Flink-读Kafka写Hive表

1. 目标

使用Flink读取Kafka数据并实时写入Hive表。

2. 环境配置

EMR环境：Hadoop 3.3.3, Hive 3.1.3, Flink 1.16.0

 

根据官网描述：

https://nightlies.apache.org/flink/flink-docs-release-1.16/docs/connectors/table/hive/overview/

当前Flink 1.16.0 支持Hive 3.1.3版本，如果是开发，则需要加入依赖有：

<dependency>
    <groupId>org.apache.flink</groupId>
    <artifactId>flink-connector-hive_2.12</artifactId>
    <version>1.16.0</version>
    <scope>provided</scope>
</dependency>

// Hive dependencies
<dependency>
    <groupId>org.apache.hive</groupId>
    <artifactId>hive-exec</artifactId>
    <version>3.1.3</version>
</dependency>

3. hive表

在读写hive表时，预先条件是注册hive catalog：

// set hive dialect
tableEnv.getConfig().setSqlDialect(SqlDialect.HIVE)

// set hive catalog
tableEnv.executeSql("CREATE CATALOG myhive WITH (" +
  "'type' = 'hive'," +
  "'default-database' = 'default'," +
  "'hive-conf-dir' = 'hiveconf'" +
  ")")

tableEnv.executeSql("use catalog myhive")

然后创建hive表：

// hive table
tableEnv.executeSql("CREATE TABLE IF NOT EXISTS hive_table (" +
  "id string," +
  "`value` float," +
  "hashdata string," +
  "num integer," +
  "token string," +
  "info string," +
  "ts timestamp " +
  ") " +
  "PARTITIONED BY (dt string, hr string) STORED AS ORC TBLPROPERTIES (" +
 // "'path'='hive-output'," +
  "'partition.time-extractor.timestamp-pattern'='$dt $hr:00:00'," +
  "'sink.partition-commit.policy.kind'='metastore,success-file'," +
  "'sink.partition-commit.trigger'='partition-time'," +
  "'sink.partition-commit.delay'='0 s'" +
  " )")

4. 消费Kafka并写入Hive表

参考官方文档：

https://nightlies.apache.org/flink/flink-docs-release-1.16/docs/connectors/datastream/kafka/

添加对应依赖：

<!-- https://mvnrepository.com/artifact/org.apache.flink/flink-connector-kafka -->
<dependency>
    <groupId>org.apache.flink</groupId>
    <artifactId>flink-connector-kafka</artifactId>
    <version>${flink.version}</version>
</dependency>

flinksql参考代码：

package com.tang.hive

import org.apache.flink.api.java.utils.ParameterTool
import org.apache.flink.streaming.api.scala.StreamExecutionEnvironment
import org.apache.flink.table.api.SqlDialect
import org.apache.flink.table.api.bridge.scala.StreamTableEnvironment

object Kafka2Hive {

  /***
   *  create hive table
   * @param tbl_env
   * @param drop
   * @param hiveConfDir
   * @param database
   * @return
   */
  def buildHiveTable(tbl_env: StreamTableEnvironment,
                     drop: Boolean,
                     hiveConfDir: String,
                     database: String,
                     tableName: String,
                     dbLocation: String) = {
    // set hive dialect
    tbl_env.getConfig().setSqlDialect(SqlDialect.HIVE)

    // set hive catalog
    tbl_env.executeSql("CREATE CATALOG myhive WITH (" +
      "'type' = 'hive'," +
      "'default-database' = '"+ database + "'," +
      "'hive-conf-dir' = '" + hiveConfDir + "'" +
      ")")
    tbl_env.executeSql("use catalog myhive")

    // whether drop hive table first
    if (drop) {
      // drop first
      tbl_env.executeSql("drop table if exists" + tableName)
    }

    val sql = "CREATE TABLE IF NOT EXISTS " + tableName + "(" +
      "id string," +
      "`value` float," +
      "hashdata string," +
      "num integer," +
      "token string," +
      "info string," +
      "ts timestamp " +
      ") " +
      "PARTITIONED BY (dt string, hr string) STORED AS ORC " +
      "LOCATION '" + dbLocation + "/" + tableName +"' TBLPROPERTIES (" +
      "'partition.time-extractor.timestamp-pattern'='$dt $hr:00:00'," +
      "'sink.partition-commit.policy.kind'='metastore,success-file'," +
      "'sink.partition-commit.trigger'='partition-time'," +
      "'sink.partition-commit.watermark-time-zone'='Asia/Shanghai'," +
      "'sink.partition-commit.delay'='0 s'," +
      "'auto-compaction'='true'" +
      " )"

    // hive table
    tbl_env.executeSql(sql)
  }

  /***
   *  create kafka table
   * @param tbl_env
   * @param drop
   * @param bootstrapServers
   * @param topic
   * @param groupId
   * @return
   */
  def buildKafkaTable(tbl_env: StreamTableEnvironment,
                      drop: Boolean,
                      bootstrapServers: String,
                      topic: String,
                      groupId: String,
                      tableName: String) = {
    // set to default dialect
    tbl_env.getConfig.setSqlDialect(SqlDialect.DEFAULT)

    if (drop) {
      tbl_env.executeSql("drop table if exists " + tableName)
    }

    // kafka table
    tbl_env.executeSql("CREATE TABLE IF NOT EXISTS "+ tableName + " (" +
      "id string," +
      "`value` float," +
      "hashdata string," +
      "num integer," +
      "token string," +
      "info string," +
      "created_timestamp bigint," +
      "ts AS TO_TIMESTAMP( FROM_UNIXTIME(created_timestamp) ), " +
      "WATERMARK FOR ts AS ts - INTERVAL '5' SECOND "+
      " )" +
      "with (" +
      " 'connector' = 'kafka'," +
      " 'topic' = '" + topic + "'," +
      " 'properties.bootstrap.servers' = '" + bootstrapServers +"'," +
      " 'properties.group.id' = '" + groupId + "'," +
      " 'scan.startup.mode' = 'latest-offset'," +
      " 'format' = 'json'," +
      " 'json.fail-on-missing-field' = 'false'," +
      " 'json.ignore-parse-errors' = 'true'" +
      ")" )

  }

    def main(args: Array[String]): Unit = {
      val senv = StreamExecutionEnvironment.getExecutionEnvironment
      val tableEnv = StreamTableEnvironment.create(senv)

      // set checkpoint
      // senv.enableCheckpointing(60000);
      //senv.getCheckpointConfig.setCheckpointStorage("file://flink-hive-chk");

      // get parameter
      val tool: ParameterTool = ParameterTool.fromArgs(args)
      val hiveConfDir = tool.get("hive.conf.dir", "src/main/resources")
      val database = tool.get("database", "default")
      val hiveTableName = tool.get("hive.table.name", "hive_tbl")
      val kafkaTableName = tool.get("kafka.table.name", "kafka_tbl")
      val bootstrapServers = tool.get("bootstrap.servers", "b-2.cdc.62vm9h.c4.kafka.ap-northeast-1.amazonaws.com:9092,b-1.cdc.62vm9h.c4.kafka.ap-northeast-1.amazonaws.com:9092,b-3.cdc.62vm9h.c4.kafka.ap-northeast-1.amazonaws.com:9092")
      val groupId = tool.get("group.id", "flinkConsumer")
      val reset = tool.getBoolean("tables.reset", false)
      val topic = tool.get("kafka.topic", "cider")
      val hiveDBLocation = tool.get("hive.db.location", "s3://tang-emr-tokyo/flink/kafka2hive/")

      buildHiveTable(tableEnv, reset, hiveConfDir, database, hiveTableName, hiveDBLocation)
      buildKafkaTable(tableEnv, reset, bootstrapServers, topic, groupId, kafkaTableName)

      // select from kafka table and write to hive table
      tableEnv.executeSql("insert into " + hiveTableName + " select id, `value`, hashdata, num, token, info, ts, DATE_FORMAT(ts, 'yyyy-MM-dd'), DATE_FORMAT(ts, 'HH') from " + kafkaTableName)

  }

}

Kafka写入数据格式：

{"id": "35f1c5a8-ec19-4dc3-afa5-84ef6bc18bd8", "value": 1327.12, "hashdata": "0822c055f097f26f85a581da2c937895c896200795015e5f9e458889", "num": 3, "token": "800879e1ef9a356cece14e49fb6949c1b8c1862107468dc682d406893944f2b6", "info": "valentine", "created_timestamp": 1690165700}

5.1. 代码配置说明

Hive表的部分配置：

"'sink.partition-commit.policy.kind'='metastore,success-file',"
=》在分区完成写入后，如何通知下游“分区数据已经可读”。目前支持metastore和success-file

"'sink.partition-commit.trigger'='partition-time',"
=》什么时候触发partition commit。Partition-time表示在watermark超过了“分区时间”+“delay”的时间后，commit partition

"'sink.partition-commit.delay'='0 s'"
=》延迟这个时间后再commit分区

'sink.partition-commit.watermark-time-zone'='Asia/Shanghai'
=》时区必须与数据时间戳一致

"'auto-compaction'='true'"
=》开启文件合并，在落盘前先合并

通过checkponit来决定落盘频率
senv.enableCheckpointing(60000);

在这个配置下，每1分钟会做一次checkpoint，即将文件写入s3。同时，还会触发自动合并的动作，最终每1分钟生成1个orc文件。

5.2. 提交job

参考flink官网：

需要移除flink-table-planner-loader-1.16.0.jar，并移入flink-table-planner_2.12-1.16.0：

cd /usr/lib/flink/lib
sudo mv flink-table-planner-loader-1.16.0.jar ../

sudo wget https://repo1.maven.org/maven2/org/apache/flink/flink-table-planner_2.12/1.16.0/flink-table-planner_2.12-1.16.0.jar

sudo chown flink:flink flink-table-planner_2.12-1.16.0.jar
sudo chmod +x flink-table-planner_2.12-1.16.0.jar

然后主节点运行：
sudo cp /usr/lib/hive/lib/antlr-runtime-3.5.2.jar /usr/lib/flink/lib
sudo cp /usr/lib/hive/lib/hive-exec-3.1.3*.jar /lib/flink/lib
sudo cp /usr/lib/hive/lib/libfb303-0.9.3.jar /lib/flink/lib
sudo cp /usr/lib/flink/opt/flink-connector-hive_2.12-1.16.0.jar /lib/flink/lib

sudo chmod 755 /usr/lib/flink/lib/antlr-runtime-3.5.2.jar
sudo chmod 755 /usr/lib/flink/lib/hive-exec-3.1.3*.jar
sudo chmod 755 /usr/lib/flink/lib/libfb303-0.9.3.jar
sudo chmod 755 /usr/lib/flink/lib/flink-connector-hive_2.12-1.16.0.jar

上传hive配置文件到hdfs：

hdfs dfs -mkdir /user/hadoop/hiveconf/
hdfs dfs -put /etc/hive/conf/hive-site.xml /user/hadoop/hiveconf/hive-site.xml

Emr主节点提交job：

flink run-application \
-t yarn-application \
-c com.tang.hive.Kafka2Hive \
-p 8 \
-D state.backend=rocksdb \
-D state.checkpoint-storage=filesystem \
-D state.checkpoints.dir=s3://tang-emr-tokyo/flink/kafka2hive/checkpoints \
-D execution.checkpointing.interval=60000 \
-D state.checkpoints.num-retained=5 \
-D execution.checkpointing.mode=EXACTLY_ONCE \
-D execution.checkpointing.externalized-checkpoint-retention=RETAIN_ON_CANCELLATION \
-D state.backend.incremental=true \
-D execution.checkpointing.max-concurrent-checkpoints=1 \
-D rest.flamegraph.enabled=true \
flink-tutorial.jar \
--hive.conf.dir hdfs:///user/hadoop/hiveconf \
--reset true

5. 测试结果

5.1. 文件数量与大小

从写入基于s3的hive表来看，基本是1分钟2个文件（因为超出了默认rolling配置的128MB文件大小，所以会额外再写1个文件）。同时，未compaction的文件对下游不可见：

5.2. hive分区注册

从hive表来看，写入数据后默认在hive元数据内注册了新分区。

S3路径：

Hive分区：

5.3. 可见的最近数据

从hive查询结果来看，下游能查询到的数据为最近1分钟之前的数据：

select current_timestamp, ts from hive_tbl order by ts desc limit 10;
》
2023-07-27 09:25:24.193    2023-07-27 09:24:24
2023-07-27 09:25:24.193    2023-07-27 09:24:24
2023-07-27 09:25:24.193    2023-07-27 09:24:24
2023-07-27 09:25:24.193    2023-07-27 09:24:24