【Trino实战】Trino下ORC与Parquet查询性能分析

Trino下ORC与Parquet查询性能分析

环境

• OS：CentOS 6.5

• JDK：1.8

• 内存：256G

• 磁盘：HDD

• CPU：Dual 8-core Intel® Xeon® CPU (32 Hyper-Threads) E5-2630 v3 @ 2.40GHz

• HDFS：2.9.2

• Hive：2.3.9

• Trino：418

借助Trino对以下格式文件的查询耗时，来分析不同格式文件的查询效率

• ORC
• Parquet
• TextFile
• RCFile

实验数据准备

1. 创建对应hive表

   ## 创建json临时表
    create table tmpjson(line string) row format delimited fields terminated by "\n";
   ## 利用hive客户端加载本地json数据文件
    LOAD DATA LOCAL INPATH '/opt/documents.json' OVERWRITE INTO TABLE test_trino.tmpjson;
   
   ## 创建国家主要城市表
    CREATE TABLE `test_trino.all_countries_orc`(
     `geonameid` bigint COMMENT '地名ID', 
     `name` string COMMENT '地名', 
     `latitude` double COMMENT '纬度',
     `longitude` double COMMENT '经度', 
     `country_code` string COMMENT '国家编码',
     `population` bigint COMMENT '城市人口数')
   COMMENT '国家城市表'
   ROW FORMAT SERDE 
     'org.apache.hadoop.hive.ql.io.orc.OrcSerde'
   STORED AS INPUTFORMAT 
     'org.apache.hadoop.hive.ql.io.orc.OrcInputFormat' 
   OUTPUTFORMAT 
     'org.apache.hadoop.hive.ql.io.orc.OrcOutputFormat';
     
    CREATE TABLE `test_trino.all_countries_parquet`(
     `geonameid` bigint COMMENT '地名ID', 
     `name` string COMMENT '地名', 
     `latitude` double COMMENT '纬度',
     `longitude` double COMMENT '经度', 
     `country_code` string COMMENT '国家编码',
     `population` bigint COMMENT '城市人口数')
   COMMENT '国家城市表' STORED AS PARQUET;
   
    CREATE TABLE `test_trino.all_countries_rcf`(
     `geonameid` bigint COMMENT '地名ID', 
     `name` string COMMENT '地名', 
     `latitude` double COMMENT '纬度',
     `longitude` double COMMENT '经度', 
     `country_code` string COMMENT '国家编码',
     `population` bigint COMMENT '城市人口数')
   COMMENT '国家城市表' STORED AS RCFILE;
   
    CREATE TABLE `test_trino.all_countries_text`(
     `geonameid` bigint COMMENT '地名ID', 
     `name` string COMMENT '地名', 
     `latitude` double COMMENT '纬度',
     `longitude` double COMMENT '经度', 
     `country_code` string COMMENT '国家编码',
     `population` bigint COMMENT '城市人口数')
   COMMENT '国家城市表' STORED AS TEXTFILE;
    
    DESCRIBE hive.test_trino.all_countries_parquet;
    DESCRIBE hive.test_trino.all_countries_orc;
   

2. 不采用任务压缩算法，将数据写入hive表

   # 不采用压缩,hive客户端会话配置
   SET hive.exec.compress.output=false;
   SET mapreduce.output.fileoutputformat.compress=false;
   
   # 数据复写到内部表all_countries_parquet中
    insert overwrite table `test_trino.all_countries_parquet` select json_tuple(line,'geonameid','name','latitude','longitude','country_code','population')as(geonameid,name,latitude,longitude,country_code,population) from test_trino.tmpjson;
    
   # 数据复写到内部表all_countries_orc中
     insert overwrite table `test_trino.all_countries_orc` select json_tuple(line,'geonameid','name','latitude','longitude','country_code','population')as(geonameid,name,latitude,longitude,country_code,population) from test_trino.tmpjson;
     
   # 数据复写到内部表all_countries_rcf中
     insert overwrite table `test_trino.all_countries_rcf` select json_tuple(line,'geonameid','name','latitude','longitude','country_code','population')as(geonameid,name,latitude,longitude,country_code,population) from test_trino.tmpjson;
     
   # 数据复写到内部表all_countries_text中
     insert overwrite table `test_trino.all_countries_text` select json_tuple(line,'geonameid','name','latitude','longitude','country_code','population')as(geonameid,name,latitude,longitude,country_code,population) from test_trino.tmpjson;
   

查询范围

验证范围

• 单列
  • long型
  • string型
  • double型
• 多列
• 聚合
• 条件聚合

查询语句

select count(1) from hive.test_trino.all_countries_parquet;
select count(1) from hive.test_trino.all_countries_orc;
select count(1) from hive.test_trino.all_countries_rcf;
select count(1) from hive.test_trino.all_countries_text;

select geonameid from hive.test_trino.all_countries_orc limit 10;
select geonameid from hive.test_trino.all_countries_parquet limit 10;
select geonameid from hive.test_trino.all_countries_rcf limit 10;
select geonameid from hive.test_trino.all_countries_text limit 10;

select name from hive.test_trino.all_countries_orc limit 10;
select name from hive.test_trino.all_countries_parquet limit 10;
select name from hive.test_trino.all_countries_rcf limit 10;
select name from hive.test_trino.all_countries_text limit 10;

select latitude from hive.test_trino.all_countries_orc limit 10;
select latitude from hive.test_trino.all_countries_parquet limit 10;
select latitude from hive.test_trino.all_countries_rcf limit 10;
select latitude from hive.test_trino.all_countries_text limit 10;

select geonameid,name from hive.test_trino.all_countries_orc limit 10;
select geonameid,name from hive.test_trino.all_countries_parquet limit 10;
select geonameid,name from hive.test_trino.all_countries_rcf limit 10;
select geonameid,name from hive.test_trino.all_countries_text limit 10;

select country_code,sum(population) from hive.test_trino.all_countries_orc group by country_code;
select country_code,sum(population) from hive.test_trino.all_countries_parquet group by country_code;
select country_code,sum(population) from hive.test_trino.all_countries_rcf group by country_code;
select country_code,sum(population) from hive.test_trino.all_countries_text group by country_code;

select country_code,sum(population) from hive.test_trino.all_countries_orc group by country_code order by country_code;
select country_code,sum(population) from hive.test_trino.all_countries_parquet group by country_code order by country_code;
select country_code,sum(population) from hive.test_trino.all_countries_rcf group by country_code order by country_code;
select country_code,sum(population) from hive.test_trino.all_countries_text group by country_code order by country_code;


select country_code,sum(population) from hive.test_trino.all_countries_orc where country_code='CN' and population>10000 group by country_code;
select country_code,sum(population) from hive.test_trino.all_countries_parquet where country_code='CN' and population>10000 group by country_code;
select country_code,sum(population) from hive.test_trino.all_countries_rcf where country_code='CN' and population>10000 group by country_code;
select country_code,sum(population) from hive.test_trino.all_countries_text where country_code='CN' and population>10000 group by country_code;

结果分析