Recycling smarter

[DGolubets]

At the moment Deadpool recycles a connection every time it gets an existing one from the pool.
This introduces additional latency.
I wonder if recycling can be smarter and only check connections which were idle for a some time?

[bikeshedder]

You can add a pre_recycle hook and discard objects there. The hooks API isn't very pretty at the moment due to #200 but it's quite functional as shown in the tests:

deadpool/tests/managed_hooks.rs

Lines 131 to 141 in f35ac2a

	let pool = Pool::<Computer>::builder(manager)
	.max_size(1)
	.pre_recycle(Hook::sync_fn(|_, metrics| {
	if metrics.recycle_count > 0 {
	Err(HookError::Continue(None))
	} else {
	Ok(())
	}
	}))
	.build()
	.unwrap();

That metrics parameters gives you access to the object Metrics

e.g. if you wanted to discard objects which weren't recycled within the last minute or so you can do it like that:

     .pre_recycle(Hook::sync_fn(|_, metrics| { 
         if metrics.last_used() > Duration::from_secs(60) {
             Err(HookError::Continue(None)) 
         } else { 
             Ok(()) 
         } 
     })) 

[bikeshedder]

Sorry. I got that wrong. You want it the other way around. Rather than fast discarding possibly broken objects you don't want to run the recycle method at all for objects which have just been returned to the pool.

[DGolubets]

Maybe pre_recycle hook can return a bool flag telling whether to recycle or not?

[bikeshedder]

(deleted some misleading comments... once again I got that wrong...)

[bikeshedder]

As a first step we could pass the Metrics object into the recycle method of the pool. That way it would be possible to write a wrapper Manager that does exactly that.

I just opened an issue about that:

• Add metrics argument to Manager::recycle method #236

[bikeshedder]

This is a requirement for #167 and will be tracked as part of this issue:

• Ahead of time object creation/recycling #167