[franklee26]

No description provided.

[github-actions]

Bluefin throughput bench (macOS) ✅ PASS

Config: 5 run(s) × 2 conn(s) on Darwin arm64

Client throttle: 25 µs sleep every 3000 sends

Retries: 1 of 5 run(s) needed retry (+2 extra attempt(s)). See per-run table.

why retries vary across runs

Hosted CI runners occasionally land on a deeply contended host where both bench peers get starved and a run produces 0 GOOD conns despite no actual code regression. To absorb this without weakening the gate, this job uses an asymmetric retry policy (current values: ON_ZERO=2, ON_PARTIAL=0, backoff 5s):

attempt outcome	extra retries granted
all 2 conns good	0 (accept)
1..1 conns good (partial)	up to ON_PARTIAL=0
0 conns good (catastrophic)	up to ON_ZERO=2

The retry budget is per-run and only shrinks across attempts: once a partial-good attempt has been observed in a run, that run is locked to the partial budget for the rest of its attempts -- it can never be promoted to the larger 0-good budget by a later attempt happening to collapse to 0. Only the final attempt's data is committed to the aggregate stats; earlier attempts' logs are still on disk under the run's attempt_* directory but are not parsed.

So the same code can show 0, 1, or 2 retries on different CI runs purely from runner-allocation luck. The attempts column in the per-run table below shows how many attempts each run took (bold = needed at least one retry).

Aggregate

metric	observed	floor	result
mean avg gb/s (good only)	0.495	0.40	✅
raw mean avg gb/s	0.366	—	—
min avg gb/s	0.072	—	—
mean peak gb/s	1.512	—	—
max peak gb/s (good only)	1.940	1.30	✅
good conns	5 / 10	3	✅

Per-run

run	attempts	good/N	mean avg gb/s (good)	max peak gb/s (good)	raw max peak gb/s
1	1	2/2	0.50	1.87	1.87
2	1	1/2	0.49	1.63	1.63
3	1	1/2	0.50	1.94	1.94
4	1	1/2	0.48	1.85	1.85
5	3	0/2	—	—	1.52

_{A "good conn" delivered ≥ 700 MB before idle timeout.
"max peak (good only)" filters out kernel-buffer-drain artifacts
that appear as huge inst-throughput readings on starved conns under
runner CPU contention. Floors are env-var overridable:
BLUEFIN_BENCH_FLOOR_MEAN_AVG_GBPS, BLUEFIN_BENCH_FLOOR_MAX_PEAK_GBPS,
BLUEFIN_BENCH_FLOOR_GOOD_CONNS, BLUEFIN_BENCH_GOOD_CONN_MIN_BYTES.
Ratchet up after stable runs land. Script: bench_ci.sh.}

[codecov]

Codecov Report

❌ Patch coverage is 97.20670% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 71.29%. Comparing base (6e8a0a4) to head (44c479d).

Files with missing lines	Patch %	Lines
bluefin-proto/src/endpoint.rs	93.70%	9 Missing ⚠️
bluefin/src/net/close_handler.rs	97.22%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #47      +/-   ##
==========================================
+ Coverage   69.13%   71.29%   +2.16%     
==========================================
  Files          24       26       +2     
  Lines        3833     3958     +125     
==========================================
+ Hits         2650     2822     +172     
+ Misses       1183     1136      -47     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.