Skip to content

Latest commit

 

History

History
353 lines (280 loc) · 20.1 KB

README.md

File metadata and controls

353 lines (280 loc) · 20.1 KB

EN

1 项目介绍

基于bahir-flink二次开发,相对bahir调整的内容有:

1.使用Lettuce替换Jedis,同步读写改为异步读写,大幅度提升了性能 
2.增加了Table/SQL API,增加select/维表join查询支持
3.增加关联查询缓存(支持增量与全量)
4.增加支持整行保存功能,用于多字段的维表关联查询
5.增加限流功能,用于Flink SQL在线调试功能
6.增加支持Flink高版本(包括1.12,1.13,1.14+)
7.统一过期策略等
8.支持flink cdc删除及其它RowKind.DELETE
9.支持select查询

因bahir使用的flink接口版本较老,所以改动较大,开发过程中参考了腾讯云与阿里云两家产商的流计算产品,取两家之长,并增加了更丰富的功能。

注:redis不支持两段提交无法实现刚好一次语义。

2 使用方法:

2.1 工程直接引用

项目依赖Lettuce(6.2.1)及netty-transport-native-epoll(4.1.82.Final),如flink环境有这两个包,则使用flink-connector-redis-1.4.3.jar, 否则使用flink-connector-redis-1.4.3-jar-with-dependencies.jar。

<dependency>
    <groupId>io.github.jeff-zou</groupId>
    <artifactId>flink-connector-redis</artifactId>
    <!-- 没有单独引入项目依赖Lettuce netty-transport-native-epoll依赖时 -->
    <!--            <classifier>jar-with-dependencies</classifier>-->
    <version>1.4.3</version>
</dependency>

2.2 自行打包

打包命令: mvn package,将生成的包放入flink lib中即可,无需其它设置。

2.3 使用示例

-- 创建redis表示例
create table redis_table (name varchar, age int) 
  with ('connector'='redis', 'host'='10.11.69.176', 'port'='6379','password'='test123', 
  'redis-mode'='single','command'='set');
-- 写入  
  insert into redis_table select * from (values('test', 1));

-- 查询  
  insert into redis_table select name,age + 1 from redis_table /*+ options('scan.key'='test') */
  
create table gen_table (age int , level int, proctime as procTime()) with ('connector'='datagen','fields.age.kind' = 'sequence',
 'fields.age.start' = '2','fields.age.end' = '2','fields.level.kind' = 'sequence','fields.level.start' = '10','fields.level.end' = '10'); 

-- 关联查询 
insert into redis_table select 'test', j.age + 10 from gen_table s left join redis_table  for system_time as of proctime as j
on j.name = 'test'

3 参数说明:

3.1 主要参数:

字段 默认值 类型 说明
connector (none) String redis
host (none) String Redis IP
port 6379 Integer Redis 端口
password null String 如果没有设置,则为 null
database 0 Integer 默认使用 db0
timeout 2000 Integer 连接超时时间,单位 ms,默认 1s
cluster-nodes (none) String 集群ip与端口,当redis-mode为cluster时不为空,如:10.11.80.147:7000,10.11.80.147:7001,10.11.80.147:8000
command (none) String 对应上文中的redis命令
redis-mode (none) Integer mode类型: single cluster sentinel
lookup.cache.max-rows -1 Integer 查询缓存大小,减少对redis重复key的查询
lookup.cache.ttl -1 Integer 查询缓存过期时间,单位为秒, 开启查询缓存条件是max-rows与ttl都不能为-1
lookup.cache.load-all false Boolean 开启全量缓存,当命令为hget时,将从redis map查询出所有元素并保存到cache中,用于解决缓存穿透问题
max.retries 1 Integer 写入/查询失败重试次数
value.data.structure column String column: value值来自某一字段 (如, set: key值取自DDL定义的第一个字段, value值取自第二个字段)
row: 将整行内容保存至value并以'\01'分割
set.if.absent false Boolean 在key不存在时才写入,只对set hset有效
io.pool.size (none) Integer Lettuce内netty的io线程池大小,默认情况下该值为当前JVM可用线程数,并且大于2
event.pool.size (none) Integer Lettuce内netty的event线程池大小 ,默认情况下该值为当前JVM可用线程数,并且大于2
scan.key (none) String 查询时redis key
scan.addition.key (none) String 查询时限定redis key,如map结构时的hashfield
scan.range.start (none) Integer 查询list结构时指定lrange start
scan.range.stop (none) Integer 查询list结构时指定lrange start
scan.count (none) Integer 查询set结构时指定srandmember count
zset.zremrangeby (none) String 执行zadd之后,是否执行zremrangeby,取值:SCORE、LEX、RANK

3.1.1 command值与redis命令对应关系:

command值 写入 查询 维表关联 删除(Flink CDC等产生的RowKind.delete)
set set get get del
hset hset hget hget hdel
get set get get del
hset hset hget hget hdel
rpush rpush lrange
lpush lpush lrange
incrBy incrByFloat incrBy incrByFloat get get 写入相对值,如:incrby 2 -> incryby -2
hincrBy hincryByFloat hincrBy hincryByFloat hget hget 写入相对值,如:hincrby 2 -> hincryby -2
zincrby zincrby zscore zscore 写入相对值,如:zincrby 2 -> zincryby -2
sadd sadd srandmember 10 srem
zadd zadd zscore zscore zrem
pfadd(hyperloglog) pfadd(hyperloglog)
publish publish
zrem zrem zscore zscore
srem srem srandmember 10
del del get get
hdel hdel hget hget
decrBy decrBy get get

注:为空表示不支持

3.1.2 value.data.structure = column(默认)

无需通过primary key来映射redis中的Key,直接由ddl中的字段顺序来决定Key,如:

create table sink_redis(username VARCHAR, passport VARCHAR)  with ('command'='set') 
其中username为key, passport为value.

create table sink_redis(name VARCHAR, subject VARCHAR, score VARCHAR)  with ('command'='hset') 
其中name为map结构的key, subject为field, score为value.

3.1.3 value.data.structure = row

整行内容保存至value并以'\01'分割

create table sink_redis(username VARCHAR, passport VARCHAR)  with ('command'='set') 
其中username为key, username\01passport为value.

create table sink_redis(name VARCHAR, subject VARCHAR, score VARCHAR)  with ('command'='hset') 
其中name为map结构的key, subject为field, name\01subject\01score为value.

3.2 sink时ttl相关参数

Field Default Type Description
ttl (none) Integer key过期时间(秒),每次sink时会设置ttl
ttl.on.time (none) String key的过期时间点,格式为LocalTime.toString(), eg: 10:00 12:12:01,当ttl未配置时才生效
ttl.key.not.absent false boolean 与ttl一起使用,当key不存在时才设置ttl

3.3 在线调试SQL时,用于限制sink资源使用的参数:

Field Default Type Description
sink.limit false Boolean 是否打开限制
sink.limit.max-num 10000 Integer taskmanager内每个slot可以写的最大数据量
sink.limit.interval 100 String taskmanager内每个slot写入数据间隔 milliseconds
sink.limit.max-online 30 * 60 * 1000L Long taskmanager内每个slot最大在线时间, milliseconds

3.4 集群类型为sentinel时额外连接参数:

字段 默认值 类型 说明
master.name (none) String 主名
sentinels.info (none) String 如:10.11.80.147:7000,10.11.80.147:7001,10.11.80.147:8000
sentinels.password (none) String sentinel进程密码

4 数据类型转换

flink type redis row converter
CHAR String
VARCHAR String
String String
BOOLEAN String String.valueOf(boolean val)
boolean Boolean.valueOf(String str)
BINARY String Base64.getEncoder().encodeToString
byte[] Base64.getDecoder().decode(String str)
VARBINARY String Base64.getEncoder().encodeToString
byte[] Base64.getDecoder().decode(String str)
DECIMAL String BigDecimal.toString
DecimalData DecimalData.fromBigDecimal(new BigDecimal(String str),int precision, int scale)
TINYINT String String.valueOf(byte val)
byte Byte.valueOf(String str)
SMALLINT String String.valueOf(short val)
short Short.valueOf(String str)
INTEGER String String.valueOf(int val)
int Integer.valueOf(String str)
DATE String the day from epoch as int
date show as 2022-01-01
TIME String the millisecond from 0'clock as int
time show as 04:04:01.023
BIGINT String String.valueOf(long val)
long Long.valueOf(String str)
FLOAT String String.valueOf(float val)
float Float.valueOf(String str)
DOUBLE String String.valueOf(double val)
double Double.valueOf(String str)
TIMESTAMP String the millisecond from epoch as long
timestamp TimeStampData.fromEpochMillis(Long.valueOf(String str))

5 使用示例:

  • 5.1 维表查询:

create table sink_redis(name varchar, level varchar, age varchar) with ( 'connector'='redis', 'host'='10.11.80.147','port'='7001', 'redis-mode'='single','password'='******','command'='hset');

-- 先在redis中插入数据,相当于redis命令: hset 3 3 100 --
insert into sink_redis select * from (values ('3', '3', '100'));
                
create table dim_table (name varchar, level varchar, age varchar) with ('connector'='redis', 'host'='10.11.80.147','port'='7001', 'redis-mode'='single', 'password'='*****','command'='hget', 'maxIdle'='2', 'minIdle'='1', 'lookup.cache.max-rows'='10', 'lookup.cache.ttl'='10', 'max-retries'='3');
    
-- 随机生成10以内的数据作为数据源 --
-- 其中有一条数据会是: username = 3  level = 3, 会跟上面插入的数据关联 -- 
create table source_table (username varchar, level varchar, proctime as procTime()) with ('connector'='datagen',  'rows-per-second'='1',  'fields.username.kind'='sequence',  'fields.username.start'='1',  'fields.username.end'='10', 'fields.level.kind'='sequence',  'fields.level.start'='1',  'fields.level.end'='10');

create table sink_table(username varchar, level varchar,age varchar) with ('connector'='print');

insert into
	sink_table
select
	s.username,
	s.level,
	d.age
from
	source_table s
left join dim_table for system_time as of s.proctime as d on
	d.name = s.username
	and d.level = s.level;
-- username为3那一行会关联到redis内的值,输出为: 3,3,100	
  • 5.2 多字段的维表关联查询

很多情况维表有多个字段,本实例展示如何利用'value.data.structure'='row'写多字段并关联查询。

-- 创建表
create table sink_redis(uid VARCHAR,score double,score2 double )
with ( 'connector' = 'redis',
			'host' = '10.11.69.176',
			'port' = '6379',
			'redis-mode' = 'single',
			'password' = '****',
			'command' = 'SET',
			'value.data.structure' = 'row');  -- 'value.data.structure'='row':整行内容保存至value并以'\01'分割
-- 写入测试数据,score、score2为需要被关联查询出的两个维度
insert into sink_redis select * from (values ('1', 10.3, 10.1));

-- 在redis中,value的值为: "1\x0110.3\x0110.1" --
-- 写入结束 --

-- create join table --
create table join_table with ('command'='get', 'value.data.structure'='row') like sink_redis

-- create result table --
create table result_table(uid VARCHAR, username VARCHAR, score double, score2 double) with ('connector'='print')

-- create source table --
create table source_table(uid VARCHAR, username VARCHAR, proc_time as procTime()) with ('connector'='datagen', 'fields.uid.kind'='sequence', 'fields.uid.start'='1', 'fields.uid.end'='2')

-- 关联查询维表,获得维表的多个字段值 --
insert
	into
	result_table
select
	s.uid,
	s.username,
	j.score, -- 来自维表
	j.score2 -- 来自维表
from
	source_table as s
join join_table for system_time as of s.proc_time as j on
	j.uid = s.uid
	
result:
2> +I[2, 1e0fe885a2990edd7f13dd0b81f923713182d5c559b21eff6bda3960cba8df27c69a3c0f26466efaface8976a2e16d9f68b3, null, null]
1> +I[1, 30182e00eca2bff6e00a2d5331e8857a087792918c4379155b635a3cf42a53a1b8f3be7feb00b0c63c556641423be5537476, 10.3, 10.1]
  • 5.3 DataStream查询方式

    示例代码路径: src/test/java/org.apache.flink.streaming.connectors.redis.datastream.DataStreamTest.java
    hset示例,相当于redis命令:hset tom math 150

        Configuration configuration = new Configuration();
        configuration.setString(REDIS_MODE, REDIS_SINGLE);
        configuration.setString(REDIS_COMMAND, RedisCommand.HSET.name());
        configuration.setInteger(TTL, 10);

        RedisSinkMapper redisMapper = new RowRedisSinkMapper(RedisCommand.HSET, configuration);

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        BinaryRowData binaryRowData = new BinaryRowData(3);
        BinaryRowWriter binaryRowWriter = new BinaryRowWriter(binaryRowData);
        binaryRowWriter.writeString(0, StringData.fromString("tom"));
        binaryRowWriter.writeString(1, StringData.fromString("math"));
        binaryRowWriter.writeString(2, StringData.fromString("152"));

        DataStream<BinaryRowData> dataStream = env.fromElements(binaryRowData, binaryRowData);

        List<String> columnNames = Arrays.asList("name", "subject", "scope");
        List<DataType> columnDataTypes =
                Arrays.asList(DataTypes.STRING(), DataTypes.STRING(), DataTypes.STRING());
        ResolvedSchema resolvedSchema = ResolvedSchema.physical(columnNames, columnDataTypes);

        FlinkConfigBase conf =
                new FlinkSingleConfig.Builder()
                        .setHost(REDIS_HOST)
                        .setPort(REDIS_PORT)
                        .setPassword(REDIS_PASSWORD)
                        .build();

        RedisSinkFunction redisSinkFunction =
                new RedisSinkFunction<>(conf, redisMapper, resolvedSchema, configuration);

        dataStream.addSink(redisSinkFunction).setParallelism(1);
        env.execute("RedisSinkTest");
  • 5.4 redis-cluster写入示例

    示例代码路径: src/test/java/org.apache.flink.streaming.connectors.redis.table.SQLInsertTest.java
    set示例,相当于redis命令: set test test11

StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
EnvironmentSettings environmentSettings = EnvironmentSettings.newInstance().useBlinkPlanner().inStreamingMode().build();
StreamTableEnvironment tEnv = StreamTableEnvironment.create(env, environmentSettings);

String ddl = "create table sink_redis(username VARCHAR, passport VARCHAR) with ( 'connector'='redis', " +
              "'cluster-nodes'='10.11.80.147:7000,10.11.80.147:7001','redis- mode'='cluster','password'='******','command'='set')" ;

tEnv.executeSql(ddl);
String sql = " insert into sink_redis select * from (values ('test', 'test11'))";
TableResult tableResult = tEnv.executeSql(sql);
tableResult.getJobClient().get()
.getJobExecutionResult()
.get();

6 解决问题联系我

img.png

7 开发环境

ide: IntelliJ IDEA

code format: google-java-format + Save Actions

flink 1.12/1.13/1.14+

jdk1.8 Lettuce 6.2.1

8 贡献

Pull Request需要提交至dev分支
提交前请使用mvn spotless:apply进行代码格式化,然后使用maven package打包确认所有测试用例能通过。

9 flink 1.12支持

请切换到分支flink-1.12(注:1.12使用jedis)

<dependency>
    <groupId>io.github.jeff-zou</groupId>
    <artifactId>flink-connector-redis</artifactId>
    <version>1.1.1-1.12</version>
</dependency>