fix(rt): bring back dynamic machines #748
Conversation
Even if we do not make use of the progress blocking, we do need to make use of the dynamic restarting of machines as far as I understand. Keeps the perf, while removing the regression from #747
|
|
||
| // Sleep for a bit longer if the scheduler state hasn't changed in a while. | ||
| if idle > 10 { | ||
| delay = (delay * 2).min(10_000); |
There was a problem hiding this comment.
Is perhaps crossbeam utils BackOff a better fit here than rolling it by hand?
There was a problem hiding this comment.
I figured if it was good enough for @stjepang it'll be good enough for me for now ;)
|
@dignifiedquire I think it's good to start a thread when you need it and close it when it's no longer needed. |
can we cache the thread? make the thread is just park instead of terminating it, then when a new thread is needed just unpark it and assume this thread is a fresh one and make sure no resource leak to the new thread (reinitialize all TLS), something like this loop {
main_loop();
invalidate_tls();
if push_to_cache(thread::current()) {
thread::park();
} else {
break;
}
} |
|
I agree we should investigate parking, but for now this is the original logic and fixes the brokeness on master |
Even if we do not make use of the progress blocking, we do need to make use of the dynamic restarting of machines as far as I understand.
Keeps the perf, while removing the regression from #747