业务实现之编写写入DIM层业务代码
一、代码逻辑和架构图
编写代码读取Kafka “KAFKA-DIM-TOPIC” topic维度数据通过Phoenix写入到HBase中，我们可以通过topic中每条数据获取该条数据对应的phoenix表名及字段名动态创建phoenix表以及插入数据，这里所有在mysql“lakehousedb.dim_tbl_config_info”中配置的维度表都会动态的写入到HBase中。这里使用Flink处理对应topic数据时如果维度数据需要清洗还可以进行清洗
二、代码编写
读取Kafka 维度数据写入HBase代码为“DimDataToHBase.scala”，主要代码逻辑如下：
代码语言：javascript复制object DimDataToHBase {
private val consumeKafkaFromEarliest: Boolean = ConfigUtil.CONSUME_KAFKA_FORMEARLIEST
private val kafkaBrokers: String = ConfigUtil.KAFKA_BROKERS
private val kafakDimTopic: String = ConfigUtil.KAFKA_DIM_TOPIC
private val phoenixURL: String = ConfigUtil.PHOENIX_URL
var ds: DataStream[String] = _
def main(args: Array[String]): Unit = {
val env: StreamExecutionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment
//1.导入隐式转换
import org.apache.flink.streaming.api.scala._
//2.设置Kafka配置
val props = new Properties()
props.setProperty("bootstrap.servers",kafkaBrokers)
props.setProperty("key.deserializer",classOf[StringDeserializer].getName)
props.setProperty("value.deserializer",classOf[StringDeserializer].getName)
props.setProperty("group.id","mygroup.id")
//3.从数据中获取Kafka DIM层 KAFKA-DIM-TOPIC 数据
/**
* 数据样例：
* {
* "gmt_create": "1646037374201",
* "commit": "true",
* "tbl_name": "mc_member_info",
* "type": "insert",
* "gmt_modified": "1646037374201",
* "member_level": "3",
* "database": "lakehousedb",
* "xid": "38450",
* "pk_col": "id",
* "balance": "10482",
* "user_id": "0uid9060",
* "phoenix_tbl_name": "DIM_MEMBER_INFO",
* "tbl_db": "lakehousedb",
* "member_points": "7568",
* "id": "10014",
* "cols": "user_id,member_growth_score,member_level,member_points,balance,gmt_create,gmt_modified",
* "table": "mc_member_info",
* "member_growth_score": "3028",
* "ts": "1646901373"
* }
*
*/
if(consumeKafkaFromEarliest){
ds = env.addSource(MyKafkaUtil.GetDataFromKafka(kafakDimTopic,props).setStartFromEarliest())
}else{
ds = env.addSource(MyKafkaUtil.GetDataFromKafka(kafakDimTopic,props))
}
ds.keyBy(line=>{
JSON.parseObject(line).getString("phoenix_tbl_name")
}).process(new KeyedProcessFunction[String,String,String] {
//设置状态，存储每个Phoenix表是否被创建
lazy private val valueState: ValueState[String] = getRuntimeContext.getState(new ValueStateDescriptor[String]("valueState",classOf[String]))
var conn: Connection = _
var pst: PreparedStatement = _
//在open方法中，设置连接Phoenix ，方便后期创建对应的phoenix表
override def open(parameters: Configuration): Unit = {
println("创建Phoenix 连接... ...")
conn = DriverManager.getConnection(phoenixURL)
}
override def processElement(jsonStr: String, ctx: KeyedProcessFunction[String, String, String]#Context, out: Collector[String]): Unit = {
val nObject: JSONObject = JSON.parseObject(jsonStr)
//从json 对象中获取对应 hbase 表名、主键、列信息
val operateType: String = nObject.getString("type")
val phoenixTblName: String = nObject.getString("phoenix_tbl_name")
val pkCol: String = nObject.getString("pk_col")
val cols: String = nObject.getString("cols")
//判断操作类型，这里只会向HBase中存入增加、修改的数据，删除等其他操作不考虑
//operateType.equals("bootstrap-insert") 这种情况主要是使用maxwell 直接批量同步维度数据时，操作类型为bootstrap-insert
if(operateType.equals("insert")||operateType.equals("update")||operateType.equals("bootstrap-insert")){
//判断状态中是否有当前表状态，如果有说明已经被创建，没有就组织建表语句，通过phoenix创建维度表
if(valueState.value() ==null){
createPhoenixTable(phoenixTblName, pkCol, cols)
//更新状态
valueState.update(phoenixTblName)
}
//向phoenix表中插入数据，同时方法中涉及数据清洗
upsertIntoPhoenixTable(nObject, phoenixTblName, pkCol, cols)
/**
* 当有维度数据更新时，那么将Redis中维度表缓存删除
* Redis中 key 为：维度表-主键值
*/
if(operateType.equals("update")){
//获取当前更新数据中主键对应的值
val pkValue: String = nObject.getJSONObject("data").getString(pkCol)
//组织Redis中的key
val key = phoenixTblName "-" pkValue
//删除Redis中缓存的此key对应数据，没有此key也不会报错
MyRedisUtil.deleteKey(key)
}
out.collect("执行成功")
}
}
private def upsertIntoPhoenixTable(nObject: JSONObject, phoenixTblName: String, pkCol: String, cols: String): Unit = {
//获取向phoenix中插入数据所有列
val colsList: ListBuffer[String] = MyStringUtil.getAllCols(cols)
//获取主键对应的值
val pkValue: String = nObject.getString(pkCol)
//组织向表中插入数据的语句
//upsert into test values ('1','zs',18);
val upsertSQL = new StringBuffer(s"upsert into ${phoenixTblName} values ('${pkValue}'")
for (col <- colsList) {
val currentColValue: String = nObject.getString(col)
println("colsList = " colsList.toString " - current col = " currentColValue)
//将列数据中的 “'”符号进行转义
upsertSQL.append(s",'${currentColValue.replace("'","\'")}'")
}
upsertSQL.append(s")")
//向表中Phoenix中插入数据
println("phoenix 插入Sql = " upsertSQL.toString)
pst = conn.prepareStatement(upsertSQL.toString)
pst.execute()
//这里如果想要批量提交，可以设置状态，当每个key 满足1000条时，commit一次，
// 另外定义定时器，每隔2分钟自动提交一次，防止有些数据没有达到2000条时没有存入Phoenix
conn.commit()
}
private def createPhoenixTable(phoenixTblName: String, pkCol: String, cols: String): Boolean = {
//获取所有列
val colsList: ListBuffer[String] = MyStringUtil.getAllCols(cols)
//组织phoenix建表语句,为了后期操作方便，这里建表语句所有列族指定为“cf",所有字段都为varchar
//create table xxx (id varchar primary key ,cf.name varchar,cf.age varchar);
val createSql = new StringBuffer(s"create table if not exists ${phoenixTblName} (${pkCol} varchar primary key,")
for (col <- colsList) {
createSql.append(s"cf.${col.replace("'","\'")} varchar,")//处理数据中带 ' 的数据
}
//将最后一个逗号替换成“) column_encoded_bytes=0” ，最后这个参数是不让phoenix对数据进行16进制编码
createSql.replace(createSql.length() - 1, createSql.length(), ") column_encoded_bytes=0")
println(s"拼接Phoenix SQL 为 = ${createSql}")
//执行sql
pst = conn.prepareStatement(createSql.toString)
pst.execute()
}
//关闭连接
override def close(): Unit = {
pst.close()
conn.close()
}
}).print()
env.execute()
}
}三、代码测试
执行代码之前首先需要启动HDFS、HBase，代码中设置读取Kafka数据从头开始读取，然后执行代码，代码执行完成后可以进入phoenix中查看对应的结果
代码语言：javascript复制# 在node4节点上启动phoenix
[root@node4 ~]# cd /software/apache-phoenix-5.0.0-HBase-2.0-bin/bin
[root@node4 bin]# ./sqlline.py 
