test: Experiment with cargo fuzz

[larseggert]

Adds four simple fuzzing targets, for packets and frames, and for fuzzed data inside client and server initials.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.23%. Comparing base (d74ad5e) to head (ed79956).
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1764      +/-   ##
==========================================
- Coverage   93.23%   93.23%   -0.01%     
==========================================
  Files         110      110              
  Lines       35669    35749      +80     
==========================================
+ Hits        33256    33330      +74     
- Misses       2413     2419       +6     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

Benchmark results

Performance differences relative to ccf0302.

• drain a timer quickly time: [428.79 ns 434.44 ns 439.72 ns]
  change: [+38.065% +40.658% +43.230%] (p = 0.00 < 0.05)
  💔 Performance has regressed.

• coalesce_acked_from_zero 1+1 entries
  time: [197.08 ns 197.60 ns 198.13 ns]
  change: [-0.3722% -0.0223% +0.3213%] (p = 0.90 > 0.05)
  No change in performance detected.

• coalesce_acked_from_zero 3+1 entries
  time: [239.13 ns 239.80 ns 240.48 ns]
  change: [-0.4614% +0.1057% +0.6371%] (p = 0.72 > 0.05)
  No change in performance detected.

• coalesce_acked_from_zero 10+1 entries
  time: [238.21 ns 239.37 ns 240.67 ns]
  change: [-1.0492% -0.1759% +0.5088%] (p = 0.71 > 0.05)
  No change in performance detected.

• coalesce_acked_from_zero 1000+1 entries
  time: [218.69 ns 218.84 ns 219.02 ns]
  change: [-5.6412% -1.7086% +0.7100%] (p = 0.52 > 0.05)
  No change in performance detected.

• RxStreamOrderer::inbound_frame()
  time: [118.91 ms 118.98 ms 119.05 ms]
  change: [-0.7655% -0.5679% -0.4288%] (p = 0.00 < 0.05)
  Change within noise threshold.

• transfer/Run multiple transfers with varying seeds
  time: [122.82 ms 123.13 ms 123.43 ms]
  thrpt: [32.406 MiB/s 32.486 MiB/s 32.568 MiB/s]
  change:
  time: [+1.5815% +1.9464% +2.3133%] (p = 0.00 < 0.05)
  thrpt: [-2.2610% -1.9092% -1.5569%]
  Change within noise threshold.

• transfer/Run multiple transfers with the same seed
  time: [123.01 ms 123.21 ms 123.43 ms]
  thrpt: [32.408 MiB/s 32.464 MiB/s 32.517 MiB/s]
  change:
  time: [+1.6664% +1.9052% +2.1581%] (p = 0.00 < 0.05)
  thrpt: [-2.1125% -1.8696% -1.6391%]
  Change within noise threshold.

• 1-conn/1-100mb-resp (aka. Download)/client
  time: [1.1182 s 1.1265 s 1.1344 s]
  thrpt: [88.149 MiB/s 88.773 MiB/s 89.426 MiB/s]
  change:
  time: [+1.6146% +2.5454% +3.3059%] (p = 0.00 < 0.05)
  thrpt: [-3.2001% -2.4822% -1.5889%]
  💔 Performance has regressed.

• 1-conn/10_000-parallel-1b-resp (aka. RPS)/client
  time: [425.12 ms 427.22 ms 429.37 ms]
  thrpt: [23.290 Kelem/s 23.407 Kelem/s 23.523 Kelem/s]
  change:
  time: [-0.8386% -0.1831% +0.4909%] (p = 0.60 > 0.05)
  thrpt: [-0.4886% +0.1834% +0.8457%]
  No change in performance detected.

• 1-conn/1-1b-resp (aka. HPS)/client
  time: [50.511 ms 50.872 ms 51.241 ms]
  thrpt: [19.516 elem/s 19.657 elem/s 19.798 elem/s]
  change:
  time: [-0.3979% +0.5834% +1.5150%] (p = 0.24 > 0.05)
  thrpt: [-1.4923% -0.5801% +0.3995%]
  No change in performance detected.

Client/server transfer results

Transfer of 134217728 bytes over loopback.

Client	Server	CC	Pacing	Mean [ms]	Min [ms]	Max [ms]	Relative
msquic	msquic			656.7 ± 243.0	385.4	1152.6	1.00
neqo	msquic	reno	on	1114.2 ± 325.5	788.1	1790.2	1.00
neqo	msquic	reno		868.6 ± 113.3	757.3	1154.1	1.00
neqo	msquic	cubic	on	983.5 ± 203.1	771.0	1360.9	1.00
neqo	msquic	cubic		940.6 ± 224.6	753.3	1389.5	1.00
msquic	neqo	reno	on	4508.7 ± 272.9	4212.4	4988.5	1.00
msquic	neqo	reno		4510.8 ± 333.3	4125.1	5103.7	1.00
msquic	neqo	cubic	on	4723.0 ± 346.5	4291.4	5293.0	1.00
msquic	neqo	cubic		4553.0 ± 215.7	4199.1	4842.5	1.00
neqo	neqo	reno	on	3778.6 ± 430.5	3292.5	4404.0	1.00
neqo	neqo	reno		3747.8 ± 225.8	3484.9	4149.0	1.00
neqo	neqo	cubic	on	4477.7 ± 538.7	3117.7	4968.8	1.00
neqo	neqo	cubic		4348.1 ± 224.8	4181.6	4952.3	1.00

⬇️ Download logs

[larseggert]

@KershawChang @martinthomson any chance for a review?

[mxinden]

I don't think the use of unsafe is necessary here.

In case you like the suggestions, you can simply merge #1789.

[martinthomson]

3766868 has a couple of comments.

I'm on vacation from tomorrow, so I trust you to do the right thing with this. This is generally OK.

(I did question whether it is the right decision to inline the write_to_corpus calls, rather than having specific code to generate that corpus, but this works.) Do you have a script for that?

[larseggert]

(I did question whether it is the right decision to inline the write_to_corpus calls, rather than having specific code to generate that corpus, but this works.) Do you have a script for that?

A corpus can get generated in a number of ways, it's basically just a collection of files. The fuzzer has commands to manage it (e.g., minify it, etc.), in Rust via cargo fuzz. For quant, I found it convenient to (re-)generate a corpus as a side effect of running tests, I just followed that approach here, too.