actually dial i/o timeout much bigger than DialTimeOut in ClusterOptions

[IcefireCgrbza]

Discussed in #2640

^{Originally posted by IcefireCgrbza July 4, 2023}
I want to test my system's downgrading abilit, so I mock redis sigstop error using kill -SIGSTOP. When redis sigstop happen, redis get and set command using go-redis becoming very slow. It cost bigger than 10 second when go-redis return dial i/o timeout error. I dont understand why timeout happen so slow, can anyone explain it?

my system is using redis cluster mode, go-redis version is v8.11.5, cluster client initialization code like below:

opt := &redis.ClusterOptions{
		Addrs:        servers,
		MaxRedirects: 3
		DialTimeout:  250 * time.Millisecond,
		ReadTimeout:  1000 * time.Millisecond,
		WriteTimeout: 1000 * time.Millisecond,
		PoolTimeout:  2000 * time.Millisecond,
		IdleTimeout:  5 * time.Minute,
		MinIdleConns: 3,
		PoolSize:     100,
		MaxRetries:   3,
}

client := redis.NewClusterClient(opt)
```</div>

[youneverknows]

MaxRetries in your configuration is 3, it means that the client will retry 4 times total, and the connect timeout every time is DialTimeout