The time for acquire a new connection is not counted when setting context timeout for a query

[liuzhaohui]

I'm using the recent version 1.0.55, disable autopipeling (with some specific type of commands designated to pipe).
I set a time limit for every call to redis (50ms), and found that sometimes I got a context timeouts which were far above the time limit I set. It seems that timeout doesn't work, but it's the time of acquire a new connection doesn't count(making a new wire if needed). It's really annoying to explain over and over again. Is there any sollution? Or taking the acquire time into account?

[rueian]

Hi @liuzhaohui,

Yes, currently acquiring a connection doesn't respect the context deadline. To respect it, we need to update:

rueidis/rueidis.go

Line 467 in f56a674

func dial(dst string, opt *ClientOption) (conn net.Conn, err error) {

the dial function to make it accept a context and use Dialer.DialContext to make a connection, and then we also need to update all the intermediate code paths that use the dial function under the hood.

Before that, you can set connection timeout at ClientOption.Dialer.Timeout:

	client, err := rueidis.NewClient(rueidis.ClientOption{
		Dialer: net.Dialer{
			Timeout: 30 * time.Millisecond,
		},
		BlockingPoolSize: 1024,
	})

Also note that you should make sure you never hit the BlockingPoolSize because once the pool is full it also gets stuck if no free connections.

[liuzhaohui]

Thank you for your advice. I'll have a try.

[rueian]

Hi @liuzhaohui, all the necessary changes are here #800

If you are interested in contributing, maybe you can help adding a few testes.

[liuzhaohui]

Thanks. I made the modification yesterday, I'll merge your code and add the testcases.

[ryanrazsa]

Also note that you should make sure you never hit the BlockingPoolSize because once the pool is full it also gets stuck if no free connections.

Is there a recommended way to detect if this is happening?