【kafka KSQL】游戏日志统计分析（1）

以游戏结算日志为例，展示利用KSQL对日志进行统计分析的过程。

启动confluent

cd ~/Documents/install/confluent-5.0.1/

bin/confluent start

查看kafka主题列表

bin/kafka-topics --list --zookeeper localhost:2181

创建接受游戏结算日志的topic

bin/kafka-topics --create --zookeeper localhost:2181 --replication-factor 1 --partitions 4 --topic score-normalized

使用生产者命令行工具往topic中写日志

bin/kafka-console-producer --broker-list localhost:9092 --topic score-normalized

> 

{"cost":7, "epoch":1512342568296,"gameId":"2017-12-04_07:09:28_高手1区_200_015_185175","gameType":"situan","gamers": [{"balance":4405682,"delta":-60,"username":"0791754000"}, {"balance":69532,"delta":-60,"username":"70837999"}, {"balance":972120,"delta":-60,"username":"abc6378303"}, {"balance":23129,"delta":180,"username":"a137671268"}],"reason":"xiayu"}

使用消费者命令行工具查看日志是否正常写入

bin/kafka-console-consumer --bootstrap-server localhost:9092 --topic score-normalized --from-beginning

;; 可以看到

{"cost":7, "epoch":1512342568296,"gameId":"2017-12-04_07:09:28_高手1区_200_015_185175","gameType":"situan","gamers": [{"balance":4405682,"delta":-60,"username":"0791754000"}, {"balance":69532,"delta":-60,"username":"70837999"}, {"balance":972120,"delta":-60,"username":"abc6378303"}, {"balance":23129,"delta":180,"username":"a137671268"}],"reason":"xiayu"}

启动KSQL客户端

bin/ksql http://localhost:8088

可以看到ksql启动后的图标，和操作终端。

ksql终端查看kafka topic列表

ksql> show topics;

打印topic中的消息

PRINT 'score-normalized';

可以看到：

Format:STRING
19-1-5 下午11时59分31秒 , NULL , {"cost":7, "epoch":1512342568296,"gameId":"2017-12-04_07:09:28_\xE9\xAB\x98\xE6\x89\x8B1\xE5\x8C\xBA_200_015_185175","gameType":"situan","gamers": [{"balance":4405682,"delta":-60,"username":"0791754000"}, {"balance":69532,"delta":-60,"username":"70837999"}, {"balance":972120,"delta":-60,"username":"abc6378303"}, {"balance":23129,"delta":180,"username":"a137671268"}],"reason":"xiayu"}

其中：

• 第一个逗号19-1-5 下午11时59分31秒表示消息时间。
• 第二个逗号NULL为消息的Key，因为是从kafka-console-producer推送的，默认为NULL。
• 后面的就是推送过来的消息内容。

从topic score-normalized创建一个Stream

CREATE STREAM SCORE_EVENT \
 (epoch BIGINT, \
  gameType VARCHAR, \
  cost INTEGER, \
  gamers ARRAY< \
              STRUCT< \
                      username VARCHAR, \
                      balance BIGINT, \
                      delta BIGINT \
                      > \
               >, \
  gameId VARCHAR, \
  tax BIGINT, \
  reason VARCHAR) \
  WITH ( KAFKA_TOPIC='score-normalized', \
         VALUE_FORMAT='JSON', \
         TIMESTAMP='epoch');

其中TIMESTAMP='epoch'表示以epoch的时间为事件的时间戳。

删除一个STREAM

DROP  STREAM stream_name ;

如果有查询语句在查询该流，则会出现错误：

Cannot drop USER_SCORE_EVENT. 
The following queries read from this source: []. 
The following queries write into this source: [CSAS_USER_SCORE_EVENT_2, InsertQuery_4, InsertQuery_5, InsertQuery_3]. 
You need to terminate them before dropping USER_SCORE_EVENT.

需要用TERMINATE命令停止这些查询语句，然后再删除流：

TERMINATE CSAS_USER_SCORE_EVENT_2;
TERMINATE InsertQuery_4;

从最早记录开始查询

ksql> SET 'auto.offset.reset' = 'earliest';

从Stream中查询所有数据

ksql> SELECT * FROM SCORE_EVENT;

可以看到：

1546702389664 | null | 1512342568296 | situan | 7 | [{USERNAME=0791754000, BALANCE=4405682, DELTA=-60}, {USERNAME=70837999, BALANCE=69532, DELTA=-60}, {USERNAME=abc6378303, BALANCE=972120, DELTA=-60}, {USERNAME=a137671268, BALANCE=23129, DELTA=180}] | 2017-12-04_07:09:28_高手1区_200_015_185175 | null | xiayu

其中：

• 第1列为记录的时间戳。
• 第2列为记录的key。
• 第3列以后就是消息中的各个字段的值，对应创建流时的顺序。
• 倒数第2列的null，是因为消息中tax字段不存在。

统计2017-12-04日的对局总数

;; 增加一个game_date字段，用于统计
CREATE STREAM SCORE_EVENT_WITH_DATE AS \
    SELECT SUBSTRING(gameId, 0, 10) AS game_date, * \
    FROM SCORE_EVENT;
    
SELECT game_date, COUNT(*) \
    FROM SCORE_EVENT_WITH_DATE \
    WHERE game_date = '2017-12-04' AND reason = 'game' \
    GROUP BY game_date;

目前KSQL还不支持类似下面的查询：

SELECT COUNT(*) \
  FROM SCORE_EVENT \
  WHERE gameId LIKE '2017-12-04_%';

统计参与对局的总玩家数（去重）

因为一条日志中包含多个玩家的对局信息，所以想法把每个玩家拆分成单独的事件

• 整合各个玩家的事件到一个统一的流USER_SCORE_EVENT：

CREATE STREAM USER_SCORE_EVENT AS \
    SELECT epoch, gameType, cost, gameId, tax, reason, gamers[0]->username AS username, gamers[0]->balance AS balance, gamers[0]->delta AS delta \
    FROM SCORE_EVENT;
    
INSERT INTO USER_SCORE_EVENT \
    SELECT epoch, gameType, cost, gameId, tax, reason, gamers[1]->username AS username, gamers[1]->balance AS balance, gamers[1]->delta AS delta \
    FROM SCORE_EVENT;
    
INSERT INTO USER_SCORE_EVENT \
    SELECT epoch, gameType, cost, gameId, tax, reason, gamers[2]->username AS username, gamers[2]->balance AS balance, gamers[2]->delta AS delta \
    FROM SCORE_EVENT;
    
INSERT INTO USER_SCORE_EVENT \
    SELECT epoch, gameType, cost, gameId, tax, reason, gamers[3]->username AS username, gamers[3]->balance AS balance, gamers[3]->delta AS delta \
    FROM SCORE_EVENT;

• 为了后续用于玩家名username的连接JOIN查询，需要重新设置Key：

CREATE STREAM USER_SCORE_EVENT_REKEY AS \ 
SELECT * FROM USER_SCORE_EVENT \
PARTITION BY username;

输出：

ksql> SELECT * FROM USER_SCORE_EVENT_REKEY;


4000 | lzc | 4000 | situan | 7 | 2017-12-04_07:09:28_高手2区_500_015_185175 | null | game | lzc | 972120 | -60
4000 | lzb | 4000 | situan | 7 | 2017-12-04_07:09:28_高手2区_500_015_185175 | null | game | lzb | 69532 | -60

注意：

实践过程中发现：只有对STREAM的field进行PARTITION BY才能生效。

• 统计各个玩家总的对局数、输赢总数、贡献的总税收，并以此创建一个表USER_SCORE_TABLE：

CREATE TABLE USER_SCORE_TABLE AS \
    SELECT username, COUNT(*) AS game_count, SUM(delta) AS delta_sum, SUM(tax) AS tax_sum \
    FROM USER_SCORE_EVENT_REKEY \
    WHERE reason = 'game' \
    GROUP BY username;

查看USER_SCORE_TABLE所有数据：

ksql> SELECT * FROM USER_SCORE_TABLE;
1546709338711 | 70837999 | 70837999 | 4 | -240 | 0
1546709352758 | 0791754000 | 0791754000 | 4 | -240 | 0
1546709338711 | a137671268 | a137671268 | 4 | 720 | 0
1546709352758 | abc6378303 | abc6378303 | 4 | -240 | 0

• 查询某个玩家的对局数、输赢总数、贡献的总税收：

ksql> SELECT * FROM USER_SCORE_TABLE WHERE username = '70837999';

输出：

1546709338711 | 70837999 | 70837999 | 4 | -240 | 0

统计玩家总数（去重）

• 添加一个傀儡列用于统计：

CREATE TABLE USER_SCORE_WITH_TAG AS \
    SELECT 1 AS tag, * FROM USER_SCORE_TABLE;

• 统计去重后的玩家总数

SELECT tag, COUNT(username) \
FROM USER_SCORE_WITH_TAG \
GROUP BY tag;

续

KSQL WINDOW 功能。