Fix VList Stream in SSR

[futursolo]

Description

This pull request fixes the rendering stream in VList.

Previously, the VList would block until a child future has finished before the child stream was returned.
This pull request allows items to be yielded to the parent stream as soon as an item is written to the child stream.

As a result, this pull request increases function router benchmark performance for ~20%.

Checklist

• I have reviewed my own code
• I have added tests

pull request:

╭─────────────────┬───────┬──────────┬──────────┬───────────┬────────────────────╮
│    Benchmark    │ Round │ Min (ms) │ Max (ms) │ Mean (ms) │ Standard Deviation │
├─────────────────┼───────┼──────────┼──────────┼───────────┼────────────────────┤
│    Baseline     │  10   │ 328.295  │ 350.396  │  343.984  │       6.029        │
│   Hello World   │  10   │ 476.104  │ 493.962  │  481.288  │       5.170        │
│ Function Router │  10   │ 1532.462 │ 1553.854 │ 1540.794  │       5.968        │
│ Concurrent Task │  10   │ 1011.433 │ 1017.915 │ 1015.075  │       2.250        │
╰─────────────────┴───────┴──────────┴──────────┴───────────┴────────────────────╯

master:

╭─────────────────┬───────┬──────────┬──────────┬───────────┬────────────────────╮
│    Benchmark    │ Round │ Min (ms) │ Max (ms) │ Mean (ms) │ Standard Deviation │
├─────────────────┼───────┼──────────┼──────────┼───────────┼────────────────────┤
│    Baseline     │  10   │ 372.443  │ 384.248  │  376.874  │       3.182        │
│   Hello World   │  10   │ 472.603  │ 498.872  │  478.182  │       7.599        │
│ Function Router │  10   │ 1816.955 │ 1861.456 │ 1834.111  │       14.948       │
│ Concurrent Task │  10   │ 1009.515 │ 1015.743 │ 1012.627  │       2.061        │
╰─────────────────┴───────┴──────────┴──────────┴───────────┴────────────────────╯

[github-actions]

Visit the preview URL for this PR (updated for commit d0ed690):

https://yew-rs-api--pr2801-fix-rendering-stream-vs03l8f3.web.app

_{(expires Sat, 13 Aug 2022 16:00:53 GMT)}

_{🔥 via Firebase Hosting GitHub Action 🌎}

[github-actions]

Size Comparison

examples	master (KB)	pull request (KB)	diff (KB)	diff (%)
boids	172.722	172.720	-0.002	-0.001%
contexts	109.580	109.579	-0.001	-0.001%
counter	86.526	86.526	0	0.000%
counter_functional	87.179	87.180	+0.001	+0.001%
dyn_create_destroy_apps	89.619	89.623	+0.004	+0.004%
file_upload	102.758	102.760	+0.002	+0.002%
function_memory_game	166.896	166.904	+0.008	+0.005%
function_router	351.738	351.736	-0.002	-0.001%
function_todomvc	161.595	161.585	-0.010	-0.006%
futures	225.433	225.390	-0.043	-0.019%
game_of_life	107.137	107.137	0	0.000%
immutable	208.692	208.695	+0.003	+0.001%
inner_html	83.687	83.685	-0.002	-0.002%
js_callback	112.752	112.750	-0.002	-0.002%
keyed_list	195.249	195.247	-0.002	-0.001%
mount_point	86.170	86.172	+0.002	+0.002%
nested_list	115.645	115.641	-0.004	-0.003%
node_refs	93.464	93.463	-0.001	-0.001%
password_strength	1546.080	1546.080	0	0.000%
portals	97.259	97.255	-0.004	-0.004%
router	321.074	321.076	+0.002	+0.001%
simple_ssr	154.248	154.251	+0.003	+0.002%
ssr_router	397.798	397.812	+0.015	+0.004%
suspense	110.486	110.486	0	0.000%
timer	89.259	89.260	+0.001	+0.001%
todomvc	142.623	142.621	-0.002	-0.001%
two_apps	87.190	87.189	-0.001	-0.001%
web_worker_fib	153.608	153.615	+0.007	+0.004%
webgl	87.539	87.538	-0.001	-0.001%

✅ None of the examples has changed their size significantly.

[github-actions]

Benchmark - SSR

Yew Master

Benchmark	Round	Min (ms)	Max (ms)	Mean (ms)	Standard Deviation
Baseline	10	352.126	368.988	360.855	5.780
Hello World	10	1210.873	1261.104	1234.832	16.158
Function Router	10	4653.621	4805.525	4719.233	46.171
Concurrent Task	10	1012.881	1016.538	1014.642	0.996

Pull Request

Benchmark	Round	Min (ms)	Max (ms)	Mean (ms)	Standard Deviation
Baseline	10	404.882	414.560	409.781	3.187
Hello World	10	1202.682	1236.553	1223.266	12.067
Function Router	10	3733.080	3819.429	3784.534	30.588
Concurrent Task	10	1011.468	1014.475	1013.286	1.009

[futursolo]

There are some additional optimisations that can be landed to improve SSR performance.
I have chosen to land them individually to make sure that each PR would improve performance when compared with master.

Most optimisations can be found in #2772.

Final Result

╭─────────────────┬───────┬──────────┬──────────┬───────────┬────────────────────╮
│    Benchmark    │ Round │ Min (ms) │ Max (ms) │ Mean (ms) │ Standard Deviation │
├─────────────────┼───────┼──────────┼──────────┼───────────┼────────────────────┤
│    Baseline     │  10   │ 339.746  │ 346.630  │  343.666  │       2.751        │
│   Hello World   │  10   │ 334.349  │ 342.953  │  338.006  │       3.302        │
│ Function Router │  10   │ 1317.307 │ 1335.086 │ 1323.627  │       6.098        │
│ Concurrent Task │  10   │ 1012.617 │ 1018.859 │ 1015.107  │       2.391        │
╰─────────────────┴───────┴──────────┴──────────┴───────────┴────────────────────╯

[ranile]

Implementation looks good. The performance improvements are great!

[WorldSEnder]

Is there any reason to use 30 as the magical cut off, or was that determined by an educated guess?

[futursolo]

Is there any reason to use 30 as the magical cut off, or was that determined by an educated guess?

This is picked by futures::future::join_all.

When the number of futures is small, it will use a select like strategy which all futures are polled when one is waken up, this is faster than FuturesUnordered which assigns wakers and operates like kqueue by only waking up futures that are ready. FuturesUnordered is used when the number of futures is big and join_all will switch to FuturesOrdered. The results don't really need to be collected into a vector, so the switching to FuturesUnordered allows to discard the result as they are ready.