You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, the timeout operation is implemented by racing given operation with a sleeping thread (BTW a similar approach is used in Cats Effect).
I was wondering if we could leverage a ScheduledExecutorService instead. It operates on a single threaded, which schedules and starts actual tasks. Netty uses this mechanism for timeouts in its IdleStateHandler. A scheduled task starts, emits an event and ends immediately. Then, Netty's internals handle the event in order to abort communication if a timeout occured.
In case of Ox, a scheduled task would start on a virtual thread (see hints in this Stack Overflow discussion). It would then call cancelation of the fork that's wrapped with timeout.
I don't know if such a solution will be more performant than racing with a sleeping thread, but I'm leaving this idea here for consideration. One certain issue is that the single-threaded pool for scheduling doesn't sound like something we want to create for each scope, but rather like a "runtime" that needs passing around.
The text was updated successfully, but these errors were encountered:
Currently, the
timeout
operation is implemented by racing given operation with a sleeping thread (BTW a similar approach is used in Cats Effect).I was wondering if we could leverage a
ScheduledExecutorService
instead. It operates on a single threaded, which schedules and starts actual tasks. Netty uses this mechanism for timeouts in its IdleStateHandler. A scheduled task starts, emits an event and ends immediately. Then, Netty's internals handle the event in order to abort communication if a timeout occured.In case of Ox, a scheduled task would start on a virtual thread (see hints in this Stack Overflow discussion). It would then call cancelation of the fork that's wrapped with timeout.
I don't know if such a solution will be more performant than racing with a sleeping thread, but I'm leaving this idea here for consideration. One certain issue is that the single-threaded pool for scheduling doesn't sound like something we want to create for each scope, but rather like a "runtime" that needs passing around.
The text was updated successfully, but these errors were encountered: