s 0.094 s
XSQL Pushdown [Elasticsearch] 0.790 s 1.478 s 2.671 s 0.135 s 0.130 s
XSQL [Elasticsearch] 338.514 896.007 2507.09 3915.32 3886.388

ES本身不支持where条件后的子查询操作,故本报告不做此处的测试。

ES本身支持的连接操作非常有限且昂贵,故本报告不做此处的测试。

结论