hyper: simplify handlers

[joshka]

• move the server header to the router and add it to all responses.
• move the response boxing to the router simplifying the return types of each handler.

https://github.com/hyperium/http-body/pull/150 will simplify the router code even futher to the following:

let mut response = match request.uri().path() {
     "/ping" => ping()?.box_body(),
     "/json" => json::get()?.box_body(),
     "/db" => single_query::get().await?.box_body(),
     "/queries" => multiple_queries::get(request.uri().query()).await?.box_body(),
     "/fortunes" => fortunes::get().await?.box_body(),
     "/plaintext" => plaintext::get()?.box_body(),
     _ => not_found_error()?.box_body(),
};

Note: includes commits from #9727. Relevant commit that's only this change is https://github.com/TechEmpower/FrameworkBenchmarks/pull/9728/commits/a50128f137861d9d995acf2a6d52b6af23f0580c

[github-actions[bot]]

The following frameworks were updated, pinging maintainers: hyper: @polachok, @seanmonstar, @steveklabnik

[joshka]

Local benchmark results (Macbook Pro M2 Max)

Before the recent hyper changes:

hyper (JSON + Plaintext):

https://www.techempower.com/benchmarks/#section=test&shareid=69e73a28-4d6c-4aa8-821b-27eb62b7e616&hw=ph&test=json

JSON responses per second, (unspecified, hostname = 48c41e9f7187)

Framework	16	32	64	128	256	512	Best	Errors	Cls	Lng	Plt	FE	Aos	IA
hyper	196,760	508,649	624,762	729,611	727,629	811,146	811,146	0	Mcr	rs	rs	hyp	lin	rea

Plaintext responses per second, (unspecified, hostname = 48c41e9f7187)

Framework	256	1,024	4,096	16,384	Best	Errors	Cls	Lng	Plt	FE	Aos	IA
hyper	5,481,306	5,887,571	5,447,194	4,462,003	5,887,571	0	Mcr	rs	rs	hyp	lin	rea

• hyper-db (Fortunes): https://www.techempower.com/benchmarks/#section=test&shareid=5502a537-4af9-4ba6-b1d5-d535309b5304&hw=ph&test=fortune

Fortunes responses per second, (unspecified, hostname = e0169ecd4a21)

Framework	16	32	64	128	256	512	Best	Errors	Cls	Lng	Plt	FE	Aos	DB	Dos	Orm	IA
hyper	80,846	177,855	129,424	141,055	147,723	149,434	177,855	0	Mcr	rs	rs	hyp	lin	Pg	lin	raw	rea

After this change:

hyper (current-thread runtime):

https://www.techempower.com/benchmarks/#section=test&shareid=0be4944f-3746-4ff5-a6d5-6537c33df272&hw=ph&test=json

JSON responses per second, (unspecified, hostname = b7f8f718c8cc)

Framework	16	32	64	128	256	512	Best	Errors	Cls	Lng	Plt	FE	Aos	IA
hyper (current-thread runtime)	207,752	470,367	598,843	692,511	821,558	872,231	872,231	0	Mcr	rs	rs	hyp	lin	rea

Database-access responses per second, single query, (unspecified, hostname = b7f8f718c8cc)

Framework	16	32	64	128	256	512	Best	Errors	Cls	Lng	Plt	FE	Aos	DB	Dos	Orm	IA
hyper (current-thread runtime)	49,527	82,542	133,047	119,479	118,664	105,070	133,047	0	Mcr	rs	rs	hyp	lin	Pg	lin	raw	rea

Responses per second, multiple queries, (unspecified, hostname = b7f8f718c8cc)

Framework	1	5	10	15	20	Errors	Cls	Lng	Plt	FE	Aos	DB	Dos	Orm	IA
hyper (current-thread runtime)	115,887	30,055	15,253	10,466	7,838	0	Mcr	rs	rs	hyp	lin	Pg	lin	raw	rea

Fortunes responses per second, (unspecified, hostname = b7f8f718c8cc)

Framework	16	32	64	128	256	512	Best	Errors	Cls	Lng	Plt	FE	Aos	DB	Dos	Orm	IA
hyper (current-thread runtime)	45,863	80,784	122,626	113,597	111,767	108,218	122,626	0	Mcr	rs	rs	hyp	lin	Pg	lin	raw	rea

Plaintext responses per second, (unspecified, hostname = b7f8f718c8cc)

Framework	256	1,024	4,096	16,384	Best	Errors	Cls	Lng	Plt	FE	Aos	IA
hyper (current-thread runtime)	4,498,684	5,150,947	5,067,568	4,265,876	5,150,947	0	Mcr	rs	rs	hyp	lin	rea

hyper (multi-thread runtime):

https://www.techempower.com/benchmarks/#section=test&shareid=e44af5ec-34be-4542-bb04-f970a375a62e&hw=ph&test=json

JSON responses per second, (unspecified, hostname = cda50185af70)

Framework	16	32	64	128	256	512	Best	Errors	Cls	Lng	Plt	FE	Aos	IA
hyper (multi-thread runtime)	136,775	205,623	280,521	351,340	517,568	681,464	681,464	0	Mcr	rs	rs	hyp	lin	rea

Database-access responses per second, single query, (unspecified, hostname = cda50185af70)

Framework	16	32	64	128	256	512	Best	Errors	Cls	Lng	Plt	FE	Aos	DB	Dos	Orm	IA
hyper (multi-thread runtime)	60,558	80,468	114,034	125,198	123,046	123,854	125,198	0	Mcr	rs	rs	hyp	lin	Pg	lin	raw	rea

Responses per second, multiple queries, (unspecified, hostname = cda50185af70)

Framework	1	5	10	15	20	Errors	Cls	Lng	Plt	FE	Aos	DB	Dos	Orm	IA
hyper (multi-thread runtime)	124,719	31,089	16,536	11,172	8,463	0	Mcr	rs	rs	hyp	lin	Pg	lin	raw	rea

Fortunes responses per second, (unspecified, hostname = cda50185af70)

Framework	16	32	64	128	256	512	Best	Errors	Cls	Lng	Plt	FE	Aos	DB	Dos	Orm	IA
hyper (multi-thread runtime)	57,991	79,421	110,869	120,830	121,191	119,472	121,191	0	Mcr	rs	rs	hyp	lin	Pg	lin	raw	rea

Plaintext responses per second, (unspecified, hostname = cda50185af70)

Framework	256	1,024	4,096	16,384	Best	Errors	Cls	Lng	Plt	FE	Aos	IA
hyper (multi-thread runtime)	4,216,747	4,716,550	4,561,864	4,090,954	4,716,550	0	Mcr	rs	rs	hyp	lin	rea

[joshka]

There's definitely a performance degradation in the results, and I haven't investigated why yet. The goal of the changes was to make the source a bit more maintainable, and then work on improving the perf aspects on top of that in ways where the code which impacts performance can be a bit more obvious and measured (with test variants). Because hyper is also the base for many of the rust frameworks, having numbers for each of the tests allows this to be used as the framework overhead in those tests (using the "versus" field in the configs", which would help show which parts of the perf are just pure hyper and which are from framework overhead (or in some cases improvements).

[msmith-techempower]

This PR now has conflicts due to the force-rm PR that was merged this morning. Much of it is likely also out of date. Closing for grooming - feel free to reopen once all issues are addressed.