v1.9.1

[notgull]

• Fix a hang that occurs when running a local queue via ticking. (Fix hangs due to tasks being stuck inside of local queues #104)

[taiki-e]

Have all the known regressions introduced in 1.9.0 (#103) been fixed? (Otherwise, we will just end up yanking it again.)

[notgull]

It was said that it fixed the regression in Bevy (albeit at a performance cost, which I find acceptable in the short term). In addition, zbus tests seem to pass when I add this to its Cargo.toml:

[patch.crates-io]
async-executor = { git = "https://github.com/smol-rs/async-executor" }

[taiki-e]

Thanks for the clarification. If the regression has been fixed, I'm okay with this.

albeit at a performance cost, which I find acceptable in the short term

The performance degradation compared to 1.9.0, which has the bug, is not a problem. However, if performance is degraded compared to 1.8.0, that is a problem. Is there a performance comparison to 1.8.0?

[notgull]

     Running benches/executor.rs (/hard_data/cargo_target/release/deps/executor-45ad0a619e1329b0)
executor::create        time:   [1.2568 µs 1.2581 µs 1.2593 µs]
                        change: [+18.276% +18.451% +18.630%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 6 outliers among 100 measurements (6.00%)
  6 (6.00%) high mild

single_thread/executor::spawn_one
                        time:   [5.2113 µs 5.3943 µs 5.5192 µs]
                        change: [+2.9134% +7.8629% +12.546%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 12 outliers among 100 measurements (12.00%)
  10 (10.00%) low severe
  2 (2.00%) high mild
single_thread/executor::spawn_many_local
                        time:   [11.058 ms 11.397 ms 11.687 ms]
                        change: [-0.7056% +3.3126% +7.2993%] (p = 0.10 > 0.05)
                        No change in performance detected.
Found 9 outliers among 100 measurements (9.00%)
  7 (7.00%) low severe
  1 (1.00%) low mild
  1 (1.00%) high mild
single_thread/executor::spawn_recursively
                        time:   [33.515 ms 33.661 ms 33.815 ms]
                        change: [-1.6246% -0.8815% -0.1761%] (p = 0.02 < 0.05)
                        Change within noise threshold.
Found 8 outliers among 100 measurements (8.00%)
  6 (6.00%) high mild
  2 (2.00%) high severe
single_thread/executor::yield_now
                        time:   [3.4690 ms 3.4755 ms 3.4820 ms]
                        change: [-11.581% -11.366% -11.136%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 9 outliers among 100 measurements (9.00%)
  1 (1.00%) low severe
  1 (1.00%) low mild
  7 (7.00%) high mild

multi_thread/executor::spawn_one
                        time:   [5.1600 µs 5.3962 µs 5.6124 µs]
                        change: [+37.545% +45.859% +55.042%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 6 outliers among 100 measurements (6.00%)
  2 (2.00%) low mild
  4 (4.00%) high mild
Benchmarking multi_thread/executor::spawn_many_local: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 6.5s, or reduce sample count to 70.
multi_thread/executor::spawn_many_local
                        time:   [55.573 ms 55.717 ms 55.867 ms]
                        change: [+159.71% +169.78% +180.36%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 1 outliers among 100 measurements (1.00%)
  1 (1.00%) high mild
Benchmarking multi_thread/executor::spawn_recursively: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 11.9s, or reduce sample count to 40.
multi_thread/executor::spawn_recursively
                        time:   [118.49 ms 118.99 ms 119.48 ms]
                        change: [-8.9313% -8.4757% -8.0163%] (p = 0.00 < 0.05)
                        Performance has improved.
Benchmarking multi_thread/executor::yield_now: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 8.0s, enable flat sampling, or reduce sample count to 50.
multi_thread/executor::yield_now
                        time:   [1.5913 ms 1.5986 ms 1.6060 ms]
                        change: [-90.748% -90.676% -90.603%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 9 outliers among 100 measurements (9.00%)
  1 (1.00%) low severe
  3 (3.00%) low mild
  3 (3.00%) high mild
  2 (2.00%) high severe

Slight regressions in the single-threaded tests, but significant improvements in multi-threaded tests.

[zeenix]

Slight regressions in the single-threaded tests, but significant improvements in multi-threaded tests.

This one looks pretty significant:

multi_thread/executor::spawn_many_local
                        time:   [55.573 ms 55.717 ms 55.867 ms]
                        change: [+159.71% +169.78% +180.36%] (p = 0.00 < 0.05)
                        Performance has regressed.

🤔

[james7132]

#104 (comment)

The contention issues from the new implementation creates a major performance regressions on Bevy's end. The tasks that are spawning Bevy's systems are taking 10x longer and average end-to-end frame times are seeing double digit percentage regressions. Given that 1.9.1 will be semver compatible, this will retroactively affect virtually every Bevy version upon running cargo update.