Use polling instead of thread::sleep
#49
Comments
jprochazk
added a commit
that referenced
this issue
Oct 24, 2024
Currently, the `ws_receiver_blocking` function would always sleep after every read. Additionally, it did not use the value from `options.delay_blocking` for the sleep delay. This would result in that thread being blocked for 10ms _after every read_ no matter what the user configured it to, meaning it would receive data at an unnecessarily slow rate. In this PR: - Instead of using `set_nonblocking` or calling `std::thread::sleep` after every read, we now use `set_read_timeout` in the non-tokio native impls. - The `delay_blocking` option is removed in favor of `read_timeout` - The default for `read_timeout` is `Some(10ms)` - The ideal thing to do here would be to not block at all and instead use native polling APIs. This work is left as a TODO for another PR. - #49
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
On native, we always block the thread using a
thread::sleep/thread::yield_nowafter a read. The reason wesleepon those threads is to not busy-loop that CPU core. We should instead set the socket to non-blocking mode and poll it using native polling APIs. This would be useful for senders as well.We use
pollingin Rerun, and it has a reasonably small footprint, so should be fine to add as a dependency here as well.The text was updated successfully, but these errors were encountered: