SeaTunnel 2.3.11 + Web 1.0.3 Docker 部署实战：Kafka 同步 Hive/ES 完整指南

优质文章学习记录

— 12 min read

SeaTunnel 2.3.11 + Web 1.0.3 Docker 部署实战：Kafka 同步 Hive/ES 完整指南

关键词：SeaTunnel 2.3.11、Docker 部署、Kafka 同步、Hive、Elasticsearch、数据集成

本文档详细介绍如何使用 Docker 部署 SeaTunnel 2.3.11 和 SeaTunnel Web 1.0.3，
并配置 Kafka 虚拟表、数据源以及 Kafka 同步到 Hive 和 Elasticsearch 的完整实战案例。

安装准备

目录结构

seatunnel-docker/ ├── docker-compose.yml # 主编排文件 ├── hive/ # Hive 配置 │ ├── hive-site.xml │ └── lib/ # 依赖 jar 包 │ └── postgresql-42.5.1.jar ├── init-sql/ # 初始化 SQL │ └── seatunnel_server_mysql.sql ├── seatunnel/ # SeaTunnel 服务端配置 │ ├── Dockerfile │ └── apache-seatunnel-2.3.11/ # 解压后的二进制包 │ └── lib/ # 依赖 jar 包 │ ├── hive-exec-3.1.3.jar │ ├── hive-metastore-3.1.3.jar │ ├── libfb303-0.9.3.jar │ ├── mysql-connector-java-8.0.28.jar │ └── seatunnel-hadoop3-3.1.4-uber.jar └── seatunnel-web/ # SeaTunnel Web 配置 ├── Dockerfile └── apache-seatunnel-web-1.0.3-bin/ # 解压后的二进制包 └── libs/ # 依赖 jar 包 └── mysql-connector-java-8.0.28.jar

下载 seatunnel

# seatunnel-2.3.11 https://dlcdn.apache.org/seatunnel/2.3.11/apache-seatunnel-2.3.11-bin.tar.gz # 源码构建seatunnel-web-1.0.3git clone https://github.com/apache/seatunnel-web.git cd seatunnel-web sh build.sh code

下载依赖包

#hive-metastore容器需要（PostgreSQL为Hive 元数据库） https://jdbc.postgresql.org/download/postgresql-42.5.1.jar #hive-metastore同步报错缺少依赖包（实际验证加前3个包即可） https://repo1.maven.org/maven2/org/apache/hive/hive-exec/3.1.3/hive-exec-3.1.3.jar https://repo1.maven.org/maven2/org/apache/hive/hive-metastore/3.1.3/hive-metastore-3.1.3.jar https://repo.maven.apache.org/maven2/org/apache/thrift/libfb303/0.9.3/libfb303-0.9.3.jar https://repo1.maven.org/maven2/org/apache/thrift/libthrift/0.12.0/libthrift-0.12.0.jar https://repo1.maven.org/maven2/org/apache/hive/hive-common/3.1.3/hive-common-3.1.3.jar

创建项目目录

将准备好的相关文件存放seatunnel-docker目录

mkdir seatunnel-docker cd seatunnel-docker

Docker部署

docker-compose.yml配置

version:'3.9'networks:seatunnel-network:driver: bridge ipam:config:-subnet: 172.16.0.0/24 services:# ===== Hive 相关服务 =====hive-metastore-db:image: postgres:15container_name: hive-metastore-db hostname: hive-metastore-db environment:POSTGRES_DB: metastore_db POSTGRES_USER: hive POSTGRES_PASSWORD: hive123456 ports:-"5432:5432"volumes:- ./hive-metastore-db-data:/var/lib/postgresql/data networks:seatunnel-network:ipv4_address: 172.16.0.2 healthcheck:# 添加健康检查test:["CMD-SHELL","pg_isready -U hive -d metastore_db"]interval: 5s timeout: 5s retries:10start_period: 10s hive-metastore:image: apache/hive:4.0.0 container_name: hive-metastore hostname: hive-metastore depends_on:hive-metastore-db:condition: service_healthy # 等待数据库健康后才启动environment:SERVICE_NAME: metastore DB_DRIVER: postgres SERVICE_OPTS:>--Djavax.jdo.option.ConnectionDriverName=org.postgresql.Driver -Djavax.jdo.option.ConnectionURL=jdbc:postgresql://hive-metastore-db:5432/metastore_db -Djavax.jdo.option.ConnectionUserName=hive -Djavax.jdo.option.ConnectionPassword=hive123456 ports:-"9083:9083"volumes:- ./hive/lib/postgresql-42.5.1.jar:/opt/hive/lib/postgresql-42.5.1.jar - ./hive/hive-site.xml:/opt/hive/conf/hive-site.xml - ./hive-warehouse:/opt/hive/data/warehouse networks:seatunnel-network:ipv4_address: 172.16.0.3 hive-server2:image: apache/hive:4.0.0 container_name: hive-server2 hostname: hive-server2 depends_on:- hive-metastore environment:HIVE_SERVER2_THRIFT_PORT:10000SERVICE_NAME: hiveserver2 IS_RESUME:"true"SERVICE_OPTS:"-Dhive.metastore.uris=thrift://hive-metastore:9083"ports:-"10000:10000"-"10002:10002"volumes:- ./hive-warehouse:/opt/hive/data/warehouse networks:seatunnel-network:ipv4_address: 172.16.0.4 # ===== MySQL =====mysql-seatunnel:image: mysql:8.0.42 container_name: mysql-seatunnel hostname: mysql-seatunnel environment:MYSQL_ROOT_PASSWORD: root123456 MYSQL_DATABASE: seatunnel MYSQL_ROOT_HOST:'%'ports:-"3806:3306"volumes:- ./mysql_data:/var/lib/mysql - ./init-sql:/docker-entrypoint-initdb.d networks:seatunnel-network:ipv4_address: 172.16.0.5 command:--default-authentication-plugin=mysql_native_password healthcheck:test:["CMD","mysqladmin","ping","-h","localhost"]interval: 10s timeout: 5s retries:5# ===== SeaTunnel =====seatunnel-master:build:context: ./seatunnel dockerfile: Dockerfile image: seatunnel:2.3.11 container_name: seatunnel-master hostname: seatunnel-master extra_hosts:-"hive-metastore:172.16.0.3"-"hive-metastore-db:172.16.0.2"environment:- SEATUNNEL_HOME=/opt/seatunnel command:> sh -c " cd /opt/seatunnel && exec bin/seatunnel-cluster.sh -r master "ports:-"5801:5801"volumes:- ./seatunnel/apache-seatunnel-2.3.11/:/opt/seatunnel/ - ./logs/master:/opt/seatunnel/logs # [修改点] 挂载 Hive 仓库目录，确保数据写入宿主机共享目录- ./hive-warehouse:/opt/hive/data/warehouse networks:seatunnel-network:ipv4_address: 172.16.0.10 seatunnel-worker1:image: seatunnel:2.3.11 container_name: seatunnel-worker1 hostname: seatunnel-worker1 extra_hosts:-"hive-metastore:172.16.0.3"-"hive-metastore-db:172.16.0.2"environment:- SEATUNNEL_HOME=/opt/seatunnel command:> sh -c " cd /opt/seatunnel && exec bin/seatunnel-cluster.sh -r worker "volumes:- ./seatunnel/apache-seatunnel-2.3.11/:/opt/seatunnel/ - ./logs/worker1:/opt/seatunnel/logs # [修改点] 挂载 Hive 仓库目录，确保数据写入宿主机共享目录- ./hive-warehouse:/opt/hive/data/warehouse depends_on:- seatunnel-master networks:seatunnel-network:ipv4_address: 172.16.0.11 seatunnel-worker2:image: seatunnel:2.3.11 container_name: seatunnel-worker2 hostname: seatunnel-worker2 extra_hosts:-"hive-metastore:172.16.0.3"-"hive-metastore-db:172.16.0.2"environment:- SEATUNNEL_HOME=/opt/seatunnel command:> sh -c " cd /opt/seatunnel && exec bin/seatunnel-cluster.sh -r worker "volumes:- ./seatunnel/apache-seatunnel-2.3.11/:/opt/seatunnel/ - ./logs/worker2:/opt/seatunnel/logs # [修改点] 挂载 Hive 仓库目录，确保数据写入宿主机共享目录- ./hive-warehouse:/opt/hive/data/warehouse depends_on:- seatunnel-master networks:seatunnel-network:ipv4_address: 172.16.0.12 seatunnel-web:build:context: ./seatunnel-web dockerfile: Dockerfile image: seatunnel-web:1.0.3 container_name: seatunnel-web hostname: seatunnel-web extra_hosts:-"hive-metastore:172.16.0.3"-"hive-metastore-db:172.16.0.2"environment:- SEATUNNEL_HOME=/opt/seatunnel - SEATUNNEL_WEB_HOME=/opt/seatunnel-web ports:-"8801:8801"volumes:- ./seatunnel/apache-seatunnel-2.3.11/:/opt/seatunnel/ - ./seatunnel-web/apache-seatunnel-web-1.0.3-bin/:/opt/seatunnel-web/ - ./logs/web:/opt/seatunnel-web/logs # [修改点] 挂载 Hive 仓库目录，保持环境一致性- ./hive-warehouse:/opt/hive/data/warehouse depends_on:- seatunnel-master networks:seatunnel-network:ipv4_address: 172.16.0.13

SeaTunnel 配置

Dockerfile

FROM eclipse-temurin:8-jdk-ubi9-minimal WORKDIR /opt/seatunnel/ # 设置环境变量 ENV SEATUNNEL_HOME=/opt/seatunnel ENV PATH=$PATH:$SEATUNNEL_HOME/bin # 暴露端口 EXPOSE 5801 # 启动命令 CMD ["sh", "bin/seatunnel-cluster.sh", "-r", "master"]

hazelcast-client.yaml 客户端配置

编辑 seatunnel/apache-seatunnel-2.3.11/config/hazelcast-client.yaml：

hazelcast-client:cluster-name: seatunnel properties:hazelcast.logging.type: log4j2 connection-strategy:connection-retry:cluster-connect-timeout-millis:3000network:cluster-members:- seatunnel-master:5801

hazelcast-master.yaml 配置

编辑 seatunnel/apache-seatunnel-2.3.11/config/hazelcast-master.yaml：

hazelcast:cluster-name: seatunnel network:rest-api:enabled:falseendpoint-groups:CLUSTER_WRITE:enabled:trueDATA:enabled:truejoin:tcp-ip:enabled:truemember-list:- seatunnel-master:5801- seatunnel-worker1:5802- seatunnel-worker2:5802port:auto-increment:falseport:5801properties:hazelcast.invocation.max.retry.count:20hazelcast.tcp.join.port.try.count:30hazelcast.logging.type: log4j2 hazelcast.operation.generic.thread.count:50hazelcast.heartbeat.failuredetector.type: phi-accrual hazelcast.heartbeat.interval.seconds:2hazelcast.max.no.heartbeat.seconds:180hazelcast.heartbeat.phiaccrual.failuredetector.threshold:10hazelcast.heartbeat.phiaccrual.failuredetector.sample.size:200hazelcast.heartbeat.phiaccrual.failuredetector.min.std.dev.millis:100

hazelcast-worker.yaml 配置

编辑 seatunnel/apache-seatunnel-2.3.11/config/hazelcast-worker.yaml：

hazelcast:cluster-name: seatunnel network:join:tcp-ip:enabled:truemember-list:- seatunnel-master:5801- seatunnel-worker1:5802- seatunnel-worker2:5802port:auto-increment:falseport:5802properties:hazelcast.invocation.max.retry.count:20hazelcast.tcp.join.port.try.count:30hazelcast.logging.type: log4j2 hazelcast.operation.generic.thread.count:50hazelcast.heartbeat.failuredetector.type: phi-accrual hazelcast.heartbeat.interval.seconds:2hazelcast.max.no.heartbeat.seconds:180hazelcast.heartbeat.phiaccrual.failuredetector.threshold:10hazelcast.heartbeat.phiaccrual.failuredetector.sample.size:200hazelcast.heartbeat.phiaccrual.failuredetector.min.std.dev.millis:100

安装连接器依赖包

配置同步任务，点击Source组件，源名称下拉框没有数据，需要安装依赖才可以显示。

cd seatunnel/apache-seatunnel-2.3.11/ sh bin/install-plugin.sh

SeaTunnel Web 配置

Dockerfile 配置

FROM eclipse-temurin:8-jdk-ubi9-minimal WORKDIR /opt/seatunnel-web/ # 设置环境变量 ENV SEATUNNEL_WEB_HOME=/opt/seatunnel-web ENV SEATUNNEL_HOME=/opt/seatunnel # 暴露端口 EXPOSE 8801 # 启动命令 CMD ["sh", "bin/seatunnel-backend-daemon.sh", "start"]

application.yml 配置

编辑 seatunnel-web/apache-seatunnel-web-1.0.3-bin/conf/application.yml：

server:port:8801spring:main:allow-circular-references:trueapplication:name: seatunnel jackson:date-format: yyyy-MM-dd HH:mm:ss datasource:driver-class-name: com.mysql.cj.jdbc.Driver url: jdbc:mysql://mysql-seatunnel:3306/seatunnel?useSSL=false&useUnicode=true&characterEncoding=utf-8&allowMultiQueries=true&allowPublicKeyRetrieval=trueusername: root password: root123456 jwt:expireTime:86400# please add key when deploy 要配置下token 32位secretKey: a3f5c8d2e1b4098765432109abcdef1234567890abcdef algorithm: HS256

hazelcast-client.yaml 客户端配置

编辑 seatunnel-web/apache-seatunnel-web-1.0.3-bin/conf/hazelcast-client.yaml：

hazelcast-client:cluster-name: seatunnel properties:hazelcast.logging.type: log4j2 connection-strategy:connection-retry:cluster-connect-timeout-millis:3000network:cluster-members:- seatunnel-master:5801

seatunnel-backend-daemon.sh

编辑 seatunnel-web/apache-seatunnel-web-1.0.3-bin/bin/seatunnel-backend-daemon.sh：
去除后台模式 去掉 nohup 和最后的 &

$JAVA_HOME/bin/java $JAVA_OPTS\-cp"$CLASSPATH"$SPRING_OPTS\ org.apache.seatunnel.app.SeatunnelApplication >>"${LOGDIR}/seatunnel.out"2>&1echo"seatunnel-web started"

plugin-mapping.properties 配置

实际验证此步骤可省略。
拷贝seatunnel/apache-seatunnel-2.3.11/connectors/plugin-mapping.properties到
seatunnel-web/apache-seatunnel-web-1.0.3-bin/conf/plugin-mapping.properties

cd seatunnel-docker cp seatunnel/apache-seatunnel-2.3.11/connectors/plugin-mapping.properties seatunnel-web/apache-seatunnel-web-1.0.3-bin/conf/plugin-mapping.properties

Hive 配置

hive-site.xml 配置

<?xml version="1.0" encoding="UTF-8"?><configuration><property><name>hive.metastore.uris</name><value>thrift://hive-metastore:9083</value></property><property><name>hive.metastore.warehouse.dir</name><value>/opt/hive/data/warehouse</value></property><property><name>metastore.metastore.event.db.notification.api.auth</name><value>false</value></property></configuration>

lib目录 依赖包

postgresql-42.5.1.jar

Mysql 配置

init-sql目录 初始SQL脚本

拷贝seatunnel-web/apache-seatunnel-web-1.0.3-bin/script/seatunnel_server_mysql.sql到
init-sql/seatunnel_server_mysql.sql

cd seatunnel-docker cp seatunnel-web/apache-seatunnel-web-1.0.3-bin/script/seatunnel_server_mysql.sql init-sql/seatunnel_server_mysql.sql

docker启动

# 启动所有服务docker compose up -d--build# 访问web ui页面 默认登录账号：admin / adminopen http://localhost:8801

运行示例

seatunnel-docker-start

登录配置语言

登录页面

login

设置

setting

配置语言

lang

配置数据源

kafka数据源

测试kafka_datasource-detail

ES数据源

测试ES_datasource-detail

Hive-metastore本地数据源

配置为thrift://hive-metastore:9083也可以。

测试本地hive-metastore_datasource-detail

配置虚拟表

虚拟表列表

virtual-table-detail

创建虚拟表流程

  1. 进入「虚拟表」菜单，点击「创建」按钮
  2. 选择数据源，配置虚拟表信息
  3. 点击「下一步」配置字段映射
  4. 点击「下一步」确认信息并保存

配置同步任务

kafka-hive 同步任务

测试kafka-hive_task-definition

任务组件配置

Source 组件配置
测试kafka-hive_task-definition-source
FieldMapper 组件配置（模型视图）
测试kafka-hive_task-definition-fieldmapper
Sink 组件配置
测试kafka-hive_task-definition-sink

Kafka-Elasticsearch 同步任务

测试kafka-ES_task-definition

任务组件配置

Source 组件配置
测试kafka-ES_task-definition-source
FieldMapper 组件配置（模型视图）
测试kafka-ES_task-definition-fieldmapper
Sink 组件配置
测试kafka-ES_task-definition-sink

创建同步任务通用流程

  1. 进入「任务」→「同步任务定义」，点击「创建」按钮
  2. 拖拽或选择 Source、FieldMapper、Sink 组件构建任务流程
  3. 双击 Source 组件，配置数据源信息（选择已配置的 Kafka 数据源）
  4. 双击 FieldMapper 组件，点击「模型」按钮配置字段映射关系
  5. 双击 Sink 组件，配置目标数据源信息（Hive 或 Elasticsearch）
  6. 保存并启动任务
    需要配置job mode ，不然保存不了，报错job env can't be empty, please change config
job-mode

hive相关操作

创建表

# 进入 HiveServer2 容器dockerexec-it hive-server2 beeline -u jdbc:hive2://localhost:10000 -e" CREATE TABLE IF NOT EXISTS default.test_user_data3 ( user_id STRING, type STRING, content STRING ) ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t' STORED AS TEXTFILE; "# 或创建 Parquet 格式（推荐）dockerexec-it hive-server2 beeline -u jdbc:hive2://localhost:10000 -e" CREATE TABLE IF NOT EXISTS default.test_user_data3 ( user_id STRING, type STRING, content STRING ) STORED AS PARQUET; "

查看表结构

dockerexec-it hive-server2 beeline -u jdbc:hive2://localhost:10000 -e" SHOW TABLES IN default; DESCRIBE default.test_user_data3; "

查询表数据

dockerexec-it hive-server2 beeline -u jdbc:hive2://localhost:10000 -e" SELECT * FROM default.test_user_data3 LIMIT 10; "

备注

seatunnel-web容器启动就退出

排查是否配置

seatunnel-backend-daemon.sh

编辑 seatunnel-web/apache-seatunnel-web-1.0.3-bin/bin/seatunnel-backend-daemon.sh：
去除后台模式 去掉 nohup 和最后的 &

$JAVA_HOME/bin/java $JAVA_OPTS\-cp"$CLASSPATH"$SPRING_OPTS\ org.apache.seatunnel.app.SeatunnelApplication >>"${LOGDIR}/seatunnel.out"2>&1echo"seatunnel-web started"

seatunnel-web启动后访问页面报错Unknown exception. secret key byte array cannot be null or empty

排查application.yml是否配置

jwt:expireTime:86400# please add key when deploysecretKey: a3f5c8d2e1b4098765432109abcdef1234567890abcdef algorithm: HS256

hive地址解析异常

seatunnel seatunnel-web ERROR [qtp2135089262-20][MetaStoreUtils.logAndThrowMetaException():166] - Got exception: java.net.URISyntaxException Illegal character inhostname at index 44: thrift://hive-metastore.seatunnel-docker_seatunnel-network:9083

docker-compose.yml对应容器加上ip绑定

extra_hosts:-"hive-metastore:172.16.0.3"-"hive-metastore-db:172.16.0.2"

Hive同步报错 error java.lang.NoClassDefFoundError

seatunnel/apache-seatunnel-2.3.11/lib存放依赖包

hive-exec-3.1.3.jar hive-metastore-3.1.3.jar libfb303-0.9.3.jar

hive同步任务显示成功，实际没有数据写入

docker-compose.yml对应容器加上hive写入本地目录的配置

volumes:# [修改点] 挂载 Hive 仓库目录，确保数据写入宿主机共享目录- ./hive-warehouse:/opt/hive/data/warehouse

查看任务执行日志 will be executed on worker

./logs/master/seatunnel-engine-master.log

 Task [TaskGroupLocation{jobId=1080750681855361026, pipelineId=1, taskGroupId=2}] will be executed on worker [[seatunnel-worker2]:5801], slotID [2], resourceProfile [ResourceProfile{cpu=CPU{core=0}, heapMemory=Memory{bytes=0}}], sequence [db6b679c-67cc-43b8-b64a-acaa85c2a4c0], assigned [1080750681855361026]

Read more

PentAGI-（AI自动化渗透）Docker环境部署

PentAGI-（AI自动化渗透）Docker环境部署

一、 深度解构：什么是 PentAGI？ PentAGI 是由 VXControl 团队开发的一款革命性开源安全项目。它代表了 AI 与网络安全 深度融合的最高水准，旨在打造一个能够自主执行任务的“数字黑客助手”。 1. 为什么它如此强大？ 不同于传统的扫描器（如 Nessus 或 OpenVAS），PentAGI 的核心是一个基于 LLM（大语言模型） 的决策引擎。它不仅能发现漏洞，更重要的是它能“理解”漏洞。 * • 自主推理：它能像人类渗透测试专家一样，根据上一步的扫描结果（如端口开放情况）动态推导下一步的攻击路径。 * • 工具编排：它能自主驱动并联动数百个安全工具（如 Nmap、Sqlmap、Nuclei、Metasploit 等），实现真正的自动化闭环。 * • 长效记忆：集成 pgvector 向量数据库，让 AI

【GitHub项目推荐--Planning with Files：基于Manus AI工作流的智能任务管理革命】⭐⭐⭐⭐⭐

简介 Planning with Files 是一个开源、跨平台的智能任务管理技能系统，由开发者OthmanAdi创建和维护。自2026年2月发布v2.15.0版本以来，这个项目以其独特的设计理念和强大的生产力提升能力，在AI辅助工作流领域引起了广泛关注。项目的核心灵感来源于Manus AI——这家在短短8个月内实现1亿美元以上收入，最终被Meta以20亿美元收购的AI代理公司。Planning with Files精准复现了Manus AI成功的核心秘诀：上下文工程（Context Engineering）。 核心价值： * 持久化工作记忆：将易失的AI上下文转化为持久的文件系统存储 * 目标防漂移机制：通过系统化检查点确保任务方向始终正确 * 错误学习循环：记录所有失败尝试，避免重复犯错 * 跨平台一致性：在14个不同IDE中提供统一的工作体验 项目定位：Planning with Files填补了AI代理的瞬时记忆与人类工作持久性之间的空白。它既不是另一个待办事项应用，也不是简单的笔记工具，而是一个完整的工作方法论实现。通过将Manus AI价值20亿美元的工作模

国产龙虾（AI Agent）全面对比及使用指南

国产龙虾（AI Agent）全面对比及使用指南 注：本文所指“龙虾”均为基于OpenClaw框架（或自主研发）的国产AI Agent工具，核心功能是通过自然语言指令实现电脑自动化操作、任务执行，以下涵盖你提及的所有型号，并补充当前主流国产型号，从核心特点、使用场景、安装步骤三方面进行详细对比，兼顾新手友好度与专业需求。 一、国产龙虾型号补充及整体对比 智谱澳龙、有道龙虾、腾讯QClaw、猎豹EasyClaw、MiniMax MaxClaw、Kimi Claw、阿里CoPaw、枫清Fabarta龙虾版（中国版龙虾，深度适配本土设备与安全需求），共8款型号，核心信息汇总如下： 型号研发主体核心定位适配系统核心优势适用人群智谱 澳龙（AutoClaw）智谱AI本地一键部署型OpenClaw，搭配GLM-5-Turbo专属龙虾模型Windows、macOS封装50+主流Skills与API，无需单独配置接口；搭配专属龙虾模型，长任务执行稳定，编程能力强开发者、技术人员、需要复杂任务执行的办公族有道 龙虾（LobsterAI）
微调模型成本太高，用RAG技术，低成本实现AI升级

微调模型成本太高，用RAG技术，低成本实现AI升级

文章目录 * 大模型 RAG 技术深度解析：从入门到进阶 * 一、大语言模型（LLM）的三大痛点 * 1.1 幻觉问题：一本正经地胡说八道 * 1.2 时效性问题：知识更新不及时 * 1.3 数据安全问题：敏感信息泄露风险 * 二、RAG 技术：检索增强生成 * 2.1 RAG 的定义 * 2.2 RAG 的架构 * 2.2.1 检索器模块 * 2.2.2 生成器模块 * 三、使用 RAG 的八大优势 * 3.1 可扩展性：减少模型大小和训练成本 * 3.