Configurable SLEEP_DURATION

[lso400]

The SLEEP_DURATION is hardcoded to one second, can we make that configurable?

[bocharov]

We could consider making SLEEP_DURATION configurable, but first I'd like to understand your use case a bit better.
Is current default of 1 second too much or too little? There is also grace_duration parameter of Synchronizer::write method, which is configurable and related to the same logic of preventing "dead readers" problem.

[lso400]

We could consider making SLEEP_DURATION configurable, but first I'd like to understand your use case a bit better. Is current default of 1 second too much or too little? There is also grace_duration parameter of Synchronizer::write method, which is configurable and related to the same logic of preventing "dead readers" problem.

Hey thanks for looking into this. One second is too long for us. I think grace_duration does not solve the issue we're seeing
E.g. We set grace_duration to 100ms. When num_readers>0 but Instant::now() < grace_expiring_at, the thread will be blocked for 1 second. We want to block for say 10ms and check num_readers again.

[bocharov]

Thanks for explaining your use case in more detail. I think we can definitely make this configurable - would you prefer it as template parameter on Synchronizer struct, which we can configure to default 1s (still configurable) and not break existing methods signature or perhaps new parameter to existing method or completely new method with new parameter?

[lso400]

template parameter on Synchronizer struct, which we can configure to default 1s (still configurable)

This sounds good to me!

[bocharov]

Perhaps, other alternative is to add tokio optional feature, which will use tokio::time::sleep instead of std::thread::sleep, which doesn't block the thread during the sleep and allows other workload to progress.

[lso400]

Perhaps, other alternative is to add tokio optional feature, which will use tokio::time::sleep instead of std::thread::sleep, which doesn't block the thread during the sleep and allows other workload to progress.

I'm afraid that this will not be helpful to applications running in sync runtime. IMO making it a template parameter on Synchronizer struct makes most sense. But of course tokio optional feature can be added on top to avoid blocking the entire thread for async runtime app

[bocharov]

Sure, I'll make template parameter change and release new package version tomorrow.

[bocharov]

@lso400 change now has been implemented and released as package version 1.0.4 via #14

[lso400]

Thanks for the prompt action, much appreciated!