FuturesUnordered Task Management

[fooxed]

This is not directly related to Tokio, was not getting responses in the users forum and someone requested asking here. As mentioned in that thread, if there is a more appropriate place to ask, please feel free to redirect. The below text is just just reproducing original text from that post:

Trying to understand how FuturesUnordered works. Getting tripped up with how the ready_to_run_queue interacts with linked list. Specifically, confused by pending_next_all, spin_next_all, link, and unlink. Does anyone have a good reference as to how this works? Specific questions include:

• The docs mention under heavy thread contention, a compare-exchange loop could become costly. What does thread contention mean here?

• How does using stub resolve this? Confused as to how using pending_next_all allows next_all as needed for iterator. When will there be a case where head_all -> stub -> next?

• More generally, would love an explanation about how ready_to_run_queue works with head_all, next_all pointers and stub. Particularly for the impl Stream block

[Darksonn]

There was some discussion on this topic here.

[quininer]

I think FuturesUnordered based on linked list is too complicated, we can simply use Lock + Queue instead, and sometimes it has good performance.

see https://github.com/quininer/taskset