altair823-org/kebab
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 46 Wiki Activity

test(eval): normalize elapsed_ms before determinism comparison (flake fix)

Browse Source 
`runner_lexical_is_deterministic_per_query_payload` 가 full-suite 첫 실행에서
간헐적으로 `elapsed_ms: 0` vs `elapsed_ms: 1` 차이로 깨지는 timing flake 가
있었음 (PR #140 회차 0 의 full-suite 실행에서 관찰).

원인: per_query 전체 JSON 을 byte-identical 비교하는데 QueryResult.elapsed_ms
가 timing 기반이라 µs-scale wall-clock jitter 가 그대로 비교에 들어감. 의도는
"timing 외에 byte-identical" — 인접 snapshot test #7 은 projection 으로
timing 을 명시적으로 제외하지만 #6 은 누락.

Fix: 비교 직전 양쪽 run 의 elapsed_ms 를 0 으로 normalize. 의도 그대로
표현하고 다른 field 의 결정성 검증은 보존. 50회 반복 stress 통과 (이전:
간헐 실패).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
altair823
2026-05-20 00:01:41 +00:00
parent 73e5b359d8
commit d715631928
1 changed files with 12 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
crates/kebab-eval/tests/runner.rs
View File

@@ -336,21 +336,29 @@ fn runner_lexical_is_deterministic_per_query_payload() {
"- id: q1\n query: ownership\n- id: q2\n query: heading\n", "- id: q1\n query: ownership\n- id: q2\n query: heading\n",
); );
let run_a = run_with_golden(&yaml, || { let mut run_a = run_with_golden(&yaml, || {
run_eval_with_config(&env.config, &lexical_opts()).unwrap() run_eval_with_config(&env.config, &lexical_opts()).unwrap()
}); });
let run_b = run_with_golden(&yaml, || { let mut run_b = run_with_golden(&yaml, || {
run_eval_with_config(&env.config, &lexical_opts()).unwrap() run_eval_with_config(&env.config, &lexical_opts()).unwrap()
}); });
// Run-level fields (`run_id`, `created_at`) intentionally diverge; // Run-level fields (`run_id`, `created_at`) intentionally diverge;
// the per-query payload (which is what the snapshot fixture pins) // the per-query payload (which is what the snapshot fixture pins)
// must be byte-identical. // must be byte-identical EXCEPT for `elapsed_ms`. Timing-sensitive
// fields aren't determinism signals — they're µs-scale wall-clock
// jitter and would otherwise make this assertion a flaky one (a 0
// vs 1 ms divergence was observed under contended-CI load). Normalize
// before comparing; see test #7 for the same exclusion done via a
// projection.
for qr in run_a.per_query.iter_mut().chain(run_b.per_query.iter_mut()) {
qr.elapsed_ms = 0;
}
let a_json = serde_json::to_string(&run_a.per_query).unwrap(); let a_json = serde_json::to_string(&run_a.per_query).unwrap();
let b_json = serde_json::to_string(&run_b.per_query).unwrap(); let b_json = serde_json::to_string(&run_b.per_query).unwrap();
assert_eq!( assert_eq!(
a_json, b_json, a_json, b_json,
"lexical-only per_query payload must be byte-identical across runs" "lexical-only per_query payload must be byte-identical across runs (timing normalized)"
); );
} }