feat: Web Stream API

[climba03003]

Closes #5280 Closes #5279

Checklist

• run npm run test and npm run benchmark
• tests and/or benchmarks are included
• documentation is changed or added
• commit message and code follows the Developer's Certification of Origin
  and the Code of conduct

[mcollina]

Could you check if this impacts our benchmarks? I fear of those long optional checks in .send().

[climba03003]

Pull Request

> fastify@4.25.2 benchmark
> concurrently -k -s first "node ./examples/benchmark/simple.js" "autocannon -c 100 -d 30 -p 10 localhost:3000/"

[1] Running 30s test @ http://localhost:3000/
[1] 100 connections with 10 pipelining factor
[1]
[1] 
[1] ┌─────────┬──────┬───────┬───────┬────────┬──────────┬──────────┬────────┐
[1] │ Stat    │ 2.5% │ 50%   │ 97.5% │ 99%    │ Avg      │ Stdev    │ Max    │
[1] ├─────────┼──────┼───────┼───────┼────────┼──────────┼──────────┼────────┤
[1] │ Latency │ 7 ms │ 18 ms │ 95 ms │ 140 ms │ 21.95 ms │ 20.46 ms │ 201 ms │
[1] └─────────┴──────┴───────┴───────┴────────┴──────────┴──────────┴────────┘
[1] ┌───────────┬─────────┬─────────┬────────┬────────┬──────────┬──────────┬─────────┐
[1] │ Stat      │ 1%      │ 2.5%    │ 50%    │ 97.5%  │ Avg      │ Stdev    │ Min     │
[1] ├───────────┼─────────┼─────────┼────────┼────────┼──────────┼──────────┼─────────┤
[1] │ Req/Sec   │ 38,143  │ 38,143  │ 44,735 │ 47,327 │ 44,534.4 │ 2,041.82 │ 38,112  │
[1] ├───────────┼─────────┼─────────┼────────┼────────┼──────────┼──────────┼─────────┤
[1] │ Bytes/Sec │ 7.17 MB │ 7.17 MB │ 8.4 MB │ 8.9 MB │ 8.37 MB  │ 384 kB   │ 7.17 MB │
[1] └───────────┴─────────┴─────────┴────────┴────────┴──────────┴──────────┴─────────┘
[1]
[1] Req/Bytes counts sampled once per second.
[1] # of samples: 30
[1]
[1] 1337k requests in 30.05s, 251 MB read
[1] autocannon -c 100 -d 30 -p 10 localhost:3000/ exited with code 0

Main Branch

> fastify@4.25.2 benchmark
> concurrently -k -s first "node ./examples/benchmark/simple.js" "autocannon -c 100 -d 30 -p 10 localhost:3000/"

[1] Running 30s test @ http://localhost:3000/
[1] 100 connections with 10 pipelining factor
[1]
[1] 
[1] ┌─────────┬──────┬───────┬───────┬────────┬──────────┬──────────┬────────┐
[1] │ Stat    │ 2.5% │ 50%   │ 97.5% │ 99%    │ Avg      │ Stdev    │ Max    │
[1] ├─────────┼──────┼───────┼───────┼────────┼──────────┼──────────┼────────┤
[1] │ Latency │ 8 ms │ 18 ms │ 95 ms │ 137 ms │ 21.96 ms │ 20.61 ms │ 213 ms │
[1] └─────────┴──────┴───────┴───────┴────────┴──────────┴──────────┴────────┘
[1] ┌───────────┬────────┬────────┬─────────┬─────────┬───────────┬─────────┬────────┐
[1] │ Stat      │ 1%     │ 2.5%   │ 50%     │ 97.5%   │ Avg       │ Stdev   │ Min    │
[1] ├───────────┼────────┼────────┼─────────┼─────────┼───────────┼─────────┼────────┤
[1] │ Req/Sec   │ 36,703 │ 36,703 │ 44,511  │ 46,975  │ 44,526.94 │ 1,874.7 │ 36,690 │
[1] ├───────────┼────────┼────────┼─────────┼─────────┼───────────┼─────────┼────────┤
[1] │ Bytes/Sec │ 6.9 MB │ 6.9 MB │ 8.37 MB │ 8.83 MB │ 8.37 MB   │ 352 kB  │ 6.9 MB │
[1] └───────────┴────────┴────────┴─────────┴─────────┴───────────┴─────────┴────────┘
[1] 
[1] Req/Bytes counts sampled once per second.
[1] # of samples: 30
[1] 
[1] 1337k requests in 30.04s, 251 MB read
[1] autocannon -c 100 -d 30 -p 10 localhost:3000/ exited with code 0

[climba03003]

@gurgunday
It is why I use Object.prototype.toString.call(payload) === '[object Response]'.

[Uzlopak]

Can we not create a symbol kResponse and set it as an attribute of the payload once, so we dont need to call toString multiple times?

[climba03003]

I don't think we can skip it in .send but we can reduce one time in onSendEnd.
Note that due to the nature of Response, it is an immutable object. Which means every mutation (like status or headers) on it requires to re-create a new Response. Setting an attribute on it will be useless somehow.

[mcollina]

lgtm

[Uzlopak]

It feels strange that this PR should not have any performance penalty. The toString call is usually slowing down.

If the Response object is immutable, what if we store the result of toString.call(payload) === '[object Response]' as an attribute kIsResponse on our Reply Instance?

[climba03003]

It feels strange that this PR should not have any performance penalty. The toString call is usually slowing down.

Do you have any reference on the toString call performance?
toString call is actually finding the Symbol.toStringTag in the object. It would be as fast as an property access if the implementation of class explicitly provide the symbol.

If the Response object is immutable, what if we store the result of toString.call(payload) === '[object Response]' as an attribute kIsResponse on our Reply Instance?

It is not an option because you would need to toString call anyway, this PR allows to change the payload to in any state which means it is not restricted to only .send API.
You need such identification of payload in .send or after onSend hook. The behavior is actually match the current implementation without an extra attribute.

[Uzlopak]

LGTM