fix(qns): fallback if server is not sending NEW_TOKEN

[mxinden]

#1676 ensured for the client to wait for a resumption token from the server if one is needed, that is when running the resumption QUIC Interop testcase. #1676 accidentally removed the fallback using a session ticked if no resumption token was available.

https://github.com/mozilla/neqo/pull/1676/files#diff-64d071700a607884926de1b2caaf1588558fceda6d3a0e899cfb3c3b7e5ae41fL1353-L1359

This commit re-introduces the fallback and enables the resumption QUIC Interop testcase on CI.

Along the way, it simplifies Runner::run and reserves the resume command line argument for the QUIC Interop testcases only. Let me know if this adds too much noise to the actual fix. Happy to extract it into a separate pull request.

---

Test failure reported in #1786 (comment).

[github-actions]

Benchmark results

Performance differences relative to 4fc4d16.

• drain a timer quickly time: [430.85 ns 437.41 ns 443.50 ns]
  change: [-0.7742% +0.5155% +1.8832%] (p = 0.43 > 0.05)
  No change in performance detected.

• coalesce_acked_from_zero 1+1 entries
  time: [202.60 ns 203.08 ns 203.55 ns]
  change: [+2.2678% +2.6124% +2.9801%] (p = 0.00 < 0.05)
  💔 Performance has regressed.

• coalesce_acked_from_zero 3+1 entries
  time: [246.26 ns 246.90 ns 247.59 ns]
  change: [+2.9003% +3.4231% +3.8724%] (p = 0.00 < 0.05)
  💔 Performance has regressed.

• coalesce_acked_from_zero 10+1 entries
  time: [245.97 ns 251.01 ns 260.31 ns]
  change: [+3.1883% +4.1610% +5.5344%] (p = 0.00 < 0.05)
  💔 Performance has regressed.

• coalesce_acked_from_zero 1000+1 entries
  time: [225.85 ns 226.04 ns 226.25 ns]
  change: [+2.4067% +3.0924% +3.7850%] (p = 0.00 < 0.05)
  💔 Performance has regressed.

• RxStreamOrderer::inbound_frame()
  time: [118.44 ms 118.63 ms 118.90 ms]
  change: [-1.0824% -0.8312% -0.5863%] (p = 0.00 < 0.05)
  Change within noise threshold.

• transfer/Run multiple transfers with varying seeds
  time: [125.00 ms 125.30 ms 125.61 ms]
  thrpt: [31.845 MiB/s 31.922 MiB/s 32.000 MiB/s]
  change:
  time: [+0.2035% +0.5363% +0.8800%] (p = 0.00 < 0.05)
  thrpt: [-0.8723% -0.5334% -0.2031%]
  Change within noise threshold.

• transfer/Run multiple transfers with the same seed
  time: [125.29 ms 125.52 ms 125.76 ms]
  thrpt: [31.806 MiB/s 31.868 MiB/s 31.927 MiB/s]
  change:
  time: [-0.0022% +0.2528% +0.4999%] (p = 0.05 < 0.05)
  thrpt: [-0.4974% -0.2521% +0.0022%]
  Change within noise threshold.

• 1-conn/1-100mb-resp (aka. Download)/client
  time: [1.1342 s 1.1404 s 1.1479 s]
  thrpt: [87.115 MiB/s 87.690 MiB/s 88.170 MiB/s]
  change:
  time: [+3.8486% +4.5461% +5.3651%] (p = 0.00 < 0.05)
  thrpt: [-5.0919% -4.3484% -3.7060%]
  💔 Performance has regressed.

• 1-conn/10_000-parallel-1b-resp (aka. RPS)/client
  time: [424.79 ms 427.13 ms 429.48 ms]
  thrpt: [23.284 Kelem/s 23.412 Kelem/s 23.541 Kelem/s]
  change:
  time: [-1.1971% -0.4592% +0.2963%] (p = 0.23 > 0.05)
  thrpt: [-0.2955% +0.4613% +1.2116%]
  No change in performance detected.

• 1-conn/1-1b-resp (aka. HPS)/client
  time: [48.905 ms 49.192 ms 49.451 ms]
  thrpt: [20.222 elem/s 20.328 elem/s 20.448 elem/s]
  change:
  time: [-3.6597% -2.8579% -2.1051%] (p = 0.00 < 0.05)
  thrpt: [+2.1504% +2.9420% +3.7987%]
  💚 Performance has improved.

Client/server transfer results

Transfer of 134217728 bytes over loopback.

Client	Server	CC	Pacing	Mean [ms]	Min [ms]	Max [ms]	Relative
msquic	msquic			760.4 ± 280.8	404.1	1169.4	1.00
neqo	msquic	reno	on	1007.7 ± 250.8	754.4	1447.9	1.00
neqo	msquic	reno		877.5 ± 167.7	747.3	1300.2	1.00
neqo	msquic	cubic	on	875.9 ± 185.7	760.6	1370.5	1.00
neqo	msquic	cubic		1030.4 ± 202.8	779.3	1460.3	1.00
msquic	neqo	reno	on	4440.8 ± 215.2	4185.3	4927.8	1.00
msquic	neqo	reno		4547.7 ± 363.1	4147.0	5152.7	1.00
msquic	neqo	cubic	on	4554.1 ± 329.7	4240.0	5388.0	1.00
msquic	neqo	cubic		4512.0 ± 276.0	4172.5	5045.0	1.00
neqo	neqo	reno	on	3485.3 ± 153.7	3238.6	3784.6	1.00
neqo	neqo	reno		3934.2 ± 401.0	3075.1	4482.2	1.00
neqo	neqo	cubic	on	4444.6 ± 234.5	4218.7	4878.3	1.00
neqo	neqo	cubic		4414.2 ± 246.2	4002.8	4749.6	1.00

⬇️ Download logs

[mxinden]

Currently debugging neqo -> msquic resumption test failure.

[mxinden]

Got it. I introduced this regression in #1676. While #1676 fixes the race condition around receiving the resumption token from the server, it removes the fallback using a session ticked 🤦.

https://github.com/mozilla/neqo/pull/1676/files#diff-64d071700a607884926de1b2caaf1588558fceda6d3a0e899cfb3c3b7e5ae41fL1353-L1359

Yet another good reason for #1785 or #1847.

Will push a patch in a bit.

//CC @larseggert

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.23%. Comparing base (4fc4d16) to head (710c500).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1837   +/-   ##
=======================================
  Coverage   93.23%   93.23%           
=======================================
  Files         110      110           
  Lines       35749    35749           
=======================================
+ Hits        33329    33330    +1     
+ Misses       2420     2419    -1     

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