updated to ack requests and to check errors on WaitFor. Flush is als…

[traetox]

…o issued

I updated to check the errors on WaitFor and to Ack request on completion. Without Acking the map that tracks the ACKs will grow indefinitely.

Testing shows a significantly lower blocking rate. I am showing throughput closer to 280Mbit/s on my i5-4690k and a cheapo SSD. No errors on testing.

[abligh]

Thanks. I'm taking it the lack of Ack() was causing the memory leak?

I'll merge this. It still won't support flush, because it the backend doesn't supply the flush flag (and it should do the same with FUA like the file driver). But that isn't much of an issue as my understanding is no file systems support AIO flush and there's no way to tell if one does in the future. So I'll put some out-of-band support for that in (aka a configuration flag).

[abligh]

Nope. This doesn't prevent the test being killed here :-(