long durations when auto-pipelining enabled

[moish83]

Hello,

We are using ioredis in our service, connecting to Amazon Elasticache service, every once in a while - happens sporadically in production, not at peak times, we get redis durations spiking (hundreds of milliseconds), the rest of the time redis response duration is at single digit milliseconds.

We've noticed that when the issue happens, there are several GET commands impacted by it, all in the same pipeline, we tried setting a timeout on the GET commands using Promise.race(), however the retry of the GET command on the same key, resulted in the same pipeline, hence we got it back at the same time with the original GET.

We've seen the commandTimeout option and would like to use that, however, this option is also affecting the connection command, which we don't want to limit to the same value as we do for GET commands, for that we use the connectTimeout option, which doesn't take precedence over the commandTimeout.

What we thought is if we can add the same mechanism like the autoPipeliningIgnoredCommands where we give a list of commands to not set the timeout, or event better a whitelist of commands to set the commandTimeout on, for example, in our case we expect the GET commands to be extremely fast, but we don't care if SET commands take a little bit longer.