Spring Boot 整合 Redis,使用 RedisTemplate 客户端

文章目录

  • 一、SpringBoot 整合 Redis
    • 1.1 整合 Redis 步骤
      • 1.1.1 添加依赖
      • 1.1.2 yml 配置文件
      • 1.1.3 Config 配置文件
      • 1.1.4 使用示例
    • 1.2 RedisTemplate 概述
      • 1.2.1 RedisTemplate 简介
      • 1.2.2 RedisTemplate 功能
  • 二、RedisTemplate API
    • 2.1 RedisTemplate 公共 API
    • 2.2 String 类型 API
      • 2.1.1 添加缓存
      • 2.1.2 删除缓存
      • 2.1.3 修改缓存
      • 2.1.4 其他操作
    • 2.3 Hash 类型 API
      • 2.3.1 添加缓存
      • 2.3.2 删除缓存
      • 2.3.3 获取缓存
      • 2.3.4 其他操作
    • 2.4 List 类型 API
      • 2.4.1 添加缓存
      • 2.4.2 删除缓存
      • 2.4.3 获取缓存
    • 2.5 Set 类型 API
      • 2.5.1 添加缓存
      • 2.5.2 删除缓存
      • 2.5.3 获取缓存
      • 2.5.4 其他操作
    • 2.6 ZSet 类型 API
      • 2.6.1 添加缓存
      • 2.6.2 删除缓存
      • 2.6.3 获取缓存
  • 三、RedisUtil 工具类封装

一、SpringBoot 整合 Redis

1.1 整合 Redis 步骤

1.1.1 添加依赖

redis 的依赖:

<dependency><groupId>org.springframework.boot</groupId><artifactId>spring-boot-starter-data-redis</artifactId>
</dependency>

1.1.2 yml 配置文件

server:port: 8070spring:# redisredis:host: localhostport: 6379password:database: 0

1.1.3 Config 配置文件

@Configuration
public class RedisConfig {@Beanpublic RedisTemplate<String, Object> redisTemplate(RedisConnectionFactory factory) {RedisTemplate<String, Object> redisTemplate = new RedisTemplate<>();redisTemplate.setConnectionFactory(factory);redisTemplate.setKeySerializer(new StringRedisSerializer());//使用Jackson2JsonRedisSerializer来序列化和反序列化redis的value值redisTemplate.setValueSerializer(new Jackson2JsonRedisSerializer<>(Object.class));redisTemplate.afterPropertiesSet();return redisTemplate;}}

1.1.4 使用示例

注入 RedisTemplate,即可操作 Redis,简单示例如下:

@RunWith(SpringRunner.class)
@SpringBootTest
public class RedisUtilTest {@Autowiredprivate RedisTemplate<String, Object> redisTemplate;@Testpublic void test() {ValueOperations<String, Object> operations = redisTemplate.opsForValue();operations.set("key1","value1");operations.set("key1","value1");operations.set("key1","value1");Object object = operations.get("key1");System.out.println(object);}

1.2 RedisTemplate 概述

1.2.1 RedisTemplate 简介

RedisTemplate 是 Spring Data Redis 项目的一部分,旨在简化在Java应用程序中使用 Redis 的过程。它提供了一组简单的方法,可以在 Redis 数据库中存储和获取各种类型的数据,包括字符串、散列、列表、集合、有序集合等。
此外,RedisTemplate 还提供了许多其他功能,如事务支持、发布/订阅、消息队列等。

以下是 RedisTemplate 的一些主要特点:

  • 简化 Redis 访问:通过提供简单的方法和模板,RedisTemplate 简化了对 Redis 数据库的访问。它抽象了 Redis 底层的一些细节,使得开发者可以专注于业务逻辑。
  • 类型安全:RedisTemplate 支持类型安全的操作,这意味着编译器可以检查操作的类型是否正确,从而减少了运行时错误。
  • 事务支持:RedisTemplate 提供了事务支持,确保在执行多个操作时,要么所有操作都成功,要么回滚。
  • 发布/订阅和消息队列:RedisTemplate 提供了发布/订阅和消息队列的功能,使得应用程序可以异步接收和处理消息。
  • 集成测试:RedisTemplate 提供了一些用于集成测试的工具和方法,使得开发者可以方便地对 Redis 数据库进行单元测试和集成测试。

RedisTemplate 部分源码:

public class RedisTemplate<K, V> extends RedisAccessor implements RedisOperations<K, V>, BeanClassLoaderAware {private boolean enableTransactionSupport = false;private boolean exposeConnection = false;private boolean initialized = false;private boolean enableDefaultSerializer = true;private @Nullable RedisSerializer<?> defaultSerializer;private @Nullable ClassLoader classLoader;// 序列化属性@SuppressWarnings("rawtypes") private @Nullable RedisSerializer keySerializer = null;@SuppressWarnings("rawtypes") private @Nullable RedisSerializer valueSerializer = null;@SuppressWarnings("rawtypes") private @Nullable RedisSerializer hashKeySerializer = null;@SuppressWarnings("rawtypes") private @Nullable RedisSerializer hashValueSerializer = null;private RedisSerializer<String> stringSerializer = RedisSerializer.string();private @Nullable ScriptExecutor<K> scriptExecutor;// 相关操作属性private final ValueOperations<K, V> valueOps = new DefaultValueOperations<>(this);private final ListOperations<K, V> listOps = new DefaultListOperations<>(this);private final SetOperations<K, V> setOps = new DefaultSetOperations<>(this);private final StreamOperations<K, ?, ?> streamOps = new DefaultStreamOperations<>(this, new ObjectHashMapper());private final ZSetOperations<K, V> zSetOps = new DefaultZSetOperations<>(this);private final GeoOperations<K, V> geoOps = new DefaultGeoOperations<>(this);private final HyperLogLogOperations<K, V> hllOps = new DefaultHyperLogLogOperations<>(this);private final ClusterOperations<K, V> clusterOps = new DefaultClusterOperations<>(this);// 构造函数public RedisTemplate() {}
}

1.2.2 RedisTemplate 功能

RedisTemplate 支持 Redis 的五种基本数据类型,并对其进行了封装,RedisTemplate 对应的封装如下:
String 类型:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();

Hash 类型:

HashOperations<String, Object, Object> hashOperations = redisTemplate.opsForHash();

List 类型:

ListOperations<String, Object> listOperations = redisTemplate.opsForList();

Set 类型:

SetOperations<String, Object> setOperations = redisTemplate.opsForSet();

ZSet 类型:

ZSetOperations<String, Object> zSetOperations = redisTemplate.opsForZSet();

另外,RedisTemplate 还提供了以下封装,用于 Redis 的额外操作:

  • redisTemplate.opsForCluster():用于操作 Redis Cluster。Redis Cluster 是一种分布式 Redis 解决方案,它允许将多个 Redis 节点组织成一个集群,并提供数据分片和故障转移等功能。
ClusterOperations<String, Object> clusterOperations = redisTemplate.opsForCluster();
  • redisTemplate.opsForGeo():Geo数据类型允许存储地理位置相关的数据,并执行基于地理位置的查询和计算。
GeoOperations<String, Object> geoOperations = redisTemplate.opsForGeo();
  • redisTemplate.opsForHyperLogLog():HyperLogLog数据类型是一种用于估算集合中元素数量的数据结构。它通过使用一种特殊的算法,可以在非常小的内存消耗下估算出集合中元素的数量。
HyperLogLogOperations<String, Object> hyperLogLogOperations = redisTemplate.opsForHyperLogLog();
  • redisTemplate.opsForStream():Stream数据类型是一种持久化的消息队列,它可以保存长期的消息数据,并且支持消费者组和消息的消费确认等功能。
StreamOperations<String, Object, Object> streamOperations = redisTemplate.opsForStream();

二、RedisTemplate API

2.1 RedisTemplate 公共 API

Boolean delete(K key):根据缓存的键 key 删除整个缓存。

  • 参数:
    • key:缓存的键。
  • 返回值:返回 Boolean 类型,删除成功为 true,删除失败为 false。

示例:

String key = "key1";
Boolean result = redisTemplate.delete(key);

Long delete(Collection<K> keys)::根据缓存的键集合 keys 批量删除整个缓存。

  • 参数:
    • keys:缓存的键集合。
  • 返回值:返回删除成功的数量,Long 类型。

示例:

List<String> keys = Arrays.asList("key1", "key2", "key3");
Long result = redisTemplate.delete(keys);

Boolean hasKey(K key):根据缓存的键 key 判断缓存是否存在。

  • 参数:
    • key:缓存的键。
  • 返回值:返回 Boolean 类型,存在为 true,不存在为 false。

示例:

String key = "key1";
Boolean result = redisTemplate.hasKey(key);

Set<K> keys(K pattern):根据匹配规则,匹配出满足条件的所有缓存键

  • 参数:
    • pattern:匹配规则。
  • 返回值:返回 Set 类型,满足匹配规则的缓存键集合。

示例:

String pattern = "key*";
Set<String> result = redisTemplate.keys(pattern);
System.out.println(result);

输出结果:

[key1, key2, key3]

void rename(K oldKey, K newKey):重命名缓存键 oldKey 为 newKey。

  • 参数:
    • oldKey:原缓存键名称。
    • newKey:新缓存键名称。

示例:

String oldKey= "key1";
String newKey= "key-new";
redisTemplate.rename(oldKey, newKey);

2.2 String 类型 API

2.1.1 添加缓存

void set(K key, V value):添加缓存。

  • 参数:
    • key:键
    • value:值

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.set("string-key-1","string-value-1");
valueOperations.set("string-key-2","string-value-2");
valueOperations.set("string-key-3","string-value-3");
System.out.println("string-key-1: "valueOperations.get("string-key-1"));
System.out.println("string-key-2: "valueOperations.get("string-key-2"));
System.out.println("string-key-3: "valueOperations.get("string-key-3"));

输出结果:

string-key-1: string-value-1
string-key-2: string-value-2
string-key-3: string-value-3

void set(K key, V value, long timeout, TimeUnit unit):添加缓存,并设定过期时间。

  • 参数:
    • key:键
    • value:值
    • timeout:过期时间
    • unit:时间单位

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.set("string-key-1", "string-value-1", 2000, TimeUnit.MILLISECONDS);
valueOperations.set("string-key-2", "string-value-2", 10, TimeUnit.SECONDS);
valueOperations.set("string-key-3", "string-value-3", 7, TimeUnit.DAYS);

default void set(K key, V value, Duration timeout):添加缓存,并设定过期时间。

  • 参数:
    • key:键
    • value:值
    • timeout:超时时间,自带有单位,部分方法如下:
      • Duration.ofDays(long days)
      • Duration.ofHours(long hours)
      • Duration.ofMinutes(long minutes)
      • Duration.ofSeconds(long seconds)
      • Duration.ofSeconds(long seconds, long nanoAdjustment)
      • Duration.ofMillis(long millis)
      • Duration.ofNanos(long nanos)
      • Duration.of(long amount, TemporalUnit unit)

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.set("string-key-1", "string-value-1", Duration.ofMillis(2000));
valueOperations.set("string-key-2", "string-value-2", Duration.ofSeconds(5));
valueOperations.set("string-key-3", "string-value-3", Duration.ofDays(7));

Boolean setIfAbsent(K key, V value):添加缓存,key 不存在则添加,否则添加失败。

  • 参数:
    • key:键
    • value:值
  • 返回值:返回 Boolean 类型,key 不存在时添加成功返回true,key 存在时添加失败返回false。

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
Boolean result1 = valueOperations.setIfAbsent("string-key-1", "string-value-1");
Boolean result2 = valueOperations.setIfAbsent("string-key-4", "string-value-4");
System.out.println(result1);
System.out.println(result2);

输出结果:

false
true

Boolean setIfAbsent(K key, V value, long timeout, TimeUnit unit):添加缓存并设定过期时间,key 不存在则添加,否则添加失败。

  • 参数:
    • key:键
    • value:值
    • timeout:过期时间
    • unit:时间单位
  • 返回值:返回 Boolean 类型,key 不存在时添加成功返回true,key 存在时添加失败返回false。

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
Boolean result1 = valueOperations.setIfAbsent("string-key-1", "string-value-1", 2000, TimeUnit.MILLISECONDS);
Boolean result2 = valueOperations.setIfAbsent("string-key-4", "string-value-4", 2000, TimeUnit.MILLISECONDS);

default Boolean setIfAbsent(K key, V value, Duration timeout):添加缓存并设定过期时间,key 不存在则添加,否则添加失败。

  • 参数:
    • key:键
    • value:值
    • timeout:超时时间,自带有单位
  • 返回值:返回 Boolean 类型,key 不存在时添加成功返回true,key 存在时添加失败返回false。

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
Boolean result1 = valueOperations.setIfAbsent("string-key-1", "string-value-1", Duration.ofMillis(2000));
Boolean result2 = valueOperations.setIfAbsent("string-key-4", "string-value-4", Duration.ofMillis(2000));

void multiSet(Map<? extends K, ? extends V> map):批量添加缓存,有重复的key直接覆盖原来的缓存内容

  • 参数:
    • map:要添加的缓存内容键值对集合

示例:

Map<String,Object> map = new HashMap<>();
map.put("string-key-1", "string-value-1");
map.put("string-key-2", "string-value-2");
map.put("string-key-3", "string-value-3");
ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.multiSet(map);

Boolean multiSetIfAbsent(Map<? extends K, ? extends V> map):批量添加缓存,当 key 不存在的时候才添加,只要有一个 key 已存在则添加失败。

  • 参数:
    • map:要添加的缓存内容键值对集合
  • 返回值:返回 Boolean 类型,添加成功返回 true,失败则为 false

示例:

Map<String,Object> map = new HashMap<>();
map.put("string-key-1", "string-value-1");
map.put("string-key-2", "string-value-2");
map.put("string-key-3", "string-value-3");
ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
Boolean result = valueOperations.multiSetIfAbsent(map);

V getAndSet(K key, V value):给指定 key 的缓存设置新值,并返回原来的旧值。如果 key 不存在,则直接添加这个缓存,但返回值为 null

  • 参数:
    • key:指定缓存键
    • value:新值
  • 返回值:返回原来的旧值。

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", "s-value");
System.out.println("s-key : " + valueOperations.get("s-key"));
Object result = valueOperations.getAndSet("s-key", "s-value-new");
System.out.println("result = " + result);
System.out.println("s-key : " + valueOperations.get("s-key"));
// 一个不存在的key
Object result1 = valueOperations.getAndSet("s-key1", "s-value-new1");
System.out.println("result1 = " + result1);
System.out.println("s-key1 : " + valueOperations.get("s-key1"));

输出结果:

s-key : s-value
result = s-value
s-key : s-value-new
result = null
s-key1 : s-value-new1

2.1.2 删除缓存

V get(Object key):根据键 key 获取缓存值,key 不存在则返回 null

  • 参数:
    • key:缓存键
  • 返回值:返回对应的值,key 不存在则返回 null

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", "s-value-new");
Object result = valueOperations.get("s-key");
System.out.println("result = " + result);
Object result1 = valueOperations.get("s-key-s");
System.out.println("result1 = " + result1);

输出结果:

result = s-value-new
result1 = null

String get(K key, long start, long end):根据键 key 获取缓存值,并根据指定的开始和结束位置截取结果

  • 参数:
    • key:缓存键
    • start:开始位置
    • end:结束位置
  • 返回值:返回结果,key 不存在则返回 null

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", "s-value-new");
Object result1 = valueOperations.get("s-key");
Object result2 = valueOperations.get("s-key", 0, -1);
Object result3 = valueOperations.get("s-key", 2, 4);
System.out.println("result1 = " + result1);
System.out.println("result2 = " + result2);
System.out.println("result3 = " + result3);

输出结果:

result1 = s-value-new
result2 = "s-value-new"
result3 = -va

List<V> multiGet(Collection<K> keys):根据键集合 keys 批量获取缓存,键不存在的对应元素返回 null

  • 参数:
    • keys:缓存键集合
  • 返回值:缓存值集合

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key1", "s-value-new1");
valueOperations.set("s-key2", "s-value-new2");
valueOperations.set("s-key3", "s-value-new3");
List<Object> result = valueOperations.multiGet(Arrays.asList("s-key1", "s-key2", "s-key3", "s-key4"));
System.out.println("result = " + result);

输出结果:

result = [s-value-new1, s-value-new2, s-value-new3, null]

2.1.3 修改缓存

void set(K key, V value, long offset):缓存内容替换,用值 value 替换缓存键 key 对应的缓存,指定替换的开始位置 offset。

  • 参数:
    • key:缓存键
    • value:替换值
    • offset:偏移位置,即开始位置

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", "aaabbbcccddd");
Object oldResult = valueOperations.get("s-key");
valueOperations.set("s-key","6666",3);
Object newResult = valueOperations.get("s-key");
System.out.println("oldResult = " + oldResult);
System.out.println("newResult = " + newResult);

输出结果:

oldResult = aaabbbcccddd
newResult = aaa6666ccddd

Integer append(K key, String value):在缓存键 key 对应的缓存内容后面追加值 value,并返回追加值后的缓存值长度。如果 key 不存在,则直接添加这个缓存,返回这个缓存的长度

  • 参数:
    • key:缓存键
    • value:追加值
  • 返回值:Integer 类型,返回操作成功后的值的长度

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", "111222");
Integer result = valueOperations.append("s-key", "aaaa");
System.out.println("result = " + result);
System.out.println("s-key = " + valueOperations.get("s-key-s"));
Integer result1 = valueOperations.append("s-key-s", "aaaa");
System.out.println("result1 = " + result1);
System.out.println("s-key-s = " + valueOperations.get("s-key-s"));

输出结果:

result = 10
s-key = 111222aaaa
result1 = 4
s-key-s = aaaa

2.1.4 其他操作

Boolean setBit(K key, long offset, boolean value):设置缓存键 key 对应缓存的指定位置的 bit 值,用于设置 Redis 中位图(bitmaps)的指定偏移量的值。位图是由位(0或1)组成的数据结构,可以用于表示某个事件的状态或者进行位运算。

  • 参数:
    • key:缓存键
    • offset:要设置的位置。
    • value:要设置的值,0 或 1
  • 返回值:返回操作结果

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
Boolean result  = valueOperations.setBit("bit-key", 2, true);

Boolean getBit(K key, long offset):获取指定位置的 bit 值

  • 参数:
    • key:缓存键
    • offset:要设置的位置。
  • 返回值:返回结果

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
Boolean result = valueOperations.getBit("bit-key", 2);

Long increment(K key):将缓存键 key 对应的缓存的值增加1,缓存的内容必须为 Integer 类型,否则会出现错误 ERR value is not an integer or out of range

  • 参数:
    • key:缓存键
  • 返回值:返回增加后的结果

示例:

ValueOperations<String, Integer> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", 1);
Long result = valueOperations.increment("s-key");
System.out.println("result = " + result);
System.out.println("s-key = " + valueOperations.get("s-key"));

Long increment(K key, long delta):将缓存键 key 对应的缓存的值增加 delta,缓存的内容必须为 Integer 类型,否则会出现错误 ERR value is not an integer or out of range

  • 参数:
    • key:缓存键
    • delta:增加的值,long 类型
  • 返回值:返回增加后的结果

示例:

ValueOperations<String, Integer> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", 1);
Long result = valueOperations.increment("s-key", 2);
System.out.println("result = " + result);
System.out.println("s-key = " + valueOperations.get("s-key"));

Double increment(K key, double delta):将缓存键 key 对应的缓存的值增加 delta,缓存的内容必须为 Integer 类型,否则会出现错误 ERR value is not an integer or out of range

  • 参数:
    • key:缓存键
    • delta:增加的值,double 类型
  • 返回值:返回增加后的结果

示例:

ValueOperations<String, Integer> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", 1);
Doubleresult = valueOperations.increment("s-key", 3.5);
System.out.println("result = " + result);
System.out.println("s-key = " + valueOperations.get("s-key"));

Long decrement(K key):将缓存键 key 对应的缓存的值减少1,缓存的内容必须为 Integer 类型,否则会出现错误 ERR value is not an integer or out of range

  • 参数:
    • key:缓存键
  • 返回值:返回减少后的结果

示例:

ValueOperations<String, Integer> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", 1);
Long result = valueOperations.decrement("s-key");
System.out.println("result = " + result);
System.out.println("s-key = " + valueOperations.get("s-key"));

Long decrement(K key, long delta):将缓存键 key 对应的缓存的值减少 delta,缓存的内容必须为 Integer 类型,否则会出现错误 ERR value is not an integer or out of range

  • 参数:
    • key:缓存键
    • delta:减少的值,long 类型
  • 返回值:返回减少后的结果

示例:

ValueOperations<String, Integer> valueOperations = redisTemplate.opsForValue();
valueOperations.set("s-key", 1);
Long result = valueOperations.decrement("s-key", 2);
System.out.println("result = " + result);
System.out.println("s-key = " + valueOperations.get("s-key"));

Long size(K key):统计缓存键 key 对应缓存的值的大小。

  • 参数:
    • key:缓存键
  • 返回值:返回统计结果

示例:

ValueOperations<String, Object> valueOperations = redisTemplate.opsForValue();
System.out.println(valueOperations.size("s-key"));
System.out.println(valueOperations.size("s-key1"));
System.out.println(valueOperations.size("s-key2"));

2.3 Hash 类型 API

2.3.1 添加缓存

void put(H key, HK hashKey, HV value):添加一个 hash 缓存,key 相同,后者会覆盖前者。

  • 参数:
    • key:缓存键
    • hashKey:缓存 map 的 key
    • value:缓存 map 的值

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
operations.put("H-KEY", "KEY1", "VALUE1");
operations.put("H-KEY", "KEY1", "VALUE11");
operations.put("H-KEY", "KEY2", "VALUE2");
operations.put("H-KEY", "KEY3", "VALUE3");

void putAll(H key, Map<? extends HK, ? extends HV> m):批量添加 hash 缓存。

  • 参数:
    • key:缓存键
    • m:缓存内容,map 形式

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
Map<String, String> map = new HashMap<String, String>() {{put("key-a", "value-a");put("key-b", "value-b");put("key-c", "value-c");
}};
operations.putAll("H-KEY1", map);

Boolean putIfAbsent(H key, HK hashKey, HV value):添加一个 hash 缓存,缓存的 hashkey 不存在则添加成功,否则添加失败,如果 key 不存在则直接作为新的 hash 缓存添加。

  • 参数:
    • key:
  • 返回值:返回操作结果,成功返回 true,失败为 false

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
Boolean result1 = operations.putIfAbsent("H-KEY", "KEY3", "VALUE33");
System.out.println(result1);
Boolean result2 = operations.putIfAbsent("H-KEY", "KEY4", "VALUE4");
System.out.println(result2);
Boolean result3 = operations.putIfAbsent("H-KEY-NEW", "KEY-NEW", "VALUE-NEW");
System.out.println(result3);

输出结果:

false
true
true

2.3.2 删除缓存

Long delete(H key, Object... hashKeys):根据缓存键 key 和缓存 map 的 hashKeys 批量删除缓存。

  • 参数:
    • key:缓存键
    • hashKeys:缓存 map 的 hashKey
  • 返回值:返回删除结果数量

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
Long delete1 = operations.delete("H-KEY","KEY1");
System.out.println("delete1 = " + delete1);
Long delete2 = operations.delete("H-KEY","KEY1","KEY2","KEY3");
System.out.println("delete1 = " + delete2);

输出结果:

delete1 = 1
delete2 = 2

2.3.3 获取缓存

HV get(H key, Object hashKey):根据缓存 key 和 hashKey 获取一个缓存值。

  • 参数:
    • key:缓存键
    • hashKey:缓存 map 的键
  • 返回值:返回获取的缓存值

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
String result = operations.get("H-KEY", "KEY1");
System.out.println("result = " + result);
String result1 = operations.get("H-KEY", "KEY99");
System.out.println("result = " + result1);

输出结果:

result = VALUE11
result = null

List<HV> multiGet(H key, Collection<HK> hashKeys)

  • 参数:根据缓存 key 和 hashKey 批量获取缓存值。
    • key:缓存键
    • hashKeys:缓存 map 的键集合
  • 返回值:返回缓存值集合

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
List<String> result = operations.multiGet("H-KEY", Arrays.asList("KEY1", "KEY2", "KEY3"));
System.out.println("result = " + result);
List<String> result1 = operations.multiGet("H-KEY", Arrays.asList("KEY1", "KEY99", "KEY98"));
System.out.println("result = " + result1);

输出结果:

result = [VALUE11, VALUE2, VALUE3]
result = [VALUE11, null, null]

Set<HK> keys(H key):获取缓存 key 对应缓存 map 的所有 hashKey

  • 参数:
    • key:缓存键
  • 返回值:缓存 map 对应的所有 hashKey 集合

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
Set<String> keys = operations.keys("H-KEY");
System.out.println("keys = " + keys);

输出结果:

keys = [KEY1, KEY2, KEY3, KEY4]

List<HV> values(H key):获取缓存 key 对应缓存 map 的所有 value

  • 参数:
    • key:缓存键
  • 返回值:缓存 map 对应的所有 value 集合

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
List<String> values = operations.values("H-KEY");
System.out.println("values = " + values);

输出结果:

values = [VALUE11, VALUE2, VALUE3, VALUE4]

Map<HK, HV> entries(H key):获取缓存 key 对应缓存 map,以键值对方式返回。

  • 参数:
    • key:缓存键
  • 返回值:缓存 map 集合

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
Map<String, String> map = operations.entries("H-KEY");
System.out.println("map = " + map);

输出结果:

map = {KEY1=VALUE11, KEY2=VALUE2, KEY3=VALUE3, KEY4=VALUE4}

2.3.4 其他操作

Boolean hasKey(H key, Object hashKey):缓存键 key 下的 hashKey 键是否存在。

  • 参数:
    • key:缓存键
  • 返回值:返回结果,存在为 true,不存在为 false

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
Boolean result1 = operations.hasKey("H-KEY", "KEY1");
System.out.println("result1 = " + result1);
Boolean result2 = operations.hasKey("H-KEY1", "key-a");
System.out.println("result2 = " + result2);

输出结果:

result1 = false
result2 = true

Long lengthOfValue(H key, HK hashKey):缓存键 key 下 hashKey 键对应的值的长度。低版本 redis 会出现错误 ERR unknown command 'HSTRLEN'

  • 参数:
    • key:缓存键
    • hashKey:缓存 map 的 hashKey
  • 返回值:返回统计结果值

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
Long result1 = operations.lengthOfValue("H-KEY", "KEY1");
System.out.println("result1 = " + result1);
Long result2 = operations.lengthOfValue("H-KEY1", "key-a");
System.out.println("result2 = " + result2);

输出结果:

result1 = 1
result2 = 3

Long size(H key):获取缓存键 key 下的缓存内容的数量。

  • 参数:
    • key:缓存键
  • 返回值:返回统计结果值

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
System.out.println(operations.size("H-KEY1"));
System.out.println(operations.size("long-key"));

输出结果:

3
1

Cursor<Map.Entry<HK, HV>> scan(H key, ScanOptions options):匹配获取缓存键 key 下的缓存内容。

  • 参数:
    • key:缓存键
    • options:匹配规则
  • 返回值:返回匹配结果值

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
Map<String,String> map = new HashMap<String, String>(){{put("key-a-a","value-a-a");put("key-a-b","value-a-b");put("key-a-c","value-a-c");put("key-b-a","value-a-a");put("key-b-b","value-b-b");put("key-b-c","value-b-c");
}};
operations.putAll("scan-key",map);
ScanOptions build = ScanOptions.scanOptions().match("key-a*").build();
Cursor<Map.Entry<String, String>> scan = operations.scan("scan-key", build);
while (scan.hasNext()){Map.Entry<String, String> next = scan.next();System.out.println(next.getKey() + " : " + next.getValue());
}
try {scan.close();
} catch (IOException e) {throw new RuntimeException(e);
}

输出结果:

key-a-a : value-a-a
key-a-c : value-a-c
key-a-b : value-a-b

Long increment(H key, HK hashKey, long delta):使缓存键 key 下的 hashKey 键对应的值以 long 类型增加 delta,如果指定的哈希表或字段不存在,那么将会创建一个新的哈希表和字段,并指定值为 delta。

  • 参数:
    • key:缓存键
    • hashKey:缓存 map 的 hashKey
    • delta:要增加的值
  • 返回值:返回增加后的结果值

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
operations.put("long-key","key1","1");
System.out.println(operations.get("long-key", "key1"));
Long result = operations.increment("long-key","key1",2);
System.out.println("result = " + result);
System.out.println(operations.get("long-key", "key1"));

输出结果:

1
result = 3
3

Double increment(H key, HK hashKey, double delta):使缓存键 key 下的 hashKey 键对应的值以 double 类型增加 delta,如果指定的哈希表或字段不存在,那么将会创建一个新的哈希表和字段,并指定值为 delta。

  • 参数:
    • key:缓存键
    • hashKey:缓存 map 的 hashKey
    • delta:要增加的值
  • 返回值:返回增加后的结果值

示例:

HashOperations<String, String, String> operations = redisTemplate.opsForHash();
operations.put("long-key","key1","1");
System.out.println(operations.get("long-key", "key1"));
Double result = operations.increment("long-key", "key1", 2.0);
System.out.println("result = " + result);
System.out.println(operations.get("long-key", "key1"));

输出结果:

1
result = 3.5
3.5

2.4 List 类型 API

2.4.1 添加缓存

Long leftPush(K key, V value):向缓存键 key 对应缓存队列的头部添加缓存。

  • 参数
    • key:缓存键
    • value:缓存值
  • 返回值:返回每次添加成功之后,缓存队列的数量

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result1 = operations.leftPush("l-key", "l-value");
Long result2 = operations.leftPush("l-key", "l-value1");
Long result3 = operations.leftPush("l-key", "l-value2");
System.out.println("result1 = " + result1);
System.out.println("result2 = " + result2);
System.out.println("result3 = " + result3);

输出结果:

result1 = 1
result2 = 2
result3 = 3

在这里插入图片描述

Long leftPush(K key, V pivot, V value):在缓存键 key 对应缓存队列中,将值 value 添加到指定值 pivot 的前面。

  • 参数
    • key:缓存键
    • pivot:指定值
    • value:缓存值
  • 返回值:返回每次添加成功之后,缓存队列的数量

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result1 = operations.leftPush("l-key", "l-value");
Long result2 = operations.leftPush("l-key", "l-value1");
Long result3 = operations.leftPush("l-key", "l-value2");
Long result = operations.leftPush("l-key", "l-value", "l-value-s");
System.out.println("result = " + result);

输出结果:

result3 = 4

在这里插入图片描述

Long leftPushAll(K key, V... values):批量向缓存键 key 对应缓存队列的头部添加缓存。

  • 参数
    • key:缓存键
    • value:缓存值列表
  • 返回值:返回每次添加成功之后,缓存队列的数量

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result1 = operations.leftPushAll("l-key1", "l-value1", "l-value2", "l-value3");
System.out.println("result = " + result);

输出结果:

result = 3

Long leftPushAll(K key, Collection<V> values):批量向缓存键 key 对应缓存队列的头部添加缓存。

  • 参数
    • key:缓存键
    • value:缓存值集合
  • 返回值:返回每次添加成功之后,缓存队列的数量

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result2 = operations.leftPushAll("l-key2", Arrays.asList("l-value1", "l-value2", "l-value3"));
System.out.println("result = " + result);

输出结果:

result = 3

Long leftPushIfPresent(K key, V value):向缓存键 key 对应缓存队列的头部添加缓存,只有缓存列表存在才成功,否则添加失败。

  • 参数
    • key:缓存键
    • value:缓存值
  • 返回值:返回执行结果,成功为 true,否则为 false

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result1 = operations.leftPushIfPresent("l-key1", "l-value1");
System.out.println("result1 = " + result1);
Long result2 = operations.leftPushIfPresent("l-key1", "l-value-s");
System.out.println("result2 = " + result2);
Long result3 = operations.leftPushIfPresent("l-key-s", "l-value1");
System.out.println("result3 = " + result3);

输出结果:

result1 = 4
result2 = 5
result3 = 0

Long rightPush(K key, V value):向缓存键 key 对应缓存队列的尾部添加缓存。

  • 参数
    • key:缓存键
    • value:缓存值
  • 返回值:返回每次添加成功之后,缓存队列的数量

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result1 = operations.rightPush("l-key", "l-value");
Long result2 = operations.rightPush("l-key", "l-value1");
Long result3 = operations.rightPush("l-key", "l-value2");
System.out.println("result1 = " + result1);
System.out.println("result2 = " + result2);
System.out.println("result3 = " + result3);

输出结果:

result1 = 1
result2 = 2
result3 = 3

在这里插入图片描述

Long rightPush(K key, V pivot, V value):在缓存键 key 对应缓存队列中,将值 value 添加到指定值 pivot 的后面。

  • 参数
    • key:缓存键
    • pivot:
    • value:缓存值
  • 返回值:返回每次添加成功之后,缓存队列的数量

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result1 = operations.rightPush("l-key", "l-value");
Long result2 = operations.rightPush("l-key", "l-value1");
Long result3 = operations.rightPush("l-key", "l-value2");
Long result = operations.rightPush("l-key", "l-value1", "l-value2-s");
System.out.println("result = " + result);

输出结果:

result = 4

在这里插入图片描述

Long rightPushAll(K key, V... values):批量向缓存键 key 对应缓存队列的尾部添加缓存。

  • 参数
    • key:缓存键
    • value:缓存值列表
  • 返回值:返回每次添加成功之后,缓存队列的数量

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result1 = operations.rightPushAll("l-key1", "l-value1", "l-value2", "l-value3");
System.out.println("result = " + result);

输出结果:

result = 3

Long rightPushAll(K key, Collection<V> values):批量向缓存键 key 对应缓存队列的尾部添加缓存。

  • 参数
    • key:缓存键
    • value:缓存值集合
  • 返回值:返回每次添加成功之后,缓存队列的数量

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result2 = operations.rightPushAll("l-key2", Arrays.asList("l-value1", "l-value2", "l-value3"));
System.out.println("result = " + result);

输出结果:

result = 3

Long rightPushIfPresent(K key, V value):向缓存键 key 对应缓存队列的尾部添加缓存,只有缓存列表存在才成功,否则添加失败

  • 参数
    • key:缓存键
    • value:缓存值
  • 返回值:返回每次添加成功之后,缓存队列的数量

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
Long result1 = operations.rightPushIfPresent("l-key1", "l-value1");
System.out.println("result1 = " + result1);
Long result2 = operations.rightPushIfPresent("l-key1", "l-value-s");
System.out.println("result2 = " + result2);
Long result3 = operations.rightPushIfPresent("l-key-s", "l-value1");
System.out.println("result3 = " + result3);

输出结果:

result1 = 4
result2 = 5
result3 = 0

void set(K key, long index, V value):缓存键 key 对应缓存队列中,替换索引 index 对应的值为 value。

  • 参数
    • key:缓存键
    • index:索引位置
    • value:缓存值

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.set("l-key",2,"new-value");

在这里插入图片描述

2.4.2 删除缓存

Long remove(K key, long count, Object value):键 key 的缓存中,移除值 value 出现的前 count 个

  • 参数
    • key:缓存键
    • count:数量
    • value:缓存值
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
Long result = operations.remove("l-key", 2, "aaa");
System.out.println("result = " + result);

输出结果:

result = 2

在这里插入图片描述

V leftPop(K key):从键 key 的缓存中,从头部移除第一个元素。

  • 参数
    • key:缓存键
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
String result = operations.leftPop("l-key");
System.out.println("result = " + result);

输出结果:

result = aaa

V leftPop(K key, long timeout, TimeUnit unit):键 key 的缓存中,从头部移除第一个元素。阻塞连接,直到元素可用或者时间超时。

  • 参数
    • key:缓存键
    • timeout:过期时间
    • unit:时间单位
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
String result = operations.leftPop("l-key",2000, TimeUnit.MILLISECONDS);
System.out.println("result = " + result);

输出结果:

result = aaa

V leftPop(K key, Duration timeout):键 key 的缓存中,从头部移除第一个元素。阻塞连接,直到元素可用或者时间超时。

  • 参数
    • key:缓存键
    • timeout:过期时间,自带单位
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
String result = operations.leftPop("l-key", Duration.ofMillis(2000));
System.out.println("result = " + result);

输出结果:

result = aaa

V rightPop(K key):从键 key 的缓存中,从尾部移除第一个元素。

  • 参数
    • key:缓存键
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
String result = operations.rightPop("l-key");
System.out.println("result = " + result);

输出结果:

result = aaa

V rightPop(K key, long timeout, TimeUnit unit):键 key 的缓存中,从尾部移除第一个元素。阻塞连接,直到元素可用或者时间超时。

  • 参数
    • key:缓存键
    • timeout:过期时间
    • unit:时间单位
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
String result = operations.rightPop("l-key", 2000, TimeUnit.MILLISECONDS);
System.out.println("result = " + result);

输出结果:

result = aaa

V rightPop(K key, Duration timeout):键 key 的缓存中,从尾部移除第一个元素。阻塞连接,直到元素可用或者时间超时。

  • 参数
    • key:缓存键
    • timeout:过期时间,自带单位
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
String result = operations.rightPop("l-key", Duration.ofMillis(2000));
System.out.println("result = " + result);

输出结果:

result = aaa

V rightPopAndLeftPush(K sourceKey, K destinationKey):从键 sourceKey 的缓存尾部移除一个元素,并将此元素添加到键 destinationKey 的缓存头部。

  • 参数
    • sourceKey:源缓存键
    • destinationKey:目标缓存键
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
String result = operations.rightPopAndLeftPush("l-key", "target-key");
System.out.println("result = " + result);

输出结果:

result = aaa

V rightPopAndLeftPush(K sourceKey, K destinationKey, long timeout, TimeUnit unit):从键 sourceKey 的缓存尾部移除一个元素,并将此元素添加到键 destinationKey 的缓存头部。阻塞连接,直到元素可用或者时间超时。

  • 参数
    • sourceKey:源缓存键
    • destinationKey:目标缓存键
    • timeout:过期时间
    • unit:时间单位
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
String result = operations.rightPopAndLeftPush("l-key", "target-key", 2000, TimeUnit.MILLISECONDS);
System.out.println("result = " + result);

输出结果:

result = aaa

V rightPopAndLeftPush(K sourceKey, K destinationKey, Duration timeout):从键 sourceKey 的缓存尾部移除一个元素,并将此元素添加到键 destinationKey 的缓存头部。阻塞连接,直到元素可用或者时间超时。

  • 参数
    • sourceKey:源缓存键
    • destinationKey:目标缓存键
    • timeout:过期时间,自带单位
  • 返回值:

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.rightPushAll("l-key", "aaa", "bbb", "ccc", "aaa", "ddd", "aaa", "eee", "aaa");
String result = operations.rightPopAndLeftPush("l-key", "target-key", Duration.ofMillis(2000));
System.out.println("result = " + result);

输出结果:

result = aaa

2.4.3 获取缓存

List<V> range(K key, long start, long end):获取键 key 的缓存,指定开始索引和结束索引。

  • 参数
    • key:缓存键
    • start:开始位置
    • end:结束位置
  • 返回值:返回取出的结果列表

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
List<String> result1 = operations.range("l-key", 0, -1);
System.out.println("result1 = " + result1);
List<String> result2 = operations.range("l-key", 0, 3);
System.out.println("result2 = " + result2);
List<String> result3 = operations.range("l-key", 2, 5);
System.out.println("result3 = " + result3);

输出结果:

result1 = [ccc, aaa, ddd, aaa, eee, aaa, aaa, bbb, ccc, aaa, ddd, aaa, eee, aaa]
result2 = [ccc, aaa, ddd, aaa]
result3 = [ddd, aaa, eee, aaa]

void trim(K key, long start, long end):截取键 key 对应的缓存,指定开始索引和结束索引。

  • 参数
    • key:缓存键
    • start:开始位置
    • end:结束位置

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
operations.trim("l-key", 0, -1);
System.out.println("result1 = " + operations.range("l-key",0,-1));
operations.trim("l-key", 0, 7);
System.out.println("result2 = " + operations.range("l-key",0,-1));
operations.trim("l-key", 2, 5);
System.out.println("result3 = " + operations.range("l-key",0,-1));

输出结果:

result1 = [ccc, aaa, ddd, aaa, eee, aaa, aaa, bbb, ccc, aaa, ddd, aaa, eee, aaa]
result2 = [ccc, aaa, ddd, aaa, eee, aaa, aaa, bbb]
result3 = [ddd, aaa, eee, aaa]

Long size(K key):统计键key对应缓存的元素个数。

  • 参数
    • key:缓存键
  • 返回值:返回统计结果

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
System.out.println("result2 = " + operations.range("l-key",0,-1));
Long size = operations.size("l-key");
System.out.println("size = " + size);

输出结果:

result2 = [ddd, aaa, eee, aaa]
size = 4

V index(K key, long index):从键 key 对应的缓存中取出索引为 index 的元素。

  • 参数
    • key:缓存键
    • index:
  • 返回值:返回获取结果

示例:

ListOperations<String, String> operations = redisTemplate.opsForList();
System.out.println(operations.range("l-key", 0, -1));
String result1 = operations.index("l-key", 0);
System.out.println("result1 = " + result1);
String result2 = operations.index("l-key", 2);
System.out.println("result2 = " + result2);
String result3 = operations.index("l-key", 4);
System.out.println("result3 = " + result3);

输出结果:

[ddd, aaa, eee, aaa]
result1 = ddd
result2 = eee
result3 = null

2.5 Set 类型 API

2.5.1 添加缓存

Long add(K key, V... values):添加缓存。

  • 参数:
    • key:缓存键
    • values:缓存值列表
  • 返回值:返回添加成功的个数

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
Long result = operations.add("s-key", "aaa", "bbb", "ccc", "ddd");
System.out.println("result = " + result);

输出结果:

result = 4

2.5.2 删除缓存

Long remove(K key, Object... values):移除缓存中的元素。

  • 参数:
    • key:缓存键
    • values:缓存元素列表
  • 返回值:返回删除成功的个数

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
Long result = operations.remove("s-key", "aaa", "vvv");
System.out.println("result = " + result);

输出结果:

result = 1

V pop(K key):从键 key 的缓存中随机移除一个元素,并返回元素的值,key 不存在则返回 null。

  • 参数:
    • key:缓存键
  • 返回值:返回移除元素的值

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
String result = operations.pop("s-key");
System.out.println("result = " + result);

输出结果:

result = ddd

List<V> pop(K key, long count):从键 key 的缓存中随机移除 count 个元素,并返回这些元素的值,redis 版本过低会报错 ERR wrong number of arguments for 'spop' command

  • 参数:
    • key:缓存键
    • count:元素个数
  • 返回值:返回移除成功的元素列表

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key","aaa");
List<String> result = operations.pop("s-key", 3);
System.out.println("result = " + result);

输出结果:

result = [aaa]

2.5.3 获取缓存

Set<V> members(K key):获取缓存键 key 下的所有缓存元素。

  • 参数:
    • key:缓存键
  • 返回值:返回元素集合

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key","aaa","bbb","ccc","ddd");
Set<String> result = operations.members("s-key");
System.out.println("result = " + result);

输出结果:

result = [bbb, aaa, ccc, ddd]

V randomMember(K key):从缓存键 key 的缓存中随机返回一个元素。

  • 参数:
    • key:缓存键
  • 返回值:返回元素

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key","aaa","bbb","ccc","ddd");
String result1 = operations.randomMember("s-key");
System.out.println("result1 = " + result1);
String result2 = operations.randomMember("s-key");
System.out.println("result2 = " + result2);

输出结果:

result1 = ccc
result2 = ddd

List<V> randomMembers(K key, long count):从缓存键 key 的缓存中随机返回 count 个元素,返回元素可重复。

  • 参数:
    • key:缓存键
    • count:元素个数
  • 返回值:返回元素集合

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key","aaa","bbb","ccc","ddd");
List<String> result = operations.randomMembers("s-key",10);
System.out.println("result = " + result);

输出结果:

result = [ddd, ddd, ddd, ccc, ccc, ccc, ccc, bbb, bbb, ddd]

Set<V> distinctRandomMembers(K key, long count):从缓存键 key 的缓存中随机返回 count 个元素,返回元素不可重复。

  • 参数:
    • key:缓存键
    • count:元素个数
  • 返回值:返回元素集合

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key","aaa","bbb","ccc","ddd");
Set<String> result = operations.distinctRandomMembers("s-key",10);
System.out.println("result = " + result);

输出结果:

result = [bbb, ddd, ccc, aaa]

Set<V> difference(K key, K otherKey):比较并返回指定键和其他键对应的缓存中不同的元素,返回的元素为指定缓存键 key 中不同的元素。

  • 参数:
    • key:缓存键
    • otherKey:其他缓存键
  • 返回值:返回结果集

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Set<String> result1 = operations.difference("s-key-1", "s-key-2");
System.out.println("result1 = " + result1);
Set<String> result2 = operations.difference("s-key-1", "s-key-3");
System.out.println("result2 = " + result2);
Set<String> result3 = operations.difference("s-key-2", "s-key-3");
System.out.println("result3 = " + result3);
Set<String> result4 = operations.difference("s-key-3", "s-key-2");
System.out.println("result4 = " + result4);

输出结果:

result1 = [ccc, ddd]
result2 = [ccc, ddd]
result3 = [eee, fff]
result4 = [ggg, hhh]

Set<V> difference(K key, Collection<K> otherKeys):比较并返回指定键和其他键对应的缓存中不同的元素,返回的元素为指定缓存键中不同的元素。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
  • 返回值:返回结果集

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Set<String> result1 = operations.difference("s-key-1", Arrays.asList("s-key-2","s-key-3"));
System.out.println("result1 = " + result1);
Set<String> result2 = operations.difference("s-key-2",  Arrays.asList("s-key-1","s-key-3"));
System.out.println("result2 = " + result2);
Set<String> result3 = operations.difference("s-key-3",  Arrays.asList("s-key-1","s-key-2"));
System.out.println("result3 = " + result3);
Set<String> result4 = operations.difference("s-key-3",  Arrays.asList("s-key-2","s-key-1"));
System.out.println("result4 = " + result4);

输出结果:

result1 = [ccc, ddd]
result2 = [eee, fff]
result3 = [ggg, hhh]
result4 = [ggg, hhh]

Set<V> difference(Collection<K> keys):比较缓存键集合中各个键对应的缓存中不同的元素,返回的元素为缓存键集合中第一个键对应的缓存中不同的元素。

  • 参数:
    • keys:缓存键集合
  • 返回值:返回结果集

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Set<String> result1 = operations.difference(Arrays.asList("s-key-1","s-key-2","s-key-3"));
System.out.println("result1 = " + result1);
Set<String> result2 = operations.difference(Arrays.asList("s-key-2","s-key-1","s-key-3"));
System.out.println("result2 = " + result2);
Set<String> result3 = operations.difference(Arrays.asList("s-key-3","s-key-1","s-key-2"));
System.out.println("result3 = " + result3);
Set<String> result4 = operations.difference(Arrays.asList("s-key-3","s-key-2","s-key-1"));
System.out.println("result4 = " + result4);

输出结果:

result1 = [ccc, ddd]
result2 = [eee, fff]
result3 = [ggg, hhh]
result4 = [ggg, hhh]

Long differenceAndStore(K key, K otherKey, K destKey):比较指定键和其他键对应的缓存中不同的元素,并将不同的元素另外缓存,缓存的元素为指定缓存键中不同的元素。

  • 参数:
    • key:缓存键
    • otherKey:其他缓存键
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Long result = operations.differenceAndStore("s-key-1", "s-key-2", "new-key");
System.out.println("result = " + result);
Set<String> members = operations.members("new-key");
System.out.println("new-key = " + members);

输出结果:

result = 2
new-key = [ccc, ddd]

Long differenceAndStore(K key, Collection<K> otherKeys, K destKey):比较指定键和其他键对应的缓存中不同的元素,并将不同的元素另外缓存,缓存的元素为指定缓存键中不同的元素。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Long result = operations.differenceAndStore("s-key-1", Arrays.asList("s-key-2","s-key-3"), "new-key");
System.out.println("result = " + result);
Set<String> members = operations.members("new-key");
System.out.println("new-key = " + members);

输出结果:

result = 2
new-key = [ccc, ddd]

Long differenceAndStore(Collection<K> keys, K destKey):比较缓存键集合中各个键对应的缓存中不同的元素,并将不同的元素另外缓存,缓存的元素为缓存键集合中第一个键对应的缓存中不同的元素。

  • 参数:
    • keys:缓存键集合
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Long result = operations.differenceAndStore( Arrays.asList("s-key-1","s-key-2","s-key-3"), "new-key");
System.out.println("result = " + result);
Set<String> members = operations.members("new-key");
System.out.println("new-key = " + members);

输出结果:

result = 2
new-key = [ccc, ddd]

Set<V> intersect(K key, K otherKey):取交集,比较并返回指定键和其他键对应的缓存中相同的元素,返回的元素为指定缓存键 key 中相同的元素。

  • 参数:
    • key:缓存键
    • otherKey:其他缓存键
  • 返回值:返回结果集

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Set<String> result= operations.intersect("s-key-1", "s-key-2");
System.out.println("result = " + result);

输出结果:

result = [aaa, bbb]

Set<V> intersect(K key, Collection<K> otherKeys):取交集,比较并返回指定键和其他键对应的缓存中相同的元素,返回的元素为指定缓存键中相同的元素。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
  • 返回值:返回结果集

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Set<String> result= operations.intersect("s-key-1", Arrays.asList("s-key-2", "s-key-3"));
System.out.println("result = " + result);

输出结果:

result = [aaa, bbb]

Set<V> intersect(Collection<K> keys):取交集,比较缓存键集合中各个键对应的缓存中相同的元素,返回的元素为缓存键集合中第一个键对应的缓存中相同的元素。

  • 参数:
    • keys:缓存键集合
  • 返回值:返回结果集

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Set<String> result= operations.intersect(Arrays.asList("s-key-1", "s-key-2", "s-key-3"));
System.out.println("result = " + result);

输出结果:

result = [aaa, bbb]

Long intersectAndStore(K key, K otherKey, K destKey):取交集,比较指定键和其他键对应的缓存中相同的元素,并将相同的元素另外缓存,缓存的元素为指定缓存键中相同的元素。

  • 参数:
    • key:缓存键
    • otherKey:其他缓存键
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Long result = operations.intersectAndStore("s-key-1", "s-key-2", "new-key");
System.out.println("result = " + result);
Set<String> members = operations.members("new-key");
System.out.println("new-key = " + members);

输出结果:

result = 2
new-key = [bbb, aaa]

Long intersectAndStore(K key, Collection<K> otherKeys, K destKey):取交集,比较指定键和其他键对应的缓存中相同的元素,并将相同的元素另外缓存,缓存的元素为指定缓存键中相同的元素。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Long result = operations.intersectAndStore("s-key-1", Arrays.asList("s-key-2", "s-key-3"), "new-key");
System.out.println("result = " + result);
Set<String> members = operations.members("new-key");
System.out.println("new-key = " + members);

输出结果:

result = 2
new-key = [bbb, aaa]

Long intersectAndStore(Collection<K> keys, K destKey):取交集,比较缓存键集合中各个键对应的缓存中相同的元素,并将相同的元素另外缓存,缓存的元素为缓存键集合中第一个键对应的缓存中相同的元素。

  • 参数:
    • keys:缓存键集合
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Long result = operations.intersectAndStore( Arrays.asList("s-key-1", "s-key-2", "s-key-3"), "new-key");
System.out.println("result = " + result);
Set<String> members = operations.members("new-key");
System.out.println("new-key = " + members);

输出结果:

result = 2
new-key = [bbb, aaa]

Set<V> union(K key, K otherKey):取并集,比较并返回指定键和其他键对应的缓存中所有的元素,会去重

  • 参数:
    • key:缓存键
    • otherKey:其他缓存键
  • 返回值:返回结果集

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Set<String> result= operations.union("s-key-1", "s-key-2");
System.out.println("result = " + result);

输出结果:

result = [hhh, eee, aaa, bbb, fff, ccc, ddd, ggg]

Set<V> union(K key, Collection<K> otherKeys):取并集,比较并返回指定键和其他键对应的缓存中所有的元素,会去重。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
  • 返回值:返回结果集

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Set<String> result= operations.union("s-key-1", Arrays.asList("s-key-2", "s-key-3"));
System.out.println("result = " + result);

输出结果:

result = [hhh, eee, aaa, bbb, fff, ccc, ddd, ggg]

Set<V> union(Collection<K> keys):取并集,比较缓存键集合中各个键对应的缓存中所有的元素,会去重。

  • 参数:
    • keys:缓存键集合
  • 返回值:返回结果集

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Set<String> result= operations.union(Arrays.asList("s-key-1", "s-key-2", "s-key-3"));
System.out.println("result = " + result);

输出结果:

result = [hhh, eee, aaa, bbb, fff, ccc, ddd, ggg]

Long unionAndStore(K key, K otherKey, K destKey):取并集,比较指定键和其他键对应的缓存中所有的元素,会去重,并将合并的元素另外缓存。

  • 参数:
    • key:缓存键
    • otherKey:其他缓存键
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Long result = operations.unionAndStore("s-key-1", "s-key-2", "new-key");
System.out.println("result = " + result);
Set<String> members = operations.members("new-key");
System.out.println("members = " + members);

输出结果:

result = 6
members = [eee, aaa, bbb, fff, ccc, ddd]

Long unionAndStore(K key, Collection<K> otherKeys, K destKey):取并集,比较指定键和其他键对应的缓存中所有的元素,会去重,并将合并的元素另外缓存。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Long result = operations.unionAndStore("s-key-1", Arrays.asList("s-key-2", "s-key-3"), "new-key");
System.out.println("result = " + result);
Set<String> members = operations.members("new-key");
System.out.println("members = " + members);

输出结果:

result = 8
members = [hhh, eee, aaa, bbb, fff, ccc, ddd, ggg]

Long unionAndStore(Collection<K> keys, K destKey):取并集,比较缓存键集合中各个键对应的缓存中所有的元素,会去重,并将合并的元素另外缓存。

  • 参数:
    • keys:缓存键集合
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key-1", "aaa", "bbb", "ccc", "ddd");
operations.add("s-key-2", "aaa", "bbb", "eee", "fff");
operations.add("s-key-3", "aaa", "bbb", "ggg", "hhh");
Long result = operations.unionAndStore(Arrays.asList("s-key-1", "s-key-2", "s-key-3"), "new-key");
System.out.println("result = " + result);
Set<String> members = operations.members("new-key");
System.out.println("members = " + members);

输出结果:

result = 8
members = [hhh, eee, aaa, bbb, fff, ccc, ddd, ggg]

Cursor<V> scan(K key, ScanOptions options):从键 key 的缓存中匹配查找元素

  • 参数:
    • key:缓存键
    • options:匹配规则
  • 返回值:返回匹配到的元素

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key", "aaa", "aab", "aac", "bba", "bbb", "bbc", "cca", "ccb", "ccc");
Cursor<String> cursor1 = operations.scan("s-key", ScanOptions.NONE);
while (cursor1.hasNext()) {System.out.print(cursor1.next() + ",");
}
System.out.println("");
Cursor<String> cursor2 = operations.scan("s-key", ScanOptions.scanOptions().match("aa*").build());
while (cursor2.hasNext()) {System.out.print(cursor2.next() + ",");
}
try {cursor1.close();cursor2.close();
} catch (IOException e) {throw new RuntimeException(e);
}

输出结果:

aaa,bbb,aab,cca,ccc,bbc,aac,ddd,bba,ccb,
aaa,aab,aac,

2.5.4 其他操作

Boolean move(K key, V value, K destKey):从键 key 的缓存中取出元素 value 并添加到目标缓存键 destKey 的缓存中。

  • 参数:
    • key:缓存键
  • 返回值:返回操作结果,成功为 true,失败为 false

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key", "aaa", "bbb", "ccc", "ddd");
System.out.println(operations.members("s-key"));
Boolean result1 = operations.move("s-key", "ccc", "target-key");
System.out.println("result1 = " + result1);
Boolean result2 = operations.move("s-key", "zzz", "target-key");
System.out.println("result2 = " + result2);
System.out.println("s-key = " + operations.members("s-key"));
System.out.println("target-key = " + operations.members("target-key"));

输出结果:

[aaa, bbb, ccc, ddd]
result1 = true
result2 = false
s-key = [aaa, bbb, ddd]
target-key = [ccc]

Long size(K key):统计键 key 的缓存元素个数。

  • 参数:
    • key:缓存键
  • 返回值:返回统计个数

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key", "aaa", "bbb", "ccc", "ddd");
Long result = operations.size("s-key");
System.out.println("result = " + result);

输出结果:

result = 4

Boolean isMember(K key, Object o):判断指定元素在键 key 的缓存是否存在。

  • 参数:
    • key:缓存键
    • o:指定缓存元素
  • 返回值:返回判断结果,存在为 true,不存在为 false

示例:

SetOperations<String, String> operations = redisTemplate.opsForSet();
operations.add("s-key", "aaa", "bbb", "ccc", "ddd");
Boolean result1 = operations.isMember("s-key","aaa");
System.out.println("result1 = " + result1);
Boolean result2 = operations.isMember("s-key","zzz");
System.out.println("result2 = " + result2);

输出结果:

result1 = true
result2 = false

2.6 ZSet 类型 API

2.6.1 添加缓存

Boolean add(K key, V value, double score):添加缓存,如果 value 已经存在,会更新分数 score,但是返回结果为 false。

  • 参数:
    • key:缓存键
    • value:缓存值,
    • score:缓存值对应的分数
  • 返回值:返回添加结果,成功为 true,失败为 false

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Boolean result1 = operations.add("zs-key", "aaa", 40);
System.out.println("result1 = " + result1);
Boolean result2 = operations.add("zs-key", "bbb", 23);
System.out.println("result2 = " + result2);
Boolean result3 = operations.add("zs-key", "ccc", 56);
System.out.println("result3 = " + result3);
Boolean result4 = operations.add("zs-key", "aaa", 66);
System.out.println("result4 = " + result4 );

输出结果:

result1 = true
result2 = true
result3 = true
result4 = false

在这里插入图片描述

Long add(K key, Set<TypedTuple<V>> tuples):添加缓存,如果 value 已经存在,会更新分数 score,但是返回结果总数量不会加1。

  • 参数:
    • key:缓存键
    • tuples:缓存内容
  • 返回值:返回添加成功数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("aaa",77.7));
Long result = operations.add("zs-key", tuples);
System.out.println("result = " + result);

输出结果:

result = 3

2.6.2 删除缓存

Long remove(K key, Object... values):删除缓存。

  • 参数:
    • key:缓存键
  • 返回值:返回删除成功数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("aaa", 77.7));
operations.add("zs-key", tuples);
Long result = operations.remove("zs-key", "aaa", "bbb", "zzz");
System.out.println("result = " + result);

输出结果:

result = 2

Long removeRange(K key, long start, long end):根据索引排序后,删除指定索引范围内的缓存,指定开始索引和结束索引。

  • 参数:
    • key:缓存键
    • start:开始索引
    • end:结束索引
  • 返回值:返回删除成功数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Long result = operations.removeRange("zs-key", 2, 4);
System.out.println("result = " + result);

输出结果:

result = 3

Long removeRangeByScore(K key, double min, double max):根据分数排序后,删除指定范围内的缓存,指定最小分数和最大分数,包含指定分数值。

  • 参数:
    • key:缓存键
    • min:最小分数
    • max:最大分数
  • 返回值:

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Long result = operations.removeRangeByScore("zs-key", 30,60);
System.out.println("result = " + result);

输出结果:

result = 3

2.6.3 获取缓存

Set<V> range(K key, long start, long end):获取缓存元素,指定开始索引和结束索引

  • 参数:
    • key:缓存键
    • start:开始索引
    • end:结束索引
  • 返回值:

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<String> result1 = operations.range("zs-key", 0, -1);
System.out.println("result1 = " + result1);
Set<String> result2 = operations.range("zs-key", 0, 4);
System.out.println("result2 = " + result2);
Set<String> result3 = operations.range("zs-key", 2, 5);
System.out.println("result3 = " + result3);

输出结果:

result1 = [ccc, bbb, ddd, aaa, fff, eee]
result2 = [ccc, bbb, ddd, aaa, fff]
result3 = [ddd, aaa, fff, eee]

Set<TypedTuple<V>> rangeWithScores(K key, long start, long end):先根据分数排序,然后从缓存取出指定索引范围内的缓存元素及元素对应的分数,指定开始索引和结束索引。

  • 参数:
    • key:缓存键
    • start:开始索引
    • end:结束索引
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<ZSetOperations.TypedTuple<String>> result1 = operations.rangeWithScores("zs-key", 0, -1);
for (ZSetOperations.TypedTuple<String> next : result1) {System.out.println(next.getValue() + " - " + next.getScore());
}
System.out.println("----------------");
Set<ZSetOperations.TypedTuple<String>> result2 = operations.rangeWithScores("zs-key", 2, 4);
for (ZSetOperations.TypedTuple<String> next : result2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

ccc - 12.9
bbb - 23.0
ddd - 42.2
aaa - 45.5
fff - 45.7
eee - 77.7
----------------
ddd - 42.2
aaa - 45.5
fff - 45.7

Set<V> rangeByScore(K key, double min, double max):先根据分数排序,然后从缓存取出指定分数范围内的缓存元素。

  • 参数:
    • key:缓存键
    • min:最小分数
    • max:最大分数
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<String> result1 = operations.rangeByScore("zs-key", 0, 100);
System.out.println("result1 = " + result1);
Set<String> result2 = operations.rangeByScore("zs-key", 45.5, 100);
System.out.println("result2 = " + result2);
Set<String> result3 = operations.rangeByScore("zs-key", 0, 45.5);
System.out.println("result3 = " + result3);

输出结果:

result1 = [ccc, bbb, ddd, aaa, fff, eee]
result2 = [aaa, fff, eee]
result3 = [ccc, bbb, ddd, aaa]

Set<V> rangeByScore(K key, double min, double max, long offset, long count)

  • 参数:
    • key:缓存键
    • min:最小分数
    • max:最大分数
    • offset:偏移量
    • count:数量
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<String> result1 = operations.rangeByScore("zs-key", 0, 100, 1, 1);
System.out.println("result1 = " + result1);
Set<String> result2 = operations.rangeByScore("zs-key", 45.5, 100, 1, 1);
System.out.println("result2 = " + result2);
Set<String> result3 = operations.rangeByScore("zs-key", 0, 45.5, 1, 1);
System.out.println("result3 = " + result3);

输出结果:

result1 = [bbb]
result2 = [fff]
result3 = [bbb]

Set<TypedTuple<V>> rangeByScoreWithScores(K key, double min, double max):先根据分数排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数。

  • 参数:
    • key:缓存键
    • min:最小分数
    • max:最大分数
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<ZSetOperations.TypedTuple<String>> result1 = operations.rangeByScoreWithScores("zs-key", 0, 100);
for (ZSetOperations.TypedTuple<String> next : result1) {System.out.println(next.getValue() + " - " + next.getScore());
}
System.out.println("----------------");
Set<ZSetOperations.TypedTuple<String>> result2 = operations.rangeByScoreWithScores("zs-key", 20, 50);
for (ZSetOperations.TypedTuple<String> next : result2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

ccc - 12.9
bbb - 23.0
ddd - 42.2
aaa - 45.5
fff - 45.7
eee - 77.7
----------------
bbb - 23.0
ddd - 42.2
aaa - 45.5
fff - 45.7

Set<TypedTuple<V>> rangeByScoreWithScores(K key, double min, double max):先根据分数排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数,指定偏移量和返回个数。

  • 参数:
    • key:缓存键
    • min:最小分数
    • max:最大分数
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<ZSetOperations.TypedTuple<String>> result1 = operations.rangeByScoreWithScores("zs-key", 0, 100,1,3);
for (ZSetOperations.TypedTuple<String> next : result1) {System.out.println(next.getValue() + " - " + next.getScore());
}
System.out.println("----------------");
Set<ZSetOperations.TypedTuple<String>> result2 = operations.rangeByScoreWithScores("zs-key", 20, 50,1,2);
for (ZSetOperations.TypedTuple<String> next : result2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

bbb - 23.0
ddd - 42.2
aaa - 45.5
----------------
ddd - 42.2
aaa - 45.5

Set<V> reverseRange(K key, long start, long end):获取缓存元素,指定开始索引和结束索引,根据索引倒序排序。

  • 参数:
    • key:缓存键
    • start:开始索引
    • end:结束索引
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<String> result1 = operations.reverseRange("zs-key", 0, -1);
System.out.println("result1 = " + result1);
Set<String> result2 = operations.reverseRange("zs-key", 0, 3);
System.out.println("result2 = " + result2);
Set<String> result3 = operations.reverseRange("zs-key", 2, 5);
System.out.println("result3 = " + result3);

输出结果:

result1 = [eee, fff, aaa, ddd, bbb, ccc]
result2 = [eee, fff, aaa, ddd]
result3 = [aaa, ddd, bbb, ccc]

Set<TypedTuple<V>> reverseRangeWithScores(K key, long start, long end):先根据分数倒序排序,然后从缓存取出指定索引范围内的缓存元素及元素对应的分数,指定开始索引和结束索引。

  • 参数:
    • key:缓存键
    • start:开始索引
    • end:结束索引
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<ZSetOperations.TypedTuple<String>> result1 = operations.reverseRangeWithScores("zs-key", 0, -1);
for (ZSetOperations.TypedTuple<String> next : result1) {System.out.println(next.getValue() + " - " + next.getScore());
}
System.out.println("----------------");
Set<ZSetOperations.TypedTuple<String>> result2 = operations.reverseRangeWithScores("zs-key", 2, 5);
for (ZSetOperations.TypedTuple<String> next : result2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

eee - 77.7
fff - 45.7
aaa - 45.5
ddd - 42.2
bbb - 23.0
ccc - 12.9
----------------
aaa - 45.5
ddd - 42.2
bbb - 23.0
ccc - 12.9

Set<V> reverseRangeByScore(K key, double min, double max):先根据分数倒序排序,然后从缓存取出指定分数范围内的缓存元素。

  • 参数:
    • key:缓存键
    • min:最小分数
    • max:最大分数
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<String> result1 = operations.reverseRangeByScore("zs-key", 0, 100);
System.out.println("result1 = " + result1);
Set<String> result2 = operations.reverseRangeByScore("zs-key", 20, 50);
System.out.println("result2 = " + result2);

输出结果:

result1 = [eee, fff, aaa, ddd, bbb, ccc]
result2 = [fff, aaa, ddd, bbb]

Set<V> reverseRangeByScore(K key, double min, double max):先根据分数倒序排序,然后从缓存取出指定分数范围内的缓存元素,指定偏移量和返回个数。

  • 参数:
    • key:缓存键
    • min:最小分数
    • max:最大分数
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<String> result1 = operations.reverseRangeByScore("zs-key", 0, 100, 1, 3);
System.out.println("result1 = " + result1);
Set<String> result2 = operations.reverseRangeByScore("zs-key", 20, 50, 1, 2);
System.out.println("result2 = " + result2);

输出结果:

result1 = [fff, aaa, ddd]
result2 = [aaa, ddd]

Set<TypedTuple<V>> reverseRangeByScoreWithScores(K key, double min, double max):先根据分数倒序排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数。

  • 参数:
    • key:缓存键
    • min:最小分数
    • max:最大分数
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<ZSetOperations.TypedTuple<String>> result1 = operations.reverseRangeByScoreWithScores("zs-key", 0, 100);
for (ZSetOperations.TypedTuple<String> next : result1) {System.out.println(next.getValue() + " - " + next.getScore());
}
System.out.println("----------------");
Set<ZSetOperations.TypedTuple<String>> result2 = operations.reverseRangeByScoreWithScores("zs-key", 20, 50);
for (ZSetOperations.TypedTuple<String> next : result2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

eee - 77.7
fff - 45.7
aaa - 45.5
ddd - 42.2
bbb - 23.0
ccc - 12.9
----------------
fff - 45.7
aaa - 45.5
ddd - 42.2
bbb - 23.0

Set<TypedTuple<V>> reverseRangeByScoreWithScores(K key, double min, double max):先根据分数倒序排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数,指定偏移量和返回个数。

  • 参数:
    • key:缓存键
    • min:最小分数
    • max:最大分数
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 45.5));
tuples.add(new DefaultTypedTuple<>("bbb", 23.0));
tuples.add(new DefaultTypedTuple<>("ccc", 12.9));
tuples.add(new DefaultTypedTuple<>("ddd", 42.2));
tuples.add(new DefaultTypedTuple<>("eee", 77.7));
tuples.add(new DefaultTypedTuple<>("fff", 45.7));
operations.add("zs-key", tuples);
Set<ZSetOperations.TypedTuple<String>> result1 = operations.reverseRangeByScoreWithScores("zs-key", 0, 100, 1, 3);
for (ZSetOperations.TypedTuple<String> next : result1) {System.out.println(next.getValue() + " - " + next.getScore());
}
System.out.println("----------------");
Set<ZSetOperations.TypedTuple<String>> result2 = operations.reverseRangeByScoreWithScores("zs-key", 20, 50, 1, 2);
for (ZSetOperations.TypedTuple<String> next : result2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

fff - 45.7
aaa - 45.5
ddd - 42.2
----------------
aaa - 45.5
ddd - 42.2

Set<V> rangeByLex(K key, Range range):从缓存中取出在范围range内的元素,并按照缓存元素value的字典顺序排序。这个排序只有在有相同分数的情况下才能使用,如果有不同的分数则返回值不确定。

  • 参数:
    • key:缓存键
    • range:范围,内容为缓存元素value的值
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 11.0));
tuples.add(new DefaultTypedTuple<>("bbb", 11.0));
tuples.add(new DefaultTypedTuple<>("ccc", 11.0));
tuples.add(new DefaultTypedTuple<>("ddd", 11.0));
tuples.add(new DefaultTypedTuple<>("eee", 11.0));
tuples.add(new DefaultTypedTuple<>("fff", 11.0));
operations.add("zs-key", tuples);
Set<String> result1 = operations.rangeByLex("zs-key", RedisZSetCommands.Range.range().lt("ccc"));
System.out.println("result1 = " + result1);
Set<String> result2 = operations.rangeByLex("zs-key", RedisZSetCommands.Range.range().gt("ccc"));
System.out.println("result2 = " + result2);

输出结果:

result1 = [aaa, bbb]
result2 = [ddd, eee, fff]

Set<V> rangeByLex(K key, Range range, Limit limit):从缓存中取出在范围range内的元素,并按照缓存元素value的字典顺序排序,限制返回元素个数。这个排序只有在有相同分数的情况下才能使用,如果有不同的分数则返回值不确定。

  • 参数:
    • key:缓存键
    • range:范围,内容为缓存元素value的值
    • limit:限制返回元素个数
  • 返回值:返回结果集合

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 11.0));
tuples.add(new DefaultTypedTuple<>("bbb", 11.0));
tuples.add(new DefaultTypedTuple<>("ccc", 11.0));
tuples.add(new DefaultTypedTuple<>("ddd", 11.0));
tuples.add(new DefaultTypedTuple<>("eee", 11.0));
tuples.add(new DefaultTypedTuple<>("fff", 11.0));
operations.add("zs-key", tuples);
Set<String> result1 = operations.rangeByLex("zs-key", RedisZSetCommands.Range.range().lt("ddd"), RedisZSetCommands.Limit.limit().offset(1).count(2));
System.out.println("result1 = " + result1);
Set<String> result2 = operations.rangeByLex("zs-key", RedisZSetCommands.Range.range().gt("ccc"), RedisZSetCommands.Limit.limit().offset(1).count(2));
System.out.println("result2 = " + result2);

输出结果:

result1 = [bbb, ccc]
result2 = [eee, fff]

Long rank(K key, Object o):根据分数排序,获取指定元素在缓存中的位置索引。

  • 参数:
    • key:缓存键
    • value:指定分数
  • 返回值:返回索引值

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples.add(new DefaultTypedTuple<>("ddd", 21.4));
tuples.add(new DefaultTypedTuple<>("eee", 61.5));
tuples.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key", tuples);
Long result1 = operations.rank("zs-key", "ccc");
System.out.println("result1 = " + result1);
Long result2 = operations.rank("zs-key", "aaa");
System.out.println("result2 = " + result2);

输出结果:

result1 = 5
result2 = 0

Long reverseRank(K key, Object o):根据分数倒序排序,获取指定元素在缓存中的位置索引。

  • 参数:
    • key:缓存键
    • value:指定分数
  • 返回值:返回索引值

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples.add(new DefaultTypedTuple<>("ddd", 21.4));
tuples.add(new DefaultTypedTuple<>("eee", 61.5));
tuples.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key", tuples);
Long result1 = operations.reverseRank("zs-key", "ccc");
System.out.println("result1 = " + result1);
Long result2 = operations.reverseRank("zs-key", "aaa");
System.out.println("result2 = " + result2);

输出结果:

result1 = 0
result2 = 5

Double score(K key, Object o):获取缓存中某个元素的分数。

  • 参数:
    • key:缓存键
    • value:指定分数
  • 返回值:返回元素的分数

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples.add(new DefaultTypedTuple<>("ddd", 21.4));
tuples.add(new DefaultTypedTuple<>("eee", 61.5));
tuples.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key", tuples);
Double result1 = operations.score("zs-key", "ccc");
System.out.println("result1 = " + result1);
Double result2 = operations.score("zs-key", "aaa");
System.out.println("result2 = " + result2);

输出结果:

result1 = 71.3
result2 = 11.1

Double incrementScore(K key, V value, double delta):将缓存中某个值的分组增加 delta,返回增加后的分数结果。

  • 参数:
    • key:缓存键
    • value:指定分数
    • delta:增加值
  • 返回值:

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples.add(new DefaultTypedTuple<>("ddd", 21.4));
tuples.add(new DefaultTypedTuple<>("eee", 61.5));
tuples.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key", tuples);
System.out.println("ccc = " + operations.score("zs-key", "ccc"));
Double result1 = operations.incrementScore("zs-key", "ccc", 22);
System.out.println("result1 = " + result1);
System.out.println("ccc = " + operations.score("zs-key", "ccc"));

输出结果:

ccc = 71.3
result1 = 93.3
ccc = 93.3

Long size(K key):获取缓存中元素的个数。

  • 参数:
    • key:缓存键
  • 返回值:返回统计结果

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples.add(new DefaultTypedTuple<>("ddd", 21.4));
tuples.add(new DefaultTypedTuple<>("eee", 61.5));
tuples.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key", tuples);
Long result1 = operations.size("zs-key");
System.out.println("result1 = " + result1);

输出结果:

result1 = 6

Long zCard(K key):获取缓存中元素的个数。

  • 参数:
    • key:缓存键
  • 返回值:返回统计结果

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples.add(new DefaultTypedTuple<>("ddd", 21.4));
tuples.add(new DefaultTypedTuple<>("eee", 61.5));
tuples.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key", tuples);
Long result1 = operations.zCard("zs-key");
System.out.println("result1 = " + result1);

输出结果:

result1 = 6

Long count(K key, double min, double max):获取指定分数范围缓存中元素的个数,指定最低分数和最高分数。

  • 参数:
    • key:缓存键
  • 返回值:返回统计结果

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples = new HashSet<>();
tuples.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples.add(new DefaultTypedTuple<>("ddd", 21.4));
tuples.add(new DefaultTypedTuple<>("eee", 61.5));
tuples.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key", tuples);
Long result1 = operations.count("zs-key",0,100);
System.out.println("result1 = " + result1);
Long result2 = operations.count("zs-key",20,50);
System.out.println("result2 = " + result2);

输出结果:

result1 = 6
result2 = 2

Long intersectAndStore(K key, K otherKey, K destKey):两个集合取交集,并将交集结果保存到指定缓存键中,交集结果中,相同的值分数会直接累加。

  • 参数:
    • key:缓存键
    • otherKey:其他缓存键
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples1 = new HashSet<>();
tuples1.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples1.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples1.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples1.add(new DefaultTypedTuple<>("ddd", 21.4));
operations.add("zs-key-1", tuples1);
Set<ZSetOperations.TypedTuple<String>> tuples2 = new HashSet<>();
tuples2.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples2.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples2.add(new DefaultTypedTuple<>("eee", 61.5));
tuples2.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-2", tuples2);Long result = operations.intersectAndStore("zs-key-1", "zs-key-2", "zs-key-new");
System.out.println("result = " + result);Set<ZSetOperations.TypedTuple<String>> tuples = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tuples) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

result = 2
aaa - 22.2
bbb - 82.4

Long intersectAndStore(K key, Collection<K> otherKeys, K destKey):多个集合取交集,并将交集结果保存到指定缓存键中,交集结果中,相同的值分数会直接累加。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples1 = new HashSet<>();
tuples1.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples1.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples1.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples1.add(new DefaultTypedTuple<>("ddd", 21.4));
operations.add("zs-key-1", tuples1);
Set<ZSetOperations.TypedTuple<String>> tuples2 = new HashSet<>();
tuples2.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples2.add(new DefaultTypedTuple<>("bbb", 21.20));
tuples2.add(new DefaultTypedTuple<>("eee", 61.5));
tuples2.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-2", tuples2);
Set<ZSetOperations.TypedTuple<String>> tuples3 = new HashSet<>();
tuples3.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples3.add(new DefaultTypedTuple<>("ccc", 41.20));
tuples3.add(new DefaultTypedTuple<>("ddd", 61.5));
tuples3.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-3", tuples3);Long result = operations.intersectAndStore("zs-key-1", Arrays.asList("zs-key-2","zs-key-3"), "zs-key-new");
System.out.println("result = " + result);Set<ZSetOperations.TypedTuple<String>> tuples = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tuples) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

result = 1
aaa - 33.3

Long intersectAndStore(K key, Collection<K> otherKeys, K destKey, Aggregate aggregate):多个集合取交集,并将交集结果保存到指定缓存键中,使用指定聚合函数获取交集元素,交集结果中,相同的值分数会直接累加。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
    • destKey:目标缓存键
    • aggregate:聚合函数
      • MAX:结果集中相同的值的分数取最大值
      • MIN:结果集中相同的值的分数取最小值
      • SUM:结果集中相同的值的分数相加
  • 返回值:返回结果集数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples1 = new HashSet<>();
tuples1.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples1.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples1.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples1.add(new DefaultTypedTuple<>("ddd", 21.4));
operations.add("zs-key-1", tuples1);
Set<ZSetOperations.TypedTuple<String>> tuples2 = new HashSet<>();
tuples2.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples2.add(new DefaultTypedTuple<>("bbb", 21.20));
tuples2.add(new DefaultTypedTuple<>("ddd", 61.5));
tuples2.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-2", tuples2);
Set<ZSetOperations.TypedTuple<String>> tuples3 = new HashSet<>();
tuples3.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples3.add(new DefaultTypedTuple<>("ccc", 41.20));
tuples3.add(new DefaultTypedTuple<>("ddd", 61.5));
tuples3.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-3", tuples3);Long result1 = operations.intersectAndStore("zs-key-1", Arrays.asList("zs-key-2","zs-key-3"), "zs-key-new",RedisZSetCommands.Aggregate.MIN);
System.out.println("result1 = " + result1);Set<ZSetOperations.TypedTuple<String>> tupleSet1 = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tupleSet1) {System.out.println(next.getValue() + " - " + next.getScore());
}Long result2 = operations.intersectAndStore("zs-key-1", Arrays.asList("zs-key-2","zs-key-3"), "zs-key-new",RedisZSetCommands.Aggregate.MAX);
System.out.println("result2 = " + result2 );Set<ZSetOperations.TypedTuple<String>> tupleSet2 = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tupleSet2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

result1 = 2
aaa - 11.1
ddd - 21.4
result2 = 2
aaa - 31.1
ddd - 61.5

Long intersectAndStore(K key, Collection<K> otherKeys, K destKey, Aggregate aggregate, Weights weights):多个集合取交集,并将交集结果保存到指定缓存键中,使用指定聚合函数和权重来计算和获取交集元素,权重值会和元素的分数相乘,交集结果中,相同的值分数会直接累加。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
    • destKey:目标缓存键
    • aggregate:聚合函数
    • weights:权重
  • 返回值:返回结果集数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples1 = new HashSet<>();
tuples1.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples1.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples1.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples1.add(new DefaultTypedTuple<>("ddd", 21.4));
operations.add("zs-key-1", tuples1);
Set<ZSetOperations.TypedTuple<String>> tuples2 = new HashSet<>();
tuples2.add(new DefaultTypedTuple<>("aaa", 31.1));
tuples2.add(new DefaultTypedTuple<>("bbb", 21.20));
tuples2.add(new DefaultTypedTuple<>("ddd", 61.5));
tuples2.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-2", tuples2);
Set<ZSetOperations.TypedTuple<String>> tuples3 = new HashSet<>();
tuples3.add(new DefaultTypedTuple<>("aaa", 21.1));
tuples3.add(new DefaultTypedTuple<>("ccc", 41.20));
tuples3.add(new DefaultTypedTuple<>("ddd", 41.5));
tuples3.add(new DefaultTypedTuple<>("fff", 61.6));
operations.add("zs-key-3", tuples3);Long result1 = operations.intersectAndStore("zs-key-1", Arrays.asList("zs-key-2", "zs-key-3"), "zs-key-new",RedisZSetCommands.Aggregate.MIN, RedisZSetCommands.Weights.of(2, 3, 5));
System.out.println("result1 = " + result1);Set<ZSetOperations.TypedTuple<String>> tupleSet1 = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tupleSet1) {System.out.println(next.getValue() + " - " + next.getScore());
}Long result2 = operations.intersectAndStore("zs-key-1", Arrays.asList("zs-key-2", "zs-key-3"), "zs-key-new",RedisZSetCommands.Aggregate.MAX, RedisZSetCommands.Weights.of(2, 3, 5));
System.out.println("result2 = " + result2);Set<ZSetOperations.TypedTuple<String>> tupleSet2 = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tupleSet2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

result1 = 2
aaa - 22.2
ddd - 42.8
result2 = 2
aaa - 105.5
ddd - 207.5

Long unionAndStore(K key, K otherKey, K destKey):两个集合取并集,并将并集结果保存到指定缓存键中。

  • 参数:
    • key:缓存键
    • otherKey:其他缓存键
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples1 = new HashSet<>();
tuples1.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples1.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples1.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples1.add(new DefaultTypedTuple<>("ddd", 21.4));
operations.add("zs-key-1", tuples1);
Set<ZSetOperations.TypedTuple<String>> tuples2 = new HashSet<>();
tuples2.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples2.add(new DefaultTypedTuple<>("bbb", 21.20));
tuples2.add(new DefaultTypedTuple<>("eee", 61.5));
tuples2.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-2", tuples2);Long result = operations.unionAndStore("zs-key-1", "zs-key-2", "zs-key-new");
System.out.println("result = " + result);Set<ZSetOperations.TypedTuple<String>> tuples = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tuples) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

result = 6
fff - 11.6
ddd - 21.4
aaa - 22.2
eee - 61.5
bbb - 62.400000000000006
ccc - 71.3

Long unionAndStore(K key, Collection<K> otherKeys, K destKey):多个集合取并集,并将并集结果保存到指定缓存键中。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
    • destKey:目标缓存键
  • 返回值:返回结果集数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples1 = new HashSet<>();
tuples1.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples1.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples1.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples1.add(new DefaultTypedTuple<>("ddd", 21.4));
operations.add("zs-key-1", tuples1);
Set<ZSetOperations.TypedTuple<String>> tuples2 = new HashSet<>();
tuples2.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples2.add(new DefaultTypedTuple<>("bbb", 21.20));
tuples2.add(new DefaultTypedTuple<>("eee", 61.5));
tuples2.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-2", tuples2);
Set<ZSetOperations.TypedTuple<String>> tuples3 = new HashSet<>();
tuples3.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples3.add(new DefaultTypedTuple<>("ccc", 41.20));
tuples3.add(new DefaultTypedTuple<>("ddd", 61.5));
tuples3.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-3", tuples3);Long result = operations.unionAndStore("zs-key-1", Arrays.asList("zs-key-2","zs-key-3"), "zs-key-new");
System.out.println("result = " + result);Set<ZSetOperations.TypedTuple<String>> tuples = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tuples) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

result = 6
fff - 23.2
aaa - 33.3
eee - 61.5
bbb - 62.400000000000006
ddd - 82.9
ccc - 112.5

Long unionAndStore(K key, Collection<K> otherKeys, K destKey, Aggregate aggregate):多个集合取并集,并将并集结果保存到指定缓存键中,使用指定聚合函数获取并集元素。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
    • destKey:目标缓存键
    • aggregate:聚合函数
  • 返回值:返回结果集数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples1 = new HashSet<>();
tuples1.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples1.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples1.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples1.add(new DefaultTypedTuple<>("ddd", 21.4));
operations.add("zs-key-1", tuples1);
Set<ZSetOperations.TypedTuple<String>> tuples2 = new HashSet<>();
tuples2.add(new DefaultTypedTuple<>("aaa", 31.1));
tuples2.add(new DefaultTypedTuple<>("bbb", 21.20));
tuples2.add(new DefaultTypedTuple<>("ddd", 61.5));
tuples2.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-2", tuples2);
Set<ZSetOperations.TypedTuple<String>> tuples3 = new HashSet<>();
tuples3.add(new DefaultTypedTuple<>("aaa", 21.1));
tuples3.add(new DefaultTypedTuple<>("ccc", 41.20));
tuples3.add(new DefaultTypedTuple<>("ddd", 41.5));
tuples3.add(new DefaultTypedTuple<>("fff", 61.6));
operations.add("zs-key-3", tuples3);Long result1 = operations.unionAndStore("zs-key-1", Arrays.asList("zs-key-2", "zs-key-3"), "zs-key-new",RedisZSetCommands.Aggregate.MIN);
System.out.println("result1 = " + result1);Set<ZSetOperations.TypedTuple<String>> tupleSet1 = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tupleSet1) {System.out.println(next.getValue() + " - " + next.getScore());
}Long result2 = operations.unionAndStore("zs-key-1", Arrays.asList("zs-key-2", "zs-key-3"), "zs-key-new",RedisZSetCommands.Aggregate.MAX);
System.out.println("result2 = " + result2);Set<ZSetOperations.TypedTuple<String>> tupleSet2 = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tupleSet2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

result1 = 6
aaa - 11.1
fff - 11.6
bbb - 21.2
ddd - 21.4
ccc - 41.2
eee - 61.5
result2 = 6
aaa - 31.1
bbb - 41.2
ddd - 61.5
eee - 61.5
fff - 61.6
ccc - 71.3

Long unionAndStore(K key, Collection<K> otherKeys, K destKey, Aggregate aggregate, Weights weights):多个集合取并集,并将并集结果保存到指定缓存键中,使用指定聚合函数和权重来计算和获取并集元素,权重值会和元素的分数相乘。

  • 参数:
    • key:缓存键
    • otherKeys:其他缓存键集合
    • destKey:目标缓存键
    • aggregate:聚合函数
    • weights:权重
  • 返回值:返回结果集数量

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples1 = new HashSet<>();
tuples1.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples1.add(new DefaultTypedTuple<>("bbb", 41.20));
tuples1.add(new DefaultTypedTuple<>("ccc", 71.3));
tuples1.add(new DefaultTypedTuple<>("ddd", 21.4));
operations.add("zs-key-1", tuples1);
Set<ZSetOperations.TypedTuple<String>> tuples2 = new HashSet<>();
tuples2.add(new DefaultTypedTuple<>("aaa", 31.1));
tuples2.add(new DefaultTypedTuple<>("bbb", 21.20));
tuples2.add(new DefaultTypedTuple<>("ddd", 61.5));
tuples2.add(new DefaultTypedTuple<>("fff", 11.6));
operations.add("zs-key-2", tuples2);
Set<ZSetOperations.TypedTuple<String>> tuples3 = new HashSet<>();
tuples3.add(new DefaultTypedTuple<>("aaa", 21.1));
tuples3.add(new DefaultTypedTuple<>("ccc", 41.20));
tuples3.add(new DefaultTypedTuple<>("ddd", 41.5));
tuples3.add(new DefaultTypedTuple<>("fff", 61.6));
operations.add("zs-key-3", tuples3);Long result1 = operations.unionAndStore("zs-key-1", Arrays.asList("zs-key-2", "zs-key-3"), "zs-key-new",RedisZSetCommands.Aggregate.MIN, RedisZSetCommands.Weights.of(2, 3, 5));
System.out.println("result1 = " + result1);Set<ZSetOperations.TypedTuple<String>> tupleSet1 = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tupleSet1) {System.out.println(next.getValue() + " - " + next.getScore());
}Long result2 = operations.unionAndStore("zs-key-1", Arrays.asList("zs-key-2", "zs-key-3"), "zs-key-new",RedisZSetCommands.Aggregate.MAX, RedisZSetCommands.Weights.of(2, 3, 5));
System.out.println("result2 = " + result2);Set<ZSetOperations.TypedTuple<String>> tupleSet2 = operations.rangeWithScores("zs-key-new", 0, -1);
for (ZSetOperations.TypedTuple<String> next : tupleSet2) {System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

result1 = 6
aaa - 22.2
fff - 34.8
ddd - 42.8
bbb - 63.599999999999994
ccc - 142.6
eee - 184.5
result2 = 6
bbb - 82.4
aaa - 105.5
eee - 184.5
ccc - 206.0
ddd - 207.5
fff - 308.0

Cursor<TypedTuple<V>> scan(K key, ScanOptions options):根据匹配规则查询结果集

  • 参数:
    • key:缓存键
  • 返回值:返回结果集

示例:

ZSetOperations<String, String> operations = redisTemplate.opsForZSet();
Set<ZSetOperations.TypedTuple<String>> tuples1 = new HashSet<>();
tuples1.add(new DefaultTypedTuple<>("aaa", 11.1));
tuples1.add(new DefaultTypedTuple<>("aab", 41.20));
tuples1.add(new DefaultTypedTuple<>("aac", 71.3));
tuples1.add(new DefaultTypedTuple<>("bba", 21.4));
tuples1.add(new DefaultTypedTuple<>("bbb", 51.4));
tuples1.add(new DefaultTypedTuple<>("bbaa", 67.3));
tuples1.add(new DefaultTypedTuple<>("bcaab", 85.2));
operations.add("zs-key", tuples1);Cursor<ZSetOperations.TypedTuple<String>> cursor1 = operations.scan("zs-key", ScanOptions.NONE);
while (cursor1.hasNext()){ZSetOperations.TypedTuple<String> next = cursor1.next();System.out.println(next.getValue() + " - " + next.getScore());
}
System.out.println("----------");
Cursor<ZSetOperations.TypedTuple<String>> cursor2 = operations.scan("zs-key", ScanOptions.scanOptions().match("aa*").build());
while (cursor1.hasNext()){ZSetOperations.TypedTuple<String> next = cursor2.next();System.out.println(next.getValue() + " - " + next.getScore());
}

输出结果:

aaa - 11.1
fff - 11.6
bba - 21.4
ddd - 21.4
aab - 41.2
bbb - 51.4
eee - 61.5
bbaa - 67.3
aac - 71.3
ccc - 71.3
bcaab - 85.2
----------

三、RedisUtil 工具类封装

@Component
public class RedisUtil {private static RedisTemplate<String, Object> redisTemplate;@Autowiredpublic RedisUtil(RedisTemplate<String, Object> redisTemplate) {RedisUtil.redisTemplate = redisTemplate;}// region redis common method/*** common method:删除缓存键key对应的缓存** @param key 缓存键*/public static void delete(String key) {try {redisTemplate.delete(key);} catch (Exception e) {e.printStackTrace();}}/*** common method:批量删除缓存键集合keys对应的缓存** @param keys 缓存键集合*/public static void delete(Collection<String> keys) {try {redisTemplate.delete(keys);} catch (Exception e) {e.printStackTrace();}}/*** common method:缓存键key对应的缓存是否存在** @param key 缓存键*/public static Boolean hasKey(String key) {try {return redisTemplate.hasKey(key);} catch (Exception e) {e.printStackTrace();return false;}}/*** common method:获取所有满足正则匹配的缓存键** @param pattern 匹配规则*/public static Set<String> keys(String pattern) {try {return redisTemplate.keys(pattern);} catch (Exception e) {e.printStackTrace();return null;}}/*** common method:重命名缓存键** @param oldKey 旧缓存键* @param newKey 新缓存键*/public static void rename(String oldKey, String newKey) {try {redisTemplate.rename(oldKey, newKey);} catch (Exception e) {e.printStackTrace();}}// endregion// region redis String/*** String 类型缓存:添加缓存** @param key   缓存键* @param value 缓存值*/public static void set(String key, Object value) {try {redisTemplate.opsForValue().set(key, value);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:添加缓存并指定过期时间** @param key        缓存键* @param value      缓存值* @param expireTime 过期时间* @param timeUnit   时间单位*/public static void set(String key, Object value, long expireTime, TimeUnit timeUnit) {try {redisTemplate.opsForValue().set(key, value, expireTime, timeUnit);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:添加缓存并指定过期时间** @param key      缓存键* @param value    缓存值* @param duration 持续时间,Java 8中的类,如 Duration.ofSeconds(5000);*/public static void set(String key, Object value, Duration duration) {try {redisTemplate.opsForValue().set(key, value, duration);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:修改缓存,从指定位置offset开始,修改键为key的缓存的值* 例如:set("kkk", "new", 4), key:value = "kkk":"this is value" -> key:value = "kkk":"thisnew value"** @param key    缓存键* @param value  缓存值* @param offset 指定位置*/public static void set(String key, Object value, long offset) {try {redisTemplate.opsForValue().set(key, value, offset);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:用于设置Redis中位图(bitmaps)的指定偏移量的值。* 位图是由位(0或1)组成的数据结构,可以用于表示某个事件的状态或者进行位运算。** @param key    缓存键* @param offset 要设置的位置,以比特为单位。* @param value  要设置的值,0 或 1。*/public static void setBit(String key, long offset, boolean value) {try {redisTemplate.opsForValue().setBit(key, offset, value);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:添加缓存,缓存键key不存在则添加,key存在则添加失败** @param key   缓存键* @param value 缓存值*/public static void setIfAbsent(String key, Object value) {try {redisTemplate.opsForValue().setIfAbsent(key, value);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:添加缓存并指定过期时间,缓存键key不存在则添加,key存在则添加失败** @param key        缓存键* @param value      缓存值* @param expireTime 过期时间* @param timeUnit   时间单位*/public static void setIfAbsent(String key, Object value, long expireTime, TimeUnit timeUnit) {try {redisTemplate.opsForValue().setIfAbsent(key, value, expireTime, timeUnit);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:添加缓存并指定过期时间,缓存键key不存在则添加,key存在则添加失败** @param key      缓存键* @param value    缓存值* @param duration 持续时间,Java 8中的类,如 Duration.ofSeconds(5000);*/public static void setIfAbsent(String key, Object value, Duration duration) {try {redisTemplate.opsForValue().setIfAbsent(key, value, duration);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:批量添加缓存** @param map 缓存内容,key:value*/public static void multiSet(Map<String, Object> map) {try {redisTemplate.opsForValue().multiSet(map);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:批量添加缓存,map中只要有一个缓存键key不存在则添加失败** @param map 缓存内容,key:value*/public static void multiSetIfAbsent(Map<String, Object> map) {try {redisTemplate.opsForValue().multiSetIfAbsent(map);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:为键key的缓存指定新值value,并获取缓存key的旧值** @param key   缓存键* @param value 新缓存值*/public static void getAndSet(String key, Object value) {try {redisTemplate.opsForValue().getAndSet(key, value);} catch (Exception e) {e.printStackTrace();}}/*** String 类型缓存:获取缓存** @param key 缓存键*/public static Object get(String key) {try {return redisTemplate.opsForValue().get(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** String 类型缓存:获取缓存的一部分,指定开始和结束位置** @param key   缓存键* @param start 开始位置* @param end   结束位置*/public static String get(String key, long start, long end) {try {return redisTemplate.opsForValue().get(key, start, end);} catch (Exception e) {e.printStackTrace();return null;}}/*** String 类型缓存:获取缓存的位图数据** @param key    缓存键* @param offset 偏移位置*/public static Boolean getBit(String key, long offset) {try {return redisTemplate.opsForValue().getBit(key, offset);} catch (Exception e) {e.printStackTrace();return null;}}// endregion// region redis hash/*** Hash 类型缓存:添加缓存** @param key     缓存键* @param hashKey hash 键* @param value   hash 值*/public static void hSet(String key, Object hashKey, Object value) {try {redisTemplate.opsForHash().put(key, hashKey, value);} catch (Exception e) {e.printStackTrace();}}/*** Hash 类型缓存:批量添加缓存** @param key 缓存键* @param map hash 键值对*/public static void hSet(String key, Map<Object, Object> map) {try {redisTemplate.opsForHash().putAll(key, map);} catch (Exception e) {e.printStackTrace();}}/*** Hash 类型缓存:添加缓存,hash 键不存在则添加,否则添加失败** @param key     缓存键* @param hashKey hash 键* @param value   hash 值*/public static void hSetIfAbsent(String key, Object hashKey, Object value) {try {redisTemplate.opsForHash().putIfAbsent(key, hashKey, value);} catch (Exception e) {e.printStackTrace();}}/*** Hash 类型缓存:获取缓存** @param key     缓存键* @param hashKey hash 键*/public static Object hGet(String key, Object hashKey) {try {return redisTemplate.opsForHash().get(key, hashKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:批量获取缓存** @param key      缓存键* @param hashKeys hash 键*/public static List<Object> hGet(String key, Collection<Object> hashKeys) {try {return redisTemplate.opsForHash().multiGet(key, hashKeys);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:获取缓存键下的所有缓存,以键值对的方式返回** @param key 缓存键*/public static Map<Object, Object> hEntries(String key) {try {return redisTemplate.opsForHash().entries(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:获取缓存键key下的所有hash键** @param key 缓存键*/public static Set<Object> hKeys(String key) {try {return redisTemplate.opsForHash().keys(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:获取缓存键key下的所有hash值** @param key 缓存键*/public static List<Object> hValues(String key) {try {return redisTemplate.opsForHash().values(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:删除缓存** @param key      缓存键* @param hashKeys hash 键列表*/public static void hDelete(String key, Object... hashKeys) {try {redisTemplate.opsForHash().delete(key, hashKeys);} catch (Exception e) {e.printStackTrace();}}/*** Hash 类型缓存:缓存键key下的hash键是否存在** @param key     缓存键* @param hashKey hash 键*/public static Boolean hHasKey(String key, Object hashKey) {try {return redisTemplate.opsForHash().hasKey(key, hashKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:缓存键key下hash键对应的hash值的长度** @param key     缓存键* @param hashKey hash 键*/public static Long hLengthOfValue(String key, Object hashKey) {try {return redisTemplate.opsForHash().lengthOfValue(key, hashKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:获取缓存键key下的所有缓存的大小** @param key 缓存键*/public static Long hSize(String key) {try {return redisTemplate.opsForHash().size(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:匹配获取键值对** @param key         缓存键* @param scanOptions hash 值*/public static Cursor<Map.Entry<Object, Object>> hScan(String key, ScanOptions scanOptions) {try {return redisTemplate.opsForHash().scan(key, scanOptions);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:使缓存键key下的hash键对应的值以long类型增加delta* 如果指定的哈希表或字段不存在,那么将会创建一个新的哈希表和字段,并指定值为delta** @param key     缓存键* @param hashKey hash 键*/public static Long hIncrement(String key, Object hashKey, long delta) {try {return redisTemplate.opsForHash().increment(key, hashKey, delta);} catch (Exception e) {e.printStackTrace();return null;}}/*** Hash 类型缓存:使缓存键key下的hash键对应的值以double类型增加delta* 如果指定的哈希表或字段不存在,那么将会创建一个新的哈希表和字段,并指定值为delta** @param key     缓存键* @param hashKey hash 键*/public static Double hIncrement(String key, Object hashKey, double delta) {try {return redisTemplate.opsForHash().increment(key, hashKey, delta);} catch (Exception e) {e.printStackTrace();return null;}}// endregion// region redis list/*** List 类型缓存:向队列头部添加缓存** @param key   缓存键* @param value 缓存值*/public static void lLeftSet(String key, Object value) {try {redisTemplate.opsForList().leftPush(key, value);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:缓存键key对应的缓存中,将值value添加到指定值pivot的前面** @param key   缓存键* @param pivot 指定值* @param value 缓存值*/public static void lLeftSet(String key, Object pivot, Object value) {try {redisTemplate.opsForList().leftPush(key, pivot, value);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:批量向队列头部添加缓存** @param key    缓存键* @param values 缓存值列表*/public static void lLeftSet(String key, Object... values) {try {redisTemplate.opsForList().leftPushAll(key, values);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:批量向队列头部添加缓存** @param key    缓存键* @param values 缓存值集合*/public static void lLeftSet(String key, Collection<Object> values) {try {redisTemplate.opsForList().leftPushAll(key, values);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:向队列头部添加缓存,只有缓存列表存在才成功,否则添加失败** @param key   缓存键* @param value 缓存值*/public static void lLeftSetIfPresent(String key, Object value) {try {redisTemplate.opsForList().leftPushIfPresent(key, value);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:向队列尾部添加缓存** @param key   缓存键* @param value 缓存值*/public static void lRightSet(String key, Object value) {try {redisTemplate.opsForList().rightPush(key, value);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:缓存键key对应的缓存中,将值value添加到指定值pivot的后面** @param key   缓存键* @param pivot 指定值* @param value 缓存值*/public static void lRightSet(String key, Object pivot, Object value) {try {redisTemplate.opsForList().rightPush(key, pivot, value);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:批量向队列尾部添加缓存** @param key    缓存键* @param values 缓存值列表*/public static void lRightSet(String key, Object... values) {try {redisTemplate.opsForList().rightPushAll(key, values);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:批量向队列尾部添加缓存** @param key    缓存键* @param values 缓存值集合*/public static void lRightSet(String key, Collection<Object> values) {try {redisTemplate.opsForList().rightPushAll(key, values);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:向队列尾部添加缓存,只有缓存列表存在才成功,否则添加失败** @param key   缓存键* @param value 缓存值*/public static void lRightSetIfPresent(String key, Object value) {try {redisTemplate.opsForList().rightPushIfPresent(key, value);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:键key的缓存中,替换索引index对应的值为value** @param key   缓存键* @param index 指定索引,从0开始* @param value 缓存值*/public static void lSet(String key, long index, Object value) {try {redisTemplate.opsForList().set(key, index, value);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:获取键key的缓存,指定开始索引和结束索引** @param key   缓存键* @param start 开始索引,从0开始* @param end   结束索引,-1为最后一个*/public static List<Object> lRange(String key, long start, long end) {try {return redisTemplate.opsForList().range(key, start, end);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:从键key对应的缓存中取出索引为index的元素** @param key   缓存键* @param index 指定索引,索引超过实际缓存大小,返回null,不会报错*/public static Object lIndex(String key, long index) {try {return redisTemplate.opsForList().index(key, index);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:键key的缓存中,移除值value出现的前count个** @param key   缓存键* @param count 出现次数,次数大于出现的实际次数,移除所有该值,不会报错* @param value 缓存值*/public static void lRemove(String key, long count, Object value) {try {redisTemplate.opsForList().remove(key, count, value);} catch (Exception e) {e.printStackTrace();}}/*** List 类型缓存:键key的缓存中,从头部取出第一个元素** @param key 缓存键*/public static Object lLeftGet(String key) {try {return redisTemplate.opsForList().leftPop(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:键key的缓存中,从头部取出第一个元素* 阻塞连接,直到元素可用或者时间超时** @param key        缓存键* @param expireTime 超时时间* @param timeUnit   时间单位*/public static Object lLeftGet(String key, long expireTime, TimeUnit timeUnit) {try {return redisTemplate.opsForList().leftPop(key, expireTime, timeUnit);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:键key的缓存中,从头部取出第一个元素* 阻塞连接,直到元素可用或者时间超时** @param key      缓存键* @param duration 持续时间*/public static Object lLeftGet(String key, Duration duration) {try {return redisTemplate.opsForList().leftPop(key, duration);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:键key的缓存中,从尾部取出第一个元素** @param key 缓存键*/public static Object lRightGet(String key) {try {return redisTemplate.opsForList().rightPop(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:键key的缓存中,从尾部取出第一个元素* 阻塞连接,直到元素可用或者时间超时** @param key        缓存键* @param expireTime 超时时间* @param timeUnit   时间单位*/public static Object lRightGet(String key, long expireTime, TimeUnit timeUnit) {try {return redisTemplate.opsForList().rightPop(key, expireTime, timeUnit);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:键key的缓存中,从尾部取出第一个元素* 阻塞连接,直到元素可用或者时间超时** @param key      缓存键* @param duration 持续时间*/public static Object lRightGet(String key, Duration duration) {try {return redisTemplate.opsForList().rightPop(key, duration);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:从键sourceKey的缓存尾部取出一个元素,并将此元素添加到键destinationKey的缓存头部** @param sourceKey      数据缓存键* @param destinationKey 目标缓存键*/public static Object lRightPopAndLeftPush(String sourceKey, String destinationKey) {try {return redisTemplate.opsForList().rightPopAndLeftPush(sourceKey, destinationKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:从键sourceKey的缓存尾部取出一个元素,并将此元素添加到键destinationKey的缓存头部* 阻塞连接,直到元素可用或者时间超时** @param sourceKey      数据缓存键* @param destinationKey 目标缓存键* @param expireTime     过期时间* @param timeUnit       时间单位*/public static Object lRightPopAndLeftPush(String sourceKey, String destinationKey, long expireTime, TimeUnit timeUnit) {try {return redisTemplate.opsForList().rightPopAndLeftPush(sourceKey, destinationKey, expireTime, timeUnit);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:从键sourceKey的缓存尾部取出一个元素,并将此元素添加到键destinationKey的缓存头部* 阻塞连接,直到元素可用或者时间超时** @param sourceKey      数据缓存键* @param destinationKey 目标缓存键* @param duration       持续时间*/public static Object lRightPopAndLeftPush(String sourceKey, String destinationKey, Duration duration) {try {return redisTemplate.opsForList().rightPopAndLeftPush(sourceKey, destinationKey, duration);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:统计键key对应缓存的元素格式** @param key 缓存键*/public static Long lSize(String key) {try {return redisTemplate.opsForList().size(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** List 类型缓存:截取键key对应的缓存,指定开始索引和结束索引** @param key   缓存键* @param start 开始索引* @param end   结束索引*/public static void lTrim(String key, long start, long end) {try {redisTemplate.opsForList().trim(key, start, end);} catch (Exception e) {e.printStackTrace();}}// endregion// region redis set/*** Set 类型缓存:添加元素** @param key    缓存键* @param values 缓存值列表*/public static void sSet(String key, Object... values) {try {redisTemplate.opsForSet().add(key, values);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:移除元素** @param key    缓存键* @param values 缓存值列表*/public static void sRemove(String key, Object... values) {try {redisTemplate.opsForSet().remove(key, values);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:从键key的缓存中随机移除一个元素,并返回元素的值** @param key 缓存键*/public static Object sPop(String key) {try {return redisTemplate.opsForSet().pop(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:从键key的缓存中随机移除count个元素,并返回这些元素的值** @param key   缓存键* @param count 元素个数*/public static List<Object> sPop(String key, long count) {try {return redisTemplate.opsForSet().pop(key, count);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:从键key的缓存中随机返回一个元素** @param key 缓存键*/public static Object sRandomMember(String key) {try {return redisTemplate.opsForSet().randomMember(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:从键key的缓存中随机返回count个元素,返回元素可重复** @param key   缓存键* @param count 元素个数*/public static List<Object> sRandomMembers(String key, long count) {try {return redisTemplate.opsForSet().randomMembers(key, count);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:从键key的缓存中随机返回count个不重复的元素* 超过元素总个数会返回所有元素,不报错** @param key   缓存键* @param count 元素个数*/public static Set<Object> sDistinctRandomMembers(String key, long count) {try {return redisTemplate.opsForSet().distinctRandomMembers(key, count);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:判断指定元素在键key的缓存是否存在** @param key   缓存键* @param value 缓存值*/public static Boolean sExists(String key, Object value) {try {return redisTemplate.opsForSet().isMember(key, value);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:统计键key的缓存元素个数** @param key 缓存键*/public static Long sSize(String key) {try {return redisTemplate.opsForSet().size(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:从键key的缓存中取出元素value添加到目标缓存键destKey的缓存中** @param key     原缓存键* @param value   缓存值* @param destKey 目标缓存键*/public static Boolean sMove(String key, Object value, String destKey) {try {return redisTemplate.opsForSet().move(key, value, destKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较并返回指定键和其他键对应的缓存中不同的元素* 返回的元素为指定缓存键中不同的元素** @param key      指定缓存键* @param otherKey 其他缓存键*/public static Set<Object> sDifference(String key, String otherKey) {try {return redisTemplate.opsForSet().difference(key, otherKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较并返回指定键和其他键对应的缓存中不同的元素* 返回的元素为指定缓存键中不同的元素** @param key       指定缓存键* @param otherKeys 其他缓存键集合*/public static Set<Object> sDifference(String key, Collection<String> otherKeys) {try {return redisTemplate.opsForSet().difference(key, otherKeys);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较缓存键集合中各个键对应的缓存中不同的元素* 返回的元素为缓存键集合中第一个键对应的缓存中不同的元素** @param keys 缓存键集合*/public static Set<Object> sDifference(Collection<String> keys) {try {return redisTemplate.opsForSet().difference(keys);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较指定键和其他键对应的缓存中不同的元素,并将不同的元素另外缓存* 缓存的元素为指定缓存键中不同的元素** @param key      指定缓存键* @param otherKey 其他缓存键* @param destKey  目标缓存键*/public static void sDifferenceAndStore(String key, String otherKey, String destKey) {try {redisTemplate.opsForSet().differenceAndStore(key, otherKey, destKey);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:比较指定键和其他键对应的缓存中不同的元素,并将不同的元素另外缓存* 缓存的元素为指定缓存键中不同的元素** @param key       指定缓存键* @param otherKeys 其他缓存键集合* @param destKey   目标缓存键*/public static void sDifferenceAndStore(String key, Collection<String> otherKeys, String destKey) {try {redisTemplate.opsForSet().differenceAndStore(key, otherKeys, destKey);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:比较缓存键集合中各个键对应的缓存中不同的元素,并将不同的元素另外缓存* 缓存的元素为缓存键集合中第一个键对应的缓存中不同的元素** @param keys    缓存键集合* @param destKey 目标缓存键*/public static void sDifferenceAndStore(Collection<String> keys, String destKey) {try {redisTemplate.opsForSet().differenceAndStore(keys, destKey);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:比较并返回指定键和其他键对应的缓存中相同的元素** @param key      指定缓存键* @param otherKey 其他缓存键*/public static Set<Object> sIntersect(String key, String otherKey) {try {return redisTemplate.opsForSet().intersect(key, otherKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较并返回指定键和其他键对应的缓存中相同的元素** @param key       指定缓存键* @param otherKeys 其他缓存键集合*/public static Set<Object> sIntersect(String key, Collection<String> otherKeys) {try {return redisTemplate.opsForSet().intersect(key, otherKeys);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较缓存键集合中各个键对应的缓存中相同的元素** @param keys 缓存键集合*/public static Set<Object> sIntersect(Collection<String> keys) {try {return redisTemplate.opsForSet().intersect(keys);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较指定键和其他键对应的缓存中相同的元素,并将相同的元素另外缓存** @param key      指定缓存键* @param otherKey 其他缓存键* @param destKey  目标缓存键*/public static void sIntersectAndStore(String key, String otherKey, String destKey) {try {redisTemplate.opsForSet().intersectAndStore(key, otherKey, destKey);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:比较指定键和其他键对应的缓存中相同的元素,并将相同的元素另外缓存** @param key       指定缓存键* @param otherKeys 其他缓存键集合* @param destKey   目标缓存键*/public static void sIntersectAndStore(String key, Collection<String> otherKeys, String destKey) {try {redisTemplate.opsForSet().intersectAndStore(key, otherKeys, destKey);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:比较缓存键集合中各个键对应的缓存中相同的元素,并将相同的元素另外缓存** @param keys    缓存键集合* @param destKey 目标缓存键*/public static void sIntersectAndStore(Collection<String> keys, String destKey) {try {redisTemplate.opsForSet().intersectAndStore(keys, destKey);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:比较并返回指定键和其他键对应的缓存中所有的元素,会去重** @param key      指定缓存键* @param otherKey 其他缓存键*/public static Set<Object> sUnion(String key, String otherKey) {try {return redisTemplate.opsForSet().union(key, otherKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较并返回指定键和其他键对应的缓存中所有的元素,会去重** @param key       指定缓存键* @param otherKeys 其他缓存键集合*/public static Set<Object> sUnion(String key, Collection<String> otherKeys) {try {return redisTemplate.opsForSet().union(key, otherKeys);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较缓存键集合中各个键对应的缓存中所有的元素,会去重** @param keys 缓存键集合*/public static Set<Object> sUnion(Collection<String> keys) {try {return redisTemplate.opsForSet().union(keys);} catch (Exception e) {e.printStackTrace();return null;}}/*** Set 类型缓存:比较指定键和其他键对应的缓存中所有的元素,会去重,并将合并的元素另外缓存** @param key      指定缓存键* @param otherKey 其他缓存键* @param destKey  目标缓存键*/public static void sUnionAndStore(String key, String otherKey, String destKey) {try {redisTemplate.opsForSet().unionAndStore(key, otherKey, destKey);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:比较指定键和其他键对应的缓存中所有的元素,会去重,并将合并的元素另外缓存** @param key       指定缓存键* @param otherKeys 其他缓存键集合* @param destKey   目标缓存键*/public static void sUnionAndStore(String key, Collection<String> otherKeys, String destKey) {try {redisTemplate.opsForSet().unionAndStore(key, otherKeys, destKey);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:比较缓存键集合中各个键对应的缓存中所有的元素,会去重,并将合并的元素另外缓存** @param keys    缓存键集合* @param destKey 目标缓存键*/public static void sUnionAndStore(Collection<String> keys, String destKey) {try {redisTemplate.opsForSet().unionAndStore(keys, destKey);} catch (Exception e) {e.printStackTrace();}}/*** Set 类型缓存:从键key的缓存中匹配查找元素** @param key         缓存键* @param scanOptions 匹配规则*/public static Cursor<Object> sScan(String key, ScanOptions scanOptions) {try {return redisTemplate.opsForSet().scan(key, scanOptions);} catch (Exception e) {e.printStackTrace();return null;}}// endregion// region redis zset/*** ZSet 类型缓存:添加缓存** @param key   缓存键* @param value 缓存值* @param score 分数*/public static void zsSet(String key, Object value, double score) {try {redisTemplate.opsForZSet().add(key, value, score);} catch (Exception e) {e.printStackTrace();}}/*** ZSet 类型缓存:批量添加缓存* 示例:* Set<ZSetOperations.TypedTuple<Object>> tuples = new HashSet<>();* ZSetOperations.TypedTuple<Object> objectTypedTuple1 = new DefaultTypedTuple<>("GGG", 37d);* ZSetOperations.TypedTuple<Object> objectTypedTuple2 = new DefaultTypedTuple<>("FFF", 92d);* tuples.add(objectTypedTuple1);* tuples.add(objectTypedTuple2);** @param key    缓存键* @param tuples 分数*/public static void zsSet(String key, Set<ZSetOperations.TypedTuple<Object>> tuples) {try {redisTemplate.opsForZSet().add(key, tuples);} catch (Exception e) {e.printStackTrace();}}/*** ZSet 类型缓存:删除缓存** @param key    缓存键* @param values 缓存值列表*/public static void zsRemove(String key, Object... values) {try {redisTemplate.opsForZSet().remove(key, values);} catch (Exception e) {e.printStackTrace();}}/*** ZSet 类型缓存:根据索引排序后,删除指定索引范围内的缓存,指定开始索引和结束索引** @param key   缓存键* @param start 开始索引* @param end   结束索引*/public static void zsRemoveRange(String key, long start, long end) {try {redisTemplate.opsForZSet().removeRange(key, start, end);} catch (Exception e) {e.printStackTrace();}}/*** ZSet 类型缓存:根据分数排序后,删除指定范围内的缓存,指定最小分数和最大分数** @param key 缓存键* @param min 最小分数* @param max 最大分数*/public static void zsRemoveRangeByScore(String key, double min, double max) {try {redisTemplate.opsForZSet().removeRangeByScore(key, min, max);} catch (Exception e) {e.printStackTrace();}}/*** ZSet 类型缓存:获取缓存元素,指定开始索引和结束索引** @param key   缓存键* @param start 开始索引* @param end   结束索引*/public static Set<Object> zsRange(String key, long start, long end) {try {return redisTemplate.opsForZSet().range(key, start, end);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:获取缓存元素,指定开始索引和结束索引,倒序排序** @param key   缓存键* @param start 开始索引* @param end   结束索引*/public static Set<Object> zsReverseRange(String key, long start, long end) {try {return redisTemplate.opsForZSet().reverseRange(key, start, end);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:从缓存中取出在范围range内的元素,并按照缓存元素value的字典顺序排序** @param key   缓存键* @param range 范围,内容为缓存元素value的值*/public static Set<Object> zsRangeByLex(String key, RedisZSetCommands.Range range) {try {return redisTemplate.opsForZSet().rangeByLex(key, range);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:从缓存中取出在范围range内的元素,并按照缓存元素value的字典顺序排序** @param key   缓存键* @param range 范围,内容为缓存元素value的值* @param limit 限制返回元素个数*/public static Set<Object> zsRangeByLex(String key, RedisZSetCommands.Range range, RedisZSetCommands.Limit limit) {try {return redisTemplate.opsForZSet().rangeByLex(key, range, limit);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数排序,然后从缓存取出指定分数范围内的缓存元素** @param key 缓存键* @param min 最小分数* @param max 最大分数*/public static Set<Object> zsRangeByScore(String key, double min, double max) {try {return redisTemplate.opsForZSet().rangeByScore(key, min, max);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数倒序排序,然后从缓存取出指定分数范围内的缓存元素** @param key 缓存键* @param min 最小分数* @param max 最大分数*/public static Set<Object> zsReverseRangeByScore(String key, double min, double max) {try {return redisTemplate.opsForZSet().reverseRangeByScore(key, min, max);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数排序,然后从缓存取出指定分数范围内的缓存元素,指定偏移量和返回个数** @param key    缓存键* @param min    最小分数* @param max    最大分数* @param offset 偏移量* @param count  返回个数*/public static Set<Object> zsRangeByScore(String key, double min, double max, long offset, long count) {try {return redisTemplate.opsForZSet().rangeByScore(key, min, max, offset, count);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数倒序排序,然后从缓存取出指定分数范围内的缓存元素,指定偏移量和返回个数** @param key    缓存键* @param min    最小分数* @param max    最大分数* @param offset 偏移量* @param count  返回个数*/public static Set<Object> zsReverseRangeByScore(String key, double min, double max, long offset, long count) {try {return redisTemplate.opsForZSet().reverseRangeByScore(key, min, max, offset, count);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数,指定偏移量和返回个数** @param key   缓存键* @param start 开始索引* @param end   结束索引*/public static Map<Object, Double> zsRangeWithScores(String key, long start, long end) {try {Set<ZSetOperations.TypedTuple<Object>> typedTuples = redisTemplate.opsForZSet().rangeWithScores(key, start, end);return handleZSet(typedTuples);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数倒序排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数,指定偏移量和返回个数** @param key   缓存键* @param start 开始索引* @param end   结束索引*/public static Map<Object, Double> zsReverseRangeWithScores(String key, long start, long end) {try {Set<ZSetOperations.TypedTuple<Object>> typedTuples = redisTemplate.opsForZSet().reverseRangeWithScores(key, start, end);return handleZSet(typedTuples);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数** @param key 缓存键* @param min 最小分数* @param max 最大分数*/public static Map<Object, Double> zsRangeByScoreWithScore(String key, double min, double max) {try {Set<ZSetOperations.TypedTuple<Object>> typedTuples = redisTemplate.opsForZSet().rangeByScoreWithScores(key, min, max);return handleZSet(typedTuples);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数倒序排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数** @param key 缓存键* @param min 最小分数* @param max 最大分数*/public static Map<Object, Double> zsReverseRangeByScoreWithScore(String key, double min, double max) {try {Set<ZSetOperations.TypedTuple<Object>> typedTuples = redisTemplate.opsForZSet().reverseRangeByScoreWithScores(key, min, max);return handleZSet(typedTuples);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数,指定偏移量和返回个数** @param key    缓存键* @param min    最小分数* @param max    最大分数* @param offset 偏移量* @param count  返回个数*/public static Map<Object, Double> zsRangeByScoreWithScore(String key, double min, double max, long offset, long count) {try {Set<ZSetOperations.TypedTuple<Object>> typedTuples = redisTemplate.opsForZSet().rangeByScoreWithScores(key, min, max, offset, count);return handleZSet(typedTuples);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:先根据分数倒序排序,然后从缓存取出指定分数范围内的缓存元素及元素对应的分数,指定偏移量和返回个数** @param key    缓存键* @param min    最小分数* @param max    最大分数* @param offset 偏移量* @param count  返回个数*/public static Map<Object, Double> zsReverseRangeByScoreWithScore(String key, double min, double max, long offset, long count) {try {Set<ZSetOperations.TypedTuple<Object>> typedTuples = redisTemplate.opsForZSet().reverseRangeByScoreWithScores(key, min, max, offset, count);return handleZSet(typedTuples);} catch (Exception e) {e.printStackTrace();return null;}}private static Map<Object, Double> handleZSet(Set<ZSetOperations.TypedTuple<Object>> typedTuples) {Map<Object, Double> map = new HashMap<>();if (Objects.nonNull(typedTuples)) {for (ZSetOperations.TypedTuple<Object> next : typedTuples) {map.put(next.getValue(), next.getScore());}}return map;}/*** ZSet 类型缓存:获取缓存中某个元素的分数** @param key   缓存键* @param value 缓存值*/public static Double zsScore(String key, Object value) {try {return redisTemplate.opsForZSet().score(key, value);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:将缓存中某个值的分组增加delta,返回增加后的分数结果** @param key   缓存键* @param value 缓存值* @param delta 增加分数值*/public static Double zsIncrementScore(String key, Object value, double delta) {try {return redisTemplate.opsForZSet().incrementScore(key, value, delta);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:获取缓存中元素的个数** @param key 缓存键*/public static Long zsSize(String key) {try {return redisTemplate.opsForZSet().size(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:获取缓存中元素的个数** @param key 缓存键*/public static Long zsCard(String key) {try {return redisTemplate.opsForZSet().zCard(key);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:获取指定范围缓存中元素的个数,指定最低分数和最高分数** @param key 缓存键* @param min 最低分数* @param max 最高分数*/public static Long zsCount(String key, long min, long max) {try {return redisTemplate.opsForZSet().count(key, min, max);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:获取指定元素在缓存中的位置索引** @param key   缓存键* @param value 最低分数*/public static Long zsRank(String key, Object value) {try {return redisTemplate.opsForZSet().rank(key, value);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:获取指定元素在缓存中的位置索引,倒序排序** @param key   缓存键* @param value 最低分数*/public static Long zsReverseRank(String key, Object value) {try {return redisTemplate.opsForZSet().reverseRank(key, value);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:两个集合取交集,并将交集结果保存到指定缓存键中** @param key      缓存键* @param otherKey 其他缓存键* @param destKey  目标缓存键* @return 返回结果集数量*/public static Long zsIntersectAndStore(String key, String otherKey, String destKey) {try {return redisTemplate.opsForZSet().intersectAndStore(key, otherKey, destKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:多个集合取交集,并将交集结果保存到指定缓存键中** @param key       缓存键* @param otherKeys 其他缓存键集合* @param destKey   目标缓存键* @return 返回结果集数量*/public static Long zsIntersectAndStore(String key, Collection<String> otherKeys, String destKey) {try {return redisTemplate.opsForZSet().intersectAndStore(key, otherKeys, destKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:多个集合取交集,并将交集结果保存到指定缓存键中,使用指定聚合函数获取交集元素* 如:key1:[{"value1" : 10}, {"value2" : 20}, {"value3" : 30}],key2:[{"value1" : 40}, {"value2" : 50}, {"value4" : 60}]* 使用聚合函数 :* -- MIN,结果为:[{"value1" : 10}, {"value2" : 20}]* -- MAX,结果为:[{"value1" : 40}, {"value2" : 50}]* -- SUM,结果为:[{"value1" : 50}, {"value2" : 70}]** @param key       缓存键* @param otherKeys 其他缓存键集合* @param destKey   目标缓存键* @param aggregate 聚合函数,如MIN、MAX、SUM* @return 返回结果集数量*/public static Long zsIntersectAndStore(String key, Collection<String> otherKeys, String destKey, RedisZSetCommands.Aggregate aggregate) {try {return redisTemplate.opsForZSet().intersectAndStore(key, otherKeys, destKey, aggregate);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:多个集合取交集,并将交集结果保存到指定缓存键中,使用指定聚合函数和权重来计算和获取交集元素,权重值会和元素的分数相乘* 如:key1:[{"value1" : 10}, {"value2" : 20}, {"value3" : 30}],key2:[{"value1" : 40}, {"value2" : 50}, {"value4" : 60}]* 设定权重[key1 = 30, key2 = 20],使用聚合函数 :* -- MIN,结果为:[{"value1" : 300}, {"value2" : 600}]* -- MAX,结果为:[{"value1" : 800}, {"value2" : 1000}]* -- SUM,结果为:[{"value1" : 1100}, {"value2" : 1600}]** @param key       缓存键* @param otherKeys 其他缓存键集合* @param destKey   目标缓存键* @param aggregate 聚合函数,如MIN、MAX、SUM* @param weights   权重,设置各个缓存集合的权重* @return 返回结果集数量*/public static Long zsIntersectAndStore(String key, Collection<String> otherKeys, String destKey, RedisZSetCommands.Aggregate aggregate, RedisZSetCommands.Weights weights) {try {return redisTemplate.opsForZSet().intersectAndStore(key, otherKeys, destKey, aggregate, weights);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:两个集合取并集,并将并集结果保存到指定缓存键中** @param key      缓存键* @param otherKey 其他缓存键* @param destKey  目标缓存键* @return 返回结果集数量*/public static Long zsUnionAndStore(String key, String otherKey, String destKey) {try {return redisTemplate.opsForZSet().intersectAndStore(key, otherKey, destKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:多个集合取并集,并将并集结果保存到指定缓存键中** @param key       缓存键* @param otherKeys 其他缓存键集合* @param destKey   目标缓存键* @return 返回结果集数量*/public static Long zsUnionAndStore(String key, Collection<String> otherKeys, String destKey) {try {return redisTemplate.opsForZSet().intersectAndStore(key, otherKeys, destKey);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:多个集合取并集,并将并集结果保存到指定缓存键中,使用指定聚合函数获取并集元素* 如:key1:[{"value1" : 10}, {"value2" : 20}, {"value3" : 30}],key2:[{"value1" : 40}, {"value2" : 50}, {"value4" : 60}]* 使用聚合函数 :* -- MIN,结果为:[{"value1" : 10}, {"value2" : 20}]* -- MAX,结果为:[{"value1" : 40}, {"value2" : 50}]* -- SUM,结果为:[{"value1" : 50}, {"value2" : 70}]** @param key       缓存键* @param otherKeys 其他缓存键集合* @param destKey   目标缓存键* @param aggregate 聚合函数,如MIN、MAX、SUM* @return 返回结果集数量*/public static Long zsUnionAndStore(String key, Collection<String> otherKeys, String destKey, RedisZSetCommands.Aggregate aggregate) {try {return redisTemplate.opsForZSet().intersectAndStore(key, otherKeys, destKey, aggregate);} catch (Exception e) {e.printStackTrace();return null;}}/*** ZSet 类型缓存:多个集合取并集,并将并集结果保存到指定缓存键中,使用指定聚合函数和权重来计算和获取并集元素,权重值会和元素的分数相乘* 如:key1:[{"value1" : 10}, {"value2" : 20}, {"value3" : 30}],key2:[{"value1" : 40}, {"value2" : 50}, {"value4" : 60}]* 设定权重[key1 = 30, key2 = 20],使用聚合函数 :* -- MIN,结果为:[{"value1" : 300}, {"value2" : 600}]* -- MAX,结果为:[{"value1" : 800}, {"value2" : 1000}]* -- SUM,结果为:[{"value1" : 1100}, {"value2" : 1600}]** @param key       缓存键* @param otherKeys 其他缓存键集合* @param destKey   目标缓存键* @param aggregate 聚合函数,如MIN、MAX、SUM* @param weights   权重,设置各个缓存集合的权重* @return 返回结果集数量*/public static Long zsUnionAndStore(String key, Collection<String> otherKeys, String destKey, RedisZSetCommands.Aggregate aggregate, RedisZSetCommands.Weights weights) {try {return redisTemplate.opsForZSet().unionAndStore(key, otherKeys, destKey, aggregate, weights);} catch (Exception e) {e.printStackTrace();return null;}}// endregion
}

本文来自互联网用户投稿,该文观点仅代表作者本人,不代表本站立场。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如若转载,请注明出处:http://www.mzph.cn/news/68005.shtml

如若内容造成侵权/违法违规/事实不符,请联系多彩编程网进行投诉反馈email:809451989@qq.com,一经查实,立即删除!

相关文章

iWatch框架设计

iWatch框架设计 一、项目框架结构设计 1、项目文件介绍 OverSeaProject&#xff1a;是IOS相关文件文件内容iWatchApp和iWatch Extension&#xff1a;是之前使用xcode14之前的xcode创建的360 app的Watch App&#xff0c;产生的文件结构&#xff0c;包含一个app和Extension的ta…

025: vue父子组件中传递方法控制:$emit,$refs,$parent,$children

第025个 查看专栏目录: VUE ------ element UI 专栏目标 在vue和element UI联合技术栈的操控下&#xff0c;本专栏提供行之有效的源代码示例和信息点介绍&#xff0c;做到灵活运用。 &#xff08;1&#xff09;提供vue2的一些基本操作&#xff1a;安装、引用&#xff0c;模板使…

如何在`Pycharm`中配置基于WSL的`Python Interpreters`,以及配置基于WSL的`Terminal`

文章目录 一、创建pycharm用户并授予sudo权限0. 启动WSL下的CentOS1. 创建pycharm用户并授予sudo权限2. 设置pycharm用户为wsl启动Linux的默认用户3. 重启并重新登录wsl下的CentOS4. 验证pycharm用户的sudo权限 二、创建基于WSL的Python Interpreter1. 添加基于WSL的Python Int…

DR IP-SoC China 2023 Day演讲预告 | 龙智Perforce专家解析芯片开发中的数字资产管理

2023年9月6日&#xff08;周三&#xff09;&#xff0c;龙智即将亮相于上海举行的D&R IP-SoC China 2023 Day&#xff0c;呈现集成了Perforce与Atlassian产品的芯片开发解决方案&#xff0c;助力企业更好、更快地进行芯片开发。 D&R IP-SoC China 2023 Day 是中国首个…

⽹络与HTTP 笔试题精讲1

OSI七层与TCP/IP 这个就是OSI参考模型,⽽实际我们现在的互联⽹世界是就是这个理论模型的落地叫做TCP/IP协议 TCP的三次握⼿与四次挥⼿ 客户端想要发送数据给服务端,在发送实际的数据之前,需要先在两端之间建⽴连接,数据发完以后也需要将该连接关闭。建⽴连接的过程就是我们…

热门框架漏洞

文章目录 一、Thinkphp5.0.23 代码执行1.thinkphp5框架2.thinkphp5高危漏洞3.漏洞特征4.THinkphp5.0 远程代码执行--poc5.TP5实验一(Windows5.0.20)a.搭建实验环境b.测试phpinfoc.写入shelld.使用菜刀连接 6.TP5实验二(Linux5.0.23)a.搭建实验环境b.测试方法c.测试phpinfod.写入…

SQL 语句继续学习之记录三

一&#xff0c;数据的插入&#xff08;insert 语句的使用方法&#xff09; 使用insert语句可以向表中插入数据(行)。原则上&#xff0c;insert语句每次执行一行数据的插入。 列名和值用逗号隔开&#xff0c;分别扩在&#xff08;&#xff09;内&#xff0c;这种形式称为清单。…

如何使用Python和正则表达式处理XML表单数据

在日常的Web开发中&#xff0c;处理表单数据是一个常见的任务。而XML是一种常用的数据格式&#xff0c;用于在不同的系统之间传递和存储数据。本文通过阐述一个技术问题并给出解答的方式&#xff0c;介绍如何使用Python和正则表达式处理XML表单数据。我们将探讨整体设计、编写思…

VB6.0 设置窗体的默认焦点位置在 TextBox 中

文章目录 VB6.0 窗体的加载过程确定指针的焦点位置添加代码效果如下未设置指定焦点已设置焦点 VB6.0 窗体的加载过程 在VB6.0中&#xff0c;窗体&#xff08;Form&#xff09;加载时会触发多个事件&#xff0c;这些事件按照特定的顺序执行。下面是窗体加载过程中常见事件的执行…

基于ETLCloud的自定义规则调用第三方jar包实现繁体中文转为简体中文

背景 前面曾体验过通过零代码、可视化、拖拉拽的方式快速完成了从 MySQL 到 ClickHouse 的数据迁移&#xff0c;但是在实际生产环境&#xff0c;我们在迁移到目标库之前还需要做一些过滤和转换工作&#xff1b;比如&#xff0c;在诗词数据迁移后&#xff0c;发现原来 MySQL 中…

常见路由跳转的几种方式

常见的路由跳转有以下四种&#xff1a; 1. <router-link to"跳转路径"> /* 不带参数 */ <router-link :to"{name:home}"> <router-link :to"{path:/home}"> // 更建议用name // router-link链接中&#xff0c;带/ 表示从根…

elementUI可拖拉宽度抽屉

1&#xff0c;需求&#xff1a; 在elementUI的抽屉基础上&#xff0c;添加可拖动侧边栏宽度的功能&#xff0c;实现效果如下&#xff1a; 2&#xff0c;在原组件上添加自定义命令 <el-drawer v-drawerDrag"left" :visible.sync"drawerVisible" direc…

Unity 顶点vertices,uv,与图片贴图,与mesh

mesh就是组成3d物体的三角形们。 mesh由顶点组成的三角形组成&#xff0c;三角形的大小 并不 需要一样&#xff0c;由顶点之间的位置决定。 mesh可以是一个或者多个面。 贴图的原点在左下角&#xff0c;uv是贴图的坐标&#xff0c;数量和顶点数一样&#xff08;不是100%确定…

Vue3 el-tooltip 根据内容控制宽度大小换行和并且内容太短不显示

el-tooltip 根据长度自适应换行以及显隐 环境 vue: "3.2.37" element-ui: "2.1.8"要求 tooltip 根据内容自动换行如果内容超出显示省略号显示&#xff0c;不超出不显示 tooltip 代码 组件 // ContentTip 组件 <template><el-tooltipv-bind&qu…

【数学建模竞赛】超详细Matlab二维三维图形绘制

二维图像绘制 绘制曲线图 g 是表示绿色 b--o是表示蓝色/虚线/o标记 c*是表示蓝绿色(cyan)/*标记 ‘MakerIndices,1:5:length(y) 每五个点取点&#xff08;设置标记密度&#xff09; 特殊符号的输入 序号 需求 函数字符结构 示例 1 上角标 ^{ } title( $ a…

初识c++

文章目录 前言一、C命名空间1、命名空间2、命名空间定义 二、第一个c程序1、c的hello world2、std命名空间的使用惯例 三、C输入&输出1、c输入&输出 四、c中缺省参数1、缺省参数概念2、缺省参数分类3、缺省参数应用 五、c中函数重载1、函数重载概念2、函数重载应用 六、…

【⑰MySQL】 变量 | 循环 | 游标 | 处理程序

前言 ✨欢迎来到小K的MySQL专栏&#xff0c;本节将为大家带来MySQL变量 | 循环 | 游标 | 处理程序的分享✨ 目录 前言1. 变量1.1系统变量1.2 用户变量 2. 定义条件与处理程序2.1 案例分析2.2 定义条件2.3 定义处理程序2.4 案例解决 3. 流程控制3.1 分支结构3.2 循环结构3.3 跳转…

设计模式-9--迭代器模式(Iterator Pattern)

一、什么是迭代器模式 迭代器模式&#xff08;Iterator Pattern&#xff09;是一种行为型设计模式&#xff0c;用于提供一种统一的方式来访问一个聚合对象中的各个元素&#xff0c;而不需要暴露该聚合对象的内部结构。迭代器模式将遍历集合的责任从集合对象中分离出来&#xf…

【深入解析spring cloud gateway】07 自定义异常返回报文

Servlet的HttpResponse对象&#xff0c;返回响应报文&#xff0c;一般是这么写的&#xff0c;通过输出流直接就可以将返回报文输出。 OutputStream out response.getOutputStream(); out.write("输出的内容"); out.flush();在filter中如果发生异常&#xff08;例如…

Java中的网络编程------基于Socket的TCP编程和基于UDP的网络编程,netstat指令

Socket 在Java中&#xff0c;Socket是一种用于网络通信的编程接口&#xff0c;它允许不同计算机之间的程序进行数据交换和通信。Socket使得网络应用程序能够通过TCP或UDP协议在不同主机之间建立连接、发送数据和接收数据。以下是Socket的基本介绍&#xff1a; Socket类型&…