[lalinsky]

Description

Upgrade dusty to 0.6, zio to 0.20.

Changes:

• faster router
• more flexible handlers/middleware
• stricter behaviors after checking the server against HTTP Garden
• TLS is now based on AWS-LC, standard entry uses libssl wrapper, tuned entry uses native protocol implementation on top of libcrypto, both are more compliant than the previously used tls.zig
• and zio brings in some performance improvements and many bug fixes

---

PR Commands — comment on this PR to trigger (requires collaborator approval):

Command	Description
/benchmark -f <framework>	Run every test the framework subscribes to
/benchmark -f <framework> -t <test>	Run one test only
/benchmark -f <framework> -t <t1>,<t2>,...	Run several tests in one job: one approval, one results commit, untouched tests keep their numbers
/benchmark -f <framework> --save	Run and save results (updates the leaderboard on merge)
/benchmark -f <framework> -t <test> --save	Run one test and save results
/benchmark -f <framework> --compare <other>	Measure the deltas against another framework instead of this one
/benchmark-multiple -f <fw1>,<fw2>,...	Benchmark several frameworks in one run — takes -t and --save too; saved results land in a single commit
/benchmark-multiple --save	No -f needed: benchmark and save every framework the PR touches
/benchmark-test -t <test>	Benchmark all enabled frameworks subscribed to <test> and save the results
/spec -f <framework>	Run the protocol-compliance suites the framework lists in spec in its meta.json
/spec -f <framework> --save	Run them and save the scores to site/data/spec/ (lands with the merge, like saved results)

For /benchmark, always specify -f <framework>; the flags combine in any order. Results come back as a comment with a per-profile table of RPS, p99, CPU and memory — one table per framework on multi runs. A new benchmark comment while a run is in flight queues behind it (one deep) instead of cancelling it. For multi-framework PRs (dependency bumps, same-language refactors) prefer /benchmark-multiple, which runs everything in a single job and commits all saved results together, so no run overwrites another. --compare works on single-framework runs only.

A /benchmark without -t also runs the compliance suites, and its --save saves their scores with the results. /spec runs the suites alone; without -f it runs them for every framework the PR touches. See Compliance.

What the deltas are measured against. By default, this framework's own results published on main - answering "did this change help?". When you are tuning a variant or a successor entry, --compare re-bases them on another entry instead:

/benchmark -f genhttp --compare genhttp-kestrel

The reply states which baseline it used, and profiles the other framework does not run show n/a rather than a delta.

---

Run benchmarks locally

You can validate and benchmark your framework locally with the lite script — no CPU pinning, fixed connection counts, all load generators run in Docker.

./scripts/validate.sh <framework>
./scripts/benchmark-lite.sh <framework> baseline
./scripts/benchmark-lite.sh --load-threads 4 <framework>

Requirements: Docker Engine on Linux. Load generators (gcannon, h2load, h2load-h3, wrk) are built as self-contained Docker images on first run.

[lalinsky]

/benchmark --save
/spec --save

[github-actions]

👋 Benchmark request received. A collaborator will review and approve the run.

[github-actions]

Benchmark Results

Framework: dusty | Test: all tests

Test	Conn	RPS	Rate	CPU/req	p99	p99.9	CPU	Mem	Δ RPS	Δ Mem
baseline	4096	3,323,203	—	—	—	—	6332.9%	231MiB	-2.8%	+15.5%
pipelined	4096	4,259,821	—	—	—	—	6440.9%	233MiB	-3.0%	+14.8%
limited-conn	4096	2,535,173	—	—	—	—	6120.5%	92MiB	+23.5%	-9.8%
json-comp	4096	549,586	—	—	—	—	6281.8%	292MiB	-1.6%	+231.8%
json-comp	16384	542,531	—	—	—	—	6558.8%	555MiB	-2.9%	+56.8%
json-tls	4096	1,147,761	—	—	—	—	6595.4%	426MiB	-4.3%	-42.7%
8gbit	512	49,351	0.9870	46.55us	117.0us	4802.0us	229.8%	89MiB	~0%	-25.8%
static-tls	1024	381,806	—	—	—	—	6569.6%	196MiB	+4.5%	-2.5%
async-db	1024	304,196	—	—	—	—	3968.2%	164MiB	+3.4%	+25.2%
fortunes	1024	276,336	—	—	—	—	5103.4%	158MiB	+0.9%	+4.6%
echo-ws	512	4,033,942	—	—	—	—	6541.7%	59MiB	-0.7%	+18.0%
echo-ws	4096	3,945,477	—	—	—	—	6538.4%	224MiB	-0.4%	+10.9%
echo-ws	16384	3,266,574	—	—	—	—	6142.1%	786MiB	-6.8%	+10.5%
echo-ws-pipeline	512	59,109,183	—	—	—	—	6454.7%	64MiB	+0.1%	+20.8%
echo-ws-pipeline	4096	57,590,595	—	—	—	—	6583.5%	226MiB	+0.6%	+4.6%
echo-ws-pipeline	16384	46,200,510	—	—	—	—	6622.7%	786MiB	-6.0%	+7.4%
echo-ws-limited	512	2,242,739	—	—	—	—	6417.1%	76MiB	+4.0%	+43.4%
echo-ws-limited	4096	2,512,674	—	—	—	—	6037.0%	206MiB	+21.5%	+96.2%
latency-1m	1024	997,337	0.9973	27.05us	202.0us	7938.0us	2941.0%	82MiB	~0%	+15.5%
latency-10k	1024	9,983	0.9983	27.34us	102.0us	153.0us	26.2%	86MiB	~0%	+26.5%
latency-500k-8cpu	1024	412,352	0.8247 ⚠️	19.33us	6002688.0us	6236160.0us	793.0%	59MiB	-2.0%	+1.7%
async	32000	2,130,544	—	—	—	—	6266.6%	1.4GiB	-4.5%	+7.7%

Composite score, main against this run, in the board's default view:

Family	main	this run	Δ
HTTP/1.1	6980.1 (5 of 107)	7102.3 (4 of 107)	+122.2
WebSocket	2855.5 (2 of 39)	2916.7 (1 of 39)	+61.2

The results on main were measured in another session, so a few points up or down can be the machine and not the change.

Compliance

Suite	Tool	Score	Failed
H/1.1 (h1)	http11probe d50d818	155 / 159	4
WebSocket (ws)	autobahn-testsuite 25.10.1	244 / 244	0

H/1.1: 4 failed

Test	Expected	Got
COMP-OPTIONS-STAR OPTIONS * is the only valid asterisk-form request (RFC 9112 §3.2.4, MUST)	2xx or 405; close/timeout = warn	404, connection left open
COMP-METHOD-CONNECT CONNECT to an origin server must be rejected (RFC 9110 §9.3.6, SHOULD)	400/405/501 or close	404, connection closed
COMP-HOST-EMPTY-VALUE Empty Host header value must be rejected (RFC 9112 §3.2, MUST)	400 or close	200, connection left open
SMUG-TE-LEADING-COMMA Transfer-Encoding with leading comma (, chunked) — RFC says empty list elements MUST be ignored (RFC 9110 §5.6.1, SHOULD)	400 or 2xx	501, connection closed

[lalinsky]

Hm, that did not auto-detect dusty-tuned

[MDA2AV]

you didnt add the -multiple

[MDA2AV]

/benchmark -f dusty-tuned --save

[github-actions]

👋 Benchmark request received. A collaborator will review and approve the run.

[github-actions]

Benchmark Results

Framework: dusty-tuned | Test: all tests

Test	Conn	RPS	Rate	CPU/req	p99	p99.9	CPU	Mem	Δ RPS	Δ Mem
baseline	4096	3,399,844	—	—	—	—	6375.6%	248MiB	-0.9%	+23.4%
pipelined	4096	4,314,403	—	—	—	—	6022.5%	234MiB	-2.4%	+14.1%
limited-conn	4096	2,535,904	—	—	—	—	6099.3%	98MiB	+23.3%	-3.9%
json-comp	4096	650,945	—	—	—	—	6610.6%	684MiB	-5.5%	+620.0%
json-comp	16384	654,739	—	—	—	—	6174.9%	1.1GiB	-3.2%	+191.8%
json-tls	4096	1,879,440	—	—	—	—	6402.0%	362MiB	+15.5%	-51.3%
8gbit	512	49,369	0.9874	41.40us	108.0us	4106.0us	185.4%	85MiB	+0.2%	-29.2%
static-tls	1024	418,559	—	—	—	—	6636.1%	137MiB	+14.6%	-31.8%
async-db	1024	349,417	—	—	—	—	3827.2%	158MiB	+5.4%	+23.4%
fortunes	1024	282,348	—	—	—	—	5081.9%	161MiB	+2.1%	+3.9%
latency-1m	1024	997,723	0.9977	26.83us	199.0us	2387.0us	2916.5%	88MiB	~0%	+25.7%
latency-10k	1024	9,979	0.9979	25.83us	104.0us	6470.0us	24.8%	77MiB	~0%	+16.7%
latency-500k-8cpu	1024	422,655	0.8453 ⚠️	18.85us	6486016.0us	7034880.0us	792.2%	67MiB	~0%	+15.5%
async	32000	2,227,512	—	—	—	—	6361.4%	1.5GiB	-1.4%	+15.4%

Composite score, main against this run, in the board's default view:

Family	main	this run	Δ
HTTP/1.1	7394.4 (1 of 107)	7745.2 (1 of 107)	+350.9
WebSocket	2854.2 (3 of 39)	2854.2 (3 of 39)	+0.0

The results on main were measured in another session, so a few points up or down can be the machine and not the change.

Compliance

Suite	Tool	Score	Failed
H/1.1 (h1)	http11probe d50d818	155 / 159	4

H/1.1: 4 failed

Test	Expected	Got
COMP-OPTIONS-STAR OPTIONS * is the only valid asterisk-form request (RFC 9112 §3.2.4, MUST)	2xx or 405; close/timeout = warn	404, connection left open
COMP-METHOD-CONNECT CONNECT to an origin server must be rejected (RFC 9110 §9.3.6, SHOULD)	400/405/501 or close	404, connection closed
COMP-HOST-EMPTY-VALUE Empty Host header value must be rejected (RFC 9112 §3.2, MUST)	400 or close	200, connection left open
SMUG-TE-LEADING-COMMA Transfer-Encoding with leading comma (, chunked) — RFC says empty list elements MUST be ignored (RFC 9110 §5.6.1, SHOULD)	400 or 2xx	501, connection closed

[lalinsky]

Oh, one more thing, I unsubscribed the tuned entry from websocket, as it doesn't add anything over the standard one, but I guess I need to delete the previous files as well?

[MDA2AV]

hmm, yes need to delete the data on websockets at the results json for tuned

[lalinsky]

ok, deleted

[lalinsky]

Sorry about the mess, I had to add one more commit. The last CI failed, the PG startup was probably slower than usual and that disabled the fortunes/async-db endpoints. Should be more stable now.