No.9 - 时序数据库随笔 - Apache IoTDB 多条时序数据联合分析-51CTO.COM

01 问题

本篇我们一起来看看Apache IoTDB如何解决要解决No8提到的网友问题，如下：

简单说就是如何处理两条时间线的数值计算？上面例子是一个 “+” 加法。

02 数据准备

要处理多条时序数据分析问题，我们首先先建立两条时间序列，我们做一下数据准备：

启动IoTDB服务实例

我们首先源码编译Master最新代码(),并启动服务实例。如下

iotdb git:(master) mvn clean install -DskipTests
 ...
 ...
[INFO] -------------------------------------------
[INFO] BUILD SUCCESS
[INFO] -------------------------------------------
[INFO] Total time:  02:41 min
[INFO] Finished at: 2021-04-07T07:13:32+08:00
[INFO] -------------------------------------------
➜  iotdb git:(master)  cd distribution/target/apache-iotdb-0.12.0-SNAPSHOT-server-bin/apache-iotdb-0.12.0-SNAPSHOT-server-bin/
➜  apache-iotdb-0.12.0-SNAPSHOT-server-bin git:(master) sbin/start-server.sh 
---------------------
Starting IoTDB
...
...
2021-04-07 07:33:50,636 [main] INFO  o.a.i.db.service.IoTDB:93 - IoTDB has started.

启动Cli

➜  iotdb git:(master)  cd distribution/target/apache-iotdb-0.12.0-SNAPSHOT-server-bin/apache-iotdb-0.12.0-SNAPSHOT-server-bin/
➜  apache-iotdb-0.12.0-SNAPSHOT-server-bin git:(master) sbin/start-cli.sh -h 127.0.0.1 -p 6667 -u root -pw root

创建存储组

IoTDB> set storage group to root.lemming

创建时间线

IoTDB> create timeseries root.lemming.device1.m1 with datatype=FLOAT,encoding=RLE
IoTDB> create timeseries root.lemming.device1.m2 with datatype=FLOAT,encoding=RLE
...
IoTDB> show timeseries;

插入时间序列数据

IoTDB> insert into root.lemming.device1(timestamp,m1, m2) values(1,3333,4444)
IoTDB> select * from root.lemming.device1;

看到这里，不知道大家是否发现，IoTDB虽然在存储层面是按时间序列进行列存储的，但是在API层面已经为大家抽象好来二维表结构，大家可以把多个时间序列当成一张二维度表的不同列,这个抽象大家就免去来入在InfluxDB中利用PIVOT进行行转列的需求，这样就更方便理解和对多时序数据的进行查询分析。好，那么下面我们看看如果实现 m1 + m2 这个需求。

03 代码数据准备

我们一个一个敲命令不如咱直接写个Java代码实例，方便反复测试，测试本篇内容，我们只需要依赖iotdb-jdbc模块，pom如下：

  <dependencies>
        <dependency>
            <groupId>org.apache.iotdb</groupId>
            <artifactId>iotdb-jdbc</artifactId>
            <version>${iotdb.version}</version>
        </dependency>
    </dependencies>

创建存储组，时间序列和准备数据的测试代码如下：

public class No9JDBCExample {

  /**
   * 第一次运行 init(),如果一切顺利，正面环境和数据准备完成
   */
  public static void main(String[] args) throws Exception {
    Class.forName("org.apache.iotdb.jdbc.IoTDBDriver");
    try (Connection connection =
            DriverManager.getConnection("jdbc:iotdb://127.0.0.1:6667/", "root", "root");
        Statement statement = connection.createStatement()) {
        init(statement);
    } catch (IoTDBSQLException e) {
      e.printStackTrace();
    }
  }

  public static void init(Statement statement) throws Exception {
    setStorageGroup(statement);
    statement.execute(
            "CREATE TIMESERIES root.lemming.device1.m1 with datatype=FLOAT,encoding=RLE");
    statement.execute(
            "CREATE TIMESERIES root.lemming.device1.m2 with datatype=FLOAT,encoding=RLE");
    statement.execute("INSERT INTO root.lemming.device1(timestamp,m1, m2) VALUES (1,3333,4444)");

    ResultSet resultSet = statement.executeQuery("SELECT timestamp, m1 as m1, m2 as m2 FROM root.lemming.device1");
    outputResult(resultSet);
  }

  public static void setStorageGroup(Statement statement) throws Exception{
    try {
      statement.execute("SET STORAGE GROUP TO root.lemming");
    }catch (Exception e){
      statement.execute("DELETE STORAGE GROUP root.lemming");
      statement.execute("SET STORAGE GROUP TO root.lemming");
    }
  }

  private static void outputResult(ResultSet resultSet) throws SQLException {
    if (resultSet != null) {
      System.out.println("--------------------------");
      final ResultSetMetaData metaData = resultSet.getMetaData();
      final int columnCount = metaData.getColumnCount();
      for (int i = 0; i < columnCount; i++) {
        System.out.print(metaData.getColumnLabel(i + 1) + " ");
      }
      System.out.println();
      while (resultSet.next()) {
        for (int i = 1; ; i++) {
          System.out.print(resultSet.getString(i));
          if (i < columnCount) {
            System.out.print(", ");
          } else {
            System.out.println();
            break;
          }
        }
      }
      System.out.println("--------------------------\n");
    }
  }
}

如果你运行得到如下结果，正面一切准备工作就绪：

04 UDF解决加法

我们接下来看看Apache IoTDB如何解决加法问题，目前IoTDB提供来自定义UDF的方式解决用户自定义的操作。如果要使用UDF我们需要增加依赖如下：

<dependency>
  <groupId>org.apache.iotdb</groupId>
  <artifactId>iotdb-server</artifactId>
  <version>0.12.0-SNAPSHOT</version>
  <scope>provided</scope>
</dependency>

完整配置查阅：https://github.com/sunjincheng121/know_how_know_why/blob/master/khkw_iotdb/No9udf/pom.xml

添加依赖之后，我们可以实现如下UDTF(User Defined Timeseries Generating Function), 这里写一下全称，不然大家会以为是User defined Table Function。当然不可否则，UDF的接口我们还可以有改进的空间，我们就目前现状与大家简单说明，如下：

public interface UDTF extends UDF {

  /**
   * This method is mainly used to customize UDTF. In this method, the user can do the following
   * things:
   *
   * <ul>
   *   <li>Use UDFParameters to get the time series paths and parse key-value pair attributes
   *       entered by the user.
   *   <li>Set the strategy to access the original data and set the output data type in
   *       UDTFConfigurations.
   *   <li>Create resources, such as establishing external connections, opening files, etc.
   * </ul>
   *
   * <p>This method is called after the UDTF is instantiated and before the beginning of the
   * transformation process.
   *
   * @param parameters used to parse the input parameters entered by the user
   * @param configurations used to set the required properties in the UDTF
   * @throws Exception the user can throw errors if necessary
   */
  @SuppressWarnings("squid:S112")
  void beforeStart(UDFParameters parameters, UDTFConfigurations configurations) throws Exception;

  /**
   * When the user specifies {@link RowByRowAccessStrategy} to access the original data in {@link
   * UDTFConfigurations}, this method will be called to process the transformation. In a single UDF
   * query, this method may be called multiple times.
   *
   * @param row original input data row (aligned by time)
   * @param collector used to collect output data points
   * @throws Exception the user can throw errors if necessary
   * @see RowByRowAccessStrategy
   */
  @SuppressWarnings("squid:S112")
  default void transform(Row row, PointCollector collector) throws Exception {}

  /**
   * When the user specifies {@link SlidingSizeWindowAccessStrategy} or {@link
   * SlidingTimeWindowAccessStrategy} to access the original data in {@link UDTFConfigurations},
   * this method will be called to process the transformation. In a single UDF query, this method
   * may be called multiple times.
   *
   * @param rowWindow original input data window (rows inside the window are aligned by time)
   * @param collector used to collect output data points
   * @throws Exception the user can throw errors if necessary
   * @see SlidingSizeWindowAccessStrategy
   * @see SlidingTimeWindowAccessStrategy
   */
  @SuppressWarnings("squid:S112")
  default void transform(RowWindow rowWindow, PointCollector collector) throws Exception {}

  /**
   * This method will be called once after all {@link UDTF#transform(Row, PointCollector) calls or
   * {@link UDTF#transform(RowWindow, PointCollector) calls have been executed. In a single UDF
   * query, this method will and will only be called once.
   *
   * @param collector used to collect output data points
   * @throws Exception the user can throw errors if necessary
   */
  @SuppressWarnings("squid:S112")
  default void terminate(PointCollector collector) throws Exception {}
}

其中最核心的是transform方法，对transform的实现逻辑就是我们业务的需求逻辑，面对我们要实现加法的逻辑，我们的核心逻辑就是transform实现，如下：

public class AddFunc implements UDTF {
...
...
    @Override
    public void transform(Row row, PointCollector collector) throws Exception {
        if (row.isNull(0) || row.isNull(1)) {
            return;
        }
        collector.putLong(
                row.getTime(), (long) (extractDoubleValue(row, 0) + extractDoubleValue(row, 1) + addend));
    }

    private double extractDoubleValue(Row row, int index) {
        double value;
        switch (row.getDataType(index)) {
            case INT32:
                value = row.getInt(index);
                break;
            case INT64:
                value = (double) row.getLong(index);
                break;
            case FLOAT:
                value = row.getFloat(index);
                break;
            case DOUBLE:
                value = row.getDouble(index);
                break;
            default:
                throw new UnSupportedDataTypeException(row.getDataType(index).toString());
        }
        return value;
    }
}

完整代码查阅：https://github.com/sunjincheng121/know_how_know_why/blob/master/khkw_iotdb/No9udf/src/main/java/org/khkw/iotdb/no9/AddFunc.java

当然开发完UDF，你还需要编写查询逻辑，如下：

public static void udf(Statement statement) {
    try{
      statement.execute("CREATE FUNCTION plus AS \"org.khkw.iotdb.no9.AddFunc\"");
      statement.execute("SHOW FUNCTIONS");
      ResultSet resultSet = statement.executeQuery("SELECT timestamp, plus(m1, m2) FROM root.lemming.device1");
      outputResult(resultSet);
    }catch (Exception e) {
      e.printStackTrace();
    }
  }

如上代码我们注册了名为 plus的UDF，然后plus(m1, m2)的方式实现了加法需求。当然，我们还不能直接运行udf方法，直接运行会报出找不到rg.khkw.iotdb.no9.AddFunc类，如下：

这是因为IoTDB配置udf的JAR加载需要指定JAR所在目录，我们利用 mvn clean package 生成JAR，然后再修改一下iotdb-engine.properties配置文件，如下：

我配置成为自己项目的jar所在路径了，如下：

配置完成之后，重新启动IoTDB服务实例。启动日志你会发现对UDF的加载：

➜  apache-iotdb-0.12.0-SNAPSHOT-server-bin git:(master) sbin/start-server.sh 
。。。
。。。
2021-04-07 08:17:53,398 [main] INFO  o.a.i.d.q.u.s.UDFClassLoaderManager:56 - UDF lib root: /Users/jincheng/work/know_how_know_why/khkw_iotdb/No9udf/target 
。。。
2021-04-07 12:34:11,622 [main] INFO  o.a.i.d.s.t.ThriftService:125 - IoTDB: start RPC ServerService successfully, listening on ip 0.0.0.0 port 6667

确保我们自定义的UDF已经被加载之后，我们运行udf，如下：

好，到目前我们利用UDF解决了加法需求，那么UDF是否可以在CLI里面使用呢？

05 CLI使用UDF

当然，我们可以在CLI里面使用UDF，我们启动一个CLI客户端,并用命令 SHOW FUNCTIONS 显示已有的FUNCTIONS，如下：

发现我们自定义的UDF已经在刚才运行Java程序时候注册了，我们可以删除让后在CLI在操作一遍（演示作用），如下：

IoTDB> CREATE FUNCTION plus AS "org.khkw.iotdb.no9.AddFunc"
Msg: 411: Failed to register non-TEMPORARY UDF PLUS(org.khkw.iotdb.no9.AddFunc), because a non-TEMPORARY UDF PLUS(org.khkw.iotdb.no9.AddFunc) with the same function name and the class name has already been registered.
IoTDB> DROP FUNCTION plus
Msg: The statement is executed successfully.
IoTDB> CREATE FUNCTION plus AS "org.khkw.iotdb.no9.AddFunc"
Msg: The statement is executed successfully.
IoTDB> SELECT timestamp, plus(m1, m2) FROM root.lemming.device1;
+-----------------------------+------------------------------------------------------+
|                         Time|plus(root.lemming.device1.m1, root.lemming.device1.m2)|
+-----------------------------+------------------------------------------------------+
|1970-01-01T08:00:00.001+08:00|                                                  7777|
+-----------------------------+------------------------------------------------------+
Total line number = 1
It costs 0.552s

OK,到这里不论是Java方式还是CLI方式我们都完成了多时序数据联合分析的需求。

06 贡献社区

上面我们发现IoTDB利用UDF的方式解决加法问题，虽然是通用做法，但是我们还可以做的更好，比如直接支持 m1 + m2 这种（加/减/乘/除）的便捷运算，所以这是你贡献的好机会，社区JIRA等你take。

https://issues.apache.org/jira/browse/IOTDB-1285

作者介绍

孙金城，51CTO社区编辑，Apache Flink PMC 成员，Apache Beam Committer，Apache IoTDB PMC 成员，ALC Beijing 成员，Apache ShenYu 导师，Apache 软件基金会成员。关注技术领域流计算和时序数据存储。