– 实时更新：支持高效的随机读写操作
– 列式存储：优化的列式存储格式
– 快速扫描：支持高性能的批量扫描
– SQL集成：与Impala深度集成
– 强一致性：支持ACID事务
– 水平扩展：支持自动分区和副本
– Schema灵活：支持表结构变更
– 高可用：支持多副本和自动故障恢复
– 压缩编码：支持多种压缩和编码算法
– 安全性：支持Kerberos认证和加密

Kudu架构组件：

组件 说明
Master 主节点，管理元数据和协调集群
Tablet Server 表服务节点，存储数据和响应请求
Tablet 表分区，数据的基本存储单元
Catalog Table 目录表，存储表和元数据信息

版本号 发布日期 说明
1.18.0 2025-07-14 最新稳定版
1.17.1 2024-11-15 稳定版，持续支持
1.16.0 2023-XX-XX 旧版
1.15.0 2022-XX-XX 旧版

Kudu 1.18.0主要更新：
– 性能优化
– 安全性增强
– Bug修复
– 兼容性改进
– 新增功能特性

官方下载地址：

Kudu官网：https://kudu.apache.org/
下载页面：https://kudu.apache.org/releases/
源码仓库：https://github.com/apache/kudu
文档中心：https://kudu.apache.org/docs/

下载Kudu源码：
$ cd /fgeudb/software
$ wget https://archive.apache.org/dist/kudu/1.18.0/apache-kudu-1.18.0.tar.gz

输出示例如下：
–2026-04-04 10:00:00– https://archive.apache.org/dist/kudu/1.18.0/apache-kudu-1.18.0.tar.gz
Resolving archive.apache.org… 163.172.17.49
Connecting to archive.apache.org|163.172.17.49|:443… connected.
HTTP request sent, awaiting response… 200 OK
Length: 12345678 (12M) [application/octet-stream]
Saving to: ‘apache-kudu-1.18.0.tar.gz’

apache-kudu-1.18.0.tar.gz 100%[======================================================================>] 11.77M 8.5MB/s in 1.4s

2026-04-04 10:00:02 (8.5 MB/s) – ‘apache-kudu-1.18.0.tar.gz’ saved [12345678/12345678]

验证下载：
$ wget https://archive.apache.org/dist/kudu/1.18.0/apache-kudu-1.18.0.tar.gz.sha512
$ sha512sum -c apache-kudu-1.18.0.tar.gz.sha512

输出示例如下：
apache-kudu-1.18.0.tar.gz: OK

解压源码：
$ tar -zxvf apache-kudu-1.18.0.tar.gz -C /fgeudb/

方式二：RPM包安装

配置Cloudera仓库：
# vi /etc/yum.repos.d/cloudera.repo

[cloudera-runtime]
name=Cloudera Runtime
baseurl=https://archive.cloudera.com/p/cdh7/7.3.1/redhat7/yum/
gpgkey=https://archive.cloudera.com/p/cdh7/7.3.1/redhat7/yum/RPM-GPG-KEY-cloudera
gpgcheck=1

安装Kudu：
# yum install -y kudu kudu-master kudu-tserver kudu-client0 kudu-client-devel

输出示例如下：
Loaded plugins: fastestmirror
Loading mirror speeds from cached hostfile
Resolving Dependencies
–> Running transaction check
—> Package kudu.x86_64 0:1.18.0-1.el7 will be installed
—> Package kudu-master.x86_64 0:1.18.0-1.el7 will be installed
—> Package kudu-tserver.x86_64 0:1.18.0-1.el7 will be installed
–> Finished Dependency Resolution

Dependencies Resolved
================================================================================
Package Arch Version Repository Size
================================================================================
Installing:
kudu x86_64 1.18.0-1.el7 cloudera-runtime 50 M
kudu-master x86_64 1.18.0-1.el7 cloudera-runtime 100 M
kudu-tserver x86_64 1.18.0-1.el7 cloudera-runtime 100 M

Transaction Summary
================================================================================
Install 3 Packages

Total download size: 250 M
Installed size: 500 M
Downloading packages:
(1/3): kudu-1.18.0-1.el7.x86_64.rpm | 50 MB 00:00:30
(2/3): kudu-master-1.18.0-1.el7.x86_64.rpm | 100 MB 00:01:00
(3/3): kudu-tserver-1.18.0-1.el7.x86_64.rpm | 100 MB 00:01:00

Complete!

方式三：源码编译安装

安装编译依赖：
# yum install -y autoconf automake cmake gcc-c++ git \
krb5-devel libtool make ncurses-devel openssl-devel \
python3-devel rsync unzip wget which zip

编译Kudu：
$ cd /fgeudb/apache-kudu-1.18.0
$ mkdir -p build/release
$ cd build/release
$ ../thirdparty/build-if-necessary.sh
$ cmake -DCMAKE_BUILD_TYPE=release ../..
$ make -j4

输出示例如下：
[ 1%] Building CXX object src/kudu/CMakeFiles/kudu.dir/…
…
[100%] Built target kudu

安装：
# make install

创建必要目录：
# mkdir -p /fgeudb/kudu/{master,data,logs}
# mkdir -p /fgeudb/kudu/master/{data,wal}
# mkdir -p /fgeudb/kudu/tserver/{data,wal}

创建kudu用户：
# groupadd kudu
# useradd -g kudu -s /sbin/nologin -M kudu

设置权限：
# chown -R kudu:kudu /fgeudb/kudu
# chmod -R 755 /fgeudb/kudu

步骤2：配置Master节点

创建Master配置文件：
# vi /etc/kudu/conf/master.gflagfile

–master_addresses=192.168.1.51:7051,192.168.1.52:7051,192.168.1.53:7051
–fs_wal_dir=/fgeudb/kudu/master/wal
–fs_data_dirs=/fgeudb/kudu/master/data
–log_dir=/fgeudb/kudu/logs
–rpc_bind_addresses=0.0.0.0:7051
–webserver_port=8051
–webserver_interface=0.0.0.0
–num_replicas=3
–default_num_replicas=3
–heartbeat_interval_ms=3000

单节点Master配置：
# vi /etc/kudu/conf/master.gflagfile

–fs_wal_dir=/fgeudb/kudu/master/wal
–fs_data_dirs=/fgeudb/kudu/master/data
–log_dir=/fgeudb/kudu/logs
–rpc_bind_addresses=0.0.0.0:7051
–webserver_port=8051

步骤3：配置Tablet Server节点

创建Tablet Server配置文件：
# vi /etc/kudu/conf/tserver.gflagfile

–tserver_master_addrs=192.168.1.51:7051,192.168.1.52:7051,192.168.1.53:7051
–fs_wal_dir=/fgeudb/kudu/tserver/wal
–fs_data_dirs=/fgeudb/kudu/tserver/data
–log_dir=/fgeudb/kudu/logs
–rpc_bind_addresses=0.0.0.0:7050
–webserver_port=8050
–webserver_interface=0.0.0.0
–heartbeat_interval_ms=3000
–memory_limit_hard_bytes=4294967296

单节点配置：
# vi /etc/kudu/conf/tserver.gflagfile

–tserver_master_addrs=192.168.1.51:7051
–fs_wal_dir=/fgeudb/kudu/tserver/wal
–fs_data_dirs=/fgeudb/kudu/tserver/data
–log_dir=/fgeudb/kudu/logs
–rpc_bind_addresses=0.0.0.0:7050
–webserver_port=8050
–memory_limit_hard_bytes=4294967296

步骤4：启动Kudu服务

启动Master：
# systemctl start kudu-master

输出示例如下：
Starting Kudu Master: [ OK ]

启动Tablet Server：
# systemctl start kudu-tserver

输出示例如下：
Starting Kudu Tablet Server: [ OK ]

设置开机自启：
# systemctl enable kudu-master
# systemctl enable kudu-tserver

查看服务状态：
# systemctl status kudu-master

输出示例如下：
● kudu-master.service – Apache Kudu Master Server
Loaded: loaded (/usr/lib/systemd/system/kudu-master.service; enabled)
Active: active (running) since Fri 2026-04-04 10:10:00 CST; 10s ago
Main PID: 12345 (kudu-master)
CGroup: /system.slice/kudu-master.service
└─12345 /usr/bin/kudu-master –flagfile=/etc/kudu/conf/master.gflagfile

查看Tablet Server状态：
# systemctl status kudu-tserver

输出示例如下：
● kudu-tserver.service – Apache Kudu Tablet Server
Loaded: loaded (/usr/lib/systemd/system/kudu-tserver.service; enabled)
Active: active (running) since Fri 2026-04-04 10:10:00 CST; 10s ago
Main PID: 12346 (kudu-tserver)

网络配置：
–rpc_bind_addresses=0.0.0.0:7051 RPC绑定地址
–webserver_port=8051 Web UI端口
–webserver_interface=0.0.0.0 Web UI绑定接口

存储配置：
–fs_wal_dir=/fgeudb/kudu/master/wal WAL日志目录
–fs_data_dirs=/fgeudb/kudu/master/data 数据目录

集群配置：
–master_addresses=192.168.1.51:7051,… Master地址列表
–num_replicas=3 默认副本数

日志配置：
–log_dir=/fgeudb/kudu/logs 日志目录
–log_level=INFO 日志级别

Tablet Server核心配置参数

网络配置：
–rpc_bind_addresses=0.0.0.0:7050 RPC绑定地址
–webserver_port=8050 Web UI端口

存储配置：
–fs_wal_dir=/fgeudb/kudu/tserver/wal WAL日志目录
–fs_data_dirs=/fgeudb/kudu/tserver/data 数据目录

内存配置：
–memory_limit_hard_bytes=4294967296 内存限制（4GB）
–block_cache_capacity_mb=512 块缓存大小

性能配置：
–num_tablet_servers_to_describe=3 描述Tablet Server数量
–heartbeat_interval_ms=3000 心跳间隔
–scanner_batch_size_rows=1024 扫描批次大小

生产环境配置优化

Master生产配置：
–master_addresses=192.168.1.51:7051,192.168.1.52:7051,192.168.1.53:7051
–fs_wal_dir=/fgeudb/kudu/master/wal
–fs_data_dirs=/fgeudb/kudu/master/data
–log_dir=/fgeudb/kudu/logs
–num_replicas=3
–default_num_replicas=3
–heartbeat_interval_ms=3000
–catalog_manager_wait_for_new_tablets_to_elect_leader_timeout_ms=60000

Tablet Server生产配置：
–tserver_master_addrs=192.168.1.51:7051,192.168.1.52:7051,192.168.1.53:7051
–fs_wal_dir=/fgeudb/kudu/tserver/wal
–fs_data_dirs=/fgeudb/kudu/tserver/data
–log_dir=/fgeudb/kudu/logs
–memory_limit_hard_bytes=17179869184
–block_cache_capacity_mb=2048
–heartbeat_interval_ms=3000
–scanner_batch_size_rows=2048
–num_scanner_threads=4
–log_level=INFO

查看集群状态：
$ kudu cluster ksck 192.168.1.51:7051

输出示例如下：
Connected to the Master
Master Summary
UUID: abc123def456
Address: 192.168.1.51:7051
State: RUNNING

Tablet Server Summary
UUID: def456ghi789
Address: 192.168.1.51:7050
State: RUNNING

创建表：
$ kudu table create 192.168.1.51:7051 fgedu_db.users ‘
{
“table_name”: “fgedu_db.users”,
“schema”: {
“columns”: [
{“name”: “id”, “type”: “INT64”, “nullable”: false},
{“name”: “name”, “type”: “STRING”, “nullable”: true},
{“name”: “email”, “type”: “STRING”, “nullable”: true},
{“name”: “created_at”, “type”: “UNIXTIME_MICROS”, “nullable”: true}
],
“key_column_names”: [“id”]
},
“partitioning”: {
“hash_partitions”: [{“columns”: [“id”], “num_buckets”: 4}],
“num_replicas”: 3
}
}’

输出示例如下：
Created table fgedu_db.users

列出表：
$ kudu table list 192.168.1.51:7051

输出示例如下：
fgedu_db.users

查看表结构：
$ kudu table describe 192.168.1.51:7051 fgedu_db.users

输出示例如下：
TABLE: fgedu_db.users
COLUMN | TYPE | NULLABLE | KEY
————-+——————-+———-+—–
id | INT64 | false | true
name | STRING | true | false
email | STRING | true | false
created_at | UNIXTIME_MICROS | true | false

使用Impala操作Kudu表

连接Impala：
$ impala-shell -i 192.168.1.51:21000

创建Kudu表：
[192.168.1.51:21000] > CREATE TABLE fgedu_db.orders (
> id BIGINT,
> user_id BIGINT,
> product_name STRING,
> amount DECIMAL(10,2),
> order_time TIMESTAMP,
> PRIMARY KEY (id)
> )
> PARTITION BY HASH(id) PARTITIONS 4
> STORED AS KUDU;

输出示例如下：
Query: create TABLE fgedu_db.orders (
id BIGINT,
user_id BIGINT,
product_name STRING,
amount DECIMAL(10,2),
order_time TIMESTAMP,
PRIMARY KEY (id)
)
PARTITION BY HASH(id) PARTITIONS 4
STORED AS KUDU

Fetched 0 row(s) in 0.25s

插入数据：
[192.168.1.51:21000] > INSERT INTO fgedu_db.orders VALUES
> (1, 100, ‘Product A’, 1000.00, NOW()),
> (2, 101, ‘Product B’, 2000.00, NOW()),
> (3, 102, ‘Product C’, 3000.00, NOW());

更新数据：
[192.168.1.51:21000] > UPDATE fgedu_db.orders
> SET amount = 1500.00
> WHERE id = 1;

删除数据：
[192.168.1.51:21000] > DELETE FROM fgedu_db.orders WHERE id = 3;

查询数据：
[192.168.1.51:21000] > SELECT * FROM fgedu_db.orders;

输出示例如下：
+—-+———+————–+———+———————+
| id | user_id | product_name | amount | order_time |
+—-+———+————–+———+———————+
| 1 | 100 | Product A | 1500.00 | 2026-04-04 10:15:00 |
| 2 | 101 | Product B | 2000.00 | 2026-04-04 10:15:00 |
+—-+———+————–+———+———————+
Fetched 2 row(s) in 0.05s

分区策略

Hash分区：
CREATE TABLE fgedu_db.events (
id BIGINT,
event_type STRING,
event_data STRING,
event_time TIMESTAMP,
PRIMARY KEY (id)
)
PARTITION BY HASH(id) PARTITIONS 8
STORED AS KUDU;

Range分区：
CREATE TABLE fgedu_db.logs (
id BIGINT,
log_date DATE,
log_message STRING,
PRIMARY KEY (id, log_date)
)
PARTITION BY RANGE(log_date) (
PARTITION VALUES < '2026-01-01', PARTITION '2026-01-01' <= VALUES < '2026-02-01', PARTITION '2026-02-01' <= VALUES < '2026-03-01', PARTITION '2026-03-01' <= VALUES ) STORED AS KUDU; Hash + Range复合分区： CREATE TABLE fgedu_db.metrics ( id BIGINT, metric_date DATE, metric_name STRING, metric_value DOUBLE, PRIMARY KEY (id, metric_date) ) PARTITION BY HASH(id) PARTITIONS 4, RANGE(metric_date) ( PARTITION VALUES < '2026-01-01', PARTITION '2026-01-01' <= VALUES < '2026-04-01', PARTITION '2026-04-01' <= VALUES ) STORED AS KUDU;

查看进程状态：
$ ps -ef | grep kudu

输出示例如下：
kudu 12345 1 5 10:10 ? 00:00:30 /usr/bin/kudu-master –flagfile=/etc/kudu/conf/master.gflagfile
kudu 12346 1 5 10:10 ? 00:00:30 /usr/bin/kudu-tserver –flagfile=/etc/kudu/conf/tserver.gflagfile

查看端口监听：
$ netstat -tlnp | grep kudu

输出示例如下：
tcp6 0 0 :::7050 :::* LISTEN 12346/kudu-tserver
tcp6 0 0 :::7051 :::* LISTEN 12345/kudu-master
tcp6 0 0 :::8050 :::* LISTEN 12346/kudu-tserver
tcp6 0 0 :::8051 :::* LISTEN 12345/kudu-master

访问Web UI：
Master Web UI：http://192.168.1.51:8051
Tablet Server Web UI：http://192.168.1.51:8050

检查集群健康：
$ kudu cluster ksck 192.168.1.51:7051

输出示例如下：
Connected to the Master
Master Summary
UUID: abc123def456
Address: 192.168.1.51:7051
State: RUNNING

Tablet Server Summary
UUID: def456ghi789
Address: 192.168.1.51:7050
State: RUNNING

Tables Summary
Name: fgedu_db.users
Replicas: 3
State: HEALTHY

Cluster Health: OK

性能测试

使用kudu-perf测试：
$ kudu perf loadgen 192.168.1.51:7051 \
–num_threads_per_client=4 \
–num_rows_per_thread=100000 \
–string_len=100

输出示例如下：
Starting load generation…
Rows inserted: 400000
Time elapsed: 45.23 seconds
Throughput: 8845.67 rows/sec

使用Impala测试查询：
[192.168.1.51:21000] > SELECT COUNT(*) FROM fgedu_db.orders;

输出示例如下：
+———-+
| count(*) |
+———-+
| 1000000 |
+———-+
Fetched 1 row(s) in 0.25s

症状：Memory limit exceeded

解决方案：
1. 增加内存限制：
–memory_limit_hard_bytes=17179869184

2. 增加块缓存：
–block_cache_capacity_mb=2048

3. 优化查询：
– 减少扫描数据量
– 使用分区裁剪
– 限制返回列

4. 监控内存使用：
访问Web UI查看内存使用情况

问题2：Tablet不平衡

症状：数据分布不均匀

解决方案：
1. 检查分区策略：
确保Hash分区键选择合理

2. 重新分区：
ALTER TABLE … ADD RANGE PARTITION …

3. 手动迁移Tablet：
$ kudu tablet move 192.168.1.51:7051 tablet_id new_server

4. 查看Tablet分布：
访问Master Web UI查看Tablet分布

问题3：写入性能低

症状：写入速度慢

解决方案：
1. 增加分区数：
PARTITION BY HASH(id) PARTITIONS 16

2. 调整内存配置：
–memory_limit_hard_bytes=17179869184

3. 批量写入：
使用批量INSERT提高吞吐量

4. 调整WAL配置：
–log_level=INFO
–fs_wal_dir使用SSD存储

Kudu管理命令

启动服务：
# systemctl start kudu-master
# systemctl start kudu-tserver

停止服务：
# systemctl stop kudu-tserver
# systemctl stop kudu-master

重启服务：
# systemctl restart kudu-master
# systemctl restart kudu-tserver

查看状态：
# systemctl status kudu-master
# systemctl status kudu-tserver

集群检查：
$ kudu cluster ksck 192.168.1.51:7051

表管理：
$ kudu table list 192.168.1.51:7051
$ kudu table describe 192.168.1.51:7051 table_name
$ kudu table delete 192.168.1.51:7051 table_name

Tablet管理：
$ kudu tablet list 192.168.1.51:7051 table_name
$ kudu tablet change_config 192.168.1.51:7051 tablet_id –replica_count=3

Master管理：
$ kudu master list 192.168.1.51:7051
$ kudu master status 192.168.1.51:7051

Tablet Server管理：
$ kudu tserver list 192.168.1.51:7051
$ kudu tserver status 192.168.1.51:7050