lettuce默认提供了线程安全的链接，大部分场景下，使用单个链接即可满足需求，但是在大并发场景，还是需要连接池，lettuce默认提供了 异步连接池配置，本文对Redis 的单链接，和连接池场景进行测试，比对性能差异和适用场景。

测试条件

Redis cluser 3主3从 6节点，

本机Intel 8核处理器，lettuce版本

       <dependency>
            <groupId>io.lettuce</groupId>
            <artifactId>lettuce-core</artifactId>
            <version>5.3.7.RELEASE</version>
        </dependency>

单链接(不用连接池)

做法可以参考我上篇文章《java 服务基于lettuce实现redis sentinel/cluster的高可用连接》

直接上测试代码

   @Autowired
   private RedisCluster redis;
  @Autowired
    protected ThreadPoolTaskScheduler taskScheduler;

    private AtomicLong count = new AtomicLong(0);
    private Long lastCount =0L;

    @PostConstruct
    public void initService() {
        LOGGER.info("initService hello"); 
      
       for(int i=0;i<10;i++)
    {
       taskScheduler.schedule(new Runnable() {
            @Override
            public void run() {
                while(true) {
                    try {
                        long start = System.currentTimeMillis();
                        String clusterRe = redis.getCMD().set("testCluster", RandomUtils.nextInt()+"-hello",
                        SetArgs.Builder.ex(10));
                      
                        start = System.currentTimeMillis();
                        String value = redis.get("testCluster");
                        count.incrementAndGet();
                    }catch (Exception e) {
                        LOGGER.error("redisSentinel test error:" + e.getMessage()); }
                    }
                 }
            },new Date(System.currentTimeMillis() + 10000));
        }
       //10秒后开始每秒统计
       taskScheduler.scheduleAtFixedRate(new Runnable() {
            @Override
            public void run() {
                      long nowCount = count.get();
                      LOGGER.info("one cluster QPS " + (nowCount -lastCount )+"/s"; }
                     lastCount = nowCount;
            },new Date(System.currentTimeMillis() + 10000),1000);

    }

设置大于1个线程，逐步增加到60线程数，循环操作，qps不稳定，并且持续下降，本机8核cpu被打满

 可以看到，一开始有20万QPS，但是持续下降，一直掉到 1000以下，猜测是由于竞争激烈导致

/**
     * 异步连接池
     */
    private BoundedAsyncPool<StatefulRedisClusterConnection<String, String>> asyncPool;

// cluster client 创建异步连接池 clusterClient 需要提前创建好
asyncPool = AsyncConnectionPoolSupport.createBoundedObjectPool(
                () -> clusterClient.connectAsync(StringCodec.UTF8).thenApply(connect->{
                    connect.setTimeout(Duration.ofSeconds(5));
                    connect.setAutoFlushCommands(false);
                    connect.setReadFrom(ReadFrom.MASTER_PREFERRED);
                    return connect;
        }), BoundedPoolConfig.builder().maxTotal(getPoolMaxTotal()).maxIdle(getPoolMaxIdle()).minIdle(getPoolMinIdle()).build());


 // 单节点，或者哨兵的client 创建异步连接池  asyncClient
        RedisClient asyncClient = RedisClient.create();
//TODO 这里需要初始化uri，需要区别单节点和哨兵
  RedisURI redisURI;
       //单节点带有密码时，会有无法正确识别协议问题，需要手工指定  https://github.com/lettuce-io/lettuce-core/issues/1543
        if (SINGLE.equals(type) && StringUtils.isNotBlank(password)) {
            asyncClient.setOptions(ClientOptions.builder().build());
        }
        asyncPool = AsyncConnectionPoolSupport.createBoundedObjectPool(
                () -> MasterReplica.connectAsync(asyncClient, StringCodec.UTF8, redisURI).thenApply(connect -> {
                    connect.setAutoFlushCommands(false);
                    connect.setReadFrom(ReadFrom.MASTER_PREFERRED);
                    return connect;
                }), BoundedPoolConfig.builder().maxTotal(getPoolMaxTotal()).maxIdle(getPoolMaxIdle()).minIdle(getPoolMinIdle()).build());