pig的一些实例（我常用的语法）

在pig中， dump和store会分别完成两个MR，不会一起进行

1：加载名用正则表达式：

LOAD'/user/wizad/data/wizad/raw/2014-0{6,7-0,7-1,7-2,7-3,8}*/3_1/adwords*'

或者定义引用：%default cleanedLog/user/wizad/data/wizad/cleaned/2014-11-{0[3-9],1[0-8]}/*/part*正确，

而%default cleanedLog/user/wizad/data/wizad/cleaned/2014-11-{0[3-9],[10-18]}/*/part*（这语法居然错了，用hadoop fs -ls/user/wizad/data/wizad/cleaned/2014-11-{0[3-9],[10-18]}/ 发现[10-18]不能使用，是错误的，所以只能用1[0-8]。原因是[]只能在10之内。我试了一年0[10-18]查的是01和08两个文件。而0[100-108] 查的10,11,18三个文件。所以只能在10之内使用。使用时格式为{[10-18]}也是一样的！）

注意：文件名读入不支持所有的正则表达式，是hadoop支持什么云可是用什么。hadoop2.0支持，

[abc]或者[^abc]

[a-z]或者[^a-z]

\c:转移字符表达，\d标示0到9的数字

{ab,cd}

2：filter的几种简单用法：

按值过滤

FILTERclickDate_all BY log_type=='2';

FILTERmapping_table BY mapping_ad_network_id=='3' AND mapping_type=='5';

test=FILTER allRow BY (ad_id=='14997' OR ad_id=='14998' OR ad_id=='14999') ANDlog_type==2;

test=FILTERallRow BY (INDEXOF(ad_id,'14997')==0 OR INDEXOF(ad_id,'14998')==0 OR INDEXOF(ad_id,'14999')==0)AND log_type==2;

配合size函数

FILTERcount_imei BY (SIZE(cimei)>14 AND SIZE(cimei)<17);

2：正则表达式

FILTERcimei2 BY NOT cimei MATCHES '^[0-9]*$';

FILTERcmac2 BY cmac MATCHES'/[A-F\d]{2}:[A-F\d]{2}:[A-F\d]{2}:[A-F\d]{2}:[A-F\d]{2}:[A-F\d]{2}/';

3：排序

ORDER province_count BY $2 DESC;

注意order多个文件，比如hdfs上part00000和part00001，order后只生成一个文件，因为合并成一个文件的操作只能用一个reduce完成，所以结果可能生成很大的文件

4：CONCAT

可用于生成独立的一列，如count了的一个数，前面加一列名称

FOREACHorigin_cleaned_data GENERATE CONCAT('<-_','->') AS cou,guid,log_type;

read_social_14=FOREACH metadata_social_14 GENERATE CONCAT('14','=='),guid_social;

all_id=FOREACH allRow GENERATE id,CONCAT('_','-') as cc;

5：过滤空值，将空值改成取值unknown。

条件表达式“（判断式）？a：b”的应用：直接对列操作

origin_historical= FOREACH origin_cleaned_data GENERATE wizad_ad_id,guid,log_type,

((province_region_id== '') ? 'unknown' : province_region_id)

另外注意：pig判断取值为null，是用is null（is not null）或者== null(!= null)

6:切分成不同子集，按值：

SPLIT geelyTuiGuang INTO android IFos_id==1,ios IF os_id==2;

SPLIT ios INTO ios6 IF(INDEXOF(os_version,'7')!=0),ios7 IF INDEXOF(os_version,'7')==0;

SPLITallCleaned INTO log_42 IF (

((chararray)$34=='1'OR (chararray)$34=='2' OR (chararray)$34=='3' OR (chararray)$34=='1' OR(chararray)$34=='4')

AND

(INDEXOF((chararray)$35,'.')>0)

AND

((chararray)$36=='1'OR (chararray)$36=='')

log_43IF (

((chararray)$34=='1'OR (chararray)$34=='2')

AND

((chararray)$35=='1'OR (chararray)$35=='2' OR (chararray)$35=='3' OR (chararray)$35=='1' OR(chararray)$35=='4')

AND

(INDEXOF((chararray)$36,'.')>0)

);

7：replace函数替换值

FOREACH ios6 GENERATE imei,mac_address ascmac,REPLACE(idfa,'null','');

8：数据流过滤

en_guid =STREAM duimei THROUGH `awk-F"," '{if($3 == "null") print$1","$2","; else print $0}'`;

9：强制转换：

cleaned_data_42=FOREACH log_42 GENERATE

(chararray)$1 AS wizad_ad_id:chararray,

(chararray)$2 AS guid:chararray,

(chararray)$6 AS log_type:chararray,

(chararray)$18AS imei:chararray,

(chararray)$22AS idfa:chararray,

(chararray)$23AS mac_address:chararray

10内置函数REGEX_EXTRACT，使用正则表达式：

allAdId=FOREACH allRow GENERATE REGEX_EXTRACT((chararray)$3,'(.*) (.*)',1) AStime,REGEX_EXTRACT((chararray)$0,'(.*)_(.*)',1) AS adn,$6 AS ad_id;

allAdId=FOREACH allRow GENERATE REGEX_EXTRACT(create_time,'(.*) (.*)',1) AStime,ad_id;

11．SUBSTRING(aa,0,n)提取0到n-1个字符：

split jn_data into same_prov if(SUBSTRING(province,0,2) == SUBSTRING(province_ad,0,2)), diff_prov if(SUBSTRING(province,0,2)

!= SUBSTRING(province_ad,0,2));

时间类型提取分钟，做计算

log_data= foreach click_log generate log_type,guid,ip,SUBSTRING(create_time,0,13) astime,SUBSTRING(create_time,14,16) as minute2,os_id,os_version,device_type;

12，ABS时间相差5分钟计算：

minute_compare= foreach join_data generatelog_type,cookie_id,guid,(int)minute1,(int)minute2,time_extract::os_version,log_data::os_version;

same_users= filter minute_compare by (ABS(minute1-minute2) <= 5);

13，统计个数

grp_diff_city= group diff_city all;

count_diff_city= foreach grp_diff_city generate COUNT_STAR($1);

dump count_same_city;

14，join by多个列（字段）

join_data= join time_extract by (ip,time,os_id), log_data by (ip,time,os_id);

从左向右依次比较