hive日志分析实战（二）

 需求

统计某游戏平台新用户渠道来源

日志格式如下：

 

Jul 23 0:00:47  [info] {SPR}gjzq{SPR}20130723000047{SPR}85493108{SPR}S1{SPR}{SPR}360wan-2j-reg{SPR}58.240.209.78{SPR}

 

 分析

问题的关键在于先找出新用户

新用户：仅在7月份登陆过平台的用户为新用户

依据map/reduce思想，可以按照如下方式找出新用户：

• 假如某用户在某月份出现过，则（qid，year，month）=1
• 按qid汇总该用户出现过的月数，即构建(qid,count(year,month))对
• 新用户的count（year，month）=1，且(year,month)=(2013,07).

找出新用户的来源渠道

来源渠道：新用户在201307可能多次登录平台，需要找出最早登陆平台所属渠道

分两步来做：

• 找出新用户所有登陆记录（qid,logintime,src）
• 针对同一qid找出logintime最小时的src

实现

1. 数据准备

    1）建表

  

create table if not exists glogin_daily (year int,month int,day int,hour int,logintime string,qid int,gkey string,skey string,loginip string,registsrc string,loginfrom string) partitioned by (dt string);

     依据日志内容及所关心的信息创建表格，按天组织分区

   2 ) 数据导入

    因日志文件存在于多处，我们先将日志汇总到一临时目录，创建临时外部表将数据加载进hive，然后通过正则匹配的方式分隔出各字段。（内部表只能load单文件，通过这种方式可以load文件夹）

   

echo "==== load data into tmp table $TMP_TABLE ==="
    ${SOFTWARE_BASE}/hive/bin/hive -e "create external table $TMP_TABLE (info string) location '${TMP_DIR}';"
    echo "==== M/R ==="
    CURR_YEAR=`echo $CURR_DOING|cut -b 1-4`
    CURR_MONTH=`echo $CURR_DOING|cut -b 5-6`
    CURR_DAY=`echo $CURR_DOING|cut -b 7-8`
    dt="${CURR_YEAR}-${CURR_MONTH}-${CURR_DAY}"
    ${SOFTWARE_BASE}/hive/bin/hive -e "add file ${SCRIPT_PATH}/${MAP_SCRIPT_FILE};set hive.exec.dynamic.partition=true;insert overwrite table glogin_daily partition (dt='${dt}') select transform (t.i) using '$MAP_SCRIPT_PARSER ./${MAP_SCRIPT_FILE}' as (y,m,d,h,t,q,g,s,ip,src,f) from (select info as i from ${TMP_TABLE}) t;"

  

其中filter_login.php:

$fr=fopen("php://stdin","r");
$month_dict = array(
    'Jan' => 1,
    'Feb' => 2,
    'Mar' => 3,
    'Apr' => 4,
    'May' => 5,
    'Jun' => 6,
    'Jul' => 7,
    'Aug' => 8,
    'Sep' => 9,
    'Oct' => 10,
    'Nov' => 11,
    'Dec' => 12,
);
while(!feof($fr))
{
    $input = fgets($fr,256);
    $input = rtrim($input);
    //Jul 23 0:00:00  [info] {SPR}xxj{SPR}20130723000000{SPR}245396389{SPR}S9{SPR}iwan-ng-mnsg{SPR}cl-reg-xxj0if{SPR}221.5.67.136{SPR}
    if(preg_match("/([^ ]+) +(\d+) (\d+):.*\{SPR\}([^\{]*)\{SPR\}(\d+)\{SPR\}(\d+)\{SPR\}([^\{]*)\{SPR\}([^\{]*)\{SPR\}(([^\{]*)\{SPR\}([^\{]*)\{SPR\})?/",$input,$matches))
    {
       $year = substr($matches[5],0,4);
       echo $year."\t".$month_dict[$matches[1]]."\t".$matches[2]."\t".$matches[3]."\t".$matches[5]."\t".$matches[6]."\t".$matches[4]."\t".$matches[7]."\t".$matches[11]."\t".$matches[8]."\t".$matches[10]."\n";
    }
}
fclose ($fr);

 

2.找出新用户

 

1)用户登陆平台记录按月消重汇总

 

create table distinct_login_monthly_tmp_07 as select qid,year,month from glogin_daily group by qid,year,month;

 
2)用户登陆平台月数

 

create table login_stat_monthly_tmp_07 as select qid,count(1) as c from distinct_login_monthly_tmp_07 where year<2013 or (year=2013 and month<=7) group by qid; 

 
平台级新用户：
1)找出登陆月数为1的用户;

2.判断这些用户是否在7月份出现，如果有出现，找出登陆所有src

 

create table new_player_monthly_07 as select distinct a.qid,b.src,b.logintime from (select qid from login_stat_monthly_tmp_07 where c=1) a join (select qid,loginfrom as src,logintime from glogin_daily where month=7 and year=2013) b on a.qid=b.qid;

 
找出最早登陆的src:

 

add file /home/game/lvbenwei/load_login/get_player_src.php;
create table new_player_src_07 as select transform (t.qid,t.src,t.logintime) using 'php ./get_player_src.php' as (qid,src,logintime) from (select * from new_player_monthly_07 order by qid,logintime) t;

 

其中get_player_src.php:

$fr=fopen("php://stdin","r");
$curr_qid = null;
$curr_src = null;
$curr_logintime=null;
while(!feof($fr))
{
	$input = fgets($fr,1024);
	$input = rtrim($input);
	$arr   = explode("\t", $input);
	$qid   = trim($arr[0]);
        if(empty($curr_qid)||$curr_qid != $qid)
        {
                $curr_qid = $qid;
                echo $input."\n";
        }
}
fclose ($fr);

 
平台级新用户数：

select count(*) from new_player_src_07;

 
平台级各渠道新用户汇总：

create table new_player_src_stat_07 as select src,count(*) from new_player_monthly_07 group by src;