Read request is not dispatched to slave nodes evenly when ReadFrom.slave

[ldwnt]

Bug Report

I have a redis cluster of 3 master nodes & 6 slave nodes. There is a key named "a" on 1 master node and 2 slave nodes. I set ReadFrom = SLAVE and try to read the "a" key from both of the 2 slave nodes in concurrently rest calls. However, when I use jmeter to call the "test" API with 10 threads, only one of the 2 slave processes has high cpu rate while the other one seem to be idle. Am I missing some config here, or is lettuce designed to read from only one slave node, even when the ReadFrom.SLAVE option is used?

Current Behavior

Only one slave node has high cpu rate.

Input Code

    private RedisClusterClient client;
    private StatefulRedisClusterConnection<String, String> connection;
    private GenericObjectPool<StatefulRedisClusterConnection<String, String>> pool;

    @RequestMapping("/lettuce-init/{ip}/{port}")
    public void lettuceOpen(@PathVariable String ip, @PathVariable int port) throws Exception {
        client = RedisClusterClient.create(RedisURI.create(ip, port));
        connection = client.connect();
        connection.setReadFrom(ReadFrom.SLAVE);
        pool = ConnectionPoolSupport.createGenericObjectPool(() -> connection, new GenericObjectPoolConfig());
    }

    @RequestMapping("/test")
    public String lettuce() throws Exception {

        StatefulRedisClusterConnection<String, String> conn = pool.borrowObject();
        String result = conn.sync().get("a");
        conn.close();
        log.debug(result);
        return result;
    }

Expected behavior/code

All slave nodes receive roughly equal amount of requests.

Environment

• Lettuce version(s): 5.1.7
• Redis version: 5.0.5

[mp911de]

Works as designed. See #834 for details. Once #1145 is fixed, you can create your own ReadFrom implementations that can randomize load. Use ReadFrom.ANY until then.

[bhavikambani]

@mp911de Is this feature available now ? I am using 5.2.2.RELEASE, can't see there at least. I also have same usecase where I have multiple SLAVES and I want to distribute read between those nodes.

[mp911de]

#834 is in place since version 5.2.0.